One Instrument to Rule Them All: The Bias and Coverage of Just-ID IV
Extracted source text · arXiv 2110.10556
\pdfoutput=1
\documentclass{article}
\usepackage[utf8]{inputenc}%
\usepackage[tracking=true, letterspace=100, expansion=false]{microtype}
\usepackage[T1]{fontenc}
\usepackage[title]{appendix}
\usepackage{amsmath}%
\usepackage{amsfonts}%
\usepackage{amsthm}%
\newtheorem{lemma}{Lemma}
\newtheorem{proposition}{Proposition}
\newtheorem{theorem}{Theorem}
\theoremstyle{definition}
\newtheorem{example}{Example}
\theoremstyle{remark}
\newtheorem{remark}{Remark}
\usepackage[margin=1.2in]{geometry}
\usepackage[onehalfspacing]{setspace}
\usepackage{mathtools}
\DeclarePairedDelimiter\indicatorfence{\{}{\}}
\DeclarePairedDelimiter\abs{\lvert}{\rvert}
\DeclarePairedDelimiter\norm{\lVert}{\rVert}
\DeclarePairedDelimiter\ip{\langle}{\rangle}
\newcommand\1{\operatorname{\mathbb{I}}\indicatorfence}
\newcommand\sdfs{\sigma_{\hat{\pi}}} %
\newcommand\sdrf{\sigma_{\hat{\delta}}} %
\newcommand\covrf{\sigma_{\hat{\delta}\hat{\pi}}} %
\usepackage{graphicx}
\DeclareMathOperator{\var}{var}
\DeclareMathOperator{\corr}{cor}
\DeclareMathOperator{\cov}{cov}
\DeclareMathOperator{\sign}{sign}
\DeclareMathOperator*{\argmin}{argmin}
\usepackage{float}
%
\usepackage[
bibencoding=utf8,%
maxbibnames=3, %
minbibnames=1, %
backend=biber,%
sortlocale=en_US,%
style=apa,%
doi=true,%
url=true,
eprint=true%
]{biblatex}
\addbibresource{iv_library.bib}
\usepackage{comment}%
\usepackage{xcolor}%
\definecolor{webbrown}{rgb}{.6,0,0}%
\usepackage{hyperref} %
\hypersetup{%
breaklinks = true,%
%
colorlinks = true,%
%
anchorcolor = webbrown,%
citecolor = webbrown,%
filecolor = webbrown,%
linkcolor = webbrown,%
menucolor = webbrown,%
urlcolor= webbrown,%
citebordercolor= 1 0 0,%
menubordercolor=1 0 0,%
urlbordercolor=1 0 0,%
runbordercolor=1 0 0,%
pdftitle={One Instrument to Rule Them All},
pdfauthor={Angrist and Kolesár}
}
\usepackage{cleveref}
\crefname{appsec}{Appendix}{Appendices}
\crefname{appsubsec}{Appendix}{Appendices}
\usepackage[nolist]{acronym}
\begin{acronym}
\acro{CI}{confidence interval}%
\acro{OVB}{omitted variable bias}%
\acro{OLS}{ordinary least squares}%
\acro{CLT}{central limit theorem}%
\acro{IV}{instrumental variables}%
\acro{ATE}{average treatment effect}%
\acro{LATE}{local average treatment effect}%
\acro{2SLS}{two-stage least squares}
\acro{just-ID}{just-identified}
\end{acronym}
\newcommand\indep{\protect\mathpalette{\protect\independenT}{\perp}}
\def\independenT#1#2{\mathrel{\rlap{$#1#2$}\mkern2mu{#1#2}}}
%
\newcommand{\overbar}[1]{\mkern 1.5mu\overline{\mkern-1.5mu#1\mkern-1.5mu}\mkern 1.5mu}
\usepackage{tikz} %
%
%
\begin{document}
\title{{One Instrument to Rule Them All: \\ The Bias and Coverage of Just-ID IV}\thanks{We thank Ahmet Gulek and Luther Yap for expert research assistance. Thanks also go to Tim Armstrong, Isaiah Andrews, Brigham Frandsen, Guido Imbens, Mike Keane, Dave Lee, Whitney Newey, and Steve Pischke for helpful discussions and insightful comments. Kolesár acknowledges support by the Sloan
Research Fellowship and by the National Science Foundation Grant
SES-22049356. The views expressed here are our own. }}
\author{Joshua Angrist\and Michal Kolesár}
\date{\today}
\maketitle
\begin{abstract}
Two-stage least squares estimates in heavily over-identified instrumental variables (IV) models can be misleadingly close to the corresponding ordinary least squares (OLS) estimates when many instruments are weak. Just-identified (just-ID) IV estimates using a single instrument are also biased, but the importance of weak-instrument bias in just-ID IV applications remains contentious. We argue that in microeconometric applications, just-ID IV estimators can typically be treated as all but unbiased and that the usual inference strategies are likely to be adequate. The argument begins with contour plots for confidence interval coverage as a function of instrument strength and explanatory variable endogeneity. These show undercoverage in excess of 5\% only for endogeneity beyond that seen even when IV and OLS estimates differ by an order of magnitude. Three widely-cited microeconometric applications are used to explain why endogeneity is likely low enough for IV estimates to be reliable. We then show that an estimator that's unbiased given a population first-stage sign restriction has bias exceeding that of IV when the restriction is imposed on the data. But screening on the sign of the estimated first stage is shown to halve the median bias of conventional IV without reducing coverage. To the extent that sign-screening is already part of empirical workflows, reported IV estimates enjoy the minimal bias of sign-screened just-ID IV.
\end{abstract}
\thispagestyle{empty}
\setcounter{page}{0}
\clearpage
\section{Introduction}
The \textcite{bekker94} and \textcite{bjb95} critiques of the heavily
over-identified \ac{2SLS} estimates reported in \textcite{angrist90} and
\textcite[AK91]{AnKr91} sparked a flood of interest in the finite-sample
behavior of \ac{IV} estimates. In the intervening three decades, attention to bias in \ac{2SLS} estimates with many weak instruments has
become a staple of applied microeconometrics. The fact that the finite-sample distribution of \ac{2SLS} estimates is shifted towards the \ac{OLS} probability limit is especially worrying. \ac{IV} is often motivated by fears that \ac{OLS} estimates are compromised by \ac{OVB}. The \ac{IV} analyst hopes, therefore, that when \ac{IV} estimates are close to \ac{OLS}, this signals minimal \ac{OVB} rather than substantial finite-sample bias in the \ac{IV} estimates.
The heavily over-identified models that prompted 1990s \ac{IV} critiques are a case of many weak instruments, roped together by \ac{2SLS} in an effort to estimate a single causal effect with acceptable precision. Strikingly, \textcite{bjb95} show that in the specifications reported in AK91 that interact quarter of birth dummies with 10 year-of-birth and 50 state-of-birth dummies to generate 180 instruments, replacing real quarter of birth dummmies with fake dummies randomly drawn yields \ac{2SLS} estimates and standard errors much like those generated by the real thing.\footnote{This ``fake instruments'' simulation was originally suggested by Alan Krueger. Although not an empirical study, \textcite{bekker94} is likewise motivated by a heavily over-identified specification in \textcite{angrist90} that uses 73 draft lottery dummies plus interaction terms as instruments for Vietnam-era veteran status. This application is featured at the end of Bekker's paper, and, originally, in an Amsterdam bar in 1992, where Paul Bekker first confronted Angrist with claims of bias.} But most studies using \ac{IV} (including \textcite{angrist90} and AK91) report \ac{just-ID} \ac{IV} estimates computed with a single instrument. Just-ID \ac{IV} estimates are less obviously biased than the estimates generated by heavily over-identified models, and the empirical relevance of bias in just-ID \ac{IV} applications remains a matter of debate. Our analysis comes in the wake of recent and renewed interest in the finite-sample properties of just-ID IV, seen in \textcite{AnAr17}, \textcite{lmmp20}, and \textcite{KeNe21}, among others.
We argue here that in typical microeconometric applications of \ac{just-ID} IV, conventional \ac{IV} estimates and $t$-tests are unlikely to be compromised by failures of conventional asymptotic theory. Our analysis builds on the (approximate) finite-sample normality of reduced-form and first-stage estimators (in the argot of classical simultaneous equations models, these are both estimated ``reduced forms''). Our modeling framework parallels that in \textcite{ass19} and earlier theoretical investigations of weak instrument problems. The normality of reduced-form estimates is justified by conventional asymptotic reasoning, as well as by the local-to-zero asymptotic sequence used in \textcite{StSt97} and \textcite{StYo05}, in which the first stage shrinks to zero at a rate inversely proportional to the square root of the sample size.
Our setup has only two free parameters: the correlation between structural and first-stage residuals (henceforth, ``endogeneity'') and the population first-stage $F$ statistic. This fact lends itself to the construction of easily-interpreted rejection contours characterizing conventional second-stage $t$-tests and confidence interval coverage rates. We see, for example, that for endogeneity less than about $0.76$, 95\% confidence interval coverage is distorted by no more than 5\% for \textit{any} population $F$. An important insight here is that, even though bias increases when the first stage gets weaker, second-stage precision falls. In contrast with the over-identified case, conventional just-ID \ac{IV} standard errors reflect this, and confidence intervals widen accordingly. This keeps coverage high unless endogeneity is extraordinarily high.\footnote{The spirit of this argument differs from that in \textcite{StYo05}, which focuses on worst-case rejection rates over all possible endogeneity values. \textcite{lmmp20} and \textcite{KeNe21}, discussed further in \Cref{sec:coverage_contours} below, consider possible restrictions on endogeneity. But these studies stress the fact that the standard definition of coverage doesn't involve restrictions on parameter values.}
Few analysts can gauge endogeneity outside of an empirical context. We therefore use three applications to calibrate endogeneity: the AK91 study, the \textcite[AE98]{AnEv98} \ac{IV} estimates using a dummy for samesex sibships as an instrument for family size, and the \textcite[AL99]{AnLa99} fuzzy regression discontinuity estimates of class size effects on student learning. These studies span a range of \ac{OVB} scenarios, from modest (for most of the AK91 estimates), to substantial (in AE98, where OLS exceeds \ac{IV} by about 50\%), to dramatic (in AL99, where \ac{IV} exceeds small, insignificant OLS estimates by an order of magnitude). Yet, the absolute value of estimated endogeneity is no more than 0.46 in these applications, and over $0.35$ only for a single specification and sample. These just-ID \ac{IV} estimates, therefore, are little-compromised by weak-\ac{IV} concerns. Although three examples do not make a theorem, we argue that the features of these studies that limit endogeneity are common to empirical strategies designed to estimate causal effects or to mitigate attenuation bias in models with measurement error.
Our theoretical case for a sanguine view of conventional just-ID \ac{IV} begins with an analysis of the \textcite{AnAr17} unbiased \ac{IV} estimator as a leading alternative. The \textcite{AnAr17} estimator, denoted $\hat{\beta}_U$, is unbiased given the normality of reduced-form estimates and a sign restriction on the population first stage. The latter restriction seems especially compelling in the context of just-ID IV\@: it is hard to imagine a contemporary analyst pursuing a just-identified \ac{IV} identification strategy without a first-stage direction in mind. Given a theoretical first-stage sign restriction, it's surely tempting to apply this to first-stage estimates. Indeed, it's hard to imagine IV empirical work that ignores perversely signed first-stage estimates. We show, however, that conditional on the sign of the \emph{estimated} first stage, the bias of $\hat{\beta}_U$ exceeds the bias of conventional just-ID \ac{IV}. As it turns out, $\hat{\beta}_U$ is unbiased by virtue of the fact that it averages two conditional estimators, each biased but in opposite directions.
This discouraging result raises the question of whether and how we might use sign restrictions to mitigate the bias of \ac{IV} in a manner consistent with empirical practice. Our answer to this question comes in the form of a novel theoretical result showing that the median bias of just-ID \ac{IV} is roughly halved if we condition the analysis on the first-stage estimate having the expected sign. %
Analysts might justifiably worry that screening on the basis of first-stage estimates distorts inference.\footnote{Recent years have seen growing awareness of the bias induced by screening on the basis of a first-stage pre-test, such as the first-stage F statistic. \textcite{HaRuWi96} appear to be the first to note that pre-testing on the first-stage F-statistic for just-ID \ac{IV} reduces confidence interval coverage; \textcite{ass19} draw on recent empirical scholarship to demonstrate the empirical relevance of pre-test bias in \ac{IV} applications.} Perhaps surprisingly, we also show that pre-screening on the \textit{sign} of the first stage is virtually costless: rejection contours for a sign-screened estimator differ little from those obtained without screening. The upshot is
that sign-screening mitigates the already-modest bias of just-ID \ac{IV} without degrading coverage. To the extent that such screening is a feature of modern empirical work, reported \ac{IV} estimates reflect the impressively minimal bias characteristic of sign-screened \ac{IV}.
The next section details the normal just-ID IV setup and derives an expression for endogeneity in terms of OLS OVB. \Cref{sec:coverage_contours} reviews the relationship between $t$-test rejection rates and the parameters that govern the normal model. This section also explains why endogeneity in applied microeconometrics is unlikely to be high enough for conventional IV inference to mislead. \Cref{sec:signrestrict} shows how and why conditioning on the sign of the estimated first stage strengthens the case for this view. \Cref{sec:conclusion} concludes. Proofs and details behind numerical calculations appear in the appendix.
\section{Setup}\label{sec:setup}
We observe a random sample of $n$ units indexed by $i$, with data on outcome variable, $Y_{i}$,
a scalar treatment variable, $D_i$, a vector of covariates, $X_i$, and a scalar instrument,
$Z_i$. Population regressions of outcome and treatment on the instrument and
covariates define the reduced form and first stage. These are written as follows:
\begin{align}
Y_i & = Z_i\delta + X_i' \psi_1 + u_i,\label{eq:rf}\\
D_i & = Z_i\pi + X_i'\psi_2 + v_i.\label{eq:fs}
\end{align}
The parameter of interest is $\beta=\frac{\delta}{\pi}$, the ratio of the reduced-form and
first-stage regression coefficients on $Z_{i}$. Provided that the instrument, $Z_i$, satisfies an exclusion restriction and is relevant (i.e.\ $\pi\neq 0$), this parameter measures the effect of $D_{i}$ on $Y_{i}$. More generally, if
treatment effects are heterogeneous and a monotonicity condition holds,
$\beta$ is a weighted average of individual causal effects
\parencite{ImAn94,AnIm95}.
Let
$\hat{\delta}=\sum_{i=1}^{n}\tilde{Z}_{i}Y_{i}/\sum_{i=1}^{n}\tilde{Z}_{i}^{2}$
and $\hat{\pi}=\sum_{i=1}^{n}\tilde{Z}_{i}D_{i}/\sum_{i=1}^{n}\tilde{Z}_{i}^{2}$
denote the \ac{OLS} estimates of $\delta$ and $\pi$, where
$\tilde{Z}_{i}$ is the residual from the regression of $Z_{i}$ on $X_{i}$. Under mild regularity conditions,
$(\hat{\delta}, \hat{\pi})$ is consistent and asymptotically normal as
$n\to\infty$, with an asymptotic covariance matrix that can be consistently
estimated. Importantly, this holds regardless of the strength of the instrument.
As in \textcite{ass19} and earlier analyses of weak instrument problems, we assume that reduced-form and first-stage estimates are distributed normally in finite samples. That is,
\begin{equation}\label{eq:normality}
\begin{pmatrix}
\hat{\delta}\\\hat{\pi}
\end{pmatrix}\sim\mathcal{N}\left(
\begin{pmatrix}
\pi\beta\\\pi
\end{pmatrix},\Sigma=
\begin{pmatrix}
\sdrf^{2} & \covrf\\
\covrf & \sdfs^{2}
\end{pmatrix}\right),
\end{equation}
with a known covariance matrix, $\Sigma$. This distributional assumption is implied by the \textcite{StSt97} weak-instrument asymptotic sequence (see
\textcite[Section 3.2]{ass19} for additional discussion and references).
\Cref{eq:normality} is our only substantive restriction; this assumption allows us to focus
on the weak instrument problem, separating this from other finite-same problems,
such as the effect of high-leverage observations on the quality of the normal approximation and the challenge of standard-error estimation with clustered data.\footnote{\textcite{young21iv} discusses these problems in an \ac{IV} context.}
With~\eqref{eq:normality} as foundation, we derive finite-sample properties of the \ac{IV}
estimator:
\begin{equation}\label{eq:iv_estimator}
\hat{\beta}_{IV} = \frac{\hat{\delta}}{\hat{\pi}},
\end{equation}
and the corresponding Wald test of a true null hypothesis. The latter is based on the $t$-statistic centered at the parameter of interest, $\beta$, divided by the estimated \ac{IV} standard error, $\hat{\sigma}_{IV}$:
\begin{equation}\label{eq:wald_CI}
t_{W}=\frac{\hat{\beta}_{IV}-\beta}{\hat{\sigma}_{IV}}; \qquad
\hat{\sigma}_{IV}^2=\frac{ \sdrf^{2}-2\covrf\hat{\beta}_{IV}+
\sdfs^{2}\hat{\beta}_{IV}^{2} }{\hat{\pi}^2},
\end{equation}
where $\hat{\sigma}_{IV}^2$ estimates the asymptotic variance of $\hat{\beta}_{IV}$ in the usual $n \rightarrow \infty$ sequence. The corresponding theoretical variance is ${\sigma}_{IV}^2=(\sdrf^{2}-2\covrf\beta+ \sdfs^{2}\beta^{2})/\pi^2$. In a homoskedastic model with constant causal effects, this simplifies to the familiar formula
\begin{equation*}
{\sigma}_{IV}^2=\frac{\sigma_{\epsilon}^{2}}{n E[\tilde{Z}_i^2]{\pi}^2},
\end{equation*}
where $\sigma_{\epsilon}^{2}$ is the variance of the residual in the structural equation,
\begin{equation}\label{eq:struc}
Y_{i}=D_{i}\beta+X_{i}'(\psi_{1}-\psi_{2}\beta)+\epsilon_{i},
\end{equation}
that motivates \ac{IV} estimation in the classic linear set-up (the structural residual is $\epsilon_{i}=u_{i}-v_{i}\beta$).
Given the assumption of a known covariance matrix for the first-stage and reduced-form estimates, both $t_W$ and $\hat{\sigma}_{IV}^2$ depend on the data only through
$(\hat{\delta},\hat{\pi})$. These have distributions determined by the two unknown parameters, $\pi$ and $\beta$.
Rather than $\pi$ and $\beta$, however, it is illuminating to characterize finite-sample behavior in terms of a pair of parameters that measures instrument strength and the degree of endogeneity (a reparameterization adopted in \textcite{StSt97} and \textcite{lmmp20}, among others). The first parameter, denoted $E[F]$, is defined as:
\begin{equation*}
E[F]=\pi^{2}/\sdfs^{2}+1.
\end{equation*}
Because $E[F]$ is the expectation of
$F=\hat{\pi}^{2}/\sdfs^{2}$, the F-statistic testing $\pi=0$, it's sometimes called the \emph{population first-stage F-statistic}, a term adopted here.
The second parameter is defined as:
\begin{equation}\label{eq:defrho}
\rho=\corr(\hat{\delta}-\hat{\pi}\beta,\hat{\pi})=
\frac{\sdfs}{
\sqrt{\sdrf^{2}-2\beta\covrf+\sdfs^{2}\beta^{2}}}\times (\covrf/\sdfs^{2}-\beta).
\end{equation}
Under independent heteroskedastic errors, $\rho$ is also given by
$\corr(\tilde{Z}_{i}\epsilon_{i},\tilde{Z}_{i}v_{i})$. When, in addition, the errors
$(u_{i}, v_{i})$ are homoskedastic, $\rho=\corr(\epsilon_{i}, v_{i})$, where
$\epsilon_{i}$ is the structural residual in \eqref{eq:struc}. We therefore
refer to $\rho$ as (the degree of) \emph{endogeneity}.\footnote{This simplification is obtained using the fact that, under homoskedasticity, the variance of $v_{i}$ is
$\sigma_{v}^{2}=\sdfs^{2} \cdot n E[\tilde{Z}_i^2]$ and the variance of $\epsilon_i$ is
$\sigma_{\epsilon}^{2}=(\sdrf^{2}-2\beta\covrf+\sdfs^{2}\beta^{2})\cdot n E[\tilde{Z}_i^2]$, with $\cov(v_i, \epsilon_i)=(\covrf-\beta\sdfs^{2})\cdot n E[\tilde{Z}_i^2]$. The homoskedastic formula for the variable of $\epsilon_i$ also leads yields the simplification of the formula for $\sigma^2_{IV}$ noted above.}
With weak instruments as well as homoskedastic error terms, $\rho$ is proportional to the bias in the \ac{OLS} estimand.
This can be seen by using the first-stage and reduced-form equations to write
the \ac{OLS} slope coefficient, $ \beta_{OLS}$, as follows:
\begin{align}
\beta_{OLS}&=\frac{E[\tilde{D}_{i} Y_i]}{E[\tilde{D}_i^2]} =\frac{E[\tilde{Z}_i^2]\pi^2\beta+E[u_{i}v_{i}]}{
E[\tilde{Z}_i^2]\pi^2+\sigma_{v}^2} \nonumber\\
&=R^2_{p}\beta+
(1-R^2_{p}) \frac{E[u_{i}v_{i}]}{\sigma_{v}^2}; &R^2_p&=\frac{E[\tilde{Z}_i^2]\pi^2}{E[\tilde{Z}_i^2]\pi^2+\sigma_{v}^2}\label{eq:def_partialR2},
\end{align}
where $\tilde{D}_i$ is the residual from a regression of $D_i$ on $X_i$, and $\sigma^2_{v}=E[v_i^2]$. The weight multiplying $\beta$ in \eqref{eq:def_partialR2}, denoted $R^2_p$, is the population partial $R^2$ generated by adding the instrument to the first-stage regression. When the instrument is weak, $R^2_p$ is close to zero, and \eqref{eq:def_partialR2} is approximately
$E[u_{i}v_{i}]/\sigma_v^2$. The OLS estimand likewise converges to $E[u_{i}v_{i}]/\sigma_v^2$ in the \textcite{StSt97} weak-instrument sequence (which takes $\pi\to 0$). This in turn equals
$\covrf/\sdfs^2$ under homoskedasticity, so the second term on the right-hand side of \eqref{eq:defrho},
\begin{equation}\label{eq:betaols_definition}
\covrf/\sdfs^{2}-\beta=\beta_{WOLS}-\beta,
\end{equation}
is the weak-instrument OVB of OLS (where we've introduced the notation $\beta_{WOLS}$ for $\covrf/\sdfs^{2})$.
