Chapter Four · failure evidence
What Fixed Effects Panel Methods got wrong, from 80 dissertations
The extracted records document common methodological challenges in fixed effects panel modeling, including negative weighting in staggered settings, incidental parameter bias in nonlinear models, and Nickell bias in dynamic panels. Researchers also frequently evaluate trade-offs between fixed and random effects specifications using Hausman tests while managing issues of variation absorption and spatial overparameterization. These records come from PhD theses at 26 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.
Standard two-way fixed effects estimators produce severe bias under staggered rollout and heterogeneous treatment effects
Standard two-way fixed effects specifications fail when treatments are adopted at different times because early-treated units act as controls for later-treated cohorts. These forbidden comparisons generate negative weights and cohort heterogeneity that distort effect estimates and cause sign reversals.
Tried and failed
Standard two-way fixed effects regression applied to Staggered treatment adoption panel data. Reason: Negative weighting from forbidden comparisons under treatment effect heterogeneity severely biased effect estimates
Birdie or Bogey? How Golf Course Construction Affects Surrounding Home Values · Harvard
Considered and rejected
Considered and rejected: Rejected standard two-way fixed effects (TWFE) and event study (TWFE-DiD) estimators due to attenuation bias towards zero from staggered rollouts comparing newly treated to already treated units
Essays on income inequality in the United States · UT Austin
Considered and rejected
Considered and rejected: Rejected using unrefined or standard two-way fixed effects estimators for main panel analyses due to bias introduced by repeated and staggered treatments.
Tried and failed
two-way fixed effects difference-in-differences estimation applied to staggered policy adoption evaluation. Reason: pervasive negative weights on early-treated cohorts and failure of treatment homogeneity assumptions
Tried and failed
two-way fixed effects with staggered rollouts applied to policy impact estimation across panel data. Reason: negative weighting issues inherent to staggered treatment timing yielded biased, counter-intuitive sign estimates
ESSAYS ON ENVIRONMENTAL POLICIES IN THE TRANSPORTATION SECTOR · Cornell
Lost to a baseline
Two-Way Fixed Effects (TWFE) estimator overestimated treatment effects compared to Callaway & Sant'Anna in the presence of dynamic treatment effects and staggered rollout.
Essays on education and environmental economics · UT Austin
Considered and rejected
Considered and rejected: Rejected standard two-way fixed effects (TWFE) difference-in-differences due to treatment effect heterogeneity across adoption cohorts and negative weighting issues.
Essays on applied labour economics: personality traits, artificial intelligence and formalisation policies · University of Nottingham Repository
Considered and rejected
Considered and rejected: Rejected standard two-way fixed-effects (TWFE) difference-in-differences models due to bias caused by dynamic/staggered treatment timing and negative weights comparing early-treated to late-treated units.
Three Essays On Economics of Marriage Law · Texas Tech
Considered and rejected
Considered and rejected: Rejected standard pooled Two-Way Fixed Effects (TWFE) due to bad comparisons and negative weights in staggered treatment settings.
Essays on the Impact and Effectiveness of Share Repurchase Regulations · Harvard
Considered and rejected
Considered and rejected: Rejected standard two-way fixed effects (TWFE) as primary causal estimator due to bias from negative weighting and dynamic treatment effect heterogeneity under staggered adoption.
Creator platform design · OpenBU
Considered and rejected
Considered and rejected: Rejected Two-Way Fixed Effects (TWFE) and dynamic TWFE models due to negative weighting biases and unreliability when comparing treated to already-treated units under staggered policy timing.
The Implications of Public Policies on Health Economics · unevada
Considered and rejected
Considered and rejected: Rejected using standard two-way fixed effects estimators for staggered policy timing due to known weighted average biases and negative weighting issues
Three Essays on Responsiveness to Higher Education Related Policies Addressing Inequality · Harvard
Considered and rejected
Considered and rejected: Rejected standard two-way fixed effects (TWFE) with staggered adoption in favor of stacked-by-event DiD due to negative weighting and dynamic treatment heterogeneity bias.
