James X. Sullivan, Northwestern University
Bruce D. Meyer, Dissertation Advisor

2002 Honorable Mention
Julie A. Kmec, University of Pennsylvania
Paula England, Dissertation Advisor

2002 Honorable Mentions
Yu-Che Chang, Indiana University
Joyce Man, Dissertation Advisor

Elizabeth A. Hoffmann, University of Wisconsin, Madison
Mark C. Suchman, Dissertation Advisor

In this dissertation I have studied the empirical content of the Butler-Heckman thesis. Using U.S. decennial census data from 1940 to 1990, I demonstrate that studies which have made inferences based on the CPS have excluded the institutionalized and incarcerated populations and thereby dramatically understated the extent of black nonemployment.

By identifying the distribution of offer wages to blacks and whites as the distribution of interest in assessing black economic progress, I discuss the economic content of alternative identifying assumptions used to recover this latent distribution.

Skill prices are estimated for nonworkers through a variety of nonparametric, semi-parametric, and parametric estimators, and the empirical content Butler-Heckman thesis is evaluated under each set of alternative estimates.

I find that matching models and their variants are inappropriate for the study of the racial wage gap as they ignore the role of unobservable characteristics in determining wages and participation. Using flexible functional forms, I find support for the traditional index sufficiency characterization of selection bias. However, I also document the extreme sensitivity of control-function estimators to the estimation samples.

Contrary to previous studies that have found little or no support for the selective-withdrawal hypothesis, my estimates indicate that a significant portion of the convergence in wages from 1940 to 1990 can be explained by ignoring the selective withdrawal of nonparticipants (over 30 percent from the period 1950-1990 and over 40 percent of the 1970-1990 convergence). Similar results are obtained for an within-cohort analysis.

Additionally, preliminary estimates find that in 1950-1970, increasing reservation wages were responsible for the withdrawal of young blacks from the labor force; however, falling skill prices explain much of the subsequent withdrawal. I find that since 1980, black males are not working because of declining skill prices. For this period, I do not find evidence for the hypothesis that rising reservation wages are causing black males to withdraw from the labor force. When this finding is combined with the results of Neal and Johnson (1996), the role of public policies that affect relative wages becomes clearer. Because less-skilled black men are not working because of low offer wages, such interventions should be made early in life and should be directed at closing the skill gap that exists between blacks and whites before they enter the labor market.

Young men today are significantly worse off than their counterparts who entered the labor market in the late 1960s. Starting wages for young men declined by nearly 10 percent between the late 1960s and late 1980s. Even college graduates experienced some decline in starting wages.

More importantly, the evidence suggests that most workers did not make up for the lost starting wages with higher wage growth. On average, young men entering the labor market during the late 1970s and 1980s experienced 15-20 percent less wage growth over the first 10 years of their careers than did young men who started working in the late 1960s.

Furthermore, this overall decline was accompanied by other changes in the distribution of wage growth that appear to have exacerbated lifetime inequality between young men with different levels of education. Among cohorts that started working in the late 1960s, workers with a high school education or less experienced 68 percent wage growth over their first 10 years, while college graduates experienced 44 percent wage growth, indicating that returns to experience ameliorated inequality in starting wages among early cohorts.

By the early 1980s, however, wage growth among college graduates had increased to between 50 and 60 percent, while wage growth among those with a high school degree or less had declined to between 40 and 45 percent. Thus, among more recent cohorts, changes in the returns to experience reinforce wage disparities across education groups.

The decline in wage growth among young men has important repercussions. Wage growth is one of the primary determinants of lifetime income, and thus these trends have the potential to impact life cycle decisions such as educational attainment, marriage, and home ownership. Moreover, lifetime earnings inequality reflects permanent differences in access to resources across individuals.

Analysis of cross-sectional earnings, the focus of many recent studies, cannot provide information on this topic; in any given year, the distribution of wages may be largely driven by transitory shocks. Thus, by focusing on the structure of wage levels in the cross section, previous studies have ignored an important economic trend of the last 30 years. Because wage growth (either by itself or through its impact on lifetime earnings) affects so many individual consumption and investment decisions, understanding these trends is particularly important for public policy.

Many public programs-ranging from the Earned Income Tax Credit to job placement services-focus on increasing employment and current earnings, while ignoring long-term considerations such as job mobility and wage growth. In doing so, these programs neglect an important incentive for individual behavior.