\section{Rejection Rates in Theory and Practice}\label{sec:coverage_contours}
We're interested in the bias of $\hat{\beta}_{IV}$ and $t$-test rejection rates when the null hypothesis is true. Beginning with the latter, the null rejection rate for a two-sided $t$-test with level $\alpha$ is the probability that the absolute value of a $t$-statistic, $\abs{t_W}$, exceeds $z_{1-\alpha/2}$, the $1-\alpha/2$ quantile of a standard normal distribution. This is:
\begin{equation*}
R_W= P_{E[F], \rho}(\abs{t_W}>z_{1-\alpha/2}),
\end{equation*}
where $P_{E[F], \rho}$ is the distribution of $t_W$ parameterized by $E[F],\rho$. As in \textcite{StYo05}, $R_W$ can be evaluated by rewriting $t_W$ in terms of the \textcite[AR]{AnRu49} statistic,
\begin{equation}\label{eq:t_AR}
t_{AR}=\frac{\hat{\delta}-\hat{\pi}\beta}{\sqrt{\sdrf^{2}-2\covrf\beta+\sdfs^{2}\beta^{2}}},
\end{equation}
and the first-stage $t$-statistic, $t_1=\hat{\pi}/\sdfs$, the square of which is the first-stage $F$ statistic.
Our interest in $t_{AR}$ is primarily computational (see \Cref{sec:t_test_rejection} for details). We note, however, that $t_{AR}$ differs from $t_W$ in that it replaces $\hat{\beta}_{IV}$ with the null value of $\beta$ in the formula for $\hat{\sigma}_{IV}^2$. When testing the null that $\beta=0$, $t_{AR}$ for a just-ID model is the $t$-statistic for the associated reduced form. The AR statistic is immune to weak instruments. Confidence intervals derived by inverting $t_{AR}$, however, have infinite length when $F<z_{1-\alpha/2}^2$ (i.e. $1.96^2$ for $95\%$ confidence intervals), and they are always wider than conventional intervals when the AR intervals are finite \parencite[see, e.g.,][]{lmmp20}. We therefore focus on the conventional $t$-statistic, $t_{W}$.\footnote{See \textcite{AnRu49} for the general form of the AR statistic in over-identified models with a fixed number of instruments and \textcite{MiSu21} for an adaptation to models with many weak instruments. }
Summarizing the behavior of a conventional 5\% nominal test, \Cref{fig:wald_unconditional} depicts rejection rates for $t_W$ as a contour plot indexed by $\rho$ and $E[F]$. The figure shows that rejection rates greatly exceed the nominal 5\% rate only if the instrument is weak (i.e., $E[F]$ is close to $1$) \emph{and} endogeneity is high. In particular, if $\abs{\rho}<0.76$, rejection rates are below 10\%, regardless of the strength of the first stage. And if $\abs{\rho}<0.565$, the nominal 5\% test under-rejects (this cutoff is also noted in \cite{lmmp20}). A simple corollary, further substantiated below, is that the coverage of conventionally-computed nominal 95\% confidence intervals for $\hat{\beta}_{IV}$ is likely to be satisfactory in most applications.
The modest over-rejection seen in the figure is explained by a signal feature of just-ID IV\@: the bias of $\hat{\beta}_{IV}$ rises as the instrument grows weaker, but precision falls apace. The \ac{IV} standard error reflects this lack of precision well enough that, unless endogeneity is egregious, inference is distorted little.\footnote{In an online note, Angrist and Pischke make a similar point (\url{http://www.mostlyharmlesseconometrics.com/2009/05/just-identified-iv/}).} This contrasts with overidentified 2SLS with many weak instruments (as in \textcite{bjb95,bekker94}), where, bias notwithstanding, the usual standard errors for 2SLS remain small enough for the $t$-statistic to be misleading.
Our conclusions here also contrast with those drawn in \textcite{StYo05} and \textcite{lmmp20} regarding the reliability of inference based on a conventional just-ID IV\@ $t$-statistic. Although \textcite{lmmp20} report a plot similar to \Cref{fig:wald_unconditional}, both studies emphasize worst-case rejection rates over $\rho$, for a given $E[F]$. As can be seen in \Cref{fig:wald_unconditional}, this worst-case rejection rate occurs at $\abs{\rho}=1$. In the same spirit, \textcite{KeNe21} highlights simulations showing that conventional just-ID \ac{IV} $t$-tests can be misleading when endogeneity is very high. \Cref{sec:rho_evidence,sec:measurement_error} explain why we are not much concerned with high values of $\rho$.
\subsection{The Anatomy of Endogeneity}\label{sec:rho_evidence}
We put endogeneity in context using three applications. These are the AK91 study that launched the modern weak instruments literature, the AE98 estimates using a dummy for samesex sibships (of first- and second-born children) as an instrument for family size, and the AL99 fuzzy regression discontinuity estimates of class size effects. The AE98 and AL99 first-stage $t$-statistics exceed those for AK91 and are arguably out of the zone where an instrument might be considered weak. With a first-stage $t$-statistic of almost 8, the AK91 quarter-of-birth instrument also seems strong enough. But all three studies can be used to calibrate endogeneity and to document contextual features that constrain it.
\Cref{IVreps} reports key statistics for specifications drawn from each
study (some estimates in the table differ slightly from those in the
original). The first row in Panel A shows estimates of the economic returns to
schooling in the AK91 sample of men born 1920--29. Here, OLS and \ac{IV} equal $0.08$
and $0.072$, respectively. These are close, so endogeneity is especially small
in this case, with an estimated $\rho$ of only $0.043$. Schooling returns estimated in the
second AK91 sample, consisting of men born 1930--39, exhibit more OVB\@. In this
sample, the \ac{IV} estimate of $0.105$ surprisingly exceeds the OLS estimate of
$0.071$ (the usual story is that omitted ability controls causes OLS estimates of the returns to schooling to be too large). Endogeneity is correspondingly larger at $\rho=-0.175$, but still well outside
the danger zone illustrated in \Cref{fig:wald_unconditional}.\footnote{Estimates of $\rho$ in the table are computed as described in \Cref{sec:estimate_rho}. These estimates may differ from $\rho$ because $\hat{\beta}_{IV}$ is biased, and because of sampling variance. In the examples analyzed here, however, the instruments are not particularly weak, so the bias in estimated endogeneity is negligible.}
The AK91, AE98, and AL99 studies span a range of OVB scenarios, from modest in the first AK91 sample, to substantial in AE98 (where OLS magnitudes consistently exceed \ac{IV} by at least 50\%), to dramatic in AL99 (where \ac{IV} exceeds small, insignificant OLS estimates, mostly by an order of magnitude, and sometimes with a sign flip). Yet, the magnitude of endogeneity exceeds $0.40$ in only one specification, that for reading scores in the AL99 discontinuity sample (which consists of classes in schools with enrollment near the cutoff that determines class size). The just-ID \ac{IV} estimates in all three of these studies are therefore unlikely to be compromised by finite-sample bias.\footnote{Estimates of $\rho$ computed under homoskedasticity in the AK91 and AE98 samples are almost identical to those reported in \Cref{IVreps}. For the AL99 samples, endogeneity parameters computed using homoskedasticity are smaller; these are reported as $\rho^*$ in the table.}
Although the consistently moderate levels of endogeneity documented in \Cref{IVreps} does not make a theorem, these applications share features with many IV-driven microeconometric investigations of causal effects. First, measured against endogenous variable standard deviations, first-stage magnitudes are small. In particular, the range of first-stage impact in \cref{eq:fs} runs from under 4\% of the endogenous variable standard deviation in AK91, to about 12\% of the endogenous variable standard deviation in AE98, with the AL99 first stage in-between.
Most importantly, endogeneity in research on causal effects is often capped by the modest size of the effects of interest. To make this point, it's helpful to write $\rho$ as a function of OVB\@. Using \cref{eq:def_partialR2,eq:betaols_definition}, we can express $\rho$ under homoskedasticity as:
\begin{align}\label{eq:defovb}
\rho&=
\frac{\sigma_{v}}{
\sigma_{\epsilon}}(\beta_{WOLS}-\beta)\\
&=\frac{\sigma_{v}}{\sigma_{\epsilon}} \left( \frac{\beta_{OLS}-\beta}{1-R^2_{p}} \right) \approx \frac{\sigma_{D}}{\sigma_{Y}}(\beta_{OLS}-\beta).\nonumber
\end{align}
We can use this expression to compute $\rho$ by replacing $\beta$ with $\hat{\beta}_{IV}$. The relevance of this representation of $\rho$ can be seen in the AE98 estimates of the effects a third child on weeks worked by women aged 21--35 in the 1980 Census. Here, the first-stage partial R-square ($R^2_{p}$) is close to zero, while the difference between the conventional OLS estimate and the corresponding \ac{IV} estimate is $3.42$. The term multiplying this, $\frac{\sigma_{v}}{\sigma_{\epsilon}}$, is well-approximated by the ratio of the endogenous variable standard deviation to the dependent variable standard deviation (denoted $\frac{\sigma_{D}}{\sigma_{Y}}$), a ratio of about $0.022$. The product of these two terms gives $0.075$, equal to the value of $\rho$ reported in the table for this sample.
\Cref{eq:defovb} suggests a number of bounds on endogeneity. For starters, in the AK91 scenario, it seems reasonable to assume that the (causal) economic returns to schooling are no more than double the OLS estimate and certainly positive. Under these restrictions, the descriptive statistics in \Cref{IVreps}, which approximate $\frac{\sigma_{v}}{\sigma_{\epsilon}}$ at around 5.2,
suggest $\abs{\rho}$ can be no more than about $0.41$. Although substantial, this is well below the $0.76$ threshold for coverage concerns. With $\beta$ bounded below by zero, large magnitudes of $\rho$ require $\beta$ to far exceed $\beta_{WOLS}$, Only when the causal effect of schooling is triple the OLS estimate (so that OLS is \textit{too small} by $0.15$) does the endogeneity danger zone become relevant.
Many microeconometric \ac{IV} applications involve linear probability models in
which causal effects are changes in probabilities. The AE98 estimates of the
effect of the birth of a third child on female labor force participation in
1980, for example, range from roughly $-0.18$ for OLS to $-0.12$ for IV\@. Labor
force participation rates for women with only two children run around 57\%.
Causal effects might therefore be as large as $-0.57$, but no larger, since
probabilities can't be negative. In this case,
$\frac{\sigma_{v}}{\sigma_{\epsilon}}$ is about 1 (again, using standard
deviations in the data rather than residuals), so $\beta_{OLS}-\beta$ can be no
larger than $-0.18+.57=0.39$, thereby bounding $\rho$ at this value. This generous bound makes no use of
the fact that selection bias is likely to make OLS estimates of family-size effects on female supply too large (in magnitude) rather
than too small. Other applications with Bernoulli outcomes admit similar sorts
of arguments.\footnote{In the same spirit, \textcite{ImRu97complier} and
\textcite{kitagawa15} use the effects of an instrument on distributions of
continuously distributed variables to construct specification tests for
just-ID \ac{IV} models. These tests use the fact that effects on distributions of outcomes are naturally bounded.}
A related argument, appropriate for models with continuous outcomes, shows endogeneity to be constrained by plausible values for causal effects measured in standard deviation units. This line of reasoning is especially apt for education research where standard-deviation-denominated effect sizes are widely reported. The influential Tennessee STAR class size experiment analyzed in \textcite{krueger99} generated a reduction of 7 students per class, roughly one standard deviation of class size in the AL99 data. The STAR experiment yielded treatment effects of about $0.2\sigma$, a reasonably typical impact for educations interventions deemed to have been effective. At the same time, education researchers often see effect sizes as large as half a standard deviation in the outcome distribution as rare, if not implausible. Using the fact that $\frac{\sigma_{v}}{\sigma_{\epsilon}}$ is about equal to $(1-R^2_{p})$ in the AL99 data, the scenario of a half-standard deviation effect size generated by a one-standard deviation reduction in class size implies $\frac{\sigma_{v}}{\sigma_{\epsilon}}\frac{\beta}{1-R^2_{p}} = -0.5$ on the second line of \cref{eq:defovb}. At the same time, OLS estimates of class size effects in AL99 are mostly zero (as is often found in class size research; see e g., \textcite{hanushek86}), so the magnitude of endogeneity is capped at $0.5$.
Contributing to all three of these arguments is the fact that endogeneity under homoskedasticity can be split into the difference between two R-squared-like terms:
\begin{align}\label{eq:defR2}
\rho \approx \frac{\sigma_{D}}{\sigma_{Y}}(\beta_{OLS}-\beta) = \frac{\sigma_{D}}{\sigma_{Y}}\beta_{OLS}-\frac{\sigma_{D}}{\sigma_{Y}}\beta.
\end{align}
The square of the first term, $(\frac{\sigma_{D}}{\sigma_{Y}}\beta_{OLS})^2$, is the variation in the dependent variable accounted for by $D_i$ in an analysis-of-variance for $Y_i$. In microeconometric applications, this term is mostly small, as is the causal analog, $(\frac{\sigma_{D}}{\sigma_{Y}}\beta)^2$. This fact limits the magnitude of the difference between them.\footnote{\textcite{KeNe21} propose to bound $\rho$ in the
context of estimates of the economic returns to schooling based on the view
that OLS returns should exceed causal effects. Although this seems defensible, it's worth noting that the literature surveyed by
\textcite{card01lecture} reports many \ac{IV} estimates in excess of the corresponding OLS estimates, a pattern first highlighted by \textcite{lang93}.}
\subsection{When Measurement Error Motivates IV}\label{sec:measurement_error}
In addition to \ac{IV} for causal effects, a second major arena for microeconometric \ac{IV} involves models with measurement error. Suppose the regression of interest is
$Y_i=D_i^{*}\beta+X_i'\gamma+\eta_{i}$, where $\eta_i$ is a residual uncorrelated with $(D^*_i, X_i)$ by definition. The regressor $D_i^{*}$ is unobserved; we see only a noisy measure, $D_i=D_i^{*}+e_i$, where the measurement error, $e_i$, is assumed to be classical, that is uncorrelated with $(D_i^{*},X_i,\eta_i)$. Replacing $D_i^{*}$ with $D_i$ yields the structural equation to be instrumented:
\begin{equation*}
\begin{split}
Y_i &=D_i\beta+X_i'\gamma+(\eta_{i}-e_i\beta) \\
&=D_i\beta+X_i'\gamma+\epsilon_i,
\end{split}
\end{equation*}
where $\epsilon_{i}=\eta_{i}-e_i\beta$ is the structural residual. Given an instrument correlated with $D_i^{*}$ and uncorrelated with $\epsilon_i$, the coefficients of interest are consistently estimated by IV\@. The first stage in this scenario can be written as in \eqref{eq:fs}, with error term $v_i$.
To calibrate endogeneity in the model, note first that, given the classical measurement error assumption, $\cov(v_i, \epsilon_i)=-\sigma^2_e\beta$. Under homoskedasticity, endogeneity squared can therefore be written:
\begin{equation}\label{eq:r}
\rho^2=\frac{\sigma^4_e\beta^2}{\sigma_v^2\sigma^2_{\epsilon}}
=\frac{\sigma^4_e\beta^2}{\sigma_v^2(\sigma^2_{\eta}+\beta^2\sigma^2_{e})}
\leq
\frac{\sigma^2_e}{\sigma_v^2}
%
= \frac{1-r}{1-R^2_{p}},
\end{equation}
where $r=\sigma^2_{\tilde{D}^*}/\sigma^2_{\tilde{D}}$ denotes the reliability or signal-to-noise ratio of mismeasured $D_i$, after partialing out covariates.\footnote{The first equality follows by definition of correlation, the middle inequality uses the fact that $\sigma^2_{\eta}$ must be non-negative, and the last equality uses the definition of partial $R^2$ in \cref{eq:def_partialR2}.}
Although we can't speak to reliability across all fields, labor economists have evidence on the reliability of a few key variables of interest. These variables include schooling, earnings, hours worked, and hourly wages. Schooling appears often on the right-hand side of wage equations, while earnings, hours, and hourly wages are used in various configurations to estimate labor supply elasticities. The \textcite{AnKr99handbook} summary of reliability estimates suggests $r \approx 0.9$ for schooling and $r \approx 0.8$ for earnings, falling to about $0.65-0.75$ for hours worked and hourly wages. The lower end of this range may be more relevant for wage reliability after partialing out covariates.\footnote{See Table 11 in \textcite{AnKr99handbook}. The reliability of variables other than wages is less sensitive to covariate adjustment.} With $r=0.65$ as a reasonably conservative value, we'd need
to see $R_p^2$ of at least $0.4$ for $\rho$ to reach $0.76$. But $E[F]=\frac{nR^2_p}{1-R^2_p}+1$, so, at this level of first-stage fit, $E[F]$ is way out of the trouble zone for any sample size that's empirically relevant. This suggests that, unless reliability is unusually low, microeconometric measurement error can be expected to generate parameter combinations for which conventional \ac{IV} inference is trouble-free.
\section{Bias Under a Good Sign}\label{sec:signrestrict}
Having made an empirical case for a sanguine view of just-ID IV, we add a novel analytical argument. This builds on the idea that \ac{IV} identification strategies are most credible when a strong institutional or theoretical foundation explains the first stage. These foundations usually imply a sign for $\pi$. In the AK91 application, for example, the quarter-of-birth first stage arises from the fact that children born later in the year enter school younger, and are therefore constrained by compulsory attendance laws to stay in school longer than those born earlier. The AE98 samesex instrument for family size is predicated on parents' preference for mixed-sex sibships. The AL99 Maimonides Rule instrument for class size is derived from regulations that determine class size as a function of enrollment. In these and many other applied micro applications, institutions or preferences determine the sign of $\pi$.
\textcite{AnAr17} use this insight to motivate an \ac{IV} estimator that is unbiased
under a first-stage sign restriction. Without loss of generality, assume
$\pi>0$, and let $\mu(x)=\frac{1-\Phi(x)}{\phi(x)}$ denote the Mills' ratio of a
standard normal random variable, where $\phi(x)$ and $\Phi(x)$ are the standard
normal density and cdf evaluated at $x$. The Andrews-Armstrong unbiased \ac{IV} estimator, denoted
$\hat{\beta}_{U}$, exploits the fact that when first-stage estimates are
normally distributed,
\begin{equation*}
E\left[\frac{\mu(t_1)}{\sigma_{\hat{\pi}}}\right]=\frac{1}{\pi}.
\end{equation*}
In other words, $\hat{\tau}\equiv\mu(t_1)/\sigma_{\hat{\pi}}$ is an unbiased estimator of the reciprocal of the first-stage coefficient, $\pi$. Define
\begin{equation}\label{eq:wted}
\hat{\beta}_{U}
\equiv \hat{\tau}(\hat{\delta}-\beta_{WOLS}\hat{\pi})+\beta_{WOLS}
=t_{1}\mu(t_{1})\hat{\beta}_{IV}+(1-t_1\mu(t_1))\beta_{WOLS}.
\end{equation}
Finally, recall that $\beta_{WOLS}$ is the slope from a regression of the estimated reduced form on the estimated first stage. $\hat{\delta}-\beta_{WOLS}\hat{\pi}$ and $\hat{\tau}$ are therefore uncorrelated, and since $E[\hat{\delta}-\beta_{WOLS}\hat{\pi}]=(\beta-\beta_{WOLS})\pi$, it follows that $\hat{\beta}_U$ is unbiased for $\beta$. Moreover, $\hat{\beta}_U$ is a linear combination of conventional IV estimates and $\beta_{WOLS}$ with coefficients, or weights, given by $t_1\mu(t_1)$ and $1-t_1\mu(t_1)$.
To interpret $\hat{\beta}_{U}$, observe that $t_1\mu(t_1)=\hat{\pi}\hat{\tau}$, so that the weights in \eqref{eq:wted} reflect the extent to which $\hat{\tau}$ differs from $1/\hat{\pi}$. If $t_1>0$, these weights are bounded by a classic Mills' ratio inequality \parencite[e.g.][p.~175]{feller68} that implies:
\begin{equation}\label{eq:mills}
0\leq 1-t_1\mu(t_1)\leq \frac{1}{t_1^2}.
\end{equation}
Thus, when the first stage is right-signed, the weights $t_{1}\mu(t_{1})$ in \cref{eq:wted} lie between $0$ and $1$, and we can interpret $\hat{\beta}_{U}$ as shrinking the conventional \ac{IV} estimate towards \ac{OLS}. The amount of shrinkage is bounded by the reciprocal of $t_{1}^{2}$, that is, by $1/F$. When $F=t_{1}^{2}=10$, the unbiased estimator shrinks $\hat{\beta}_{IV}$ at most 10\% of the way towards $\beta_{WOLS}$.
The shrinkage interpretation of $\hat{\beta}_{U}$ seems surprising:
since $\hat{\beta}_{IV}$ is biased towards OLS, shrinkage towards
OLS increases bias. This counterintuitive fact is reconciled with the unbiasedness of $\hat{\beta}_{U}$ by the following theorem:
%
\begin{theorem}\label{theorem:bias_betaU}
Consider the model in \eqref{eq:normality}, and suppose that $\pi>0$. Let
$\lambda=\pi/\sdfs=\sqrt{E[F]-1}$. Then the relative mean bias of $\hat{\beta}_{U}$
conditional on $t_{1}>0$ can be written:
\begin{equation*}
\frac{E[\hat{\beta}_{U}-\beta\mid t_{1}>0]}{\beta_{WOLS}-\beta}%
%
= \sqrt{\pi/2}\frac{\phi(\lambda)}{\Phi(\lambda)},
\end{equation*}
while, conditional on $t_{1}<0$, relative mean bias is:
\begin{equation*}
\frac{ E[\hat{\beta}_{U}-\beta\mid t_{1}< 0]}{
\beta_{WOLS}-\beta}=
-\sqrt{\pi/2}\frac{\phi(\lambda)}{(1-\Phi(\lambda))}.
\end{equation*}
\end{theorem}
\noindent The estimator $\hat{\beta}_{U}$ is therefore unbiased because it averages conditional positive bias when $t_{1}>0$ and conditional negative bias when $t_{1}<0$. As in
\textcite{StYo05}, the theorem scales mean bias by the weak-\ac{IV} OVB of the OLS estimand, defined in \cref{eq:betaols_definition}. This simplifies the bias formulas, but the relationship between conditional and unconditional bias stands without this rescaling.\footnote{\textcite{StYo05} focus on relative mean bias for 2SLS models with over-identifying restrictions.}
It is hard to imagine an analyst who is prepared to sign the population first stage while ignoring the sign of the estimated first stage. Such conditioning, however, strips $\hat{\beta}_{U}$ of it its appeal. The next section shows that, conditional on the sign of $t_1$, the bias of $\hat{\beta}_{U}$ exceeds that of $\hat{\beta}_{IV}$.
At the same time, first-stage sign conditioning has salutary effects on conventional IV estimates.