Considered and rejected
Considered and rejected: Standard Two-Way Fixed Effects (TWFE) event-study regression rejected due to dynamic treatment rollout, negative weighting, and cross-lag contamination issues under heterogeneous effects
THREE ESSAYS IN LABOR ECONOMICS · Cornell
Considered and rejected
Considered and rejected: Rejected relying primarily on two-way fixed effects (TWFE) / party and year fixed effects in main manifesto models due to interpretability issues and bias under treatment effect heterogeneity.
Dovish Reputation Theory: When Fighting To Demonstrate Resolve Backfires · Penn
Considered and rejected
Considered and rejected: Rejected relying solely on two-way fixed-effects OLS for staggered policy rollouts due to vulnerability to treatment effect heterogeneity and sign-reversals.
ESSAYS ON THE HUMAN CAPITAL AND OCCUPATIONAL CHOICES OF ADOLESCENTS · Cornell
Considered and rejected
Considered and rejected: Standard two-way fixed effects (TWFE) event study regressions for 3G rollout (rejected due to negative weighting and cohort treatment effect heterogeneity).
Considered and rejected
Considered and rejected: Rejected conventional two-way fixed effects (TWFE) for staggered solar policy adoption due to forbidden comparisons and negative weighting under heterogeneous effects.
Institutions and Operational Frictions in the Energy Transition · Georgia Tech
Considered and rejected
Considered and rejected: Rejected standard two-way fixed effects (TWFE) / Mean Treatment Effect DiD regressions as primary estimators due to severe bias under heterogeneous treatment effects and staggered timing.
Beyond Trade Losses: Sanctions and the Resilience of Productive Economies · Texas Tech
Considered and rejected
Considered and rejected: Rejected standard Two-Way Fixed Effects event study because leads/lags mechanically eliminate never-treated units and fail under treatment effect heterogeneity.
Technology, Violent Conflict, and the Determinants of Migration · Georgia Tech
Considered and rejected
Considered and rejected: Rejected standard Two-Way Fixed Effects (TWFE) with staggered treatment due to time-varying treatment effects leading to biased comparisons with already-treated units, adopting stacked difference-in-differences instead
Essays on infrastructure and urban development in developing countries · OpenBU
Considered and rejected
Considered and rejected: Standard two-way fixed effects (TWFE) model as primary causal estimator, rejected because it forces a common treatment effect across provinces and time when treatment effects vary
The causal effect of carbon pricing on industrial energy consumption in Canada · MSpace - University of Manitoba
Considered and rejected
Considered and rejected: Rejected standard two-way fixed-effects staggered DID due to bias from heterogeneous treatment effects across units/time and insufficient state-level sample sizes for date-specific cohorts.
Evaluation and implementation science methods for behavioral health interventions · JScholarship
Considered and rejected
Considered and rejected: Rejected relying solely on standard Two-Way Fixed Effects (TWFE) without verifying robustness across modern heterogeneous treatment effect DiD estimators due to forbidden comparison bias.
Random effects models are rejected in favor of fixed effects due to correlation with unobserved unit heterogeneity
Hausman specification tests consistently reject random effects models across panel datasets because explanatory variables correlate with unobserved time-invariant entity characteristics. Researchers select fixed effects estimators instead to prevent omitted variable bias and ensure consistent parameter estimation.
Lost to a baseline
Random Effects and Pooled OLS were rejected in favor of Fixed Effects for OECD regressions via Hausman tests (p = 0.0051, 0.0039, 0.0069).
Essays in international trade and finance · Texas Tech
Tried and failed
random effects panel regression applied to executive compensation and institutional assets. Reason: fails to control for unobserved entity-level characteristics correlated with both outcome and explanatory variables
Essays in Applied Microeconomics · Cornell
Considered and rejected
Considered and rejected: Random-effects model rejected in favor of fixed-effects panel regression based on the Hausman test.
Effect of income diversification on Canadian credit union performance. · HARVEST
Considered and rejected
Considered and rejected: Rejected random effects models in favor of fixed effects specifications because Hausman tests rejected the orthogonality assumption across all models.
Considered and rejected
Considered and rejected: Random effects models were rejected in favor of fixed effects to avoid omitted variable bias across panel state units based on Hausman test results.