My dissertation documents the trends in wage growth and explores the implications of the observed changes, as well as possible factors that have contributed to the trends. The first chapter examines whether the decline in employment among low-skilled young men since the late 1960s can be explained by the decline in wage growth, which has reduced the value of work considerably. The second chapter details the increases in lifetime earnings inequality over the past 30 years and shows that they have been at least as great as the much more widely studied increases in cross-sectional inequality. Finally, the third chapter investigates the extent to which skill-biased technical change has been responsible for the observed changes in the structure of wage growth.

Govert Bijwaard, Free University, Amsterdam
Geert Ridder, Dissertation Advisor

Recently, social experiments have gained popularity as a method for evaluating social and labor market programs. High-profile evaluations such as the National JTPA Study in the United States and the Self-Sufficiency Project in Canada have brought about real changes in the views and the actions of policymakers. Most of the literature relates to these two countries, in which there is a long-standing tradition of evaluating labor market programs.

In this study, we also use data collected from an experiment in the United States. Indeed, in the United States, there is a requirement for public authorities to evaluate their programs. Few European countries have carried out rigorous evaluations; the Netherlands are not an exception. There, the most common method of "evaluation" consists of simply monitoring the labor market status and earnings of recipients for a brief period following their participation in the program.

Although this kind of exercise provides useful information, it cannot answer the vital question as to whether the program fulfills its aim. Ideally, the evaluator would like to know what the outcome would have been for a program participant if the person had not participated; the fundamental difficulty is that a person is never observed in both states. The observable target of estimation is typically the average effect, defined as the average difference between treated (in the program) and untreated outcomes across all persons in a population or in some subpopulation.

In a non-experimental evaluation, statistical techniques are used to adjust the outcomes of individuals who choose not to participate in the program to resemble what the participants would have experienced had they not participated. By contrast, a random experiment directly produces the counterfactual of what would have happened to the participant had they not participated by forcing some potential participants not to participate.

In theory, data from a randomized experiment produce an unbiased estimate of the effect of an intervention or program on an outcome variable. A simple comparison of the average outcomes of the treatment group (consisting of those who participate in the program) and the control group (those who are excluded from participation) produces a consistent estimate of the impact of the program on its participants.

In practice, a randomized experiment may suffer from the same problems that affect behavioral studies. In particular, the random assignment of the intervention is often compromised by noncompliance with the assigned intervention, i.e., members of the treatment sample may drop out of the program and members of the control group may participate. Noncompliance complicates the analysis of data from a randomized experiment in the same way as does nonresponse in (random) sample surveys and panel attrition in longitudinal studies. If the noncompliance is selective (i.e., is correlated with the outcome variable), then the difference of the average outcomes is a biased estimate of the average effect of the intervention.

Sample selectivity is a familiar problem for economists, and over the years a number of approaches have been suggested to reduce selectivity bias. Since Heckman's work, the dominant approach has been to model the selection process. This is the natural approach if the selection process is of independent interest and the econometrician understands the process well enough to propose a reasonably accurate model.

The first generation of these models required an assumption about the joint distribution of the response variable and the (latent) variable that determined participation in the program. In the second generation, this assumption is replaced by an elaborate model of the selection process under the assumption that an unbiased estimate of the intervention effect is obtained by comparing units with an (approximately) equal probability of participation.

Often - and the application considered in this thesis is a good example - there is not enough information to specify a model of the selection process. Moreover, the available characteristics of the individuals, although significantly correlated with compliance, do not explain compliance well enough to enable a comparison between members of the treatment and control groups in a subsample with the same probability of compliance. Under these circumstances an approach that does not require a model of the selection process is preferable. The method of instrumental variables (IV) gives an unbiased estimate of the intervention effect and does not require a model of participation.

This method assumes that the treatment assignment results from a two-stage process, where in the first stage the sample is divided randomly in two (or more) groups, and in the second stage, units are free to decide whether to participate in the program or not.

In the clinical literature, this experimental design is called the intention-to-treat (ITT) design. Most of the evaluation literature has focused on static interventions, i.e., interventions that are administered at a particular point in time or in a particular time interval. If the outcome is a waiting time (e.g., the time until reemployment), the intervention can be dynamic; that is, it can be switched on and off over time. Examples are the unemployment insurance experiments in which the unemployed receive a cash bonus if they find a job in a specified period. Another example is a temporary cut in unemployment benefits of unemployed individuals who do not expend sufficient effort to find a job.