\subsection{Sign-Screened Bias and Coverage}\label{sec:sign_screening}
Suppose we assume $\pi$ is positive and report second-stage estimates only when first-stage estimates are positive as well (equivalently, when $t_1>0$). In contrast with $\hat{\beta}_{U}$, sign-screening reduces the bias of $\hat{\beta}_{IV}$ markedly. Moreover, screening on the estimated first-stage sign has no downside in terms of coverage: in contrast with procedures that screen on the \emph{magnitude} of the first-stage $F$ statistic, screening on the sign of the corresponding $t$-statistic is shown here to have little effect on rejection rates for a conventional second-stage $t$-test.
Because the expectation of a just-ID IV estimator is undefined, our investigation of sign-screening looks at median bias (the expectation of 2SLS exists only for over-identified models). As in \Cref{theorem:bias_betaU}, we scale bias by the weak-\ac{IV} OVB of the OLS estimand. We are interested in the difference between relative median bias computed unconditionally and conditional on $\hat{\pi}>0$.
Surprisingly, \Cref{th:median_bias} below shows that the worst-case relative
bias obtains in the limit as $\abs{\rho}\to 0$ (this is not the same as relative bias when
$\rho=0$; with no endogeneity, both \ac{IV} and OLS are unbiased, so that relative
bias is discontinuous in $\rho$). The relationship between $\rho$ and relative median bias
therefore contrasts with that in \Cref{sec:coverage_contours}, which shows
higher endogeneity leading to worse coverage. This reversal reflects the fact
that, although the bias of $\hat{\beta}_{IV}$ increases with endogeneity, OVB
increases faster. Because modest endogeneity is empirically relevant, median
relative bias is reasonably characterized by the worst case over
$\rho$. This leads us to the following theoretical result:
%
\begin{theorem}\label{th:median_bias}
Consider the model in \cref{eq:normality}, and suppose that $\pi>0$. Let
$\lambda=\pi/\sdfs=\sqrt{E[F]-1}$. Then, for a given $\lambda$,
unconditional relative median bias of $\hat{\beta}_{IV}$ is characterized by
\begin{equation*}
\sup_{\rho}\abs*{\frac{\operatorname{median}_{E[F], \rho}(\hat{\beta}_{IV}-\beta)}{\beta_{WOLS}-\beta}}=
\frac{\phi(\lambda)}{\lambda[\Phi(\lambda)-1/2]+\phi(\lambda)}.
\end{equation*}
Moreover, if $\lambda\geq 0.84$, the relative median
bias of $\hat{\beta}_{IV}$ conditional on $\hat{\pi}>0$ satisfies
\begin{equation*}
\sup_{\rho}\abs*{\frac{\operatorname{median}_{E[F], \rho}(\hat{\beta}_{IV}-\beta\mid \hat{\pi}>0)}{\beta_{WOLS}-\beta}}
=\frac{\phi(\lambda)}{
\lambda\Phi(\lambda)+\phi(\lambda)}.
\end{equation*}
Equivalently, these expressions give the limit of relative unconditional and conditional
median bias as $\abs{\rho}\to 0$.
\end{theorem}
\Cref{fig:bias} plots the two bias expressions in the theorem. We see that (i) unconditional relative bias falls rapidly as first-stage strength increases, and (ii) unless $E[F]$ is exceedingly low, conditional median bias is only about half as large as the corresponding unconditional bias. This is a consequence of the fact that the ratio of conditional to unconditional median bias bounds is:
\begin{equation*}
1 - \frac{0.5\lambda}{\lambda\Phi(\lambda)+\phi(\lambda)}.
\end{equation*}
For $\lambda$ greater than about 1.5, this is within 1 percentage point of 0.50,
since the normal cdf is then close to one and the normal density is close to zero. The shaded regions in \Cref{fig:bias} delineate the range of variation in relative median bias as a function of $\rho$ (see \Cref{sec:medi-bias-comparison} for computational details). We see that the relative median bias varies surprisingly little with $\rho$.
The conditional median bias of $\hat{\beta}_{IV}$ is substantially smaller than both the mean and the median conditional bias of $\hat{\beta}_{U}$. The ratio of the conditional median bias of $\hat{\beta}_{IV}$ to the conditional mean bias of $\hat{\beta}_{U}$ given in \Cref{theorem:bias_betaU} is given by $\sqrt{2/\pi}\cdot \Phi(\lambda)/(\lambda\Phi(\lambda)+\phi(\lambda))\leq 1$.
This is decreasing in $\lambda$, and smaller than $0.5$ once $\lambda$ is greater than about 1.5. \Cref{sec:medi-bias-comparison} shows that the conditional median bias of $\hat{\beta}_{IV}$ is always smaller than that of $\hat{\beta}_{U}$, and at least 50\% smaller once $\lambda\geq 1$.\footnote{In contrast, \textcite{AnAr17} show numerically that the unconditional median bias of $\hat{\beta}_{U}$ is smaller than that of $\hat{\beta}_{IV}$ when $E[F]$ is small, but the bias ordering reverses for larger $E[F]$. They also note that the median absolute deviation of $\hat{\beta}_{U}$ is always smaller than that of $\hat{\beta}_{IV}$. Our numerical calculations indicate that this is no longer true for all parameter values conditional on the estimated first stage sign.}
Screening on the sign of the first-stage estimate clearly mitigates bias. Since many analysts likely pursue a just-ID \ac{IV} identification strategy only when the first-stage estimates are signed as expected, the inner line in \Cref{fig:bias} may be a better guide to the bias in published \ac{IV} estimates. But perhaps this sort of screening is costly. Econometricians have long warned that screening on the first-stage $F$-statistic risks a form of pretest bias and so may do more harm than good. Intuitively, when $\pi$ is truly zero, large $F$-statistics signal realizations in which the in-sample correlation between instruments and structural errors is largest. Consequently, when instruments are weak, large $F$s come with especially misleading $\hat{\beta}_{IV}$ realizations. Perhaps sign-screening runs a similar risk.
By way of evidence on this point, \Cref{fig:wald_conditional} plots rejection contours for a conventional (second-stage) $t$-test conditional on $\hat{\pi}>0$. That is, the figure plots contours for:
\begin{equation*}
R^{c}_{W}= P_{E[F], \rho}(\abs{t_W}>z_{1-\alpha/2}\mid \hat{\pi}>0).
\end{equation*}
As in the unconditional case explored in \Cref{sec:coverage_contours}, this is computed by numerical evaluation of a one-dimensional integral (see \Cref{sec:t_test_rejection} for details).
\Cref{fig:wald_conditional} shows that screening on the first-stage sign affects rejection rates little. For instance, the endogeneity cutoff required to keep rejections rates below
10\% is $\abs{\rho}\leq 0.75$, rather than the unconditional $0.76$. This result
is explained by the fact that, when the instrument is very weak, screening has
two effects. On one hand, the bias of $\hat{\beta}_{IV}$ is reduced. At the same time, screening out wrong-signed first-stage estimates leads to an overestimate of first-stage strength, on average. These two effects are just about offsetting, so that the rejection contours in \Cref{fig:wald_conditional} are much like those in \Cref{fig:wald_unconditional}.
What practical lesson should we draw from this? An empiricist hewing to conventional \ac{IV} reporting strategy---showing reduced-form, first-stage, and \ac{IV} estimates and the associated standard errors---can be reassured that, provided the first stage has the anticipated sign, bias is likely to be minimal and coverage adequate. Only when endogeneity exceeds a level that applied microeconomists rarely encounter is inference likely to be misleading. The careful analyst likewise judges statistical power by standard errors rather than $t$-statistics, and hesitates to declare a finding conclusive based on $\hat{\beta}_{IV}$ alone when the reduced form is statistically uninformative. Just-ID \ac{IV} is a strong hand, but the reduced form, a regression coefficient with anodyne statistical properties, is the \ac{IV} analyst's ace in the hole.
In the context of the AK91, AE98, and AL99 studies, first-stage sign screening adds no action items to the empirical agenda. The first-stage estimates in these applications are robustly right-signed. The reduced forms, reported in detail, are consistent with second-stage estimates in clearly showing that the instrument moves outcomes in a manner implied by plausible first-stage values. In applications with weaker instruments than these, an empirical strategy that begins by examining the first-stage sign would seem to have no downside. Claims of credible causal evidence requires more than this, however. In AK91, for instance, the quarter-of-birth story holds water because schooling can be seen to move sharply up and down with quarter of birth as predicted by compulsory attendance laws, across 30 birth cohorts in three data sets, and because graduate degree completion that should be changed little by compulsory attendance, indeed moves little with quarter of birth. This coherence is part of what gives the AK91 first stage its strength.
\section{Summary and Conclusions}\label{sec:conclusion}
Assuming reduced-form and first-stage estimates are normally distributed, rejection rates for conventional $t$-tests in just-ID IV models are distorted little unless endogeneity is extraordinarily large. A corollary is that conventional IV standard errors are likely to yield confidence intervals with good coverage. Three widely-cited applications, two of which demonstrate considerable OVB in OLS estimates, are characterized by modest endogeneity and consequently fall well inside the low-distortion comfort zone.
We've argued that these three examples should be seen as representative rather than idiosyncratic: the structure of much applied micro research naturally bounds endogeneity.
We've also introduced a new theoretical reason to be unconcerned with the bias of just-ID IV\@. This builds on the \textcite{AnAr17} argument that in credible applications of just-ID \ac{IV}, the analyst is rarely agnostic about the direction of the first stage. Unlike \textcite{AnAr17}, however, we impose the same sign restriction on the estimated as well as the theoretical first stage. Such conditioning roughly halves the median bias of the \ac{IV} estimator. Moreover, in contrast with screening on the first-stage $F$, sign-screening generates no pretest bias. Since most analysts likely impose an estimated first-stage sign screen as a matter of course, the bias reduction sign-conditioning engenders is already reflected in published empirical work.
%
\printbibliography
\newpage
\begin{figure}[htp!]
\centering
\input{wald_unconditional_rej.tex}
\caption{Contour plot of the rejection rate $R_{W}$ of conventional $t$-test with nominal level
$\alpha=0.05$ as function of $E[F]$ and $\rho$.}\label{fig:wald_unconditional}
\end{figure}
\begin{figure}[htp!]
\centering \input{iv_bias_shaded.tex}
\caption{Bound on median bias of $\hat{\beta}_{IV}$ in units of OLS bias, i.e.\ as a fraction of $\abs{\beta_{WOLS}-\beta}$. Dashed line: unconditional median bias. Solid line: median bias conditional on $\hat{\pi}>0$. The yellow and blue shaded areas delineate the range of variation in unconditional and conditional median bias, respectively, over possible values of $\rho$.}\label{fig:bias}
\end{figure}
\begin{figure}[htp]
\centering \input{wald_conditional_rej.tex}
\caption{Contour plot of the rejection rate $R^{c}_{W}$ of conventional $t$-test with nominal level
$\alpha=0.05$, conditional on $\pi>0$, as function of
$E[F]$ and $\rho$.}\label{fig:wald_conditional}
\end{figure}
\begin{table}[H]
\caption{Estimates and Endogeneity in Three \ac{IV} Applications\label{IVreps}}
{\centering
\includegraphics[width = .95\textwidth]{JAtbl1}}
\footnotesize{} Notes: This table reports \ac{IV} and OLS estimates replicating the AK91, AE98, and AL99 studies discussed in the text. For each study, the table reports estimates from multiple samples, as well as the corresponding first-stage estimate and its standard error. The endogeneity parameter appears below columns showing OLS and IV\@. Endogeneity estimates for AL99 marked with an asterisk were computed assuming homoskedasticity. The 2nd, 3rd, and 4th columns report standard deviations in parentheses. Other columns show standard errors in parentheses. These are robust for AK91 and AE98, and clustered on school for AL99.
\end{table}
\begin{appendices}
\crefalias{section}{appsec}
\crefalias{subsection}{appsubsec}
\numberwithin{equation}{section}
\numberwithin{figure}{section}
\section{Derivations and Proofs}\label{sec:derivations}
The appendix uses the notation
$\tilde{\beta}_{IV}=(\hat{\beta}_{IV}-\beta)/\abs{\beta_{WOLS}-\beta}$ and
$\tilde{\beta}_{U}=(\hat{\beta}_{IV}-\beta)/\abs{\beta_{WOLS}-\beta}$ to denote
the IV and the unbiased estimator, after centering and scaling by the weak-IV OVB of OLS\@.
Also, we let $\lambda=\pi/\sdfs=\sqrt{E[F]-1}$, and $s=\rho/\sqrt{1-\rho^{2}}$.
\subsection{Estimating \texorpdfstring{$\rho$}{rho}}\label{sec:estimate_rho}
We estimate $\rho$ (defined in \cref{eq:defrho}) using first-stage and \ac{IV} estimates, and the associated first-stage, reduced-form and \ac{IV} standard errors. To see how this works, rewrite \cref{eq:wald_CI} as
\begin{equation}\label{eq:covrf_formula}
\covrf=
\frac{\sdfs^{2}\hat{\beta}_{IV}^{2} - \hat{\pi}^{2}\hat{\sigma}^{2}_{IV}+\sdrf^{2}}{2\hat{\beta}_{IV}}.
\end{equation}
With this in hand for $\covrf$, endogeneity can be computed as the sample analog of \cref{eq:defrho}, replacing $\beta$ with $\hat{\beta}_{IV}$. The resulting estimator is:
\begin{equation*}
\hat{\rho}=\frac{\sdfs}{\abs{\hat{\pi}} \hat{\sigma}_{IV}}\times
(\covrf/\sdfs^{2}-\hat{\beta}_{IV}).
\end{equation*}
\subsection{\texorpdfstring{$t$}{t}-Test Rejection Rates}\label{sec:t_test_rejection}
This section writes the rejection probabilities of the $t$-test as an integral
indexed by $(E[F],\rho)$. \textcite{StYo05} use Monte Carlo methods to compute
unconditional rejection probabilities in a similar setup. The calculation
described here is faster and more accurate. More importantly, it allows us to
easily compute both the unconditional rejection rates, and rejection rates
conditional on sign-screening.
Using \cref{eq:defrho}, and the fact that $\beta_{WOLS}-\beta$ and $\rho$ have
the same sign, we may write $t_{AR}$ as
\begin{equation}\label{eq:ar_test}
t_{AR}%
%
=\frac{(\hat{\delta}-\hat{\pi}\beta)\abs{\rho}}{\sdfs\abs{\beta_{WOLS}-\beta}}.
\end{equation}
Consequently,
\begin{equation}\label{eq:rewrite-IV}
\tilde{\beta}_{IV}=\frac{\hat{\delta}-\beta\hat{\pi}}{\sdfs t_{1}\abs{\beta_{WOLS}-\beta}}
=\frac{t_{AR}}{\abs{\rho}t_{1}}.
\end{equation}
Thus,
\begin{equation}\label{eq:rewrite-wald}
t_{W}%
%
=\frac{\sign(t_{1})t_{AR}}{
\sqrt{\frac{\sdrf^{2}/\sdfs^{2}-2\beta_{WOLS}\beta+\beta^{2}}{(\beta_{WOLS}-\beta)^{2}}
\rho^{2} +\frac{t_{AR}^{2}}{t_{1}^{2}}-2\rho\frac{t_{AR}}{t_{1}}
} }=\frac{\sign(t_{1})t_{AR}}{
\sqrt{1 +t_{AR}^{2}/t_{1}^{2}-2\rho t_{AR}/t_{1}
} }
\end{equation}
where the first equality uses \cref{eq:rewrite-IV} and the definition of
$\beta_{WOLS}$, and the second equality uses \cref{eq:defrho}. This expression
for $t_{W}$ implies that conditional on $t_1$, the rejection region
$\{\abs{t_{W}}\geq z_{1-\alpha/2}\}$ is quadratic in $t_{AR}$.
%
%
%
%
Solving this quadratic inequality implies that the rejection region is given by
\begin{equation*}
t_{AR}\in \begin{cases}
\emptyset & \text{if $t_1^{2}\leq (1-\rho^{2})z^{2}_{1-\alpha/2}$,} \\
[a_{1},a_{2}] & \text{if $ (1-\rho^{2})z^{2}_{1-\alpha/2}\leq t_1^{2}\leq z^{2}_{1-\alpha/2}$.}\\
(-\infty, a_{2})\cup (a_{1},\infty) & \text{if $t_1^{2}\geq z^{2}_{1-\alpha/2}$}
\end{cases}
\end{equation*}
where
\begin{align*}
a_{1}&=\frac{\rho z^{2}_{1-\alpha/2}t_1-
\abs{t_1}z_{1-\alpha/2}\sqrt{t_1^{2}-(1-\rho^{2})z^{2}_{1-\alpha/2}
}
}{z^{2}_{1-\alpha/2}-t_1^{2}},\\
a_{2}&=\frac{\rho z^{2}_{1-\alpha/2}t_1+
\abs{t_1}z_{1-\alpha/2}\sqrt{t_1^{2}-(1-\rho^{2})z^{2}_{1-\alpha/2}
}
}{z^{2}_{1-\alpha/2}-t_1^{2}}.
\end{align*}
Note that $\corr(t_{AR},t_{1})=\rho$, so that
\begin{equation}\label{eq:ar_given_z}
P(t_{AR}\leq x\mid t_{1})=\Phi((x-\rho (t_{1}-\lambda))/\sqrt{1-\rho^{2}}).
\end{equation}
Thus, conditional on $t_{1}$, the rejection probability is
given by
\begin{equation}\label{eq:conditional_rejection}
\begin{split}
P(\abs{t_{W}}\geq z_{1-\alpha}\mid t_{1})&= (P(t_{AR}\leq a_{2}\mid
t_{1})-P(t_{AR}\leq a_{1}\mid t_{1}))\1{t_{1}^{2}\geq
z_{1-\alpha/2}^{2}(1-\rho^{2})}\\
&\qquad +\1{t_{1}^{2}\geq z_{1-\alpha/2}^{2}}\\
& = f(t_{1};\lambda,\rho)\1{t_{1}^{2}\geq(1-\rho^{2})
z_{1-\alpha/2}^{2}}+\1{t_{1}^{2}\geq z_{1-\alpha/2}^{2}},
\end{split}
\end{equation}
where
\begin{equation*}
f(t_{1};\lambda,\rho)=
\Phi\left(\frac{a_{2}-\rho
(t_{1}-\lambda)}{\sqrt{1-\rho^{2}}}\right)- \Phi\left(\frac{a_{1}-\rho
(t_{1}-\lambda)}{\sqrt{1-\rho^{2}}}\right).
\end{equation*}
Since $t_{1}\sim\mathcal{N}(\lambda, 1)$, the rejection probability conditional
on $t_{1}\geq c$ is therefore given by
\begin{equation*}
P(\abs{t_{W}}\geq z_{1-\alpha}\mid t_{1}\geq c)=\frac{
\int_{c}^{\infty} (\1{t_{1}^{2}\geq (1-\rho^{2})z_{1-\alpha}}f(t_{1};\lambda,\rho)+\1{t_{1}^{2}\geq z_{1-\alpha/2}^{2}})
\phi(t_{1}-\lambda)dt_{1}}{
\Phi(\lambda-c)}.
\end{equation*}
The unconditional rejection probability $R_{W}$ obtains by setting $c=-\infty$.
The rejection probability conditional on sign screening,
$R^{c}_{W}$, obtains by setting $c=0$. The coverage contours in
\Cref{fig:wald_unconditional,fig:wald_conditional} evaluate the above expression
as a function of $(\rho,\lambda)$ by numerical integration.
\subsection{Proof of Theorem~\ref{theorem:bias_betaU}}\label{sec:proof-betaU}
We may write
\begin{equation}\label{eq:tildebetaU}
\tilde{\beta}_{U}=t_{1}\mu(t_{1})\tilde{\beta}_{IV}+(1-t_1\mu(t_1))\sign(\rho)
=\mu(t_{1})\frac{t_{AR}}{\abs{\rho}}+(1-t_1\mu(t_1))\sign(\rho)
\end{equation}
where the first equality follows from \cref{eq:wted}, and the fact that $\beta_{WOLS}-\beta$ and $\rho$ have the
same sign, and the second equality applies \cref{eq:rewrite-IV}.
Since $E[t_{AR}\mid t_{1}]=\rho(t_{1}-\lambda)$, the relative bias conditional
on $t_{1}$ is given by
\begin{equation*}
E[\tilde{\beta}_{U}\mid t_{1}]=\sign(\rho)\left[
1-\lambda\mu(t_{1})\right].
\end{equation*}
By arguments analogous to those in the proof of Lemma 2.1 in \textcite{AnAr17},
we have
\begin{equation*}
\begin{split}
E[\lambda\mu(t_{1})\mid t_{1}>0] &
=\frac{1}{\Phi(\lambda)}\int_{t=0}^{\infty}\lambda\frac{1-\Phi(t)}{\phi(t)}\phi(t-\lambda)dt
=
\frac{e^{-\lambda^{2}/2}}{\Phi(\lambda)}\int_{t=0}^{\infty}(1-\Phi(t))\cdot \lambda e^{\lambda t}dt\\
&=\frac{e^{-\lambda^{2}/2}}{\Phi(\lambda)}\left\{ \left[e^{\lambda
t}(1-\Phi(\lambda))\right]_{t=0}^{\infty}
+\int_{t=0}^{\infty}\phi(t)e^{\lambda t}dt \right\}\\
&= \frac{1}{\Phi(\lambda)}\left[-\frac{1}{2}e^{-\lambda^{2}/2}
+\int_{t=0}^{\infty}\phi(t-\lambda) dt\right]
=-\frac{1}{2}\frac{e^{-\lambda^{2}/2}}{\Phi(\lambda)}+1,
\end{split}
\end{equation*}
where the first line uses the definition of a Mills' ratio, second line uses
integration by parts, and the third follows by ``completing the square''. It
therefore follows that
\begin{equation*}
\frac{E[\hat{\beta}_{U}-\beta\mid t_{1}>0]}{\beta_{WOLS}-\beta}=
\frac{1}{2}\frac{e^{-\lambda^{2}/2}}{\Phi(\lambda)}
= \sqrt{\pi/2}\frac{\phi(\lambda)}{\Phi(\lambda)}.
\end{equation*}
The second claim follows by an analogous argument.
\subsection{Proof of Theorem~\ref{th:median_bias}}\label{sec:proof-median_bias}
The distribution of $\tilde{\beta}_{IV}$ conditional on $t_{1}$
can then be written as
\begin{equation}\label{eq:tilde_beta_distro}
\begin{split}
P(\tilde{\beta}_{IV}\leq x\mid t_{1};s)&
=\begin{cases} P(t_{AR} \leq t _{1}x\abs{\rho}\mid t_{1}) =
\Phi(s[\lambda-(1-\sign(s) x) t_{1}])
& \text{if $t_{1}\geq 0$,} \\
P(t_{AR}\geq t_{1} x\abs{s}\mid t_{1})=
\Phi(-s[\lambda-(1-\sign(s) x) t_{1}])& \text{if $t_{1}<0$.}
\end{cases}\\
&=\Phi\left(s[\sign(t_{1})\lambda-(1-\sign(s) x) \abs{t_{1}}]\right).
