Chapter Four · failure evidence
What Instrumental Variable Estimation got wrong, from 47 dissertations
The records evaluate applications and methodological limits of instrumental variable estimation across diverse empirical settings. Researchers frequently encountered weak instruments, violations of the exclusion restriction, absent instruments in observational data, and estimation instability under misspecified models. These records come from PhD theses at 14 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Instruments suffered from weak first stages and low statistical power
Multiple studies reported that instruments failed to strongly predict endogenous variables, often yielding first-stage F-statistics below standard thresholds. As a consequence, instrumental variable estimates lost statistical precision, produced biased results, or resulted in negative R-squared values.
Tried and failed
instrumental variables with lagged or spatial instruments applied to endogenous peer composition effects. Outcome: no signal. Reason: instruments were weak, producing imprecise estimates compared to ordinary least squares
Tried and failed
two-stage least squares instrumental variables applied to linear-in-means social peer effects estimation. Outcome: no signal. Reason: Weak instruments caused estimates to lose statistical significance and precision.
Peer Effects and Human Capital Accumulation · Georgia Tech
Tried and failed
Two-stage least squares instrumental variable regression applied to text sentiment effect on continuous response. Outcome: worse than baseline. Reason: Severe model misspecification or weak instruments leading to highly negative R-squared
Tried and failed
two-stage least squares instrumental variable estimation applied to neighborhood-level crime shock analysis. Outcome: no signal. Reason: The instrument had a weak first stage, failing to sufficiently predict the endogenous variable.
Tried and failed
instrumental variable estimation with regional instruments applied to time-varying technical inefficiency models. Outcome: no signal. Reason: instruments were weak with first-stage F-statistic below threshold and failed to establish endogeneity
Tried and failed
two-stage least squares instrumental variables applied to linear peer effects on clique networks. Outcome: unstable. Reason: weak instruments violate identification relevance in disjoint clique topologies
Weak Identification and Network Measurement Error in Peer Effects Estimation · MIT
Tried and failed
unweighted instrumental variable estimation applied to subsample causal inference. Outcome: no signal. Reason: weak instruments with first-stage F-statistic below 10 caused biased and insignificant estimates
Three essays on economics of crime and immigration · Texas Tech
Tried and failed
historical asset value ratio as instrumental variable applied to agricultural finance econometric estimation. Outcome: no signal. Reason: the instrument was weak with a statistically insignificant first stage relationship
Tried and failed
instrumental variable with assignment probability instruments applied to estimating health plan utilization and costs. Outcome: no signal. Reason: instrument lacked sufficient statistical power across subgroups to detect significant differences
Medicaid in ACA Marketplaces : the welfare impact of New Hampshire’s Medicaid experiment · UT Austin
Considered and rejected
Considered and rejected: Rejected difference GMM due to vulnerability to weak instrumental variable problems.
Essays on Residential Segregation and the Spatial Interaction of Housing and Labor Markets · DSpace at SUNY Buffalo
Considered and rejected
Considered and rejected: Instrumental variable (IV) approach using private company hacks as an instrument for county IT spending ratio to explain bond yield spreads due to being underpowered
Considered and rejected
Considered and rejected: Handpicking instrumental variables or naively using the full set of 1,236 network instruments was rejected in favor of square-root Lasso selection to prevent weak first-stage identification.
Exclusion restrictions were violated by direct effects or unobserved confounding
Several analyses failed or were rejected because candidate instruments were correlated with measurement errors, advising channels, or omitted geographic and historical confounders. These direct relationships with outcomes violated the core exclusion restriction and led to implausible non-zero effect estimates in placebo tests.
Tried and failed
two-stage least squares with dichotomized noisy proxy applied to causal compliance effect estimation. Reason: measurement error correlated with the instrumental variable, inducing bias and inconsistency in effect estimates
Considered and rejected
Considered and rejected: Rejected simultaneous equation / instrumental variable causal inference modeling due to the lack of a defensible exclusion restriction.
Rising powers, subordinate monopolization, and major interstate war · Oxford
Tried and failed
instrumental variable selection using exclusion restriction applied to linear outcome equation in selection model. Reason: omitted historical and geographic confounders violated the exclusion restriction
Tried and failed
instrumental variable regression with placebo test applied to health facility access causal estimation. Reason: instrument violated exclusion restriction yielding implausible non-zero effect on predetermined adult height
Essays on maternal and child health in India · UT Austin
Considered and rejected
Considered and rejected: Rejected Instrumental Variables (IV) approach using universal FAFSA policy adoption as an instrument for college enrollment because counselor advising violates the exclusion restriction.