In general, the intervention may even depend on information that accumulates during the unemployment spell. With such a time-varying intervention, the effect of the intervention becomes dependent on the outcome.

A basic quantity, fundamental in duration analysis, is the hazard rate. The hazard rate is, roughly speaking, the probability of finding a job after some time spent in unemployment, given that the individual was still unemployed at that time. It has a direct relation to the density of a random variable.

Economic models for durations, e.g., search models, often have direct implications for the hazard rate. One advantage of a hazard rate model is that incorporating time-varying interventions is fairly easy and natural. Another reason to consider the effect of an intervention on the hazard rate is that duration data are usually censored. Censoring limits the observation period but is not a feature of the program. Hence, the estimated effect should be independent of the censoring time. Because the hazard rate is invariant to censoring, it is natural to relate the intervention to this quantity.

Two competing approaches to the estimation of the effect of a time-varying treatment on survival have been the (Mixed) Proportional Hazard ([M]PH) model and the Accelerated Failure Time (AFT) model. The MPH models have developed popularity among econometricians, and the AFT is commonly used by biostatisticians and medical statisticians. In the MPH model, the hazard is written as the product of the baseline hazard, a non-negative regression function, and a non-negative random variable that represents the covariates that are omitted from the regression function.

There is a direct relation between hazard models (in particular the PH and MPH models) and transformation models (i.e., regression models in which the dependent variable is transformed). The simplest transformation model is the AFT model in which the dependent variable is the logarithm of the duration.

If all the covariates in the model are exogenous, many possible procedures to estimate the parameters of the AFT or MPH models exist. However, if some of the covariates are endogenous, all these methods fail and it is not straightforward how to generalize the standard IV methods to these nonlinear duration models. This thesis is one of the first attempts to provide instrumental methods for these models with possible endogenous covariates. It also gives a new solution to well-known inference problems of the MPH models. The estimation methods are all based on extensions of the (inverse of the) log-rank statistic. In Chapter 2, we introduce a two-stage IV method for the MPH models to estimate the effect of a possible endogenous intervention on the hazard. This rank estimation method requires that the members of the control group are excluded from participation. The unobserved heterogeneity component of the MPH models is notoriously hard to estimate and empirically difficult to distinguish from the duration dependence, which can lead to misleading conclusions with regard to the regression parameters.

In Chapter 3, we introduce the Generalized Accelerated Failure Time (GAFT) model, which is a generalization of both the AFT and the MPH models. We discuss a semi-parametric estimation procedure of the parameters of this model that is independent of the shape of the unobserved heterogeneity. In Chapter 3 we only consider exogenous interventions.

The analysis of endogenous interventions in the GAFT is dealt with in Chapter 4. In that chapter, we develop an IV method for the GAFT models.

This dissertation comprises four essays in labor economics and microeconometrics. In particular, I am concerned with three issues: firms' hiring decisions and compensation structures; immigration and welfare; and earnings dynamics. I also propose new econometric methods for the empirical analyses.

Chapter 1. Who Gets Good Jobs? The Hiring Decisions and Compensation Structures of Large Firms

The firm-specific human capital theory implies that large firms prefer to hire younger workers because they invest more in workers than small firms do and because those investments are fixed costs. In this paper, I use data from the Benefits Supplement to the Current Population Survey (CPS) to demonstrate that large firms indeed hire younger workers than small firms, especially for white-collar occupations. I present a simple model of firm cost minimization within an employee search framework, which is consistent with large firms' propensity to hire younger workers and has additional testable implications regarding large firms' compensation structures. The findings in this paper suggest the importance of human capital accumulation in workers' labor market success.

Chapter 2. Use of Means-Tested Transfer Programs by Immigrants, Their Children, and Their Children's Children (co-written with Kristin Butcher)

The Personal Responsibility and Work Opportunity Reconciliation Act of 1996 radically changed the welfare system in the United States. While the changes brought by this law affect all welfare recipients, noncitizens were singled out and their access to federally funded programs was especially curtailed. We present a brief summary of the changes in the welfare rules pertaining to noncitizens and highlight the fact that the new rules lead to especially high variability across states for noncitizens. We then use the most recent data available prior to these law changes (1994 1996 Current Population Surveys) to examine immigrants' participation in a wide range of transfer programs, providing a baseline against which to compare immigrants' post-reform participation. Next, we examine the participation in transfer programs of the native-born children of immigrants (the second generation). We combine information from the 1990s on the second generation with information on immigrants in the 1970s to analyze intergenerational correlation in welfare use for immigrants and their children.