\end{split}
\end{equation}
where the first equality uses~\cref{eq:rewrite-IV}, and the second equality
follows from \cref{eq:ar_given_z}. Observe that since
$P(\tilde{\beta}_{IV}\leq x\mid t_{1};-s)=1-P(\tilde{\beta}_{IV}\leq -x\mid
t_{1};-s)$, the distribution is symmetric in $s$. It therefore suffices to
consider $s>0$.
We first prove the claims concerning the distribution of $\tilde{\beta}_{IV}$
conditional on $t_{1}>0$. By~\cref{eq:tilde_beta_distro}, this distribution is
given by
\begin{equation}\label{eq:cdf_conditional}
P(\tilde{\beta}_{IV}\leq x\mid t_{1}>0;s)
=\frac{1}{\Phi(\lambda)} \int_{-\lambda}^{\infty}
\Phi\left(s[\lambda-(1-x)t_{1}]\right)\phi(z)dz.
\end{equation}
Observe that the conditional median, denoted $m_{c}=m_{c}(s)$, of
$\tilde{\beta}_{IV}$ is smaller than $1$, since
\begin{equation}\label{eq:median_lt1}
P(\tilde{\beta}_{IV}\leq 1\mid t_{1}>0;s)
=\frac{\Phi\left(s\lambda\right)}{\Phi(\lambda)} \int_{-\lambda}^{\infty}
\phi(z)dz=\Phi(s\lambda)> 1/2.
\end{equation}
Next, by the mean value theorem, for some $\tilde{s}=\tilde{s}(x, s)\in[0,s]$,
\begin{multline*}
P(\tilde{\beta}_{IV}\leq x\mid t_{1}>0;s)= \Phi\left(0\right)
+\frac{s}{\Phi(\lambda)} \int_{-\lambda}^{\infty} ((x-1)z+x\lambda)
\phi\left(s ( (x-1)z+x\lambda)\right)\phi(z)dz\\
=\frac{1}{2} +\frac{s}{\tilde{s}^{2}(1-x)\Phi(\lambda)} \int_{-\infty}^{\tilde{s}\lambda} y
\phi\left(y\right)\phi(a+by)dy,
\end{multline*}
where the second line uses the change of variables
$y=\tilde{s}x\lambda-\tilde{s}(1-x)z$, and we let $a=x\lambda/(1-x)$,
$b=-\frac{1}{\tilde{s}(1-x)}$. By line 111 of Table 1 in \textcite{owen80},
\begin{equation}\label{eq:owen111}
\int
x\phi(x)\phi(a+bx)=\frac{\phi(a/t)}{t^{2}}\left[
-\phi(tx+ab/t)-\frac{ab}{t}\Phi(tx+ab/t)\right],\quad t=\sqrt{1+b^{2}}.
\end{equation}
Applying this result to the preceding display then yields
\begin{multline*}
P(\tilde{\beta}_{IV}\leq x\mid t_{1}>0;s) =\frac{1}{2}
+\frac{s}{\tilde{s}^{2}(1-x)\Phi(\lambda)} \frac{\phi(a/t)}{1+b^{2}}\left[
-\phi\left(t\tilde{s}\lambda+ab/t\right)
-\frac{ab}{\sqrt{1+b^{2}}}\Phi(t\tilde{s}\lambda+ab/t)\right]\\
= \frac{1}{2} +\frac{s}{\Phi(\lambda)}
\frac{\phi(a/t)(1-x)}{\tilde{s}^{2}(1-x)^{2}+1}\left[
\frac{x}{\abs{1-x}}\frac{\lambda}{\tilde{g}(x, \tilde{s})}\Phi(\lambda
g(x, \tilde{s})) -\phi\left(\lambda g(x, \tilde{s})\right) \right] ,
\end{multline*}
%
%
%
%
%
%
where
$g(x, \tilde{s})=\frac{\tilde{s}^{2}\abs{1-x}+\sign(1-x)}{
\sqrt{\tilde{s}^{2}(1-x)^{2}+1}}$, and
$\tilde{g}(x, \tilde{s})=\sqrt{\tilde{s}^{2}(1-x)^{2}+1}$. When evaluated at
$x=m_{c}$, the expression in square brackets must equal zero by definition of
the median. Therefore, $m_{c}>0$, and since we also know
from~\cref{eq:median_lt1} that $m_{c}<1$, the conditional median must satisfy
\begin{equation}\label{eq:conditional_median_bias}
m_{c}=\frac{1}{\frac{\lambda}{\tilde{g}(m_{c}, \tilde{s}(m_{c}, s))}
\frac{\Phi\left(\lambda g(m_{c}, \tilde{s}(m_{c},s))\right)}{
\phi(\lambda
g(m_{c}, \tilde{s}(m_{c},s)))}+1},
\end{equation}
We have
\begin{equation*}
\frac{\lambda}{\tilde{g}}\frac{\Phi\left(\lambda g\right)}{
\phi(\lambda
g)}\geq \frac{\lambda}{\tilde{g}}\frac{\Phi\left(\lambda \tilde{g}\right)}{
\phi(\lambda\tilde{g})},\quad\text{and}\quad
\frac{\lambda}{\tilde{g}}\frac{\Phi\left(\lambda \tilde{g}\right)}{
\phi(\lambda\tilde{g})}\geq
\lambda
\frac{\Phi\left(\lambda\right)}{
\phi(\lambda)} \quad\text{if $\lambda\geq 0.84$}.
\end{equation*}
Here the first inequality follows because $\Phi(x)/\phi(x)$ is increasing in
$x$, and $g\geq \tilde{g}$, and the second inequality follows because
$\frac{\Phi(x)}{x\phi(x)}$ is increasing for $x\geq 0.84$, and $\tilde{g}\geq 1$.
%
Therefore,
\begin{equation*}
m_{c}\leq \frac{\phi(\lambda)}{\lambda \Phi(\lambda)+\phi(\lambda)}=\lim_{s\downarrow 0}m_{c}(s),
\end{equation*}
where the equality follows since the right-hand side of
\cref{eq:conditional_median_bias} converges to $\frac{\phi(\lambda)}{\lambda
\Phi(\lambda)+\phi(\lambda)}$ as $s\to 0$.
We now prove the claims concerning the unconditional distribution of
$\tilde{\beta}_{IV}$. From~\cref{eq:tilde_beta_distro}, we obtain
\begin{multline}\label{eq:cdf_unconditional}
P(\tilde{\beta}_{IV}\leq x;s)=
1-\Phi(\lambda) \\ +\int_{-\lambda}^{\infty}
\Phi\left(s (x\lambda- (1-x)z)\right)\phi(z)dz
-\int_{-\infty}^{-\lambda}
\Phi\left(s (x\lambda- (1-x)z)\right)\phi(z)dz.
\end{multline}
Observe that the median of $\tilde{\beta}_{IV}$, denoted $m_{u}=m_{u}(s)$ is
smaller than $1$, since
\begin{multline}\label{eq:umedian_lt1}
P(\tilde{\beta}_{IV}\leq 1;s)= 1-\Phi(\lambda)+ \Phi\left(s
\lambda\right)\left[\int_{-\lambda}^{\infty} \phi(z)dz
-\int_{-\infty}^{-\lambda}
\phi(z)dz\right]\\
=1-\Phi(\lambda)+\Phi\left(s \lambda\right)[\Phi(\lambda)-1+\Phi(\lambda)]
>1-\Phi(\lambda)+[\Phi(\lambda)-1+\Phi(\lambda)]=\Phi(\lambda)>1/2.
\end{multline}
By arguments as in the conditional case, for $x<1$,
\begin{equation*}
\begin{split}
P(\tilde{\beta}_{IV}\leq x;\rho)&= \frac{1}{2}
+\frac{s}{\tilde{s}^{2}(1-x)}\left[\int_{-\infty}^{\tilde{s}\lambda}
y\phi\left(y\right)\phi(a+by)dy
-\int_{\tilde{s}\lambda}^{\infty}y\phi\left(y\right)\phi(a+by)dy\right]\\
&= \frac{1}{2} +\frac{s}{\tilde{s}^{2}(1-x)}\frac{\phi(a/t)}{1+b^{2}}\left[
-2\phi(t\tilde{s}\lambda+ab/t)-2\frac{ab}{t}\Phi(t\tilde{s}\lambda+ab/t)
+\frac{ab}{t}\right]\\
&= \frac{1}{2} +\frac{s}{\tilde{s}^{2}(1-x)}\frac{\phi(a/t)}{1+b^{2}}\left[
-2\phi(\lambda g(\tilde{s},x )) +2\frac{x}{1-x}\frac{\lambda}{\tilde{g}}
\Phi( \lambda g(\tilde{s}, x))
-\frac{x}{1-x}\frac{\lambda}{\tilde{g}}\right].
\end{split}
\end{equation*}
Here the first line follows by the mean value theorem, where
$\tilde{s}=\tilde{s}(x, s)\in[0,s]$, the second line uses~\cref{eq:owen111}, and
the last line follows by algebra. When evaluated at $x=m_{c}$, the expression in
square brackets must equal zero by definition of the median. Therefore,
$m_{u}>0$, and it must satisfy
\begin{equation}\label{eq:unconditional_median_bias}
m_{u}= \frac{ 1} {
\frac{\lambda}{\tilde{g}}\frac{\Phi( \lambda
g)-1/2}{\phi(\lambda g)}+1
}
\end{equation}
Now,
\begin{equation*}
\frac{\lambda}{\tilde{g}}\frac{\Phi( \lambda
g)-1/2}{\phi(\lambda g)}\geq
\frac{\lambda}{\tilde{g}}\frac{\Phi( \lambda
\tilde{g})-1/2}{\phi(\lambda \tilde{g})}
\geq \lambda\frac{\Phi( \lambda)-1/2}{\phi(\lambda)}.
\end{equation*}
Here the first inequality follows because $\Phi(x)/\phi(x)$ is increasing in
$x$, and $g\geq \tilde{g}$, and the second inequality follows because
$\frac{\Phi(x)-1/2}{x\phi(x)}$ is increasing for $x>0$. As a result,
\begin{equation*}
m_{u}\leq \frac{\phi(\lambda)}{\lambda(\Phi( \lambda)-1/2+\phi(\lambda))}=\lim_{s\downarrow 0}m_{u}(s),
\end{equation*}
where the equality follows since the right-hand side of
\cref{eq:unconditional_median_bias} converges to
$\frac{\phi(\lambda)}{\lambda(\Phi( \lambda)-1/2+\phi(\lambda))}$ as $s\to 0$.
\subsection{Median Bias Comparisons}\label{sec:medi-bias-comparison}
To evaluate the relative median bias of $\hat{\beta}_{IV}$ as a function of both $E[F]$
and $\rho$ conditional on $t_{1}\geq c$, we first evaluate
\begin{equation}\label{eq:median_iv_formula}
P(\tilde{\beta}_{IV}\leq x\mid t_{1}\geq c;\rho,\lambda)
=\frac{1}{\Phi(\lambda-c)}
\int_{c-\lambda}^{\infty}f_{IV}(z;x,\lambda,s)\phi(z)dz
\end{equation}
by numerical integration. Here we use the formula
$f_{IV}(z;x,\rho,\lambda)=\Phi\left(s[\sign(z+\lambda)\lambda-(1-x)\abs{z+\lambda}]\right)$
from \cref{eq:tilde_beta_distro} for the cdf conditional on $z=t_{1}-\lambda$.
We then numerically solve for the median. For unconditional median, we set
$c=-\infty$, and for the median conditional on sign screening, we set $c=0$. The
shaded regions in \Cref{fig:bias} correspond to the range of the absolute value
of the relative median bias as $\rho$ varies between $-1$ and $1$.
To compare the relative median bias to that of $\hat{\beta}_{U}$, it suffices to
consider $\rho>0$, since the distributions of $\hat{\beta}_{U}$ and
$\hat{\beta}_{IV}$ are symmetric in $\rho$. By \cref{eq:tildebetaU}, it follows
that for $t_{1}>0$,
\begin{equation*}
P(\tilde{\beta}_{U}\leq x\mid t_{1};s)=
P\left( \tilde{\beta}_{IV}
\leq x-(1-x)\frac{(1-t_{1}\mu(t_{1}))}{t_{1}\mu(t_{1})}\mid t_{1};s\right),
\end{equation*}
which for $x<1$ is smaller than $P(\tilde{\beta}_{IV}\leq x\mid t_{1};s)$. Since
the median of $\tilde{\beta}_{IV}$ conditional on $t_{1}>0$ is smaller than $1$,
it follows that the conditional median bias of $\tilde{\beta}_{IV}$ is always
smaller than that of $\tilde{\beta}_{U}$.
To compare the relative magnitudes of the median biases, we compute the relative
median bias of $\tilde{\beta}_{U}$ analogously to that of $\tilde{\beta}_{IV}$,
except we replace $f_{IV}$ in \cref{eq:median_iv_formula} with
$f_{U}(z;x,\lambda,\rho)=\Phi(s[\lambda-(1-\sign(s)x)/\mu(\lambda+z)])$ (it
follows from \cref{eq:tildebetaU,eq:ar_given_z} that this is the cdf
$\tilde{\beta}_{U}$ conditional on $z=t_{1}-\lambda$). We then compute the ratio
$\operatorname{median}_{\lambda,\rho}(\tilde{\beta}_{U}\mid
t_{1}>0)/\operatorname{median}_{\lambda,\rho}(\tilde{\beta}_{IV}\mid t_{1}>0)$
of the median biases on a fine grid of values of $(\rho,\lambda)$. This ratio is
greater than $2$ if $E[F]\geq 2$, and greater than $3$ if $E[F]\geq 3$, regardless of the value of $\rho$.
\end{appendices}
\end{document}
% Created by tikzDevice version 0.12.3.1 on 2021-10-26 21:51:57
% !TEX encoding = UTF-8 Unicode
\begin{tikzpicture}[x=1pt,y=1pt]
\definecolor{fillColor}{RGB}{255,255,255}
\path[use as bounding box,fill=fillColor,fill opacity=0.00] (0,0) rectangle (390.26,252.94);
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{255,255,255}
\definecolor{fillColor}{RGB}{255,255,255}
\path[draw=drawColor,line width= 0.6pt,line join=round,line cap=round,fill=fillColor] ( 0.00, 0.00) rectangle (390.26,252.94);
\end{scope}
\begin{scope}
\path[clip] ( 41.55, 30.40) rectangle (385.26,247.95);
\definecolor{fillColor}{RGB}{255,255,255}
\path[fill=fillColor] ( 41.55, 30.40) rectangle (385.26,247.95);
\definecolor{drawColor}{RGB}{208,208,208}
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55, 86.42) --
(385.26, 86.42);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55,133.50) --
(385.26,133.50);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55,161.66) --
(385.26,161.66);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55,181.44) --
(385.26,181.44);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55, 40.75) --
(385.26, 40.75);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55,247.95) --
(385.26,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55, 30.40) --
( 41.55,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 73.40, 30.40) --
( 73.40,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (106.47, 30.40) --
(106.47,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (266.19, 30.40) --
(266.19,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (306.66, 30.40) --
(306.66,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (385.26, 30.40) --
(385.26,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (385.26, 30.40) --
(385.26,247.95);
\definecolor{drawColor}{RGB}{119,119,119}
\path[draw=drawColor,line width= 0.6pt,line join=round] (385.26, 42.49) --
(376.61, 42.15) --
(368.08, 41.85) --
(359.65, 41.58) --
(351.33, 41.34) --
(343.71, 41.17) --
(343.13, 41.16) --
(335.03, 41.00) --
(327.05, 40.88) --
(319.17, 40.80) --
(311.41, 40.76) --
(303.75, 40.76) --
(296.21, 40.79) --
(288.77, 40.86) --
(281.45, 40.97) --
(274.23, 41.12) --
(272.64, 41.17) --
(267.13, 41.31) --
(260.13, 41.54) --
(253.25, 41.80) --
(246.47, 42.11) --
(239.81, 42.45) --
(233.26, 42.83) --
(226.81, 43.24) --
(222.20, 43.57) --
(220.48, 43.69) --
(214.26, 44.18) --
(208.14, 44.70) --
(202.14, 45.26) --
(196.25, 45.86) --
(195.16, 45.98) --
(190.47, 46.48) --
(184.79, 47.14) --
(179.23, 47.84) --
(175.16, 48.38) --
(173.78, 48.56) --
(168.44, 49.32) --
(163.20, 50.10) --
(158.93, 50.79) --
(158.08, 50.92) --
(153.07, 51.76) --
(148.17, 52.63) --
(145.18, 53.19) --
(143.38, 53.52) --
(138.70, 54.44) --
(134.13, 55.38) --
(133.11, 55.59) --
(129.67, 56.33) --
(125.31, 57.32) --
(122.43, 58.00) --
(121.07, 58.31) --
(116.94, 59.32) --
(112.92, 60.35) --
(112.73, 60.40) --
(109.01, 61.38) --
(105.21, 62.43) --
(103.91, 62.81) --
(101.52, 63.48) --
( 97.94, 64.54) --
( 95.76, 65.21) --
( 94.47, 65.60) --
( 91.11, 66.65) --
( 88.18, 67.62) --
( 87.87, 67.72) --
( 84.73, 68.76) --
( 81.70, 69.81) --
( 81.11, 70.02) --
( 78.78, 70.84) --
( 75.97, 71.86) --
( 74.46, 72.43) --
( 73.27, 72.86) --
( 70.68, 73.84) --
( 68.20, 74.81) --
( 68.15, 74.83) --
( 65.84, 75.73) --
( 63.58, 76.65) --
( 62.18, 77.23) --
( 61.43, 77.54) --
( 59.39, 78.38) --
( 57.47, 79.20) --
( 56.47, 79.64) --
( 55.65, 79.99) --
( 53.94, 80.73) --
( 52.34, 81.44) --
( 51.00, 82.04) --
( 50.86, 82.11) --
( 49.48, 82.72) --
( 48.21, 83.29) --
( 47.06, 83.83) --
( 46.01, 84.32) --
( 45.73, 84.45) --
( 45.07, 84.75) --
( 44.25, 85.13) --
( 43.53, 85.47) --
( 42.93, 85.75) --
( 42.43, 86.00) --
( 42.04, 86.17) --
( 41.77, 86.31) --
( 41.60, 86.38) --
( 41.55, 86.41);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55,133.49) --
( 41.60,133.50) --
( 41.77,133.51) --
( 42.04,133.52) --
( 42.43,133.55) --
( 42.93,133.58) --
( 43.53,133.62) --
( 44.25,133.67) --
( 45.07,133.73) --
( 46.01,133.81) --
( 47.06,133.89) --
( 48.21,133.99) --
( 49.48,134.10) --
( 50.86,134.23) --
( 52.34,134.38) --
( 53.94,134.55) --
( 55.65,134.74) --
( 57.30,134.94) --
( 57.47,134.96) --
( 59.39,135.20) --
( 61.43,135.47) --
( 63.58,135.77) --
( 65.84,136.11) --
( 68.20,136.49) --
( 70.68,136.92) --
( 73.03,137.34) --
( 73.27,137.39) --
( 75.97,137.90) --
( 78.78,138.47) --
( 81.70,139.11) --
( 84.48,139.75) --
( 84.73,139.81) --
( 87.87,140.57) --
( 91.11,141.41) --
( 93.81,142.15) --
( 94.47,142.34) --
( 97.94,143.34) --
(101.52,144.44) --
(101.90,144.56) --
(105.21,145.63) --
(109.01,146.95) --
(109.05,146.96) --
(112.92,148.37) --
(115.54,149.37) --
(116.94,149.93) --
(121.07,151.61) --
(121.47,151.77) --
(125.31,153.43) --
(127.00,154.17) --
(129.67,155.41) --
(132.18,156.58) --
(134.13,157.54) --
(137.03,158.98) --
(138.70,159.86) --
(141.63,161.39) --
(143.38,162.35) --
(146.04,163.79) --
(148.17,165.02) --
(150.25,166.20) --
(153.07,167.89) --
(154.29,168.60) --
(158.08,170.95) --
(158.17,171.00) --
(161.99,173.41) --
(163.20,174.22) --
(165.68,175.81) --
(168.44,177.68) --
(169.27,178.22) --
(172.79,180.62) --
(173.78,181.32) --
(176.27,183.03) --
(179.23,185.14) --
(179.65,185.43) --
(183.05,187.84) --
(184.79,189.12) --
(186.38,190.24) --
(189.68,192.64) --
(190.47,193.23) --
(192.99,195.05) --
(196.23,197.45) --
(196.25,197.47) --
(199.54,199.86) --
(202.14,201.80) --
(202.78,202.26) --
(206.07,204.67) --
(208.14,206.22) --
(209.32,207.07) --
(212.60,209.47) --
(214.26,210.72) --
(215.88,211.88) --
(219.16,214.28) --
(220.48,215.27) --
(222.45,216.69) --
(225.74,219.09) --
(226.81,219.89) --
(229.06,221.50) --
(232.36,223.90) --
(233.26,224.56) --
(235.71,226.31) --
(239.03,228.71) --
(239.81,229.28) --
(242.40,231.11) --
(245.75,233.52) --
(246.47,234.05) --
(249.15,235.92) --
(252.51,238.33) --
(253.25,238.86) --
(255.94,240.73) --
(259.33,243.14) --
(260.13,243.71) --
(262.78,245.54) --
(266.21,247.95);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55,161.66) --
( 41.60,161.68) --
( 41.77,161.76) --
( 42.04,161.89) --
( 42.43,162.06) --
( 42.93,162.30) --
( 43.53,162.58) --
( 44.25,162.93) --
( 45.07,163.34) --
( 45.97,163.79) --
( 46.01,163.81) --
( 47.06,164.34) --
( 48.21,164.93) --
( 49.48,165.61) --
( 50.53,166.20) --
( 50.86,166.37) --
( 52.34,167.21) --
( 53.94,168.15) --
( 54.70,168.60) --
( 55.65,169.18) --
( 57.47,170.32) --
( 58.52,171.00) --
( 59.39,171.58) --
( 61.43,172.98) --
( 62.04,173.41) --
( 63.58,174.53) --
( 65.29,175.81) --
( 65.84,176.24) --
( 68.20,178.14) --
( 68.30,178.22) --
( 70.68,180.25) --
( 71.11,180.62) --
( 73.27,182.62) --
( 73.71,183.03) --
( 75.97,185.25) --
( 76.15,185.43) --
( 78.38,187.84) --
( 78.78,188.30) --
( 80.49,190.24) --
( 81.70,191.75) --
( 82.43,192.64) --
( 84.24,195.05) --
( 84.73,195.75) --
( 85.95,197.45) --
( 87.51,199.86) --
( 87.87,200.46) --
( 88.99,202.26) --
( 90.35,204.67) --
( 91.11,206.17) --
( 91.61,207.07) --
( 92.81,209.47) --
( 93.90,211.88) --
( 94.47,213.25) --
( 94.94,214.28) --
( 95.94,216.69) --
( 96.87,219.09) --
( 97.73,221.50) --
( 97.94,222.13) --
( 98.60,223.90) --
( 99.44,226.31) --
(100.25,228.71) --
(101.03,231.11) --
(101.52,232.67) --
(101.82,233.52) --
(102.63,235.92) --
(103.42,238.33) --
(104.18,240.73) --
(104.94,243.14) --
(105.21,244.04) --
(105.72,245.54) --
(106.51,247.95);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55,181.42) --
( 41.60,181.47) --
( 41.77,181.60) --
( 42.04,181.82) --
( 42.43,182.14) --
( 42.93,182.55) --
( 43.49,183.03) --
( 43.53,183.07) --
( 44.25,183.67) --
( 45.07,184.38) --
( 46.01,185.23) --
( 46.24,185.43) --
( 47.06,186.18) --
( 48.21,187.27) --
( 48.79,187.84) --
( 49.48,188.51) --
( 50.86,189.92) --
( 51.17,190.24) --
( 52.34,191.51) --
( 53.37,192.64) --
( 53.94,193.30) --
( 55.41,195.05) --
( 55.65,195.35) --
( 57.29,197.45) --
( 57.47,197.69) --
( 59.13,199.86) --
( 59.39,200.24) --
( 60.67,202.26) --
( 61.43,203.52) --
( 62.14,204.67) --
( 63.47,207.07) --
( 63.58,207.28) --
( 64.72,209.47) --
( 65.81,211.88) --
( 65.84,211.94) --
( 66.84,214.28) --
( 67.73,216.69) --
( 68.20,218.16) --
( 68.52,219.09) --
( 69.24,221.50) --
( 69.85,223.90) --
( 70.37,226.31) --
( 70.68,228.00) --
( 70.83,228.71) --
( 71.25,231.11) --
( 71.62,233.52) --
( 71.95,235.92) --
( 72.26,238.33) --
( 72.55,240.73) --
( 72.84,243.14) --
( 73.12,245.54) --
( 73.27,246.88) --
( 73.41,247.95);
\definecolor{drawColor}{RGB}{255,255,255}