The impact of the Affordable Care Act contraceptive mandate on fertility and abortion rates · JScholarship
Considered and rejected
Considered and rejected: Random effects modeling was considered and rejected based on a Hausman test (p < .001) in favor of fixed effects
Considered and rejected
Considered and rejected: Rejected Random-Effects (RE) models in favor of Fixed-Effects (FE) estimators due to Hausman test results indicating unobserved firm-level time-invariant heterogeneity.
An Investigation into the Constructive Role of Task-based Conflict within the Agency Theory · Publikationssystem UB Tuebingen
Considered and rejected
Considered and rejected: Rejected random effects models across all performance and fee specifications based on significant Hausman test results favoring fixed effects.
The Impact of Audit Committees on Audit Quality and Firm Performance - The Case of Jordanian Listed Companies · De Montfort Open Research Archive (DORA)
Considered and rejected
Considered and rejected: Rejected random-effects panel regression due to Hausman test confirming inconsistency versus fixed-effects estimator
Fragile Statehood and Military Aid (In)Effectiveness: An Assessment of United States Experiences in Lake Chad Basin · Carleton University Institutional Repository
Considered and rejected
Considered and rejected: Random effects models rejected in favor of fixed effects models based on Hausman test results (p < 0.001).
Pobreza y gasto público en México a nivel estatal, 2008-2016 · Repositorio Institucional BUAP
Considered and rejected
Considered and rejected: Random effects model rejected in favor of fixed effects regression for calculating asset-based income based on Hausman test results.
Dynamics in poverty and livelihood choices in rural Southeast Asia · Leibniz Universität Hannover Repository
Considered and rejected
Considered and rejected: Rejected random effects in favor of fixed effects panel specification based on Hausman test results.
Counterterrorist Financing and the Evolution and Adaptation of Terrorist Tactics · Carleton University Institutional Repository
Considered and rejected
Considered and rejected: Rejected random effects (RE) estimation in favor of fixed effects (FE) based on the Hausman specification test (p < 0.05).
Board Gender Diversity, Sustainability Governance and Corporate Sustainability Disclosure: Evidence from MENA Region · De Montfort Open Research Archive (DORA)
Considered and rejected
Considered and rejected: Rejected random-effects models in favor of fixed-effects models based on Hausman specification tests (p < 0.001).
Nonlinear panel models suffer from incidental parameter bias and computational non-convergence under fixed effects
Estimating fixed effects directly within nonlinear models like logit, probit, and stochastic frontier specifications produces persistent bias when panels have short time dimensions. The large number of individual parameters causes standard error underestimation, numerical non-convergence, or curse of dimensionality problems in optimization.
Considered and rejected
Considered and rejected: Rejected using Logit/Probit models with fixed effects due to potential inconsistency from incidental parameter problems, selecting OLS with clustered robust standard errors.
Experimental Approaches to Strategy and Innovation · Harvard
Considered and rejected
Considered and rejected: Rejected estimating a fixed-effects probit model with an ordered dependent variable using panel data due to large and persistent bias of the fixed-effects estimator.
Three essays on post-retirement labor · Texas Tech
Considered and rejected
Considered and rejected: Rejected using logistic regression with fixed effects for binary panel outcomes due to incidental parameter bias, selecting OLS linear probability models instead.
Considered and rejected
Considered and rejected: Rejected logistic regression in favor of linear probability regression due to biased estimation properties in the presence of fixed effects
Incorporating Equity into Healthcare Decision Making Around New Technologies · ResearchWorks
Considered and rejected
Considered and rejected: Rejected non-linear fixed effects models (e.g., fixed-effects Logit/Probit) for primary specifications due to the incidental parameter problem, preferring linear probability models and OLS.
ESSAYS ON THE HUMAN CAPITAL AND OCCUPATIONAL CHOICES OF ADOLESCENTS · Cornell
Considered and rejected
Considered and rejected: Rejected standard logistic regression for the primary state-level policy evaluation in favor of linear probability fixed-effects models, because logistic estimators underestimate standard errors under high-dimensional fixed effects.
