Survey data underlie most empirical work in economics, yet economists typically have little familiarity with survey sample design and its effects on inference. This paper describes how sample designs depart from the simple random sampling model implicit in most econometrics textbooks, points out where the effects of this departure are likely to be greatest, and describes the relationship between design-based estimators developed by survey statisticians and related econometric methods for regression. Its intent is to provide empirical economists with enough background in survey methods to make informed use of design-based estimators. It emphasizes surveys of households (the source of most public-use files), but also considers how surveys of businesses differ. Examples from the National Longitudinal Survey of Youth of 1979 and the Current Population Survey illustrate practical aspects of design-based estimation.
-
Estimation and Inference in Regression Discontinuity Designs with Clustered Sampling
August 2015
Working Paper Number:
carra-2015-06
Regression Discontinuity (RD) designs have become popular in empirical studies due to their attractive properties for estimating causal effects under transparent assumptions. Nonetheless, most popular procedures assume i.i.d. data, which is not reasonable in many common applications. To relax this assumption, we derive the properties of traditional non-parametric estimators in a setting that incorporates potential clustering at the level of the running variable, and propose an accompanying optimal-MSE bandwidth selection rule. Simulation results demonstrate that falsely assuming data are i.i.d. when selecting the bandwidth may lead to the choice of bandwidths that are too small relative to the optimal-MSE bandwidth. Last, we apply our procedure using person-level microdata that exhibits clustering at the census tract level to analyze the impact of the Low-Income Housing Tax Credit program on neighborhood characteristics and low-income housing supply.
View Full
Paper PDF
-
Who are the people in my neighborhood? The 'contextual fallacy' of measuring individual context with census geographies
February 2018
Working Paper Number:
CES-18-11
Scholars deploy census-based measures of neighborhood context throughout the social sciences and epidemiology. Decades of research confirm that variation in how individuals are aggregated into geographic units to create variables that control for social, economic or political contexts can dramatically alter analyses. While most researchers are aware of the problem, they have lacked the tools to determine its magnitude in the literature and in their own projects. By using confidential access to the complete 2010 U.S. Decennial Census, we are able to construct'for all persons in the US'individual-specific contexts, which we group according to the Census-assigned block, block group, and tract. We compare these individual-specific measures to the published statistics at each scale, and we then determine the magnitude of variation in context for an individual with respect to the published measures using a simple statistic, the standard deviation of individual context (SDIC). For three key measures (percent Black, percent Hispanic, and Entropy'a measure of ethno-racial diversity), we find that block-level Census statistics frequently do not capture the actual context of individuals within them. More problematic, we uncover systematic spatial patterns in the contextual variables at all three scales. Finally, we show that within-unit variation is greater in some parts of the country than in others. We publish county-level estimates of the SDIC statistics that enable scholars to assess whether mis-specification in context variables is likely to alter analytic findings when measured at any of the three common Census units.
View Full
Paper PDF
-
SYNTHETIC DATA FOR SMALL AREA ESTIMATION IN THE AMERICAN COMMUNITY SURVEY
April 2013
Working Paper Number:
CES-13-19
Small area estimates provide a critical source of information used to study local populations. Statistical agencies regularly collect data from small areas but are prevented from releasing detailed geographical identifiers in public-use data sets due to disclosure concerns. Alternative data dissemination methods used in practice include releasing summary/aggregate tables, suppressing detailed geographic information in public-use data sets, and accessing restricted data via Research Data Centers. This research examines an alternative method for disseminating microdata that contains more geographical details than are currently being released in public-use data files. Specifically, the method replaces the observed survey values with imputed, or synthetic, values simulated from a hierarchical Bayesian model. Confidentiality protection is enhanced because no actual values are released. The method is demonstrated using restricted data from the 2005-2009 American Community Survey. The analytic validity of the synthetic data is assessed by comparing small area estimates obtained from the synthetic data with those obtained from the observed data.