Chapter 3. Estimating a Censored Dynamic Panel Data Model with an Application to Earnings Dynamics

There is a large literature on the black/white wage differential and its convergence after the antidiscrimination law of 1964. However, less is known about the racial difference in earnings stability and its change over time. This chapter studies earnings dynamics using a particular data set, namely, the matched data from the CPS and Social Security Administration (SSA) earnings records. The main empirical finding is that although linear GMM estimation yields no difference in earnings dynamics by race, the earnings process for white men appears to be more persistent than that for black men (conditional on individual heterogeneity) after censoring is taken into account.

Chapter 4. Estimation of Cross-Sectional and Panel Data Censored Regression Models with Endogeneity (co-written with Bo Honor‚)

This chapter is an extension of the econometric theory considered in Chapter 3. It proposes methods to estimate panel data censored regression models with general predetermined variables rather than a lagged dependent variable, which was considered in Chapter 3. The insight behind the estimation methods can also be applied to cross-sectional censored regression models with endogenous explanatory variables.

2000 Honorable Mentions

Darren Lubotsky, University of California, Berkeley
David Card and Hilary Hoynes, Dissertation Advisors

In this dissertation, I analyze two distinct issues. In the first part, I use a new data source to address an old and rather controversial topic in labor economics: how well immigrants fare in the U.S. labor market. The second part is motivated by the recent overhaul of the federal welfare system and examines whether increased labor market participation by welfare recipients will displace employment or reduce the earnings of other low-skilled workers in the labor market. This is investigated through a study of the 1991 elimination of the General Assistance program in Michigan.

Chapter 1. The Economic Performance of Immigrants in the U.S. Labor Market

The first contribution of this work is to clarify how selective outmigration by some immigrants influences measures of immigrant earnings growth in repeated cross-sectional and longitudinal data. The final section of this part of the dissertation assesses the extent to which rising wage inequality in the United States during the past two decades has reduced the relative earnings of immigrants.

Chapter 2. The Labor Market Effects of Welfare Reform

Part 2 of this dissertation analyzes how increased labor supply by former welfare recipients may affect the labor market outcomes of other low-skilled workers. The 1996 reform of the federal welfare system was meant to encourage self-sufficiency among recipients by providing strong incentives for them to leave welfare and enter the workforce. This increased labor market participation may, however, exert downward pressure on wages or displace employment of others already in the labor market. Since there have been limited changes in eligibility for federal welfare programs from which to draw inferences, the magnitude of these effects are uncertain. This study therefore analyzes an earlier, state-level welfare reform, the elimination of the General Assistance program in Michigan in 1991, that may provide useful evidence on the effect of the 1996 federal reform.

Despite recent declines, crime remains at the forefront of the problems facing society. In a Gallup poll conducted earlier this year that asked respondents to identify the most important problem facing their community, crime was the problem cited by 12 percent of those surveyed, ranking second behind only education. This anxiety about crime is not unjustified; an estimated 31 million crimes were committed in 1998, translating to a victimization rate of roughly 11.5 crimes per 100 residents. Because crime is a significant concern, it is important to understand the causes and consequences of criminal activity. This thesis focuses on the relationship between work and crime.

Chapter 1. Wages and Youth Arrests

This chapter uses exogenous variation in teenage wages to address the question of whether youth crime is responsive to economic conditions. I compare state-level changes in the mean log wage of teenagers and the change in teenage arrest rates between 1989 and 1992. I employ an instrumental variables strategy that uses the 1990 and 1991 increases in the federal minimum wage as a source of exogenous variation in the wages of teenagers. Two-stage least squares estimates show that while arrest rates for violent crimes do not respond strongly to changes in wages, participation in burglary, motor vehicle theft, vandalism, and robbery is negatively related to market wages.

Chapter 2. The Effect of Criminal Victimization on Employment and Income

In this chapter, I focus on one aspect of the indirect cost of crime: decreases in the earnings of victims of crime. The effect of victimization on the employment status of victims is estimated using a longitudinal version of the NCVS that contains data on the employment outcomes and victimization history of a representative sample of U.S. households. Multiple observations for each individual allow estimation of models that control for both observed and unobserved differences between victims and non-victims of crime.