Tried and failed
fixed effects logit regression applied to short panel discrete choice data. Outcome: no signal. Reason: Incidental parameters bias and severe overfitting in panels with short time dimensions.
Considered and rejected
Considered and rejected: Rejected standard panel fixed effects and maximum likelihood logit with fixed effects on the full dataset due to computational intractability and incidental parameters bias from high freelancer growth relative to panel length
Considered and rejected
Considered and rejected: Rejected True Fixed Effects Stochastic Frontier model due to severe incidental parameter bias when panel length is short (T < 10) relative to cross-sectional units.
Essays on vulnerability, finance and livelihoods of rural households in Southeast Asia · Leibniz Universität Hannover Repository
Considered and rejected
Considered and rejected: Decided against standard fixed effects SFA due to incidental parameter bias contaminating variance parameters in short panels.
Eficiencia de las sucursales bancarias: el caso de un banco comercial español · accedaCRIS
Considered and rejected
Considered and rejected: Rejected Greene (2005) True Fixed Effects (TFE) stochastic frontier model due to excessive parameters causing ML non-convergence on unbalanced panel data and incidental parameter bias.
Performance and risk in the Chinese regional ranking sector: a meta-frontier approach · University of Nottingham Repository
Tried and failed
estimating high-dimensional fixed effects in nonlinear GMM applied to random-coefficient discrete choice demand estimation. Outcome: unstable. Reason: severe curse of dimensionality and inaccurate parameter estimates from excessive parameters
ESSAYS ON INDUSTRIAL ORGANIZATION AND FIRM DYNAMICS · Cornell
Considered and rejected
Considered and rejected: Rejected standard county/regional fixed effects in main nonlinear dynamic ZOIB due to incidental parameters bias and bias in dynamic panels with lagged variables.
Fixed effects absorb cross-sectional variation and prevent the estimation of time-invariant or slow-moving variables
Saturating models with unit or granular spatial fixed effects absorbs the variation of key predictors that change slowly or remain constant across time. This lack of within-unit variation leads to imprecise or statistically insignificant coefficients and leaves models unable to identify historical or geographic effects.
Tried and failed
unit fixed effects in panel regression applied to labor turnover and mobility models. Outcome: no signal. Reason: Adding individual fixed effects absorbed the variation, making the estimated effect statistically indistinguishable from zero
Considered and rejected
Considered and rejected: Rejected unit fixed effects in pooled time-series cross-sectional models due to bias introduced when combining slow-moving covariates (GDP, regime type, fragility) with lagged dependent variables.
Offshoring Militarism: U.S. Military Aid and the Limits of American Foreign Policy · ResearchWorks
Considered and rejected
Considered and rejected: Rejected Fixed Effects regression for organizational panel data because it absorbs time-invariant predictors like follower counts and follower/following ratios
Tried and failed
fixed effect and first-difference panel regression applied to slowly changing or time-invariant panel variables. Outcome: no signal. Reason: lack of within-panel variation led to statistically insignificant or wrongly signed coefficient estimates
Human capital, decision-making and performance · Georgia Tech
Tried and failed
instrumental variable regression with fine spatial fixed-effects applied to estimating historical property rights effects. Outcome: no signal. Reason: historical instrument lacked sufficient spatial variation to survive fine geographic fixed effects
Tried and failed
subgroup-specific linear regression with fixed effects applied to environmental hedonic property valuation. Outcome: no signal. Reason: insufficient within-group variance in the primary explanatory variable
Examining Implicit Price Variation for Lake Water Quality · Virginia Tech
Considered and rejected
Considered and rejected: Rejected using year fixed effects interacted with subdivision-specific trends because recreational utility varied at the zip code level, leaving insufficient variation across repeat sales
Three essays on the economics of water pollution control · UT Austin
Considered and rejected
Considered and rejected: Rejected fixed effects probit models because they omit or fail to estimate effects of non-time-varying or slowly varying characteristics (e.g., age, income, wealth) unless categorized.
Three essays on aging, risk tolerance, financial literacy, and financial satisfaction · Texas Tech
Static fixed effects models suffer from dynamic endogeneity and Nickell bias in dynamic panel settings
Standard static fixed effects estimators cannot accommodate lagged dependent variables and persistent time series dynamics. In short panels, eliminating individual effects via transformations introduces severe Nickell downward bias and fails to account for dynamic productivity shocks.