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.51, 42.45) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (165.20,171.61) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 88.75,194.72) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 70.12,211.34) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.13, 43.02) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.82,172.18) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 88.37,195.29) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.74,211.91) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (208.56, 43.40) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.24,172.56) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 87.80,195.68) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.16,212.30) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.88, 43.54) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.57,172.70) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 87.12,195.81) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 68.49,212.43) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.21, 43.40) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (162.89,172.56) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 86.45,195.68) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 67.82,212.30) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.64, 43.02) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (162.32,172.18) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 85.88,195.29) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 67.24,211.91) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.25, 42.45) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (161.94,171.61) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 85.49,194.72) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 66.86,211.34) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.12, 41.77) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (161.80,170.93) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 85.36,194.05) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 66.73,210.67) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.25, 41.10) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (161.94,170.26) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 85.49,193.37) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 66.86,209.99) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.64, 40.53) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (162.32,169.68) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 85.88,192.80) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 67.24,209.42) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.21, 40.14) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (162.89,169.30) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 86.45,192.42) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 67.82,209.04) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.88, 40.01) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.57,169.17) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 87.12,192.28) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 68.49,208.90) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (208.56, 40.14) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.24,169.30) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 87.80,192.42) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.16,209.04) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.13, 40.53) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.82,169.68) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 88.37,192.80) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.74,209.42) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.51, 41.10) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (165.20,170.26) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 88.75,193.37) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 70.12,209.99) {0.2};
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.65, 41.77) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (165.33,170.93) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 88.89,194.05) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 70.25,210.67) {0.2};
\definecolor{drawColor}{RGB}{0,0,0}
\node[text=drawColor,rotate= -5.11,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.88, 41.77) {0.05};
\node[text=drawColor,rotate= 32.56,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.57,170.93) {0.1};
\node[text=drawColor,rotate= 54.67,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 87.12,194.05) {0.15};
\node[text=drawColor,rotate= 65.63,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 68.49,210.67) {0.2};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55, 30.40) --
( 41.55,247.95);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05, 83.66) {0.661};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05,130.75) {0.76};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05,158.91) {0.819};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05,178.68) {0.86};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05, 38.00) {0.565};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05,245.19) {1};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05, 86.42) --
( 41.55, 86.42);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05,133.50) --
( 41.55,133.50);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05,161.66) --
( 41.55,161.66);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05,181.44) --
( 41.55,181.44);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05, 40.75) --
( 41.55, 40.75);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05,247.95) --
( 41.55,247.95);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55, 30.40) --
(385.26, 30.40);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55, 27.90) --
( 41.55, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 73.40, 27.90) --
( 73.40, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (106.47, 27.90) --
(106.47, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (266.19, 27.90) --
(266.19, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (306.66, 27.90) --
(306.66, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (385.26, 27.90) --
(385.26, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (385.26, 27.90) --
(385.26, 30.40);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 41.55, 20.39) {1};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 73.40, 20.39) {1.83};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (106.47, 20.39) {2.7};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (266.19, 20.39) {6.88};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (306.66, 20.39) {7.94};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (385.26, 20.39) {10};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (385.26, 20.39) {10};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 1.00] at (213.40, 6.94) {$E[F]$};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,rotate= 90.00,anchor=base,inner sep=0pt, outer sep=0pt, scale= 1.00] at ( 11.89,139.17) {$|\rho|$};
\end{scope}
\end{tikzpicture}
% Created by tikzDevice version 0.12.3.1 on 2021-10-26 22:23:45
% !TEX encoding = UTF-8 Unicode
\begin{tikzpicture}[x=1pt,y=1pt]
\definecolor{fillColor}{RGB}{255,255,255}
\path[use as bounding box,fill=fillColor,fill opacity=0.00] (0,0) rectangle (390.26,245.72);
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{255,255,255}
\definecolor{fillColor}{RGB}{255,255,255}
\path[draw=drawColor,line width= 0.6pt,line join=round,line cap=round,fill=fillColor] ( 0.00, 0.00) rectangle (390.26,245.72);
\end{scope}
\begin{scope}
\path[clip] ( 37.55, 30.40) rectangle (385.26,240.72);
\definecolor{fillColor}{RGB}{255,255,255}
\path[fill=fillColor] ( 37.55, 30.40) rectangle (385.26,240.72);
\definecolor{drawColor}{RGB}{238,228,218}
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 37.55, 30.40) --
(385.26, 30.40);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 37.55, 40.90) --
(385.26, 40.90);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 37.55, 51.41) --
(385.26, 51.41);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 37.55, 72.42) --
(385.26, 72.42);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 37.55,135.45) --
(385.26,135.45);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 37.55,240.51) --
(385.26,240.51);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 37.55, 30.40) --
( 37.55,240.72);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 81.08, 30.40) --
( 81.08,240.72);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (115.47, 30.40) --
(115.47,240.72);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (154.50, 30.40) --
(154.50,240.72);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (385.26, 30.40) --
(385.26,240.72);
\definecolor{fillColor}{RGB}{86,180,233}
\path[fill=fillColor,fill opacity=0.20] ( 37.55,240.51) --
( 37.59,232.64) --
( 37.69,224.99) --
( 37.87,217.55) --
( 38.12,210.32) --
( 38.44,203.30) --
( 38.83,196.48) --
( 39.29,189.86) --
( 39.82,183.45) --
( 40.42,177.22) --
( 41.10,171.19) --
( 41.84,165.35) --
( 42.66,159.69) --
( 43.55,154.21) --
( 44.50,148.91) --
( 45.53,143.78) --
( 46.63,138.82) --
( 47.80,134.03) --
( 49.04,129.41) --
( 50.36,124.94) --
( 51.74,120.63) --
( 53.20,116.48) --
( 54.72,112.47) --
( 56.32,108.61) --
( 57.98,104.90) --
( 59.72,101.32) --
( 61.53, 97.88) --
( 63.41, 94.57) --
( 65.36, 91.40) --
( 67.39, 88.34) --
( 69.48, 85.41) --
( 71.64, 82.60) --
( 73.88, 79.91) --
( 76.18, 77.33) --
( 78.56, 74.85) --
( 81.01, 72.49) --
( 83.53, 70.22) --
( 86.12, 68.06) --
( 88.78, 65.99) --
( 91.51, 64.02) --
( 94.31, 62.14) --
( 97.19, 60.34) --
(100.13, 58.63) --
(103.15, 57.00) --
(106.23, 55.45) --
(109.39, 53.97) --
(112.62, 52.57) --
(115.92, 51.23) --
(119.29, 49.97) --
(122.73, 48.77) --
(126.24, 47.63) --
(129.83, 46.55) --
(133.48, 45.53) --
(137.20, 44.56) --
(141.00, 43.65) --
(144.87, 42.79) --
(148.81, 41.98) --
(152.81, 41.21) --
(156.89, 40.48) --
(161.04, 39.80) --
(165.27, 39.16) --
(169.56, 38.55) --
(173.92, 37.99) --
(178.36, 37.45) --
(182.86, 36.95) --
(187.44, 36.48) --
(192.09, 36.04) --
(196.81, 35.63) --
(201.59, 35.25) --
(206.45, 34.89) --
(211.39, 34.55) --
(216.39, 34.24) --
(221.46, 33.94) --
(226.61, 33.67) --
(231.82, 33.42) --
(237.11, 33.18) --
(242.46, 32.96) --
(247.89, 32.75) --
(253.39, 32.56) --
(258.96, 32.39) --
(264.60, 32.23) --
(270.31, 32.07) --
(276.10, 31.93) --
(281.95, 31.80) --
(287.87, 31.69) --
(293.87, 31.58) --
(299.94, 31.47) --
(306.07, 31.38) --
(312.28, 31.29) --
(318.56, 31.21) --
(324.91, 31.14) --
(331.33, 31.07) --
(337.83, 31.01) --
(344.39, 30.96) --
(351.02, 30.91) --
(357.73, 30.86) --
(364.50, 30.81) --
(371.35, 30.78) --
(378.27, 30.74) --
(385.26, 30.71) --
(385.26, 30.51) --
(378.27, 30.53) --
(371.35, 30.54) --
(364.50, 30.56) --
(357.73, 30.58) --
(351.02, 30.60) --
(344.39, 30.62) --
(337.83, 30.65) --
(331.33, 30.67) --
(324.91, 30.70) --
(318.56, 30.74) --
(312.28, 30.77) --
(306.07, 30.81) --
(299.94, 30.86) --
(293.87, 30.91) --
(287.87, 30.96) --
(281.95, 31.02) --
(276.10, 31.08) --
(270.31, 31.15) --
(264.60, 31.23) --
(258.96, 31.31) --
(253.39, 31.40) --
(247.89, 31.50) --
(242.46, 31.61) --
(237.11, 31.72) --
(231.82, 31.85) --
(226.61, 31.99) --
(221.46, 32.14) --
(216.39, 32.30) --
(211.39, 32.48) --
(206.45, 32.67) --
(201.59, 32.88) --
(196.81, 33.11) --
(192.09, 33.35) --
(187.44, 33.61) --
(182.86, 33.90) --
(178.36, 34.21) --
(173.92, 34.54) --
(169.56, 34.90) --
(165.27, 35.28) --
(161.04, 35.70) --
(156.89, 36.14) --
(152.81, 36.62) --
(148.81, 37.14) --
(144.87, 37.69) --
(141.00, 38.28) --
(137.20, 38.92) --
(133.48, 39.60) --
(129.83, 40.33) --
(126.24, 41.11) --
(122.73, 41.94) --
(119.29, 42.83) --
(115.92, 43.78) --
(112.62, 44.79) --
(109.39, 45.87) --
(106.23, 47.02) --
(103.15, 48.25) --
(100.13, 49.55) --
( 97.19, 50.93) --
( 94.31, 52.40) --
( 91.51, 53.96) --
( 88.78, 55.61) --
( 86.12, 57.36) --
( 83.53, 59.22) --
( 81.01, 61.18) --
( 78.56, 63.25) --
( 76.18, 65.44) --
( 73.88, 67.76) --
( 71.64, 70.20) --
( 69.48, 72.78) --
( 67.39, 75.49) --
( 65.36, 78.35) --
( 63.41, 81.36) --
( 61.53, 84.52) --
( 59.72, 87.84) --
( 57.98, 91.33) --
( 56.32, 94.99) --
( 54.72, 98.83) --
( 53.20,102.85) --
( 51.74,107.06) --
( 50.36,111.47) --
( 49.04,116.08) --
( 47.80,120.90) --
( 46.63,125.93) --
( 45.53,131.19) --
( 44.50,136.67) --
( 43.55,142.38) --
( 42.66,148.33) --
( 41.84,154.53) --
( 41.10,160.97) --
( 40.42,167.68) --
( 39.82,174.65) --
( 39.29,181.89) --
( 38.83,189.41) --
( 38.44,197.22) --
( 38.12,205.32) --
( 37.87,213.73) --
( 37.69,222.47) --
( 37.59,231.66) --
( 37.55,240.51) --
cycle;
\path[] ( 37.55,240.51) --
( 37.59,232.64) --
( 37.69,224.99) --
( 37.87,217.55) --
( 38.12,210.32) --
( 38.44,203.30) --
( 38.83,196.48) --
( 39.29,189.86) --
( 39.82,183.45) --
( 40.42,177.22) --
( 41.10,171.19) --
( 41.84,165.35) --
( 42.66,159.69) --
( 43.55,154.21) --
( 44.50,148.91) --
( 45.53,143.78) --
( 46.63,138.82) --
( 47.80,134.03) --
( 49.04,129.41) --
( 50.36,124.94) --
( 51.74,120.63) --
( 53.20,116.48) --
( 54.72,112.47) --
( 56.32,108.61) --
( 57.98,104.90) --
( 59.72,101.32) --
( 61.53, 97.88) --
( 63.41, 94.57) --
( 65.36, 91.40) --
( 67.39, 88.34) --
( 69.48, 85.41) --
( 71.64, 82.60) --
( 73.88, 79.91) --
( 76.18, 77.33) --
( 78.56, 74.85) --
( 81.01, 72.49) --
( 83.53, 70.22) --
( 86.12, 68.06) --
( 88.78, 65.99) --
( 91.51, 64.02) --
( 94.31, 62.14) --
( 97.19, 60.34) --
(100.13, 58.63) --
(103.15, 57.00) --
(106.23, 55.45) --
(109.39, 53.97) --
(112.62, 52.57) --
(115.92, 51.23) --
(119.29, 49.97) --
(122.73, 48.77) --
(126.24, 47.63) --
(129.83, 46.55) --
(133.48, 45.53) --
(137.20, 44.56) --
(141.00, 43.65) --
(144.87, 42.79) --
(148.81, 41.98) --
(152.81, 41.21) --
(156.89, 40.48) --
(161.04, 39.80) --
(165.27, 39.16) --
(169.56, 38.55) --
(173.92, 37.99) --
(178.36, 37.45) --
(182.86, 36.95) --
(187.44, 36.48) --
(192.09, 36.04) --
(196.81, 35.63) --
(201.59, 35.25) --
(206.45, 34.89) --
(211.39, 34.55) --
(216.39, 34.24) --
(221.46, 33.94) --
(226.61, 33.67) --
(231.82, 33.42) --
(237.11, 33.18) --
(242.46, 32.96) --
(247.89, 32.75) --
(253.39, 32.56) --
(258.96, 32.39) --
(264.60, 32.23) --
(270.31, 32.07) --
(276.10, 31.93) --
(281.95, 31.80) --
(287.87, 31.69) --
(293.87, 31.58) --
(299.94, 31.47) --
(306.07, 31.38) --
(312.28, 31.29) --
(318.56, 31.21) --
(324.91, 31.14) --
(331.33, 31.07) --
(337.83, 31.01) --
(344.39, 30.96) --
(351.02, 30.91) --
(357.73, 30.86) --
(364.50, 30.81) --
(371.35, 30.78) --
(378.27, 30.74) --
(385.26, 30.71);
\path[] (385.26, 30.51) --
(378.27, 30.53) --
(371.35, 30.54) --
(364.50, 30.56) --
(357.73, 30.58) --
(351.02, 30.60) --
(344.39, 30.62) --
(337.83, 30.65) --
(331.33, 30.67) --
(324.91, 30.70) --
(318.56, 30.74) --
(312.28, 30.77) --
(306.07, 30.81) --
(299.94, 30.86) --
(293.87, 30.91) --
(287.87, 30.96) --
(281.95, 31.02) --
(276.10, 31.08) --
(270.31, 31.15) --
(264.60, 31.23) --
(258.96, 31.31) --
(253.39, 31.40) --
(247.89, 31.50) --
(242.46, 31.61) --
(237.11, 31.72) --
(231.82, 31.85) --
(226.61, 31.99) --
(221.46, 32.14) --
(216.39, 32.30) --
(211.39, 32.48) --
(206.45, 32.67) --
(201.59, 32.88) --
(196.81, 33.11) --
(192.09, 33.35) --
(187.44, 33.61) --
(182.86, 33.90) --
(178.36, 34.21) --
(173.92, 34.54) --
(169.56, 34.90) --
(165.27, 35.28) --
(161.04, 35.70) --
(156.89, 36.14) --
(152.81, 36.62) --
(148.81, 37.14) --
(144.87, 37.69) --
(141.00, 38.28) --
(137.20, 38.92) --
(133.48, 39.60) --
(129.83, 40.33) --
(126.24, 41.11) --
(122.73, 41.94) --
(119.29, 42.83) --
(115.92, 43.78) --
(112.62, 44.79) --
(109.39, 45.87) --
(106.23, 47.02) --
(103.15, 48.25) --
(100.13, 49.55) --
( 97.19, 50.93) --
( 94.31, 52.40) --
( 91.51, 53.96) --
( 88.78, 55.61) --
( 86.12, 57.36) --
( 83.53, 59.22) --
( 81.01, 61.18) --
( 78.56, 63.25) --
( 76.18, 65.44) --
( 73.88, 67.76) --
( 71.64, 70.20) --
( 69.48, 72.78) --
( 67.39, 75.49) --
( 65.36, 78.35) --
( 63.41, 81.36) --
( 61.53, 84.52) --
( 59.72, 87.84) --
( 57.98, 91.33) --
( 56.32, 94.99) --
( 54.72, 98.83) --
( 53.20,102.85) --
( 51.74,107.06) --
( 50.36,111.47) --
( 49.04,116.08) --
( 47.80,120.90) --
( 46.63,125.93) --
( 45.53,131.19) --
( 44.50,136.67) --
( 43.55,142.38) --
( 42.66,148.33) --
( 41.84,154.53) --
( 41.10,160.97) --
( 40.42,167.68) --
( 39.82,174.65) --
( 39.29,181.89) --
( 38.83,189.41) --
( 38.44,197.22) --
( 38.12,205.32) --
( 37.87,213.73) --
( 37.69,222.47) --
( 37.59,231.66) --
( 37.55,240.51);
\definecolor{fillColor}{RGB}{230,159,0}
\path[fill=fillColor,fill opacity=0.20] ( 37.55,240.51) --
( 37.59,240.32) --
( 37.69,239.74) --
( 37.87,238.78) --
( 38.12,237.45) --
( 38.44,235.76) --
( 38.83,233.71) --
( 39.29,231.33) --
( 39.82,228.63) --
( 40.42,225.62) --
( 41.10,222.33) --
( 41.84,218.79) --
( 42.66,215.00) --
( 43.55,210.99) --
( 44.50,206.79) --
( 45.53,202.42) --
( 46.63,197.90) --
( 47.80,193.25) --
( 49.04,188.51) --
( 50.36,183.69) --
( 51.74,178.80) --
( 53.20,173.88) --
( 54.72,168.95) --
( 56.32,164.01) --
( 57.98,159.07) --
( 59.72,154.18) --
( 61.53,149.33) --
( 63.41,144.53) --
( 65.36,139.81) --
( 67.39,135.16) --
( 69.48,130.61) --
( 71.64,126.15) --
( 73.88,121.80) --
( 76.18,117.56) --
( 78.56,113.43) --
( 81.01,109.42) --
( 83.53,105.53) --
( 86.12,101.77) --
( 88.78, 98.13) --
( 91.51, 94.63) --
( 94.31, 91.25) --
( 97.19, 87.99) --
(100.13, 84.87) --
(103.15, 81.87) --
(106.23, 79.00) --
(109.39, 76.25) --
(112.62, 73.62) --
(115.92, 71.10) --
(119.29, 68.70) --
(122.73, 66.42) --
(126.24, 64.24) --
(129.83, 62.17) --
(133.48, 60.21) --
(137.20, 58.34) --
(141.00, 56.57) --
(144.87, 54.90) --
(148.81, 53.31) --
(152.81, 51.81) --
(156.89, 50.39) --
(161.04, 49.05) --
(165.27, 47.79) --
(169.56, 46.60) --
(173.92, 45.49) --
(178.36, 44.43) --
(182.86, 43.44) --
(187.44, 42.52) --
(192.09, 41.65) --
(196.81, 40.83) --
(201.59, 40.06) --
(206.45, 39.35) --
(211.39, 38.68) --
(216.39, 38.06) --
(221.46, 37.47) --
(226.61, 36.93) --
(231.82, 36.42) --
(237.11, 35.95) --
(242.46, 35.51) --
(247.89, 35.10) --
(253.39, 34.73) --
(258.96, 34.38) --
(264.60, 34.05) --
(270.31, 33.75) --
(276.10, 33.47) --
(281.95, 33.21) --
(287.87, 32.97) --
(293.87, 32.75) --
(299.94, 32.55) --
(306.07, 32.36) --
(312.28, 32.19) --
(318.56, 32.03) --
(324.91, 31.89) --
(331.33, 31.75) --
(337.83, 31.63) --
(344.39, 31.52) --
(351.02, 31.41) --
(357.73, 31.32) --
(364.50, 31.23) --
(371.35, 31.16) --
(378.27, 31.08) --
(385.26, 31.02) --
(385.26, 30.63) --
(378.27, 30.66) --
(371.35, 30.69) --
(364.50, 30.72) --
(357.73, 30.76) --
(351.02, 30.80) --
(344.39, 30.85) --
(337.83, 30.90) --
(331.33, 30.95) --
(324.91, 31.01) --
(318.56, 31.08) --
(312.28, 31.15) --
(306.07, 31.23) --
(299.94, 31.32) --
(293.87, 31.41) --
(287.87, 31.52) --
(281.95, 31.63) --
(276.10, 31.76) --
(270.31, 31.90) --
(264.60, 32.05) --
(258.96, 32.21) --
(253.39, 32.39) --
(247.89, 32.59) --
(242.46, 32.80) --
(237.11, 33.03) --
(231.82, 33.28) --
(226.61, 33.56) --
(221.46, 33.85) --
(216.39, 34.18) --
(211.39, 34.52) --
(206.45, 34.90) --
(201.59, 35.31) --
(196.81, 35.75) --
(192.09, 36.22) --
(187.44, 36.74) --
(182.86, 37.29) --
(178.36, 37.89) --
(173.92, 38.53) --
(169.56, 39.21) --
(165.27, 39.95) --
(161.04, 40.75) --
(156.89, 41.60) --
(152.81, 42.51) --
(148.81, 43.48) --
(144.87, 44.52) --
(141.00, 45.63) --
(137.20, 46.81) --
(133.48, 48.07) --
(129.83, 49.42) --
(126.24, 50.84) --
(122.73, 52.36) --
(119.29, 53.97) --
(115.92, 55.67) --
(112.62, 57.47) --
(109.39, 59.38) --
(106.23, 61.39) --
(103.15, 63.51) --
(100.13, 65.74) --
( 97.19, 68.09) --
( 94.31, 70.56) --
( 91.51, 73.15) --
( 88.78, 75.87) --
( 86.12, 78.71) --
( 83.53, 81.68) --
( 81.01, 84.78) --
( 78.56, 88.01) --
( 76.18, 91.37) --
( 73.88, 94.87) --
( 71.64, 98.49) --
( 69.48,102.25) --
( 67.39,106.13) --
( 65.36,110.14) --
( 63.41,114.28) --
( 61.53,118.54) --
( 59.72,122.91) --
( 57.98,127.40) --
( 56.32,132.00) --
( 54.72,136.70) --
( 53.20,141.49) --
( 51.74,146.37) --
( 50.36,151.33) --
( 49.04,156.35) --
( 47.80,161.44) --
( 46.63,166.57) --
( 45.53,171.75) --
( 44.50,176.94) --
( 43.55,182.14) --
( 42.66,187.34) --
( 41.84,192.51) --
( 41.10,197.65) --
( 40.42,202.73) --
( 39.82,207.73) --
( 39.29,212.63) --
( 38.83,217.41) --
( 38.44,222.05) --
( 38.12,226.52) --
( 37.87,230.81) --
( 37.69,234.98) --
( 37.59,239.56) --
( 37.55,240.51) --
cycle;
\path[] ( 37.55,240.51) --
( 37.59,240.32) --
( 37.69,239.74) --
( 37.87,238.78) --
( 38.12,237.45) --
( 38.44,235.76) --
( 38.83,233.71) --
( 39.29,231.33) --
( 39.82,228.63) --
( 40.42,225.62) --
( 41.10,222.33) --
( 41.84,218.79) --
( 42.66,215.00) --
( 43.55,210.99) --
( 44.50,206.79) --
( 45.53,202.42) --
( 46.63,197.90) --