Chapter 3. Efficient Bootstrapping for GMM (with Bryan W. Brown and Whitney K. Newey)

It is well known that the usual asymptotic theory can be a poor approximation to the distribution of the estimators, particularly when there are many overidentifying restrictions or when the parameters of interest are not well identified. The bootstrap provides one approach to improvements in the approximation; this chapter describes a relatively efficient bootstrap method for GMM in cross-section and panel data.

Despite overall economic growth, pockets of severe blight persist in our nation's cities and rural areas. There has been considerable debate in the economic development and regional science literature as to whether targeting policy initiatives at these distressed areas is the appropriate policy response. Nevertheless, over 40 states have proceeded to implement targeted policy in the form of enterprise zone legislation since the early 1980s. Furthermore, the federal government has recently passed an enterprise zone program. In December of 1994, President Clinton named six urban, three rural, and two supplemental empowerment zones and 60 urban, 30 rural, and four enhanced enterprise communities.

This dissertation examines the impact of state urban enterprise zones on business and housing market outcomes at the ZIP code level in six states: California, Florida, New Jersey, New York, Pennsylvania, and Virginia. In order to obtain consistent estimates of zone impacts, I limit the analysis to relatively similar subsamples of zone and non-zone areas. The estimated probability of zone designation is used to create comparison groups that control for differences in pre- designation characteristics. I find that, on average, zones have little impact on business or housing market outcomes. I find that new businesses create significantly more jobs in zones, but that shrinking business establishments offset this growth.

The data used to evaluate the six zone programs come from a number of sources. Detailed information about the programs was compiled from various documents provided by each state's program office and from U.S. Department of Housing and Urban Development publications. Outcome data come from three sources. Housing, demographic, income and unemployment information come from the 1980 and 1990 Censuses. Employment data come form an unofficial Census Bureau tabulation of the Standard Statistical Establishment List (SSEL). Establishment- level manufacturing panel data come from the U.S. Bureau of Census' Longitudinal Research Database (LRD).

Use of the LRD data to examine business outcomes represents an important contribution to the study of enterprise zones. By using establishment-level data, changes in employment levels and other variables can be attributed to new firms, ongoing firms, or firms that have closed. I find the distinction to be very important and show that enterprise zones have different impacts on the different types of firms. Additional contributions include the use of objective, non-survey data to measure outcomes and the use of multiple states in order to be able to draw wider implications of the findings.

This dissertation explores three prominent U.S. labor market developments of the 1980s and 90s: the rapid advent of workplace computerization, the historic rise in earnings inequality, and the unprecedented growth of temporary help supply employment. Essay 1, joint with Lawrence F. Katz and Alan B. Krueger, asks whether computerization is in part responsible for the dramatic growth of U.S. earnings inequality between 1980 and 1996. Using a supply-demand framework and extensive technology measures, the essay presents evidence that computerization contributed to an observed acceleration in the growth of relative demand for college graduates thereby leading, in combinations with supply shifts, to increased earnings differentials by education. In this framework, approximately 30 to 40 percent of the observed acceleration in skill upgrading within detailed industries since 1970 is explained by measures of computer capital and computer investment.

Essay 2 investigates the poorly understood labor market role of Temporary Help Supply (THS) agencies by exploring an unusual industry practice: offering nominally free, non-contracted training in general, portable skills - particularly computer skills - to temporary help workers, seemingly in defiance of the competitive model of training. Drawing on a confidential Bureau of Labor Statistics industry survey, the essay presents theory and evidence suggesting that in addition to skills formation, training serves an informational role at THS firms by helping to elicit private information that workers hold about ability - and that THS firms may in turn play this informational role in the broader labor market.

Building on the review of THS as an information broker, Essay 3 explores whether the erosion of the common law doctrine of employment at will that permitted employers unlimited discretion to terminate workers has contributed to the recent growth of THS by generating increasing demand for screening and arms-length contracting. By exploiting variation in the timing of state court decisions recognizing exceptions to the at will doctrine, the empirical analysis demonstrates that adoption of one class of exception, the "implied contract" exception, has had a sizable and robust impact on THS employment, resulting in an estimated 291,000 to 399,000 additional workers in THS on a daily basis as of 1998.

This dissertation investigates the relationship between labor market institutions and low income populations. Chapter 2 examines how changes in the compositions of educational groups has affected changes in the return to schooling. Most researchers have ignored the possibility that increases in educational attainment could result in decreases in ability within educational groups. Schooling ranks - cohort-specific relative rankings in educational attainment - are used to control for shifts in the distribution of ability within educational groups (using the intuition that relative ranks proxy for the same level of ability in all years and for all cohorts). Using data from the 1960-1990 Decennial Census, about half of the increase between 1969 and 1989 in the college/high school differential disappears once schooling ranks are held constant.