View Full
Paper PDF
-
Evaluating the Use of Commercial Data to Improve Survey Estimates of Property Taxes
August 2016
Working Paper Number:
carra-2016-06
While commercial data sources offer promise to statistical agencies for use in production of official statistics, challenges can arise as the data are not collected for statistical purposes. This paper evaluates the use of 2008-2010 property tax data from CoreLogic, Inc. (CoreLogic), aggregated from county and township governments from around the country, to improve 2010 American Community Survey (ACS) estimates of property tax amounts for single-family homes. Particularly, the research evaluates the potential to use CoreLogic to reduce respondent burden, to study survey response error and to improve adjustments for survey nonresponse. The research found that the coverage of the CoreLogic data varies between counties as does the correspondence between ACS and CoreLogic property taxes. This geographic variation implies that different approaches toward using CoreLogic are needed in different areas of the country. Further, large differences between CoreLogic and ACS property taxes in certain counties seem to be due to conceptual differences between what is collected in the two data sources. The research examines three counties, Clark County, NV, Philadelphia County, PA and St. Louis County, MO, and compares how estimates would change with different approaches using the CoreLogic data. Mean county property tax estimates are highly sensitive to whether ACS or CoreLogic data are used to construct estimates. Using CoreLogic data in imputation modeling for nonresponse adjustment of ACS estimates modestly improves the predictive power of imputation models, although estimates of county property taxes and property taxes by mortgage status are not very sensitive to the imputation method.
View Full
Paper PDF
-
Gradient Boosting to Address Statistical Problems Arising from Non-Linkage of Census Bureau Datasets
June 2024
Working Paper Number:
CES-24-27
This article introduces the twangRDC package, which contains functions to address non-linkage in US Census Bureau datasets. The Census Bureau's Person Identification Validation System facilitates data linkage by assigning unique person identifiers to federal, third party, decennial census, and survey data. Not all records in these datasets can be linked to the reference file and as such not all records will be assigned an identifier. This article is a tutorial for using the twangRDC to generate nonresponse weights to account for non-linkage of person records across US Census Bureau datasets.
View Full
Paper PDF
-
MISCLASSIFICATION IN BINARY CHOICE MODELS
May 2013
Working Paper Number:
CES-13-27
We derive the asymptotic bias from misclassification of the dependent variable in binary choice models. Measurement error is necessarily non-classical in this case, which leads to bias in linear and non-linear models even if only the dependent variable is mismeasured. A Monte Carlo study and an application to food stamp receipt show that the bias formulas are useful to analyze the sensitivity of substantive conclusions, to interpret biased coefficients and imply features of the estimates that are robust to misclassification. Using administrative records linked to survey data as validation data, we examine estimators that are consistent under misclassification. They can improve estimates if their assumptions hold, but can aggravate the problem if the assumptions are invalid. The estimators differ
in their robustness to such violations, which can be improved by incorporating additional information. We propose tests for the presence and nature of misclassification that can help to choose an estimator.
View Full
Paper PDF
-
The Need to Account for Complex Sampling Features when Analyzing Establishment Survey Data: An Illustration using the 2013 Business Research and Development and Innovation Survey (BRDIS)
January 2017
Working Paper Number:
CES-17-62
The importance of correctly accounting for complex sampling features when generating finite population inferences based on complex sample survey data sets has now been clearly established in a variety of fields, including those in both statistical and non statistical domains. Unfortunately, recent studies of analytic error have suggested that many secondary analysts of survey data do not ultimately account for these sampling features when analyzing their data, for a variety of possible reasons (e.g., poor documentation, or a data producer may not provide the information in a publicuse data set). The research in this area has focused exclusively on analyses of household survey data, and individual respondents. No research to date has considered how analysts are approaching the data collected in establishment surveys, and whether published articles advancing science based on analyses of establishment behaviors and outcomes are correctly accounting for complex sampling features. This article presents alternative analyses of real data from the 2013 Business Research and Development and Innovation Survey (BRDIS), and shows that a failure to account for the complex design features of the sample underlying these data can lead to substantial differences in inferences about the target population of establishments for the BRDIS.