( 47.80,193.25) --
( 49.04,188.51) --
( 50.36,183.69) --
( 51.74,178.80) --
( 53.20,173.88) --
( 54.72,168.95) --
( 56.32,164.01) --
( 57.98,159.07) --
( 59.72,154.18) --
( 61.53,149.33) --
( 63.41,144.53) --
( 65.36,139.81) --
( 67.39,135.16) --
( 69.48,130.61) --
( 71.64,126.15) --
( 73.88,121.80) --
( 76.18,117.56) --
( 78.56,113.43) --
( 81.01,109.42) --
( 83.53,105.53) --
( 86.12,101.77) --
( 88.78, 98.13) --
( 91.51, 94.63) --
( 94.31, 91.25) --
( 97.19, 87.99) --
(100.13, 84.87) --
(103.15, 81.87) --
(106.23, 79.00) --
(109.39, 76.25) --
(112.62, 73.62) --
(115.92, 71.10) --
(119.29, 68.70) --
(122.73, 66.42) --
(126.24, 64.24) --
(129.83, 62.17) --
(133.48, 60.21) --
(137.20, 58.34) --
(141.00, 56.57) --
(144.87, 54.90) --
(148.81, 53.31) --
(152.81, 51.81) --
(156.89, 50.39) --
(161.04, 49.05) --
(165.27, 47.79) --
(169.56, 46.60) --
(173.92, 45.49) --
(178.36, 44.43) --
(182.86, 43.44) --
(187.44, 42.52) --
(192.09, 41.65) --
(196.81, 40.83) --
(201.59, 40.06) --
(206.45, 39.35) --
(211.39, 38.68) --
(216.39, 38.06) --
(221.46, 37.47) --
(226.61, 36.93) --
(231.82, 36.42) --
(237.11, 35.95) --
(242.46, 35.51) --
(247.89, 35.10) --
(253.39, 34.73) --
(258.96, 34.38) --
(264.60, 34.05) --
(270.31, 33.75) --
(276.10, 33.47) --
(281.95, 33.21) --
(287.87, 32.97) --
(293.87, 32.75) --
(299.94, 32.55) --
(306.07, 32.36) --
(312.28, 32.19) --
(318.56, 32.03) --
(324.91, 31.89) --
(331.33, 31.75) --
(337.83, 31.63) --
(344.39, 31.52) --
(351.02, 31.41) --
(357.73, 31.32) --
(364.50, 31.23) --
(371.35, 31.16) --
(378.27, 31.08) --
(385.26, 31.02);
\path[] (385.26, 30.63) --
(378.27, 30.66) --
(371.35, 30.69) --
(364.50, 30.72) --
(357.73, 30.76) --
(351.02, 30.80) --
(344.39, 30.85) --
(337.83, 30.90) --
(331.33, 30.95) --
(324.91, 31.01) --
(318.56, 31.08) --
(312.28, 31.15) --
(306.07, 31.23) --
(299.94, 31.32) --
(293.87, 31.41) --
(287.87, 31.52) --
(281.95, 31.63) --
(276.10, 31.76) --
(270.31, 31.90) --
(264.60, 32.05) --
(258.96, 32.21) --
(253.39, 32.39) --
(247.89, 32.59) --
(242.46, 32.80) --
(237.11, 33.03) --
(231.82, 33.28) --
(226.61, 33.56) --
(221.46, 33.85) --
(216.39, 34.18) --
(211.39, 34.52) --
(206.45, 34.90) --
(201.59, 35.31) --
(196.81, 35.75) --
(192.09, 36.22) --
(187.44, 36.74) --
(182.86, 37.29) --
(178.36, 37.89) --
(173.92, 38.53) --
(169.56, 39.21) --
(165.27, 39.95) --
(161.04, 40.75) --
(156.89, 41.60) --
(152.81, 42.51) --
(148.81, 43.48) --
(144.87, 44.52) --
(141.00, 45.63) --
(137.20, 46.81) --
(133.48, 48.07) --
(129.83, 49.42) --
(126.24, 50.84) --
(122.73, 52.36) --
(119.29, 53.97) --
(115.92, 55.67) --
(112.62, 57.47) --
(109.39, 59.38) --
(106.23, 61.39) --
(103.15, 63.51) --
(100.13, 65.74) --
( 97.19, 68.09) --
( 94.31, 70.56) --
( 91.51, 73.15) --
( 88.78, 75.87) --
( 86.12, 78.71) --
( 83.53, 81.68) --
( 81.01, 84.78) --
( 78.56, 88.01) --
( 76.18, 91.37) --
( 73.88, 94.87) --
( 71.64, 98.49) --
( 69.48,102.25) --
( 67.39,106.13) --
( 65.36,110.14) --
( 63.41,114.28) --
( 61.53,118.54) --
( 59.72,122.91) --
( 57.98,127.40) --
( 56.32,132.00) --
( 54.72,136.70) --
( 53.20,141.49) --
( 51.74,146.37) --
( 50.36,151.33) --
( 49.04,156.35) --
( 47.80,161.44) --
( 46.63,166.57) --
( 45.53,171.75) --
( 44.50,176.94) --
( 43.55,182.14) --
( 42.66,187.34) --
( 41.84,192.51) --
( 41.10,197.65) --
( 40.42,202.73) --
( 39.82,207.73) --
( 39.29,212.63) --
( 38.83,217.41) --
( 38.44,222.05) --
( 38.12,226.52) --
( 37.87,230.81) --
( 37.69,234.98) --
( 37.59,239.56) --
( 37.55,240.51);
\definecolor{drawColor}{RGB}{0,0,0}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 37.55,240.51) --
( 37.59,232.64) --
( 37.69,224.99) --
( 37.87,217.55) --
( 38.12,210.32) --
( 38.44,203.30) --
( 38.83,196.48) --
( 39.29,189.86) --
( 39.82,183.45) --
( 40.42,177.22) --
( 41.10,171.19) --
( 41.84,165.35) --
( 42.66,159.69) --
( 43.55,154.21) --
( 44.50,148.91) --
( 45.53,143.78) --
( 46.63,138.82) --
( 47.80,134.03) --
( 49.04,129.41) --
( 50.36,124.94) --
( 51.74,120.63) --
( 53.20,116.48) --
( 54.72,112.47) --
( 56.32,108.61) --
( 57.98,104.90) --
( 59.72,101.32) --
( 61.53, 97.88) --
( 63.41, 94.57) --
( 65.36, 91.40) --
( 67.39, 88.34) --
( 69.48, 85.41) --
( 71.64, 82.60) --
( 73.88, 79.91) --
( 76.18, 77.33) --
( 78.56, 74.85) --
( 81.01, 72.49) --
( 83.53, 70.22) --
( 86.12, 68.06) --
( 88.78, 65.99) --
( 91.51, 64.02) --
( 94.31, 62.14) --
( 97.19, 60.34) --
(100.13, 58.63) --
(103.15, 57.00) --
(106.23, 55.45) --
(109.39, 53.97) --
(112.62, 52.57) --
(115.92, 51.23) --
(119.29, 49.97) --
(122.73, 48.77) --
(126.24, 47.63) --
(129.83, 46.55) --
(133.48, 45.53) --
(137.20, 44.56) --
(141.00, 43.65) --
(144.87, 42.79) --
(148.81, 41.98) --
(152.81, 41.21) --
(156.89, 40.48) --
(161.04, 39.80) --
(165.27, 39.16) --
(169.56, 38.55) --
(173.92, 37.99) --
(178.36, 37.45) --
(182.86, 36.95) --
(187.44, 36.48) --
(192.09, 36.04) --
(196.81, 35.63) --
(201.59, 35.25) --
(206.45, 34.89) --
(211.39, 34.55) --
(216.39, 34.24) --
(221.46, 33.94) --
(226.61, 33.67) --
(231.82, 33.42) --
(237.11, 33.18) --
(242.46, 32.96) --
(247.89, 32.75) --
(253.39, 32.56) --
(258.96, 32.39) --
(264.60, 32.23) --
(270.31, 32.07) --
(276.10, 31.93) --
(281.95, 31.80) --
(287.87, 31.69) --
(293.87, 31.58) --
(299.94, 31.47) --
(306.07, 31.38) --
(312.28, 31.29) --
(318.56, 31.21) --
(324.91, 31.14) --
(331.33, 31.07) --
(337.83, 31.01) --
(344.39, 30.96) --
(351.02, 30.91) --
(357.73, 30.86) --
(364.50, 30.81) --
(371.35, 30.78) --
(378.27, 30.74) --
(385.26, 30.71);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 2pt off 2pt ,line join=round] ( 37.55,240.51) --
( 37.59,240.32) --
( 37.69,239.74) --
( 37.87,238.78) --
( 38.12,237.45) --
( 38.44,235.76) --
( 38.83,233.71) --
( 39.29,231.33) --
( 39.82,228.63) --
( 40.42,225.62) --
( 41.10,222.33) --
( 41.84,218.79) --
( 42.66,215.00) --
( 43.55,210.99) --
( 44.50,206.79) --
( 45.53,202.42) --
( 46.63,197.90) --
( 47.80,193.25) --
( 49.04,188.51) --
( 50.36,183.69) --
( 51.74,178.80) --
( 53.20,173.88) --
( 54.72,168.95) --
( 56.32,164.01) --
( 57.98,159.07) --
( 59.72,154.18) --
( 61.53,149.33) --
( 63.41,144.53) --
( 65.36,139.81) --
( 67.39,135.16) --
( 69.48,130.61) --
( 71.64,126.15) --
( 73.88,121.80) --
( 76.18,117.56) --
( 78.56,113.43) --
( 81.01,109.42) --
( 83.53,105.53) --
( 86.12,101.77) --
( 88.78, 98.13) --
( 91.51, 94.63) --
( 94.31, 91.25) --
( 97.19, 87.99) --
(100.13, 84.87) --
(103.15, 81.87) --
(106.23, 79.00) --
(109.39, 76.25) --
(112.62, 73.62) --
(115.92, 71.10) --
(119.29, 68.70) --
(122.73, 66.42) --
(126.24, 64.24) --
(129.83, 62.17) --
(133.48, 60.21) --
(137.20, 58.34) --
(141.00, 56.57) --
(144.87, 54.90) --
(148.81, 53.31) --
(152.81, 51.81) --
(156.89, 50.39) --
(161.04, 49.05) --
(165.27, 47.79) --
(169.56, 46.60) --
(173.92, 45.49) --
(178.36, 44.43) --
(182.86, 43.44) --
(187.44, 42.52) --
(192.09, 41.65) --
(196.81, 40.83) --
(201.59, 40.06) --
(206.45, 39.35) --
(211.39, 38.68) --
(216.39, 38.06) --
(221.46, 37.47) --
(226.61, 36.93) --
(231.82, 36.42) --
(237.11, 35.95) --
(242.46, 35.51) --
(247.89, 35.10) --
(253.39, 34.73) --
(258.96, 34.38) --
(264.60, 34.05) --
(270.31, 33.75) --
(276.10, 33.47) --
(281.95, 33.21) --
(287.87, 32.97) --
(293.87, 32.75) --
(299.94, 32.55) --
(306.07, 32.36) --
(312.28, 32.19) --
(318.56, 32.03) --
(324.91, 31.89) --
(331.33, 31.75) --
(337.83, 31.63) --
(344.39, 31.52) --
(351.02, 31.41) --
(357.73, 31.32) --
(364.50, 31.23) --
(371.35, 31.16) --
(378.27, 31.08) --
(385.26, 31.02);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{130,106,80}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 37.55, 30.40) --
( 37.55,240.72);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{85,85,85}
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 33.05, 27.64) {0};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 33.05, 38.15) {0.05};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 33.05, 48.65) {0.1};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 33.05, 69.66) {0.2};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 33.05,132.70) {0.5};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 33.05,237.75) {1};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{130,106,80}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 35.05, 30.40) --
( 37.55, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 35.05, 40.90) --
( 37.55, 40.90);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 35.05, 51.41) --
( 37.55, 51.41);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 35.05, 72.42) --
( 37.55, 72.42);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 35.05,135.45) --
( 37.55,135.45);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 35.05,240.51) --
( 37.55,240.51);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{130,106,80}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 37.55, 30.40) --
(385.26, 30.40);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{130,106,80}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 37.55, 27.90) --
( 37.55, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 81.08, 27.90) --
( 81.08, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (115.47, 27.90) --
(115.47, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (154.50, 27.90) --
(154.50, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (385.26, 27.90) --
(385.26, 30.40);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{85,85,85}
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.55, 20.39) {1};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 81.08, 20.39) {2.13};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (115.47, 20.39) {3.02};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (154.50, 20.39) {4.03};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (385.26, 20.39) {10};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{85,85,85}
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 1.00] at (211.40, 6.94) {$E[F]$};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,245.72);
\definecolor{drawColor}{RGB}{85,85,85}
\node[text=drawColor,rotate= 90.00,anchor=base,inner sep=0pt, outer sep=0pt, scale= 1.00] at ( 11.89,135.56) {Median bias relative to OLS OVB};
\end{scope}
\end{tikzpicture}
% Created by tikzDevice version 0.12.3.1 on 2021-10-26 21:51:51
% !TEX encoding = UTF-8 Unicode
\begin{tikzpicture}[x=1pt,y=1pt]
\definecolor{fillColor}{RGB}{255,255,255}
\path[use as bounding box,fill=fillColor,fill opacity=0.00] (0,0) rectangle (390.26,252.94);
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{255,255,255}
\definecolor{fillColor}{RGB}{255,255,255}
\path[draw=drawColor,line width= 0.6pt,line join=round,line cap=round,fill=fillColor] ( 0.00, 0.00) rectangle (390.26,252.94);
\end{scope}
\begin{scope}
\path[clip] ( 41.55, 30.40) rectangle (385.26,247.95);
\definecolor{fillColor}{RGB}{255,255,255}
\path[fill=fillColor] ( 41.55, 30.40) rectangle (385.26,247.95);
\definecolor{drawColor}{RGB}{208,208,208}
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55, 40.79) --
(385.26, 40.79);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55,128.02) --
(385.26,128.02);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55,160.72) --
(385.26,160.72);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55,181.58) --
(385.26,181.58);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 76.16, 30.40) --
( 76.16,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (113.81, 30.40) --
(113.81,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (270.98, 30.40) --
(270.98,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (289.87, 30.40) --
(289.87,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] ( 41.55, 30.40) --
( 41.55,247.95);
\path[draw=drawColor,line width= 0.6pt,dash pattern=on 4pt off 4pt ,line join=round] (385.26, 30.40) --
(385.26,247.95);
\definecolor{drawColor}{RGB}{119,119,119}
\path[draw=drawColor,line width= 0.6pt,line join=round] (385.26, 43.06) --
(376.61, 42.69) --
(368.08, 42.35) --
(360.10, 42.05) --
(359.65, 42.04) --
(351.33, 41.76) --
(343.13, 41.52) --
(335.03, 41.32) --
(327.05, 41.13) --
(319.17, 41.00) --
(311.41, 40.87) --
(303.75, 40.81) --
(296.21, 40.77) --
(288.77, 40.76) --
(281.45, 40.78) --
(274.23, 40.83) --
(267.13, 40.91) --
(260.13, 41.02) --
(253.25, 41.17) --
(246.47, 41.34) --
(239.81, 41.55) --
(233.26, 41.78) --
(226.81, 42.04) --
(226.60, 42.05) --
(220.48, 42.33) --
(214.26, 42.65) --
(208.14, 42.99) --
(202.14, 43.36) --
(196.25, 43.75) --
(190.47, 44.16) --
(186.92, 44.45) --
(184.79, 44.62) --
(179.23, 45.08) --
(173.78, 45.57) --
(168.44, 46.07) --
(163.20, 46.61) --
(161.05, 46.84) --
(158.08, 47.16) --
(153.07, 47.73) --
(148.17, 48.31) --
(143.38, 48.92) --
(141.07, 49.24) --
(138.70, 49.55) --
(134.13, 50.19) --
(129.67, 50.85) --
(125.31, 51.53) --
(124.69, 51.63) --
(121.07, 52.21) --
(116.94, 52.92) --
(112.92, 53.64) --
(110.90, 54.02) --
(109.01, 54.37) --
(105.21, 55.12) --
(101.52, 55.88) --
( 99.04, 56.42) --
( 97.94, 56.65) --
( 94.47, 57.43) --
( 91.11, 58.23) --
( 88.78, 58.81) --
( 87.87, 59.04) --
( 84.73, 59.85) --
( 81.70, 60.68) --
( 79.89, 61.21) --
( 78.78, 61.52) --
( 75.97, 62.37) --
( 73.27, 63.23) --
( 72.19, 63.60) --
( 70.68, 64.10) --
( 68.20, 64.98) --
( 65.84, 65.84) --
( 65.47, 65.99) --
( 63.58, 66.77) --
( 61.43, 67.68) --
( 59.88, 68.39) --
( 59.39, 68.60) --
( 57.47, 69.53) --
( 55.65, 70.47) --
( 55.10, 70.78) --
( 53.94, 71.42) --
( 52.34, 72.38) --
( 51.11, 73.18) --
( 50.86, 73.34) --
( 49.48, 74.34) --
( 48.21, 75.34) --
( 47.95, 75.57) --
( 47.06, 76.34) --
( 46.01, 77.36) --
( 45.46, 77.97) --
( 45.07, 78.39) --
( 44.25, 79.42) --
( 43.62, 80.36) --
( 43.53, 80.48) --
( 42.93, 81.54) --
( 42.43, 82.63) --
( 42.39, 82.75) --
( 42.04, 83.72) --
( 41.77, 84.84) --
( 41.72, 85.15) --
( 41.60, 85.96) --
( 41.55, 87.10);
\path[draw=drawColor,line width= 0.6pt,line join=round] (270.97,247.95) --
(267.69,245.55) --
(267.13,245.15) --
(264.44,243.16) --
(261.16,240.76) --
(260.13,240.02) --
(257.93,238.37) --
(254.68,235.97) --
(253.25,234.93) --
(251.47,233.58) --
(248.27,231.19) --
(246.47,229.87) --
(245.07,228.79) --
(241.90,226.40) --
(239.81,224.85) --
(238.72,224.00) --
(235.58,221.61) --
(233.26,219.87) --
(232.42,219.22) --
(229.31,216.82) --
(226.81,214.94) --
(226.16,214.43) --
(223.08,212.03) --
(220.48,210.07) --
(219.94,209.64) --
(216.87,207.25) --
(214.26,205.26) --
(213.75,204.85) --
(210.68,202.46) --
(208.14,200.52) --
(207.57,200.06) --
(204.50,197.67) --
(202.14,195.88) --
(201.38,195.28) --
(198.30,192.88) --
(196.25,191.34) --
(195.17,190.49) --
(192.04,188.09) --
(190.47,186.92) --
(188.88,185.70) --
(185.69,183.30) --
(184.79,182.66) --
(182.49,180.91) --
(179.23,178.53) --
(179.21,178.52) --
(175.92,176.12) --
(173.78,174.63) --
(172.54,173.73) --
(169.09,171.33) --
(168.44,170.90) --
(165.61,168.94) --
(163.20,167.36) --
(161.99,166.55) --
(158.26,164.15) --
(158.08,164.04) --
(154.46,161.76) --
(153.07,160.93) --
(150.51,159.36) --
(148.17,158.00) --
(146.41,156.97) --
(143.38,155.28) --
(142.13,154.58) --
(138.70,152.73) --
(137.67,152.18) --
(134.13,150.37) --
(132.99,149.79) --
(129.67,148.17) --
(128.06,147.39) --
(125.31,146.13) --
(122.83,145.00) --
(121.07,144.24) --
(117.22,142.61) --
(116.94,142.49) --
(112.92,140.87) --
(111.23,140.21) --
(109.01,139.38) --
(105.21,138.01) --
(104.66,137.82) --
(101.52,136.75) --
( 97.94,135.60) --
( 97.39,135.42) --
( 94.47,134.54) --
( 91.11,133.58) --
( 89.08,133.03) --
( 87.87,132.71) --
( 84.73,131.92) --
( 81.70,131.22) --
( 78.98,130.63) --
( 78.78,130.59) --
( 75.97,130.03) --
( 73.27,129.55) --
( 70.68,129.13) --
( 68.20,128.78) --
( 65.84,128.49) --
( 63.58,128.27) --
( 63.32,128.24) --
( 61.43,128.06) --
( 59.39,127.97) --
( 57.47,127.91) --
( 55.65,127.89) --
( 53.94,127.93) --
( 52.34,128.02) --
( 50.86,128.16) --
( 50.25,128.24) --
( 49.48,128.34) --
( 48.21,128.57) --
( 47.06,128.84) --
( 46.01,129.16) --
( 45.07,129.52) --
( 44.25,129.93) --
( 43.53,130.38) --
( 43.22,130.63) --
( 42.93,130.87) --
( 42.43,131.40) --
( 42.04,131.98) --
( 41.77,132.61) --
( 41.67,133.03) --
( 41.60,133.28) --
( 41.55,133.98);
\path[draw=drawColor,line width= 0.6pt,line join=round] (114.20,247.95) --
(112.93,245.55) --
(112.92,245.54) --
(112.07,243.16) --
(111.20,240.76) --
(110.31,238.37) --
(109.39,235.97) --
(109.01,235.01) --
(108.50,233.58) --
(107.61,231.19) --
(106.69,228.79) --
(105.75,226.40) --
(105.21,225.08) --
(104.81,224.00) --
(103.87,221.61) --
(102.89,219.22) --
(101.85,216.82) --
(101.52,216.11) --
(100.81,214.43) --
( 99.73,212.03) --
( 98.56,209.64) --
( 97.94,208.44) --
( 97.36,207.25) --
( 96.10,204.85) --
( 94.73,202.46) --
( 94.47,202.04) --
( 93.32,200.06) --
( 91.74,197.67) --
( 91.11,196.74) --
( 90.19,195.28) --
( 88.46,192.88) --
( 87.87,192.11) --
( 86.64,190.49) --
( 84.73,188.16) --
( 84.67,188.09) --
( 82.60,185.70) --
( 81.70,184.73) --
( 80.38,183.30) --
( 78.78,181.70) --
( 78.00,180.91) --
( 75.97,179.01) --
( 75.44,178.52) --
( 73.27,176.62) --
( 72.70,176.12) --
( 70.68,174.48) --
( 69.74,173.73) --
( 68.20,172.57) --
( 66.52,171.33) --
( 65.84,170.86) --
( 63.58,169.32) --
( 62.99,168.94) --
( 61.43,167.95) --
( 59.39,166.73) --
( 59.07,166.55) --
( 57.47,165.64) --
( 55.65,164.68) --
( 54.57,164.15) --
( 53.94,163.85) --
( 52.34,163.11) --
( 50.86,162.48) --
( 49.48,161.96) --
( 48.90,161.76) --
( 48.21,161.52) --
( 47.06,161.17) --
( 46.01,160.91) --
( 45.07,160.73) --
( 44.25,160.63) --
( 43.53,160.61) --
( 42.93,160.66) --
( 42.43,160.79) --
( 42.04,160.99) --
( 41.77,161.27) --
( 41.60,161.62) --
( 41.59,161.76) --
( 41.55,162.03);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55,181.70) --
( 41.60,181.52) --
( 41.77,181.43) --
( 42.04,181.44) --
( 42.43,181.54) --
( 42.93,181.76) --
( 43.53,182.08) --
( 44.25,182.52) --
( 45.07,183.07) --
( 45.39,183.30) --
( 46.01,183.75) --
( 47.06,184.56) --
( 48.21,185.52) --
( 48.42,185.70) --
( 49.48,186.62) --
( 50.86,187.90) --
( 51.05,188.09) --
( 52.34,189.36) --
( 53.43,190.49) --
( 53.94,191.04) --
( 55.59,192.88) --
( 55.65,192.95) --
( 57.47,195.14) --
( 57.58,195.28) --
( 59.39,197.65) --
( 59.41,197.67) --
( 61.10,200.06) --
( 61.43,200.58) --
( 62.66,202.46) --
( 63.58,204.02) --
( 64.08,204.85) --
( 65.39,207.25) --
( 65.84,208.15) --
( 66.59,209.64) --
( 67.68,212.03) --
( 68.20,213.32) --
( 68.67,214.43) --
( 69.58,216.82) --
( 70.38,219.22) --
( 70.68,220.26) --
( 71.11,221.61) --
( 71.77,224.00) --
( 72.36,226.40) --
( 72.87,228.79) --
( 73.27,230.89) --
( 73.33,231.19) --
( 73.80,233.58) --
( 74.23,235.97) --
( 74.64,238.37) --
( 75.03,240.76) --
( 75.42,243.16) --
( 75.79,245.55) --
( 75.97,246.75) --
( 76.17,247.95);
\definecolor{drawColor}{RGB}{255,255,255}