Tried and failed
Static panel OLS and fixed effects estimation applied to translog cost function estimation. Outcome: worse than baseline. Reason: omits dynamic unobserved productivity shocks and endogeneity, causing severe downward bias in returns to scale
Understanding the costs of urban transportation using causal inference methods · Imperial
Considered and rejected
Considered and rejected: Rejected standard Ordinary Least Squares (OLS) and static fixed-effects models due to inability to handle dynamic endogeneity and omitted variable bias.
Corporate governance and firm performance: evidence from India. · Cranfield
Tried and failed
first-differencing to remove fixed effects applied to dynamic panel data estimation. Reason: introduces transformation bias from omitted dynamics unless the autoregressive parameter is exactly one
High-Dimensional Statistics for Causal Inference and Panel Data · MIT
Considered and rejected
Considered and rejected: Rejected OLS, Fixed Effects, and Difference-GMM estimators due to severe weak instrument bias and failure to address dynamic panel endogeneity in persistent growth time series.
Foreign Investment and Institutional Effectiveness in Resource-Dependent Economies · DSpace at SUNY Buffalo
Considered and rejected
Considered and rejected: Rejected static OLS/fixed-effects specifications as the primary model due to failure to accommodate dynamic lagged dependent variable inertia and Nickell bias.
Foreign Direct Investment Inflows, Financial Development and Political Risk: Evidence from Developing Countries · De Montfort Open Research Archive (DORA)
Tried and failed
static panel regression omitting lagged outcomes applied to panel data with dynamic effects. Reason: omitting past outcomes introduces severe dynamic bias, even under randomized treatment assignment
High-Dimensional Statistics for Causal Inference and Panel Data · MIT
Considered and rejected
Considered and rejected: Standard individual fixed effects estimation was rejected due to severe Nickell (1981) downward dynamic panel bias in short time horizons (T=3), using instead pre-treatment ability quintiles and IV/PEMU estimators.
Essays on Labor and Risk · Penn
High-dimensional and spatial fixed effects cause overparameterization and instability across geographic units
Adding fine spatial fixed effects or spatiotemporal trends inflates parameter counts and variance when data only exhibit simpler location heterogeneity. These overparameterized specifications reduce test power, cause intercept instability that flips coefficient signs, or leave findings vulnerable to specific subregions.
Considered and rejected
Considered and rejected: Rejected using two-way fixed effects (FElt) and location-level fixed effects with a trend (FEl,Trend) when data only contains location heterogeneity, due to inefficiency and inflated variance from overparameterization.
Evaluating Meta-Regression Models with Simulation Studies and Machine Learning Driven Imputation · Virginia Tech
Tried and failed
study fixed effects with linear time trend applied to meta-regression under spatio-temporal heterogeneity. Reason: produced consistently biased and inaccurate trend estimates when joint location-time or location-only heterogeneity exists
Evaluating Meta-Regression Models with Simulation Studies and Machine Learning Driven Imputation · Virginia Tech
Tried and failed
longitudinal fixed-effects ordinary least squares regression applied to spatiotemporal panel data across geographic units. Outcome: unstable. Reason: extreme intercept variability across spatial units produced unexpected sign reversal
Tried and failed
hedonic regression with spatial fixed effects and interactions applied to spatial implicit price heterogeneity estimation. Outcome: no signal. Reason: Joint hypothesis tests lacked power to detect administrative boundary heterogeneity despite true underlying regional variation
Examining Implicit Price Variation for Lake Water Quality · Virginia Tech
Tried and failed
Hedonic regression with geographic fixed effects applied to Environmental amenity property valuation. Outcome: no signal. Reason: Effect significance depended entirely on a single high-sample geographic subregion
Examining Implicit Price Variation for Lake Water Quality · Virginia Tech
Fixed effects difference-in-differences designs fail when parallel trends assumptions are violated
Event-study and difference-in-differences specifications with two-way fixed effects fail when pre-treatment trends diverge between treated and control groups. Statistically significant pre-treatment leads and failed placebo falsification tests indicate that underlying differential trends confound causal attribution.
