Econometrics of Mismeasured Data

Paper Session

Saturday, Jan. 7, 2017 1:00 PM – 3:00 PM

Hyatt Regency Chicago, Toronto
Hosted By: American Economic Association
  • Chair: Rusty Tchernis, Georgia State University

What Leads to Measurement Error? Evidence from Reports of Program Participation in Three Surveys

Pablo Celhay
,
University of Chicago
Bruce D. Meyer
,
University of Chicago and NBER
Nikolas Mittag
,
CERGE-EI

Abstract

Measurement error is often a large source of bias in survey data. Lack of knowledge of the determinants of such errors makes it difficult for data producers to reduce the extent of errors and for data users to assess the validity of analyses using the data. We study the causes of survey error using high quality administrative data on government transfers linked to three major U.S. surveys. The differences between survey and administrative records show that up to six out of ten cash welfare recipients are missed by the surveys. We find that misreporting by respondents, survey design features, and imputation of missing data induce substantial error. Our results for non-imputed respondents confirm several theories of misreporting, e.g. that errors are related to event recall, forward and backward telescoping, salience of receipt, respondent’s degree of cooperation, and the stigma of reporting participation in welfare programs. Our results provide guidance on the conditions under which survey data are likely to be accurate and suggest different ways to reduce survey errors.

On the Estimation of Average Treatment Effects With Endogenous Misreporting

Pierre Nguimkeu
,
Georgia State University
Augustine Denteh
,
Georgia State University
Rusty Tchernis
,
Georgia State University

Abstract

Participation in social programs is often misreported in survey data, complicating the estimation of the effects of those programs. In this paper we propose a model to estimate treatment effect under endogenous participation and endogenous misreporting. We show that failure to account for endogenous misreporting can result in the estimates of the treatment effect having opposite sign from the true effect. We present an expression for the asymptotic bias of both OLS and IV estimators and discuss the conditions under which sign reversal may occur. We provide a method of eliminating this bias when researchers have access to information related to both participation and misreporting. We establish the consistency and asymptotic normality of our estimator and present its small sample performance through Monte Carlo simulations.

Are Proxy Earnings Reports Reliable? Evidence From the Current Population Survey

Christopher Bollinger
,
University of Kentucky
Barry Hirsch
,
Georgia State University
Charles Hokayem
,
Centre College
James Ziliak
,
University of Kentucky

Abstract

The Current Population Survey (CPS) typically relies on a single householder to provide responses for all household members. Roughly half of all earnings reports are from these “proxy” respondents, making their reliability a question of some importance. Cross-section estimates of proxy effects on reported earnings suggest a modest negative effect of about 2%, but these results mask what are large differences for non-spousal versus spousal proxy reports and, to a lesser degree, among wives’ reports for husbands and husbands’ reports for wives. Our research examines the accuracy of proxy reports using two types of data and approaches. Longitudinal evidence from CPS public use files identify proxy reporting effects based on workers who self-report in one year and have a proxy report in a second year. Using internal Current Population Survey (CPS ASEC) records matched to administrative data on earnings, we replicate CPS cross-section results using the administrative earnings data regressed on the proxy status from the survey. If the coefficient on proxy reflects response error in the CPS, there should be no correlation between proxy and earnings in the administrative data. Both CPS panel data and administrative data lead to the conclusion that proxy response is correlated with earnings, conditioning on covariates, but this correlation largely reflects unmeasured worker heterogeneity and not misreporting of earnings. Broadly speaking, proxy earnings reports are reliable, the exception being modest underreporting of husbands’ earnings by their wives.
Discussant(s)
Charlie Brown
,
University of Michigan
Kei Hirano
,
University of Arizona
Dan Black
,
University of Chicago
JEL Classifications
  • C1 - Econometric and Statistical Methods and Methodology: General
  • C8 - Data Collection and Data Estimation Methodology; Computer Programs