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.60, 40.73) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (166.45,165.58) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 92.46,191.82) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 70.88,209.05) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.22, 41.30) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (166.07,166.16) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 92.08,192.39) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 70.50,209.62) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (208.65, 41.68) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (165.50,166.54) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 91.51,192.77) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.92,210.01) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.97, 41.82) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.82,166.67) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 90.83,192.91) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.25,210.14) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.30, 41.68) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.15,166.54) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 90.16,192.77) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 68.57,210.01) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.73, 41.30) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.58,166.16) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 89.59,192.39) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 68.00,209.62) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.34, 40.73) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.20,165.58) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 89.20,191.82) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 67.62,209.05) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.21, 40.05) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.06,164.91) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 89.07,191.14) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 67.48,208.38) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.34, 39.38) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.20,164.23) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 89.20,190.47) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 67.62,207.70) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (206.73, 38.81) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (163.58,163.66) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 89.59,189.90) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 68.00,207.13) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.30, 38.42) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.15,163.28) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 90.16,189.51) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 68.57,206.75) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.97, 38.29) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.82,163.14) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 90.83,189.38) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.25,206.61) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (208.65, 38.42) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (165.50,163.28) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 91.51,189.51) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.92,206.75) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.22, 38.81) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (166.07,163.66) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 92.08,189.90) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 70.50,207.13) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.60, 39.38) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (166.45,164.23) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 92.46,190.47) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 70.88,207.70) {0.2};
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (209.74, 40.05) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (166.59,164.91) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 92.60,191.14) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 71.01,208.38) {0.2};
\definecolor{drawColor}{RGB}{0,0,0}
\node[text=drawColor,rotate= -3.35,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (207.97, 40.05) {0.05};
\node[text=drawColor,rotate= 33.44,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at (164.82,164.91) {0.1};
\node[text=drawColor,rotate= 53.75,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 90.83,191.14) {0.15};
\node[text=drawColor,rotate= 64.57,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.85] at ( 69.25,208.38) {0.2};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55, 30.40) --
( 41.55,247.95);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05, 38.04) {0.563};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05,125.26) {0.747};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05,157.97) {0.816};
\node[text=drawColor,anchor=base east,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 37.05,178.83) {0.860};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05, 40.79) --
( 41.55, 40.79);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05,128.02) --
( 41.55,128.02);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05,160.72) --
( 41.55,160.72);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 39.05,181.58) --
( 41.55,181.58);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55, 30.40) --
(385.26, 30.40);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{144,144,144}
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 76.16, 27.90) --
( 76.16, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (113.81, 27.90) --
(113.81, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (270.98, 27.90) --
(270.98, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (289.87, 27.90) --
(289.87, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] ( 41.55, 27.90) --
( 41.55, 30.40);
\path[draw=drawColor,line width= 0.6pt,line join=round] (385.26, 27.90) --
(385.26, 30.40);
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 76.16, 20.39) {1.91};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (113.81, 20.39) {2.89};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (270.98, 20.39) {7.01};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (289.87, 20.39) {7.5};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at ( 41.55, 20.39) {1};
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 0.80] at (385.26, 20.39) {10};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,anchor=base,inner sep=0pt, outer sep=0pt, scale= 1.00] at (213.40, 6.94) {$E[F]$};
\end{scope}
\begin{scope}
\path[clip] ( 0.00, 0.00) rectangle (390.26,252.94);
\definecolor{drawColor}{RGB}{68,68,68}
\node[text=drawColor,rotate= 90.00,anchor=base,inner sep=0pt, outer sep=0pt, scale= 1.00] at ( 11.89,139.17) {$|\rho|$};
\end{scope}
\end{tikzpicture}
% $ biblatex auxiliary file $
% $ biblatex bbl format version 3.1 $
% Do not modify the above lines!
%
% This is an auxiliary file used by the 'biblatex' package.
% This file may safely be deleted. It will be recreated by
% biber as required.
%
\begingroup
\makeatletter
\@ifundefined{ver@biblatex.sty}
{\@latex@error
{Missing 'biblatex' package}
{The bibliography requires the 'biblatex' package.}
\aftergroup\endinput}
{}
\endgroup
\refsection{0}
\datalist[entry]{apa/global//global/global}
\entry{AnRu49}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=6f63a006d64453372efea778c0d93799}{%
family={Anderson},
familyi={A\bibinitperiod},
given={Theodore\bibnamedelima W.},
giveni={T\bibinitperiod\bibinitdelim W\bibinitperiod},
givenun=0}}%
{{un=1,uniquepart=given,hash=62a17f3760779bdde9c10c546249bb56}{%
family={Rubin},
familyi={R\bibinitperiod},
given={Herman},
giveni={H\bibinitperiod},
givenun=1}}%
}
\strng{namehash}{2a9dff1173788cc90856d13c09f2b601}
\strng{fullhash}{2a9dff1173788cc90856d13c09f2b601}
\strng{bibnamehash}{2a9dff1173788cc90856d13c09f2b601}
\strng{authorbibnamehash}{2a9dff1173788cc90856d13c09f2b601}
\strng{authornamehash}{2a9dff1173788cc90856d13c09f2b601}
\strng{authorfullhash}{2a9dff1173788cc90856d13c09f2b601}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{The Annals of Mathematical Statistics}
\field{month}{3}
\field{number}{1}
\field{title}{Estimation of the Parameters of a Single Equation in a Complete System of Stochastic Equations}
\field{volume}{20}
\field{year}{1949}
\field{pages}{46\bibrangedash 63}
\range{pages}{18}
\verb{doi}
\verb 10.1214/aoms/1177730090
\endverb
\endentry
\entry{AnAr17}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=fffe81ec7237c0735d37245f7d43b67d}{%
family={Andrews},
familyi={A\bibinitperiod},
given={Isaiah},
giveni={I\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=134f3762ae0758b041f7a008ccf00188}{%
family={Armstrong},
familyi={A\bibinitperiod},
given={Timothy\bibnamedelima B.},
giveni={T\bibinitperiod\bibinitdelim B\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{a0bb277b8b8e0554bc637700cf9e38bf}
\strng{fullhash}{a0bb277b8b8e0554bc637700cf9e38bf}
\strng{bibnamehash}{a0bb277b8b8e0554bc637700cf9e38bf}
\strng{authorbibnamehash}{a0bb277b8b8e0554bc637700cf9e38bf}
\strng{authornamehash}{a0bb277b8b8e0554bc637700cf9e38bf}
\strng{authorfullhash}{a0bb277b8b8e0554bc637700cf9e38bf}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Quantitative Economics}
\field{month}{7}
\field{number}{2}
\field{title}{Unbiased Instrumental Variables Estimation under Known First-Stage Sign}
\field{volume}{8}
\field{year}{2017}
\field{pages}{479\bibrangedash 503}
\range{pages}{25}
\verb{doi}
\verb 10.3982/QE700
\endverb
\endentry
\entry{ass19}{article}{}
\name{author}{3}{}{%
{{un=0,uniquepart=base,hash=fffe81ec7237c0735d37245f7d43b67d}{%
family={Andrews},
familyi={A\bibinitperiod},
given={Isaiah},
giveni={I\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=2d01c3b46f4061430cc9eb2a1f6408a4}{%
family={Stock},
familyi={S\bibinitperiod},
given={James\bibnamedelima H.},
giveni={J\bibinitperiod\bibinitdelim H\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=3011926b49b3aa7b3e8913ca855e42fc}{%
family={Sun},
familyi={S\bibinitperiod},
given={Liyang},
giveni={L\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{4992cf8f84e27555b4308aeedeb0edfd}
\strng{fullhash}{55595dd48f58b4787421db14a31a0b98}
\strng{bibnamehash}{55595dd48f58b4787421db14a31a0b98}
\strng{authorbibnamehash}{55595dd48f58b4787421db14a31a0b98}
\strng{authornamehash}{4992cf8f84e27555b4308aeedeb0edfd}
\strng{authorfullhash}{55595dd48f58b4787421db14a31a0b98}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Annual Review of Economics}
\field{month}{8}
\field{number}{1}
\field{title}{Weak Instruments in Instrumental Variables Regression: Theory and Practice}
\field{volume}{11}
\field{year}{2019}
\field{pages}{727\bibrangedash 753}
\range{pages}{27}
\verb{doi}
\verb 10.1146/annurev-economics-080218-025643
\endverb
\endentry
\entry{angrist90}{article}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=ee326f34eea24952d04c41cef1771bfc}{%
family={Angrist},
familyi={A\bibinitperiod},
given={Joshua\bibnamedelima D.},
giveni={J\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{ee326f34eea24952d04c41cef1771bfc}
\strng{fullhash}{ee326f34eea24952d04c41cef1771bfc}
\strng{bibnamehash}{ee326f34eea24952d04c41cef1771bfc}
\strng{authorbibnamehash}{ee326f34eea24952d04c41cef1771bfc}
\strng{authornamehash}{ee326f34eea24952d04c41cef1771bfc}
\strng{authorfullhash}{ee326f34eea24952d04c41cef1771bfc}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{American Economic Review}
\field{month}{6}
\field{number}{3}
\field{title}{Lifetime Earnings and the {Vietnam} Era Draft Lottery: Evidence from Social Security Administrative Records}
\field{volume}{80}
\field{year}{1990}
\field{pages}{313\bibrangedash 336}
\range{pages}{24}
\verb{urlraw}
\verb https://www.jstor.org/stable/10.2307/2006669
\endverb
\verb{url}
\verb https://www.jstor.org/stable/10.2307/2006669
\endverb
\endentry
\entry{AnEv98}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=ee326f34eea24952d04c41cef1771bfc}{%
family={Angrist},
familyi={A\bibinitperiod},
given={Joshua\bibnamedelima D.},
giveni={J\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=7a13f4aa754b50c84a0858a07c923975}{%
family={Evans},
familyi={E\bibinitperiod},
given={William\bibnamedelima N.},
giveni={W\bibinitperiod\bibinitdelim N\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{acf1e644ff609fa5a5ed8de13f026542}
\strng{fullhash}{acf1e644ff609fa5a5ed8de13f026542}
\strng{bibnamehash}{acf1e644ff609fa5a5ed8de13f026542}
\strng{authorbibnamehash}{acf1e644ff609fa5a5ed8de13f026542}
\strng{authornamehash}{acf1e644ff609fa5a5ed8de13f026542}
\strng{authorfullhash}{acf1e644ff609fa5a5ed8de13f026542}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{American Economic Review}
\field{month}{6}
\field{number}{3}
\field{title}{Children and Their Parents' Labor Supply: Evidence from Exogenous Variation in Family Size}
\field{volume}{88}
\field{year}{1998}
\field{pages}{450\bibrangedash 477}
\range{pages}{28}
\endentry
\entry{AnIm95}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=ee326f34eea24952d04c41cef1771bfc}{%
family={Angrist},
familyi={A\bibinitperiod},
given={Joshua\bibnamedelima D.},
giveni={J\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=9f4fa8d00c0311b9cdad0f20ad429e57}{%
family={Imbens},
familyi={I\bibinitperiod},
given={Guido\bibnamedelima W.},
giveni={G\bibinitperiod\bibinitdelim W\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{fb6fa0691b2637fa874fbf6303f66473}
\strng{fullhash}{fb6fa0691b2637fa874fbf6303f66473}
\strng{bibnamehash}{fb6fa0691b2637fa874fbf6303f66473}
\strng{authorbibnamehash}{fb6fa0691b2637fa874fbf6303f66473}
\strng{authornamehash}{fb6fa0691b2637fa874fbf6303f66473}
\strng{authorfullhash}{fb6fa0691b2637fa874fbf6303f66473}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{month}{6}
\field{number}{430}
\field{title}{Two-Stage Least Squares Estimation of Average Causal Effects in Models With Variable Treatment Intensity}
\field{volume}{90}
\field{year}{1995}
\field{pages}{431\bibrangedash 442}
\range{pages}{12}
\verb{doi}
\verb 10.1080/01621459.1995.10476535
\endverb
\endentry
\entry{AnKr91}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=ee326f34eea24952d04c41cef1771bfc}{%
family={Angrist},
familyi={A\bibinitperiod},
given={Joshua\bibnamedelima D.},
giveni={J\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=34526c3bf548080921683822747fad7e}{%
family={Krueger},
familyi={K\bibinitperiod},
given={Alan\bibnamedelima B.},
giveni={A\bibinitperiod\bibinitdelim B\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{fullhash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{bibnamehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{authorbibnamehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{authornamehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{authorfullhash}{f88e0b2c89794fe8175a868a7e38c174}
\field{extraname}{1}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{The Quarterly Journal of Economics}
\field{month}{11}
\field{number}{4}
\field{title}{Does Compulsory School Attendance Affect Schooling and Earnings?}
\field{volume}{106}
\field{year}{1991}
\field{pages}{979\bibrangedash 1014}
\range{pages}{36}
\verb{doi}
\verb 10.2307/2937954
\endverb
\verb{urlraw}
\verb https://www.jstor.org/stable/116844
\endverb
\verb{url}
\verb https://www.jstor.org/stable/116844
\endverb
\endentry
\entry{AnKr99handbook}{inbook}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=ee326f34eea24952d04c41cef1771bfc}{%
family={Angrist},
familyi={A\bibinitperiod},
given={Joshua\bibnamedelima D.},
giveni={J\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=34526c3bf548080921683822747fad7e}{%
family={Krueger},
familyi={K\bibinitperiod},
given={Alan\bibnamedelima B.},
giveni={A\bibinitperiod\bibinitdelim B\bibinitperiod},
givenun=0}}%
}
\name{editor}{2}{}{%
{{hash=38104046316aaf803651690c49e29684}{%
family={Ashenfelter},
familyi={A\bibinitperiod},
given={Orley\bibnamedelima C.},
giveni={O\bibinitperiod\bibinitdelim C\bibinitperiod}}}%
{{hash=a9e88233ec6ce30d2d31e294961e3e76}{%
family={Card},
familyi={C\bibinitperiod},
given={David},
giveni={D\bibinitperiod}}}%
}
\list{location}{1}{%
{Amsterdam}%
}
\list{publisher}{1}{%
{Elsevier}%
}
\strng{namehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{fullhash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{bibnamehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{authorbibnamehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{authornamehash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{authorfullhash}{f88e0b2c89794fe8175a868a7e38c174}
\strng{editorbibnamehash}{ec08e6a0ac81747a350eaa2fe11c56cb}
\strng{editornamehash}{ec08e6a0ac81747a350eaa2fe11c56cb}
\strng{editorfullhash}{ec08e6a0ac81747a350eaa2fe11c56cb}
\field{extraname}{2}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{booktitle}{Handbook of Labor Economics}
\field{chapter}{23}
\field{title}{Empirical Strategies in Labor Economics}
\field{volume}{3A}
\field{year}{1999}
\field{pages}{1277\bibrangedash 1366}
\range{pages}{90}
\verb{doi}
\verb 10.1016/S1573-4463(99)03004-7
\endverb
\endentry
\entry{AnLa99}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=ee326f34eea24952d04c41cef1771bfc}{%
family={Angrist},
familyi={A\bibinitperiod},
given={Joshua\bibnamedelima D.},
giveni={J\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=c0b275458848c41e8da821625f9a27c1}{%
family={Lavy},
familyi={L\bibinitperiod},
given={Victor},
giveni={V\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{390181b3a152836e1c2a752ea08eb79d}
\strng{fullhash}{390181b3a152836e1c2a752ea08eb79d}
\strng{bibnamehash}{390181b3a152836e1c2a752ea08eb79d}
\strng{authorbibnamehash}{390181b3a152836e1c2a752ea08eb79d}
\strng{authornamehash}{390181b3a152836e1c2a752ea08eb79d}
\strng{authorfullhash}{390181b3a152836e1c2a752ea08eb79d}
\field{sortinit}{A}
\field{sortinithash}{a3dcedd53b04d1adfd5ac303ecd5e6fa}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{The Quarterly Journal of Economics}
\field{month}{5}
\field{number}{2}
\field{title}{Using {Maimonides}' Rule to Estimate the Effect of Class Size on Scholastic Achievement}
\field{volume}{114}
\field{year}{1999}
\field{pages}{533\bibrangedash 575}
\range{pages}{43}
\verb{doi}
\verb 10.1162/003355399556061
\endverb
\endentry
\entry{bekker94}{article}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=aea203c99218a041e6d53b0709fab796}{%
family={Bekker},
familyi={B\bibinitperiod},
given={Paul\bibnamedelima A.},
giveni={P\bibinitperiod\bibinitdelim A\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{aea203c99218a041e6d53b0709fab796}
\strng{fullhash}{aea203c99218a041e6d53b0709fab796}
\strng{bibnamehash}{aea203c99218a041e6d53b0709fab796}
\strng{authorbibnamehash}{aea203c99218a041e6d53b0709fab796}
\strng{authornamehash}{aea203c99218a041e6d53b0709fab796}
\strng{authorfullhash}{aea203c99218a041e6d53b0709fab796}
\field{sortinit}{B}
\field{sortinithash}{8de16967003c7207dae369d874f1456e}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Econometrica}
\field{month}{5}
\field{number}{3}
\field{title}{Alternative Approximations to the Distributions of Instrumental Variable Estimators}
\field{volume}{62}
\field{year}{1994}
\field{pages}{657\bibrangedash 681}
\range{pages}{25}
\verb{doi}
\verb 10.2307/2951662
\endverb
\endentry
\entry{bjb95}{article}{}
\name{author}{3}{}{%
{{un=0,uniquepart=base,hash=206a431f3c3d7796c2f18f2c019cee45}{%
family={Bound},
familyi={B\bibinitperiod},
given={John},
giveni={J\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=dc0934d6a00d8a633f4e0d41b74e700e}{%
family={Jaeger},
familyi={J\bibinitperiod},
given={David\bibnamedelima A.},