Tried and failed
difference-in-differences with two-way fixed effects applied to evaluating policy effects on crime rates. Reason: violation of parallel trends assumption demonstrated by failed pre-trend placebo tests
Three essays on economics of crime and immigration · Texas Tech
Tried and failed
event-study difference-in-differences with high-dimensional fixed effects applied to quarterly policy impact on innovation counts. Outcome: no signal. Reason: violation of parallel trends assumptions and failure on placebo falsification tests
Tried and failed
two-way fixed effects event study design applied to estimating policy effects on charter enrollment. Reason: statistically significant pre-treatment leads violated parallel trends assumption
The Effects of Centralized School Assignment on Public School Enrollment · MIT
Tried and failed
two-stage least squares with two-way fixed effects applied to panel data with differential pre-trends. Reason: unobserved regional catch-up growth trends confounded the instrumental variable estimates
Fixed effects estimators are rejected in favor of random effects when Hausman tests support efficiency
When Hausman diagnostics confirm that random effects orthogonality assumptions hold, fixed effects estimators are rejected to preserve degrees of freedom. In such settings, fixed effects prove inefficient and fail to capture combined cross-sectional and temporal variation as accurately as random effects.
Tried and failed
fixed effects panel regression applied to spatial energy burden variation. Reason: Hausman test indicated inability to accurately capture cross-sectional and temporal variations compared to random effects
Considered and rejected
Considered and rejected: Fixed effects estimation rejected in favour of random effects based on Hausman test diagnostics.
The determinants of intrafamily ideological differentiation: Western European social democracy between 1990 and 2019 · IRIS - LUISS - prod
Considered and rejected
Considered and rejected: Fixed effects model rejected in favor of random effects model based on the Hausman test for panel analysis.
Considered and rejected
Considered and rejected: Rejected standard fixed-effects panel modeling after Hausman tests indicated random effects estimators were appropriate and to avoid consuming degrees of freedom with time-invariant factors.
Does CEO Compensation Encourage Risk-Taking? Empirical Evidence from FTSE350. · De Montfort Open Research Archive (DORA)
Left open by the authors
Problems the authors named and did not get to.
Left open
Incorporate additional fixed-effects specifications into the event-study difference-in-differences regression model of clean energy patent output. Blocker: Unclear which specific additional fixed effects or control variables are intended to improve causality
Left open
Incorporate additional fixed effects into the event-study difference-in-differences regression model for clean energy patent output. Blocker: Unspecified which additional fixed effects or specifications are needed
Left open
Estimate patent citation network regressions with lagged dependent variables using autoregression estimators instead of fixed effects estimators. Blocker: None
Mean field games: theory, approximations, and applications · Imperial
Left open
Analyze 2SLS identification and estimation rates under increasing-domain asymptotics with community fixed effects eliminated via differencing. Blocker: None
Weak Identification and Network Measurement Error in Peer Effects Estimation · MIT
Left open
Estimate two-way fixed effects regressions comparing the impact of foreign aid on life expectancy and death rates versus immunization rates. Blocker: None
Left open
Evaluate the effect of lagged hospital financial metrics on alternative patient outcome metrics using fixed effects regression. Blocker: No specific alternative patient metrics are defined and hospital datasets typically require restricted access agreements.
How Lagging Financial Metrics Affect Next Year Hospital Patient Metrics · Penn
Left open
Reformulate the hospital panel regression models to eliminate linear dependencies among the time fixed effect dummy variables. Blocker: Requires access to the specific hospital financial and patient metrics dataset used in the thesis
How Lagging Financial Metrics Affect Next Year Hospital Patient Metrics · Penn
Left open
Estimate whether public and private welfare expansions act as complements or substitutes under heightened electoral competition using fixed effects panel regressions. Blocker: None
When Voice Leads to Exit: Democracy, Development, and Private Provision · MIT
Left open
Extend the directed network jackknife bias correction method to multi-period panel networks with time fixed effects and dynamic splitting. Blocker: None
Essays in Econometrics · MIT
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.