Essays in the Economics of Education · Harvard
Considered and rejected
Considered and rejected: Rejected conventional Instrumental Variables (LATE) for persuasive ad experiments due to implausibility of the exclusion restriction.
Messy Measurement: Approaches to Causal Inference With Unobserved Variables · MIT
Considered and rejected
Considered and rejected: Standard instrumental variable (IV) method rejected for lack of suitable exclusion restriction instruments, opting instead for identification by functional form and non-parametric bounding
Essays on food security in the United States · Texas Tech
Researchers lacked plausible and valid candidate instruments in the data
Empirical projects rejected instrumental variable strategies because defensible instruments were absent, unobserved, or affected by severe missing data. In several settings, authors could not locate exogenous assignment mechanisms or indisputable links to endogenous behaviors.
Considered and rejected
Considered and rejected: Rejected instrumental variable (IV) approach due to difficulty finding a valid instrument indisputably linking women's micro-business participation to poverty outcomes without failing significance tests.
An Analysis of Women-Owned Micro Business Performance and the Effect on Poverty in the Wolaita Zone, Southern Ethiopia · Research Repository UCD
Considered and rejected
Considered and rejected: Instrumental variables were rejected because potential instruments that satisfied exogeneity either had substantial missing data or were absent
Considered and rejected
Considered and rejected: Rejected the instrumental variables (IV) approach due to the non-availability/limitations of finding valid instruments for endogenous director compensation and shareholding variables.
DIRECTORS’ COMPENSATION & FINANCIAL STATEMENT FRAUD: A COMPARATIVE STUDY OF CHINA AND THE US · University of Nottingham Repository
Considered and rejected
Considered and rejected: Rejected using instrumental variables to address selection into divorce/separation because no plausible instrument was available in the data.
Considered and rejected
Considered and rejected: Rejected regression-based and instrumental variable approaches for FoodAPS analysis due to lack of valid instruments and reliance on extrapolation/model-based assumptions.
Healthy Food Access and Consumption: Informing Interventions Through Analytics · MIT
Tried and failed
instrumental variable and regression discontinuity designs applied to estimating infection exposure effects. Outcome: data insufficient. Reason: Lack of valid instruments and threshold assignment mechanisms in the observational data
Considered and rejected
Considered and rejected: Decided against Instrumental Variable and Regression Discontinuity designs for Covid-19 primary care analysis due to absent valid instruments and assignment rules.
Estimators failed due to model misspecification and restrictive structural assumptions
Standard linear two-stage least squares produced unbounded and inefficient estimates when applied to binary outcomes, while joint monotonicity assumptions resulted in empty identified sets. Additionally, incorporating principal component factors introduced bias toward two-stage least squares, and deep instrumental variable methods were outperformed by polynomial baselines.
Tried and failed
partial identification using monotone instrumental variables applied to estimating returns to education. Reason: joint monotonicity and instrumental variable assumptions produced empty identified sets refuted by the empirical data
Three essays on applied econometric methods · Iowa State
Tried and failed
Linear two-stage least squares instrumental variables applied to binary endogenous variables with binary outcomes. Outcome: unstable. Reason: Produced unbounded and inefficient estimates when modelling binary treatment and binary outcome relationships
Considered and rejected
Considered and rejected: Rejected two-stage least squares (TSLS) with instrumental variables in favor of OLS based on Hausman tests showing OLS was more consistent.
Are We Done Fighting Traffic? Planning Congestion Resilient Regions · Penn
Considered and rejected
Considered and rejected: Rejected ordinary least squares (OLS) regression for the first stage of the instrumental variable model in favor of a binomial regression with logit link to ensure non-negative predicted online shopping percentages
Essays on Sustainability in Agriculture and Food Systems · MIT
Tried and failed
direct PCA factor projection applied to instrumental variable panel data. Reason: estimator bias increases toward standard two-stage least squares as more factor components are included
Tried and failed
iterative PCA factor structure estimation applied to panel instrumental variables causal estimation. Outcome: worse than baseline. Reason: increasing the estimated factor count introduced bias towards standard two-stage least squares
Lost to a baseline
DeepIV was consistently outperformed by Poly2SLS across low-dimensional instrumental variable regression benchmarks.
