Jeffrey A. Dubin

Selection Bias in Linear Regression, Logit and Probit Models

Missing data are common in observational studies due to self-selection of subjects. Missing data can bias estimates of linear regression and related models. The nature of selection bias and econometric methods for correcting it are described. The econometric approach relies upon a specification of the selection mechanism. We extend this approach to binary logit and probit models and provide a simple test for selection bias in these models. An analysis of candidate preference in the 1984 U.S. presidential election illustrates the technique.

A Micro-Econometric Analysis of Risk-Aversion and the Decision to Self-Insure

This study estimates a von Neumann-Morgenstern utility function using market data and microeconometric methods. We investigate the decision whether to purchase insurance against the risk of telephone line trouble in the home. Using the choices of approximately 10,000 residential customers, we determine the shape of the utility function and the degree of risk aversion. We find that risk aversion varies systematically in the population and varies with the level of income and that the observed choice behavior is consistent with expected utility maximization.

Criminal Investigation Enforcement Activities and Taxpayer Noncompliance

This article tests empirically whether measurable activities of the IRS Criminal Investigation Division (CI) affect taxpayer compliance. The analysis is based on a state-level cross-section for the time period 1988 through 2001. First, it finds that CI activities have a measurable and significant effect on voluntary compliance. Second, it concludes that the mix of sentenced cases (for tax and money laundering violations) is not a significant determinant of tax compliance. Third, it finds that incarceration and probation (rather than fines) have the most influence on taxpayers. Simulations using the estimated models show that the direct effect of doubling the audit rate on assessed tax collections (reported amounts and additional taxes and penalties) is

Experimental estimates of the impact of wage subsidies

21.7 billion. Doubling CI tax and money laundering sentences is forecast to increase assessed collections by

Initial Virological and Immunologic Response to Highly Active Antiretroviral Therapy Predicts Long-Term Clinical Outcome

16.0 billion. It estimates the general deterrence or spillover effects from either audit or CI activities to be approximately 95 percent.

The Demand for Tax Return Preparation Services

The effects of a wage subsidy program on the duration of insured unemployment are investigated using experimental data. Participation in the experiment was voluntary and about one third of the subjects refused to take the subsidy voucher offered to them. Because subsidies appear to have stigmatic effects which tend to lower participation rates by high-skilled workers, experimental participants have longer average durations of unemployment than non-participants. However, correcting for self-selection, we find that wage subsidies can substantially increase a participants probability of reemployment. Subsidies are also compared to a search bonus proposal which is also cost effective, but, due to differences in participation patterns, has rather different effects.

Comparing absentee and precinct voters: A view over time

Little is known about the long-term clinical outcomes for human immunodeficiency virus (HIV)-infected patients who have received highly active antiretroviral therapy (HAART). Determining factors associated with long-term clinical outcomes early in the course of treatment may allow modifications to be made for patients who are at a greater risk of treatment failure. To evaluate these factors, we studied 213 HIV-infected patients who had received HAART for at least 115 weeks. In the univariate analysis, virological response, which was measured as the change in virus load from baseline at month 3 of treatment, was the single best predictor of clinical outcome (relative hazard, 0.722; P=.001), independent of virological suppression. In the multivariate analysis, virological response and immunologic response, which was measured as an increase in CD4 cell count of >200 cells/mm(3), resulted in better prediction of clinical outcomes than did use of either variable alone (P=.02). Our results indicate that changes in virus load and immunologic response together are good predictors of clinical outcome and can be assessed after the initiation of HAART, which would allow clinicians to identify patients early in the course of therapy who are at greater risk of negative outcome.

State Income Tax Amnesties: Causes

The authors analyze taxpayer choices of return preparation services. They distinguish between two types of nonpaid preparers, six types of paid third parties, and self-preparation. Among other things, the authors find significant differences in the factors that explain the demand for paid third parties who are and are not able to represent clients before the IRS. Among these factors are increases in IRS audit rates and the frequency of IRS penalties. Coauthors are Michael J. Graetz, Michael A. Udell, and Louis L. Wilde. Copyright 1992 by MIT Press.

The Distributional Effects of the Federal Energy Tax Act

This paper examines the trend in absentee voting over the last 30 years in California. With the liberalization of absentee voting laws and practices, an increase in the numbers of absentee voters quickly followed. Absentee voters have already demonstrated their ability to influence the outcomes of local elections. An open question is what will become of absentee voters in the future. If they are the model for “voting at home,” and if technological advances allow such, then the behavior of current absentee voters may be indicative of the future electorate. The increasing trend of voters opting for absentee ballots is analyzed by using GLS on a random-effects time-series cross-section model with county-level data. The focus is on identifying structural factors such as changing voter demographics that have influenced the decision of voters to cast absentee ballots. Thirty-three recent statewide elections in California are the basis for this analysis, covering the statewide primary and general elections from November 1962 through November 1994. We find that the impact of demographics and time trends on absentee voting differ between general and primary elections. In addition, we find that a 1977 liberalization law in California had the effect of accelerating the usage of the absentee format. Finally, we conclude that absentee and precinct voting are substitutes in general elections but complements in primary elections.

The Demand for Branded and Unbranded Products—An Econometric Method for Valuing Intangible Assets

This paper analyzes empirically for the years 1980-1988 the factors that led states with state income taxes to run tax amnesty programs. We find that the potential yield from an amnesty is more important than the fiscal status of a state. Furthermore, we estimate that if the 1RS audit rate had remained constant during the 1980–1988 period (instead of falling by almost one half), then the cumulative probability that an average state would have had a tax amnesty by 1988 would have fallen by just over 25 percent.