giveni={D\bibinitperiod\bibinitdelim A\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=0f0cb6ed3d059568affba03fb9cb537f}{%
family={Baker},
familyi={B\bibinitperiod},
given={Regina\bibnamedelima M.},
giveni={R\bibinitperiod\bibinitdelim M\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{fc78cd8458bd8920657b315050767c80}
\strng{fullhash}{7aabe9008b7b461a7c4ebe01368760bf}
\strng{bibnamehash}{7aabe9008b7b461a7c4ebe01368760bf}
\strng{authorbibnamehash}{7aabe9008b7b461a7c4ebe01368760bf}
\strng{authornamehash}{fc78cd8458bd8920657b315050767c80}
\strng{authorfullhash}{7aabe9008b7b461a7c4ebe01368760bf}
\field{sortinit}{B}
\field{sortinithash}{8de16967003c7207dae369d874f1456e}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Journal of the American Statistical Association}
\field{month}{6}
\field{number}{430}
\field{title}{Problems with Instrumental Variables Estimation When the Correlation Between the Instruments and the Endogenous Explanatory Variable Is Weak}
\field{volume}{90}
\field{year}{1995}
\field{pages}{443\bibrangedash 450}
\range{pages}{8}
\verb{doi}
\verb 10.1080/01621459.1995.10476536
\endverb
\endentry
\entry{card01lecture}{article}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=a9e88233ec6ce30d2d31e294961e3e76}{%
family={Card},
familyi={C\bibinitperiod},
given={David},
giveni={D\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{a9e88233ec6ce30d2d31e294961e3e76}
\strng{fullhash}{a9e88233ec6ce30d2d31e294961e3e76}
\strng{bibnamehash}{a9e88233ec6ce30d2d31e294961e3e76}
\strng{authorbibnamehash}{a9e88233ec6ce30d2d31e294961e3e76}
\strng{authornamehash}{a9e88233ec6ce30d2d31e294961e3e76}
\strng{authorfullhash}{a9e88233ec6ce30d2d31e294961e3e76}
\field{sortinit}{C}
\field{sortinithash}{4c244ceae61406cdc0cc2ce1cb1ff703}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Econometrica}
\field{month}{9}
\field{number}{5}
\field{title}{Estimating the Return to Schooling: Progress on Some Persistent Econometric Problems}
\field{volume}{69}
\field{year}{2001}
\field{pages}{1127\bibrangedash 1160}
\range{pages}{34}
\verb{doi}
\verb 10.1111/1468-0262.00237
\endverb
\endentry
\entry{feller68}{book}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=0dc1084389f176f601af7787031a22f1}{%
family={Feller},
familyi={F\bibinitperiod},
given={William},
giveni={W\bibinitperiod},
givenun=0}}%
}
\list{location}{1}{%
{New York, NY}%
}
\list{publisher}{1}{%
{Wiley}%
}
\strng{namehash}{0dc1084389f176f601af7787031a22f1}
\strng{fullhash}{0dc1084389f176f601af7787031a22f1}
\strng{bibnamehash}{0dc1084389f176f601af7787031a22f1}
\strng{authorbibnamehash}{0dc1084389f176f601af7787031a22f1}
\strng{authornamehash}{0dc1084389f176f601af7787031a22f1}
\strng{authorfullhash}{0dc1084389f176f601af7787031a22f1}
\field{sortinit}{F}
\field{sortinithash}{fb0c0faa89eb6abae8213bf60e6799ea}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{edition}{3}
\field{title}{An Introduction to Probability Theory and Its Application}
\field{volume}{1}
\field{year}{1968}
\endentry
\entry{HaRuWi96}{article}{}
\name{author}{3}{}{%
{{un=0,uniquepart=base,hash=cfdcf786df1126e7367e70af3572b572}{%
family={Hall},
familyi={H\bibinitperiod},
given={Alastair\bibnamedelima R.},
giveni={A\bibinitperiod\bibinitdelim R\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=cb141fc12492e5c288adce13d9da5468}{%
family={Rudebusch},
familyi={R\bibinitperiod},
given={Glenn\bibnamedelima D.},
giveni={G\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=34b2b43aa099d76f590fafeec4e312ee}{%
family={Wilcox},
familyi={W\bibinitperiod},
given={David\bibnamedelima W.},
giveni={D\bibinitperiod\bibinitdelim W\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{c14f66345e4f4701a212e26e54d1d90f}
\strng{fullhash}{2cafd72f8f96e94c2f36c9e29ba1005d}
\strng{bibnamehash}{2cafd72f8f96e94c2f36c9e29ba1005d}
\strng{authorbibnamehash}{2cafd72f8f96e94c2f36c9e29ba1005d}
\strng{authornamehash}{c14f66345e4f4701a212e26e54d1d90f}
\strng{authorfullhash}{2cafd72f8f96e94c2f36c9e29ba1005d}
\field{sortinit}{H}
\field{sortinithash}{6db6145dae8dc9e1271a8d556090b50a}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{International Economic Review}
\field{month}{5}
\field{number}{2}
\field{title}{Judging Instrument Relevance in Instrumental Variables Estimation}
\field{volume}{37}
\field{year}{1996}
\field{pages}{283}
\range{pages}{1}
\verb{doi}
\verb 10.2307/2527324
\endverb
\endentry
\entry{hanushek86}{article}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=ab739a255c8b5c754c14f9fa8eefafdf}{%
family={Hanushek},
familyi={H\bibinitperiod},
given={Eric\bibnamedelima A},
giveni={E\bibinitperiod\bibinitdelim A\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{ab739a255c8b5c754c14f9fa8eefafdf}
\strng{fullhash}{ab739a255c8b5c754c14f9fa8eefafdf}
\strng{bibnamehash}{ab739a255c8b5c754c14f9fa8eefafdf}
\strng{authorbibnamehash}{ab739a255c8b5c754c14f9fa8eefafdf}
\strng{authornamehash}{ab739a255c8b5c754c14f9fa8eefafdf}
\strng{authorfullhash}{ab739a255c8b5c754c14f9fa8eefafdf}
\field{sortinit}{H}
\field{sortinithash}{6db6145dae8dc9e1271a8d556090b50a}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Journal of Economic Literature}
\field{month}{9}
\field{number}{3}
\field{title}{The Economics of Schooling: Production and Efficiency in Public Schools}
\field{volume}{24}
\field{year}{1986}
\field{pages}{1141\bibrangedash 1177}
\range{pages}{37}
\verb{urlraw}
\verb https://www.jstor.org/stable/2725865
\endverb
\verb{url}
\verb https://www.jstor.org/stable/2725865
\endverb
\endentry
\entry{ImAn94}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=9f4fa8d00c0311b9cdad0f20ad429e57}{%
family={Imbens},
familyi={I\bibinitperiod},
given={Guido\bibnamedelima W.},
giveni={G\bibinitperiod\bibinitdelim W\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=ee326f34eea24952d04c41cef1771bfc}{%
family={Angrist},
familyi={A\bibinitperiod},
given={Joshua\bibnamedelima D.},
giveni={J\bibinitperiod\bibinitdelim D\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{f50c6335d6f592bc253fb31d4b8ee774}
\strng{fullhash}{f50c6335d6f592bc253fb31d4b8ee774}
\strng{bibnamehash}{f50c6335d6f592bc253fb31d4b8ee774}
\strng{authorbibnamehash}{f50c6335d6f592bc253fb31d4b8ee774}
\strng{authornamehash}{f50c6335d6f592bc253fb31d4b8ee774}
\strng{authorfullhash}{f50c6335d6f592bc253fb31d4b8ee774}
\field{sortinit}{I}
\field{sortinithash}{9417e9a1288a9371e2691d999083ed39}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Econometrica}
\field{month}{3}
\field{number}{2}
\field{title}{Identification and Estimation of Local Average Treatment Effects}
\field{volume}{62}
\field{year}{1994}
\field{pages}{467\bibrangedash 475}
\range{pages}{9}
\verb{doi}
\verb 10.2307/2951620
\endverb
\endentry
\entry{ImRu97complier}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=9f4fa8d00c0311b9cdad0f20ad429e57}{%
family={Imbens},
familyi={I\bibinitperiod},
given={Guido\bibnamedelima W.},
giveni={G\bibinitperiod\bibinitdelim W\bibinitperiod},
givenun=0}}%
{{un=1,uniquepart=given,hash=49a367b6280de8a239f8bbdd573f996f}{%
family={Rubin},
familyi={R\bibinitperiod},
given={Donald\bibnamedelima B.},
giveni={D\bibinitperiod\bibinitdelim B\bibinitperiod},
givenun=1}}%
}
\strng{namehash}{b14854f46d89a760d43091c58ecd121f}
\strng{fullhash}{b14854f46d89a760d43091c58ecd121f}
\strng{bibnamehash}{b14854f46d89a760d43091c58ecd121f}
\strng{authorbibnamehash}{b14854f46d89a760d43091c58ecd121f}
\strng{authornamehash}{b14854f46d89a760d43091c58ecd121f}
\strng{authorfullhash}{b14854f46d89a760d43091c58ecd121f}
\field{sortinit}{I}
\field{sortinithash}{9417e9a1288a9371e2691d999083ed39}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Review of Economic Studies}
\field{month}{10}
\field{number}{4}
\field{title}{Estimating Outcome Distributions for Compliers in Instrumental Variables Models}
\field{volume}{64}
\field{year}{1997}
\field{pages}{555\bibrangedash 574}
\range{pages}{20}
\verb{doi}
\verb 10.2307/2971731
\endverb
\endentry
\entry{KeNe21}{report}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=6c10c2a5a5db6618cebbdad9d5824748}{%
family={Keane},
familyi={K\bibinitperiod},
given={Michael},
giveni={M\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=c8f20e682983b4d4ccfb1ad599c835f5}{%
family={Neal},
familyi={N\bibinitperiod},
given={Timothy},
giveni={T\bibinitperiod},
givenun=0}}%
}
\list{institution}{1}{%
{SSRN}%
}
\strng{namehash}{5d8391efa3bb6a34e08e736384387a32}
\strng{fullhash}{5d8391efa3bb6a34e08e736384387a32}
\strng{bibnamehash}{5d8391efa3bb6a34e08e736384387a32}
\strng{authorbibnamehash}{5d8391efa3bb6a34e08e736384387a32}
\strng{authornamehash}{5d8391efa3bb6a34e08e736384387a32}
\strng{authorfullhash}{5d8391efa3bb6a34e08e736384387a32}
\field{sortinit}{K}
\field{sortinithash}{d3edc18d54b9438a72c24c925bfb38f4}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{month}{8}
\field{title}{A Practical Guide to Weak Instruments}
\field{type}{Working Paper}
\field{year}{2021}
\verb{doi}
\verb 10.2139/ssrn.3846841
\endverb
\endentry
\entry{kitagawa15}{article}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=2590fe940059a91685b684a80919d063}{%
family={Kitagawa},
familyi={K\bibinitperiod},
given={Toru},
giveni={T\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{2590fe940059a91685b684a80919d063}
\strng{fullhash}{2590fe940059a91685b684a80919d063}
\strng{bibnamehash}{2590fe940059a91685b684a80919d063}
\strng{authorbibnamehash}{2590fe940059a91685b684a80919d063}
\strng{authornamehash}{2590fe940059a91685b684a80919d063}
\strng{authorfullhash}{2590fe940059a91685b684a80919d063}
\field{sortinit}{K}
\field{sortinithash}{d3edc18d54b9438a72c24c925bfb38f4}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Econometrica}
\field{month}{9}
\field{number}{5}
\field{title}{A Test for Instrument Validity}
\field{volume}{83}
\field{year}{2015}
\field{pages}{2043\bibrangedash 2063}
\range{pages}{21}
\verb{doi}
\verb 10.3982/ECTA11974
\endverb
\endentry
\entry{krueger99}{article}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=34526c3bf548080921683822747fad7e}{%
family={Krueger},
familyi={K\bibinitperiod},
given={Alan\bibnamedelima B.},
giveni={A\bibinitperiod\bibinitdelim B\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{34526c3bf548080921683822747fad7e}
\strng{fullhash}{34526c3bf548080921683822747fad7e}
\strng{bibnamehash}{34526c3bf548080921683822747fad7e}
\strng{authorbibnamehash}{34526c3bf548080921683822747fad7e}
\strng{authornamehash}{34526c3bf548080921683822747fad7e}
\strng{authorfullhash}{34526c3bf548080921683822747fad7e}
\field{sortinit}{K}
\field{sortinithash}{d3edc18d54b9438a72c24c925bfb38f4}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{The Quarterly Journal of Economics}
\field{month}{5}
\field{number}{2}
\field{title}{Experimental Estimates of Education Production Functions}
\field{volume}{114}
\field{year}{1999}
\field{pages}{497\bibrangedash 532}
\range{pages}{36}
\verb{doi}
\verb 10.1162/003355399556052
\endverb
\endentry
\entry{lang93}{unpublished}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=ffc8200dad00583c74b36973725aadcb}{%
family={Lang},
familyi={L\bibinitperiod},
given={Kevin},
giveni={K\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{ffc8200dad00583c74b36973725aadcb}
\strng{fullhash}{ffc8200dad00583c74b36973725aadcb}
\strng{bibnamehash}{ffc8200dad00583c74b36973725aadcb}
\strng{authorbibnamehash}{ffc8200dad00583c74b36973725aadcb}
\strng{authornamehash}{ffc8200dad00583c74b36973725aadcb}
\strng{authorfullhash}{ffc8200dad00583c74b36973725aadcb}
\field{sortinit}{L}
\field{sortinithash}{dad3efd0836470093a7b4a7bb756eb8c}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{month}{3}
\field{note}{Unpublished manuscript, Boston University}
\field{title}{Ability Bias, Discount Rate Bias, and the Return to Education}
\field{year}{1993}
\endentry
\entry{lmmp20}{online}{}
\name{author}{4}{}{%
{{un=0,uniquepart=base,hash=95a248b56768f250923ee04ec474cf6c}{%
family={Lee},
familyi={L\bibinitperiod},
given={David\bibnamedelima S.},
giveni={D\bibinitperiod\bibinitdelim S\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=74747235a7c978a6735c95127004b4a4}{%
family={McCrary},
familyi={M\bibinitperiod},
given={Justin},
giveni={J\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=f3be8c74e3db1754509e79a188c19077}{%
family={Moreira},
familyi={M\bibinitperiod},
given={Marcelo\bibnamedelima J.},
giveni={M\bibinitperiod\bibinitdelim J\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=ba467abd4949a5168c1da3e1c06a060b}{%
family={Porter},
familyi={P\bibinitperiod},
given={Jack},
giveni={J\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{2f5293e54027ce2e810a77ad701e3940}
\strng{fullhash}{1a99d6b2ed95d6c8362a58f742a1560d}
\strng{bibnamehash}{2f5293e54027ce2e810a77ad701e3940}
\strng{authorbibnamehash}{2f5293e54027ce2e810a77ad701e3940}
\strng{authornamehash}{2f5293e54027ce2e810a77ad701e3940}
\strng{authorfullhash}{1a99d6b2ed95d6c8362a58f742a1560d}
\field{sortinit}{L}
\field{sortinithash}{dad3efd0836470093a7b4a7bb756eb8c}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{eprinttype}{arxiv}
\field{month}{10}
\field{title}{Valid T-Ratio {{Inference}} for {{IV}}}
\field{year}{2020}
\verb{eprint}
\verb 2010.05058
\endverb
\endentry
\entry{MiSu21}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=4f40e04107ec35e92de59ac29cefad50}{%
family={Mikusheva},
familyi={M\bibinitperiod},
given={Anna},
giveni={A\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=3011926b49b3aa7b3e8913ca855e42fc}{%
family={Sun},
familyi={S\bibinitperiod},
given={Liyang},
giveni={L\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{fb861c2a11bf5727dbc71493784d7cdc}
\strng{fullhash}{fb861c2a11bf5727dbc71493784d7cdc}
\strng{bibnamehash}{fb861c2a11bf5727dbc71493784d7cdc}
\strng{authorbibnamehash}{fb861c2a11bf5727dbc71493784d7cdc}
\strng{authornamehash}{fb861c2a11bf5727dbc71493784d7cdc}
\strng{authorfullhash}{fb861c2a11bf5727dbc71493784d7cdc}
\field{sortinit}{M}
\field{sortinithash}{2e5c2f51f7fa2d957f3206819bf86dc3}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Review of Economic Studies}
\field{title}{Inference with Many Weak Instruments}
\field{year}{2021}
\endentry
\entry{owen80}{article}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=a7ee17826784acae06da892c0d6974b8}{%
family={Owen},
familyi={O\bibinitperiod},
given={Donald\bibnamedelima B.},
giveni={D\bibinitperiod\bibinitdelim B\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{a7ee17826784acae06da892c0d6974b8}
\strng{fullhash}{a7ee17826784acae06da892c0d6974b8}
\strng{bibnamehash}{a7ee17826784acae06da892c0d6974b8}
\strng{authorbibnamehash}{a7ee17826784acae06da892c0d6974b8}
\strng{authornamehash}{a7ee17826784acae06da892c0d6974b8}
\strng{authorfullhash}{a7ee17826784acae06da892c0d6974b8}
\field{sortinit}{O}
\field{sortinithash}{ff8d4eeb5101e3cf3809959b3592d942}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Communications in Statistics - Simulation and Computation}
\field{month}{1}
\field{number}{4}
\field{title}{A Table of Normal Integrals: A Table}
\field{volume}{9}
\field{year}{1980}
\field{pages}{389\bibrangedash 419}
\range{pages}{31}
\verb{doi}
\verb 10.1080/03610918008812164
\endverb
\endentry
\entry{StSt97}{article}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=effff03798f6c2f47e9bbb095efbf892}{%
family={Staiger},
familyi={S\bibinitperiod},
given={Douglas},
giveni={D\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=2d01c3b46f4061430cc9eb2a1f6408a4}{%
family={Stock},
familyi={S\bibinitperiod},
given={James\bibnamedelima H.},
giveni={J\bibinitperiod\bibinitdelim H\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{acd85da75cbda2752eef64952064e29b}
\strng{fullhash}{acd85da75cbda2752eef64952064e29b}
\strng{bibnamehash}{acd85da75cbda2752eef64952064e29b}
\strng{authorbibnamehash}{acd85da75cbda2752eef64952064e29b}
\strng{authornamehash}{acd85da75cbda2752eef64952064e29b}
\strng{authorfullhash}{acd85da75cbda2752eef64952064e29b}
\field{sortinit}{S}
\field{sortinithash}{c319cff79d99c853d775f88277d4e45f}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{journaltitle}{Econometrica}
\field{month}{5}
\field{number}{3}
\field{title}{Instrumental Variables Regression with Weak Instruments}
\field{volume}{65}
\field{year}{1997}
\field{pages}{557\bibrangedash 586}
\range{pages}{30}
\verb{doi}
\verb 10.2307/2171753
\endverb
\endentry
\entry{StYo05}{inbook}{}
\name{author}{2}{}{%
{{un=0,uniquepart=base,hash=2d01c3b46f4061430cc9eb2a1f6408a4}{%
family={Stock},
familyi={S\bibinitperiod},
given={James\bibnamedelima H.},
giveni={J\bibinitperiod\bibinitdelim H\bibinitperiod},
givenun=0}}%
{{un=0,uniquepart=base,hash=e9a1907a19f18dcbfb0a2e56ce35939c}{%
family={Yogo},
familyi={Y\bibinitperiod},
given={Motohiro},
giveni={M\bibinitperiod},
givenun=0}}%
}
\name{editor}{2}{}{%
{{hash=819f147f61363909f87ab9d5fdf6c835}{%
family={Andrews},
familyi={A\bibinitperiod},
given={Donald\bibnamedelimb W.\bibnamedelimi K.},
giveni={D\bibinitperiod\bibinitdelim W\bibinitperiod\bibinitdelim K\bibinitperiod}}}%
{{hash=2d01c3b46f4061430cc9eb2a1f6408a4}{%
family={Stock},
familyi={S\bibinitperiod},
given={James\bibnamedelima H.},
giveni={J\bibinitperiod\bibinitdelim H\bibinitperiod}}}%
}
\list{location}{1}{%
{Cambridge, UK}%
}
\list{publisher}{1}{%
{Cambridge University Press}%
}
\strng{namehash}{d2afa551707a38446f1e5d11ef0e997b}
\strng{fullhash}{d2afa551707a38446f1e5d11ef0e997b}
\strng{bibnamehash}{d2afa551707a38446f1e5d11ef0e997b}
\strng{authorbibnamehash}{d2afa551707a38446f1e5d11ef0e997b}
\strng{authornamehash}{d2afa551707a38446f1e5d11ef0e997b}
\strng{authorfullhash}{d2afa551707a38446f1e5d11ef0e997b}
\strng{editorbibnamehash}{d9472e0cd64f94d5890340aad9855fb0}
\strng{editornamehash}{d9472e0cd64f94d5890340aad9855fb0}
\strng{editorfullhash}{d9472e0cd64f94d5890340aad9855fb0}
\field{sortinit}{S}
\field{sortinithash}{c319cff79d99c853d775f88277d4e45f}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{booktitle}{Identification and Inference for Econometric Models: Essays in Honor of {Thomas} {Rothenberg}}
\field{chapter}{5}
\field{title}{Testing for Weak Instruments in Linear {IV} Regression}
\field{year}{2005}
\field{pages}{80\bibrangedash 108}
\range{pages}{29}
\verb{doi}
\verb 10.1017/CBO9780511614491.006
\endverb
\endentry
\entry{young21iv}{unpublished}{}
\name{author}{1}{}{%
{{un=0,uniquepart=base,hash=c81085988f0d561d9ee3a775d3e0f22f}{%
family={Young},
familyi={Y\bibinitperiod},
given={Alwyn},
giveni={A\bibinitperiod},
givenun=0}}%
}
\strng{namehash}{c81085988f0d561d9ee3a775d3e0f22f}
\strng{fullhash}{c81085988f0d561d9ee3a775d3e0f22f}
\strng{bibnamehash}{c81085988f0d561d9ee3a775d3e0f22f}
\strng{authorbibnamehash}{c81085988f0d561d9ee3a775d3e0f22f}
\strng{authornamehash}{c81085988f0d561d9ee3a775d3e0f22f}
\strng{authorfullhash}{c81085988f0d561d9ee3a775d3e0f22f}
\field{sortinit}{Y}
\field{sortinithash}{6ae2ff5458ff901249c2745238c951b7}
\field{extradatescope}{labelyear}
\field{labeldatesource}{year}
\true{uniqueprimaryauthor}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{month}{6}
\field{note}{Unpublished working paper, LSE}
\field{title}{Leverage, Heteroskedasticity and Instrumental Variables in Practical Application}
\field{year}{2021}
\verb{urlraw}
\verb https://personal.lse.ac.uk/YoungA/LeverageandIV.pdf
\endverb
\verb{url}
\verb https://personal.lse.ac.uk/YoungA/LeverageandIV.pdf
\endverb
\endentry
\enddatalist
\endrefsection
\endinput