Adversarial Machine Learning Methods for Causal Inference under Unmeasured Confounding · Cornell
Estimates exhibited severe instability and sensitivity to measurement error and specification
Instrumental variable models proved sensitive to instrument choice, sample composition, and controls for baseline outcome rates. In network and panel contexts, unmodeled measurement error and missing data induced non-vanishing asymptotic bias and frequent sign reversals in parameter estimates.
Tried and failed
ordinary least squares and instrumental variable regression applied to data with measurement error and missingness. Outcome: unstable. Reason: unmodeled measurement error and missing data caused severe bias and frequent sign reversals in parameter estimates
Essays on Econometrics, Causal Inference, and Machine Learning · MIT
Tried and failed
shift-share instrumental variables applied to reduced-form discrete choice regression. Outcome: unstable. Reason: estimates were sensitive to instrument choice and composition of years and industries selected
Essays in Urban Economics · Harvard
Tried and failed
instrumental variable estimation applied to childcare labor employment responses. Outcome: unstable. Reason: empirical results failed standard sensitivity and robustness checks
Tried and failed
two-stage least squares instrumental variable regression applied to estimating neighborhood collective action spillover effects. Outcome: did not generalise. Reason: estimates were not robust to controlling for baseline outcome rates
Managing the "Water Pressure": The Political Economy of Urban Service Provision · Harvard
Tried and failed
two-stage least squares instrumental variables estimation applied to dense network peer effect models. Reason: simultaneous weak identification and graph measurement error cause non-vanishing asymptotic bias
Weak Identification and Network Measurement Error in Peer Effects Estimation · MIT
Tried and failed
alternative instrumental variable specification applied to estimating debt effect on earnings. Outcome: no signal. Reason: the estimated causal effect lost statistical significance and diminished towards zero under the alternative instrument
Essays on student debt, unemployment, and labor market outcomes · UT Austin
Left open by the authors
Problems the authors named and did not get to.
Left open
Identify an exogenous instrument for labor market participation to estimate earnings impacts of teen childbearing via MTE. Blocker: Requires discovering a novel, valid instrumental variable for endogenous labor market selection
ESSAYS ON THE INTERGENERATIONAL TRANSMISSION OF HUMAN CAPITAL AND SOCIOECONOMIC STATUS · Cornell
Left open
Estimate the causal effect of education on lifetime completed fertility using instrumental variables on older women census or survey cohorts. Blocker: None
Three essays on factors influencing changes in labor and marriage markets · Iowa State
Left open
Replicate promise program expectation findings using a longer longitudinal panel or instrumental variables strategy to formally test parallel trends. Blocker: Requires access to longitudinal education survey or administrative data tracking promise program eligibility and expectations.
Three Essays In Economics, Education Policy, And Inequality · Penn
Left open
Develop an instrumental variable strategy to resolve endogeneity and reverse causality in micro-neighborhood ethnic demography spatial econometric models. Blocker: Finding a theoretically valid, exogenous instrument for micro-neighborhood level ethnic residential sorting in Kenya is an open econometric identification challenge.
Safety in Separation? Ethnic Demography and Violence in Kenya · Cornell
Left open
Perform an instrumental variables analysis instrumenting the decision to share news diet summaries with disclosure treatment status using experiment data. Blocker: Requires private field experiment data collected on Twitter/X by the thesis author.
Left open
Estimate heterogeneous treatment effects of inpatient physician consults across patient, provider, and setting subgroups using quasi-experimental instrumental variable methods. Blocker: Requires private hospital electronic health record or claims data used in the thesis.
Essays on Physician Consults · Penn
Left open
Develop robust statistical inference methods for arbitrarily weak and invalid instrumental variables with high-dimensional covariates. Blocker: Lacks a specific mathematical approach or formulation for handling arbitrary weakness alongside invalidity
Statistical Inference For High-Dimensional Linear Models · Penn
Left open
Construct an exogenous instrumental variable for investor-state dispute filing rates to improve the first-stage F-statistic above weak instrument thresholds. Blocker: Requires designing a valid economic identification strategy and theoretical instrument for dispute filings.
Left open
Disentangle the causal health impacts of concurrent pollutants like O3, SO2, NO2, and CO from PM2.5 using distributed lag instrumental variable models. Blocker: Lack of a specified econometric identification strategy or valid instrumental variables for multiple simultaneous pollutants.
Left open
Implement a delta-method standard error and a fully joint two-stage difference-in-differences instrumental variable (2SDD-IV) event-study estimator. Blocker: None
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.