View Full
Paper PDF
-
Public Use Microdata: Disclosure And Usefulness
September 1988
Working Paper Number:
CES-88-03
Official statistical agencies such as the Census Bureau and the Bureau of Labor Statistics collect enormous quantities of microdata in statistical surveys. These data are valuable for economic research and market and policy analysis. However, the data cannot be released to the public because of confidentiality commitments to individual respondents. These commitments, coupled with the strong research demand for microdata, have led the agencies to consider various proposals for releasing public use microdata. Most proposals for public use microdata call for the development of surrogate data that disguise the original data. Thus, they involve the addition of measurement errors to the data. In this paper, we examine disclosure issues and explore alternative masking methods for generating panels of useful economic microdata that can be released to researchers. While our analysis applies to all confidential microdata, applications using the Census Bureau's Longitudinal Research Data Base (LRD) are used for illustrative purposes throughout the discussion.
View Full
Paper PDF
-
Neighborhood Effects on High-School Drop-Out Rates and Teenage Childbearing: Tests for Non-Linearities, Race-Specific Effects, Interactions with Family Characteristics, and Endogenous Causation using Geocoded California Census Microdata
May 2008
Working Paper Number:
CES-08-12
This paper examines the relationship between neighborhood characteristics and the likelihood that a youth will drop out of high school or have a child during the teenage years. Using a dataset that is uniquely wellsuited to the study of neighborhood effects, the impact of the neighborhood poverty rate and the percentage of professionals in the local labor force on youth outcomes in California is examined. The first section of the paper tests for non-linearities in the relationship between indicators of neighborhood distress and youth outcomes. Some evidence is found for a break-point at low levels of poverty. Suggestive but inconclusive evidence is also found for a second breakpoint, at very high levels of poverty, for African-American youth only. The second part of the paper examines interactions between family background characteristics and neighborhood effects, and finds that White youth are most sensitive to neighborhood effects, while the effect of parental education depends on the neighborhood measure in question. Among White youth, those from single-parent households are more vulnerable to neighborhood conditions. The third section of the paper finds that for White youth and Hispanic youth, the relevant neighborhood variables appear to be the own-race poverty rates and the percentage of professionals of youths' own race. The final section of the paper estimates a tract-fixed effects model, using the results from the third section to define multiple relevant poverty rates within each tract. The fixed-effects specification suggests that for White and Hispanic youth in California, neighborhood effects remain significant, even with the inclusion of controls for any unobserved family and neighborhood characteristics that are constant within tracts.
View Full
Paper PDF
-
A METHOD OF CORRECTING FOR MISREPORTING APPLIED TO THE FOOD STAMP PROGRAM
May 2013
Working Paper Number:
CES-13-28
Survey misreporting is known to be pervasive and bias common statistical analyses. In this paper, I first use administrative data on SNAP receipt and amounts linked to American Community Survey data from New York State to show that survey data can misrepresent the program in important ways. For example, more than 1.4 billion dollars received are not reported in New York State alone. 46 percent of dollars received by house- holds with annual income above the poverty line are not reported in the survey data, while only 19 percent are missing below the poverty line. Standard corrections for measurement error cannot remove these biases. I then develop a method to obtain consistent estimates by combining parameter estimates from the linked data with publicly available data. This conditional density method recovers the correct estimates using public use data only, which solves the problem that access to linked administrative data is usually restricted. I examine the degree to which this approach can be used to extrapolate across time and geography, in order to solve the problem that validation data is often based on a convenience sample. I present evidence from within New York State that the extent of heterogeneity is small enough to make extrapolation work well across both time and geography. Extrapolation to the entire U.S. yields substantive differences to survey data and reduces deviations from official aggregates by a factor of 4 to 9 compared to survey aggregates.
View Full
Paper PDF