Chapter 3 (joint with Bruce Meyer) analyzes the dramatic changes in the tax and welfare systems between 1984 and 1996, which increased the incentives for low income single mothers to work. In particular, there were large expansions of the Earned Income Tax Credit (EITC) and Medicaid, and changes in the Aid to Families with Dependent Children (AFDC) program and related training and child care programs. Moreover, the employment of single mothers increased sharply during this period. We present evidence that suggests that a large share of the increase in work by single mothers can be attributed to the EITC, with smaller shares for welfare benefit reductions, and other changes in welfare programs.

Chapter 4 (joint with Bruce Meyer) examines the extent to which UI insures disadvantaged workers against unforeseen events or subsidizes firms and workers engaged in temporary layoffs. Using a five-year panel of UI administrative records from five states, we find that most claimants receive UI only once during this period. However, those individuals with three or more years of receipt during the five-year period account for forty percent of claims. Most repeat recipients are concentrated in seasonal industries and are laid off by the same employer each time. Middle-aged and high-paid workers are more likely to be repeat recipients, suggesting that workers in bad jobs do not repeatedly receive UI.

This dissertation contains three econometric studies of long-run earnings inequality. Two studies are empirical analyses of long-run earnings inequality in the United States; one contributes to the well-established literature on male earnings inequality and the other extends the analysis to the under-studied level of families. These empirical chapters focus on documenting trends in long-run earnings inequality and evaluating the potential causes. The third study develops an econometric technique that is necessary to complete the empirical analyses. Specifically, the chapter demonstrates how to use a Generalized Method of Moments estimator with an incomplete data set. Because the estimator is becoming increasingly popular and researchers often face the prospects of using incomplete date, the study will be useful to the applied researcher. Each of these studies is contained in a separate chapter of my dissertation.

Earnings Instability and Earnings Inequality in the United States: 1967-91

In this chapter, I examine the relationship among annul earnings inequality, and earnings instability. Many previous researchers have found that annual earnings inequality in the United States increased substantially during the past twenty years. Recently, economists have pointed out that this increase could have come from either of two sources: lifetime earnings could have become more unequal among individuals and/or the receipt of lifetime earnings could have become more erratic for each individual. Implicitly, researchers assumed the former was true. Distinguishing between the two possibilities will be important both for evaluating the hypotheses that have been put forward for increasing annual earnings inequality and to inform the welfare evaluation of increasing annual earnings inequality.

Using the Panel Study of Income Dynamics (PSID), I first examine these issues by comparing the distribution of long-run earnings for two time periods. I find that long-run earnings inequality has increased substantially. I then examine the same issues with a parametric approach that is common in the earnings dynamics literature. In particular, I specify a model to lifetime earnings inequality and earnings instability. I estimate the parameters of the model using a Generalized Method of Moments (GMM) framework. The increased complexity of the parametric approach is useful because I am able to examine the timing of changes. I find that lifetime inequality increased almost exclusively during the 1980s and that earnings instability increased during the 1970s. These results contribute to the existing literature by encompassing most previous findings and importantly demonstrating the inconsistency of other findings.

After establishing the changes in earnings inequality, I turn to explore what could have been the underlying causes. I find that the increase in lifetime inequality is associated with an increase in persistent wage inequality rather than an increase in persistent hours inequality. Furthermore, I find that about one-third of the increase in the persistent component is due to increasing returns to education. These findings suggest that we should look to demand shifts that bid up the relative wages of highly paid individuals to explain the increase of annual earnings inequality in the 1980s. Explanations that are consistent with these findings include skill-biased technological change and changes in international trade. Finally, I find that earnings instability increased during the 1970s because wages and hours became more unstable and earnings instability is strongly counter- cyclical because hours inequality is strongly counter-cyclical. Explanations that are consistent with these findings include changes in job stability and unionization.

Generalized Method of Moments with Incomplete Data

This chapter is an econometric theory study that develops an estimation technique used in the Earnings Instability chapter. Much of the theoretical work on GMM estimation presumes the availability of a data set with only complete observations. Completer observations are those that have values available for all variables. Applied researchers often face the prospect of using a data set with incomplete observations or "missing data." Incomplete observations might arise because of item non-responses in survey data or sample attrition in panel data. Previously, many researchers have simply discarded the incomplete observations, so that the usual GMM formulas could be applied to the remaining data set. Such an approach can lead to large efficiency losses, particularly when many observations must be discarded. Other researchers confronted with incomplete data sets have devised procedures that exploit the incomplete observations. However, their procedures have not been sufficiently general to be applicable to a wide range of incomplete data problems or do not have desirable small sample properties.

In this chapter, I develop a GMM estimator that is applicable to a broad class of incomplete data problems and that includes the specific applications previously developed. The estimator relies on a "ignorability" assumption and makes explicit the conditions under which the incomplete data problem is ignorable; a similar assumption is needed for the previous strategies used for GMM with incomplete data. With this assumption, moment functions can be constructed that satisfy the usual GMM assumptions. Thus, I show that the incomplete data estimator is consistent for the population parameter value and asymptotically normally distributed by relying on the usual theorems. A consistent estimator for standard errors is readily available. Because the estimator is relatively easy to implement, it should be useful to applied researchers.

The Long-Run Earnings Inequality of Families

Although there exists a growing body of research that examines whether long-run inequality is changing for individuals, very little research has examined changes at the family level. Extending such an analysis to families is interesting because many important consumption decisions are made at the family level and because there have been many significant labor market and demographic changes during the last three decades. For example, male long-run earnings inequality increased but female long-run earnings inequality declined. In addition, the labor supply of wives increased substantially, but the increase has been concentrated among wives of high-wage, high-labor supply males. Finally, childbearing among couples has declined, childbearing among single mothers has increased, and the divorce rate has increased. Each of these changes could have important implications concerning the pool of married couples and how labor supply decisions are made within these couples, which will in turn affect the distribution of family earnings.

Also using the PSID, I first directly examine changes in the long-run inequality of family earnings. I find that long-run family inequality has increased and that most of the increase is due to the increasing inequality of husbands' earnings; the earnings of wives were actually equalizing. Another important factor contributing to the increase in long-run family inequality was an increase in the correlation of spousal earnings; this increase was particularly strong for young couples. Overall, my results indicate that increasing husbands' inequality was responsible for about 75 percent of the gross increase in family inequality and the increasing correlation coefficient was responsible for about 25 percent of the gross increase; decreasing wives' inequality served to offset about 30 percent of the gross increase.

Although these changes in family earnings inequality will represent important changes in the labor market and are important to policy makers, focusing on earnings overlooks the distinction of whether there were accompanying changes in the distribution of wages and/or hours. Much of the literature on family earnings inequality fails to make this distinction. I next examine whether the changes in the earnings distribution were due to changes in the distribution of hours or the distribution of wages. My findings indicate that most of the change in husbands' earnings inequality was due to changes in the distribution of wages and theat wives' earnings inequality declined despite an increase in wage inequality. However, In find that much of the change in the correlation of spousal earnings was due to the wives of high-wage husbands dramatically increasing their labor supply. Overall, changes in the distribution of wages were much more important than changes in the distribution of hours for the increase in family earnings inequality.

Employer provided health insurance covers the majority of Americans under the age of 65. A beneficial aspect of employers providing health insurance is that the group to be insured is selected by a criterion other than the demand for health insurance. Hence, the problem of adverse selection is sidestepped. Recently, however, the disadvantages of the link between employment and health insurance, especially those pertaining to the labor market, have received a great deal of attention in academic and policy circles.

One problem with this system of employer provided health insurance is that the strong link between employment and the provision of health insurance implies that if wages do not perfectly offset differences in the valuation of health insurance across jobs, individual may change jobs even when new jobs with higher match specific productivity are available. This phenomenon, called "job-lock," may result in a welfare loss.

Another problem lies in the small group market for health insurance, (A small group is typically defined by insurance companies as a firm that employs less than 25 individuals.) where the problem of adverse selection remains despite the link between employment and health insurance. Almost sixty percent of non-elderly adults without health insurance are employed. Nearly two- thirds of uninsured working adults are employed in small firms.

Small firms that provide health insurance to their employees struggle to find and keep affordable health insurance, since a single expensive illness or accident may lead to health insurance cancellations or prohibitive price increases. The impact of the small group health insurance market on small firm behavior is an area of much speculation, but little research.

Policy makers have attempted to improve the system of employer-provided health insurance. Concerns about the health insurance system played a prominent role in the health care reform debate during 1993-1994. The Clinton Health Security Act, for example, called for workers to obtain health insurance through their employers, with small employers receiving government subsidies to lower the effective price of health benefits. This comprehensive restructuring of the American health care system was rejected in 1994. However, health insurance reform is still alive, mostly in the form of smaller scale state and federal laws.

Most states have implemented health insurance reform since 1990 that modifies the terms and conditions under which health insurance is offered to small employers. At the federal level, in 1985, Congress passed the Consolidated Omnibus Budget Reconciliation Act (COBRA) that enhanced portability by allowing the employee the opportunity to purchase health insurance from the form for up to 18 months after termination of employment. The Health Insurance Portability and Accountability federal legislation that covers all employer-sponsored health insurance plans, was enacted in August 21996. This federal legislation mandates portability, but does not place any restriction on health insurance premiums.

Despite the academic and policy interest in these problems with the system of employer provided health insurance, the extent or the problems and the effect of the policy measures have not been fully established. The literature on job-lock has reached no consensus on the importance of portability on job mobility. There is little to no research examining how small firms are affected by aspects of the small group health insurance market. Furthermore, there is little research that evaluates the impact of recently enacted state-level health legislation. This dissertation addresses this gap between policy and research.

As the nation's largest cash welfare program, Aid to Families with Dependent Children (AFDC), is replaced by state welfare programs funded by federal block grants, the effects of reforms on the welfare-participation and labor-supply choices of low-income individuals remain a topic of considerable interest.

This dissertation uses two different approaches a descriptive approach and a formal structural discrete-choice approach to analyze the effects of changes in welfare-program benefit formulas and eligibility rules on the economic well-being of and the choices made by low-income individuals.

The Personal Responsibility and Work Opportunity Act of 1996, which ended the AFDC program was the culmination of a fierce public policy debate over welfare reform. The current welfare system, and especially the AFDC program, was widely regarded as a flawed program that encouraged dependence on cash handouts from the government and discouraged self-sufficiency.

The current welfare-reform debate has raised old issues concerning the long-term efficacy of welfare programs on alleviating poverty and the disincentive effects of the benefit formula on labor supply. As states design new welfare programs to replace AFDC, many plan to expand earnings disregards so that more recipients can work without losing eligibility, and many plan to raise limits on allowable assets, so that more families may own cars and maintain savings accounts without losing eligibility.

Empirical evidence of the effects of such changes is available from the 1981 AFDC reforms, which sought to reduce caseloads by restricting limits on earnings and assets. The 1981 Omnibus Budget Reconciliation Act (OBRA81), designed to reign in the explosive growth in caseloads that had occurred over the previous decade, dramatically altered the treatment of earnings and assets in the AFDC program.

The 1981 Act eliminated the disregard of one-third of earnings after four months of work, capped the disregards given by caseworkers for work expenses and child care expenses, and imposed new eligibility restrictions. OBRA81 denied eligibility to families whose pre-disregard income was over 150 percent of the state need standard, and imposed a new asset limit of $1,000 per family. The overall result of these changes was that drastically fewer recipients could maintain eligibility while working, and hence the work disincentive for AFDC recipients was significantly increased.

Current state TANF proposals to expand earnings disregards and asset limits would directly reverse several of the federal regulations implemented under the 1981 Act, which was designed principally to reduce caseloads by tightening eligibility. This dissertation presents an analysis of the effects of the 1981 welfare reforms on female-headed households with children.

Although there is a consensus that there has been a dramatic improvement in the economic position of black Americans since the dismal prognosis of Gunnar Myrdal in An American Dilemma 50 years ago, there is little consensus on the magnitude of these gains, their underlying causes, and whether racial parity has been reached. My dissertation empirically examines the evolution of the economic status of blacks in the U.S. labor market over the last 40 years.

I use three different strategies to identify and estimate shifts in the economic status of African Americans. In Chapter 1, I use a treatment-and-control group methodology to evaluate the effects of the 1972 Equal Employment Opportunity Act (EEOA). In particular, the 1972 EEOA, in conjunction with preexisting state fair employment practice laws, provides a "natural experiment" in which differences across industries and states in treatment status are used to identify the impact of civil rights policy.

In Chapter 2, I reevaluate the impact of Title VII of the Civil Rights Act of 1964 using a unique data source which contains longitudinal information on individual earnings. An evaluation strategy is proposed which uses the longitudinal structure of the earnings data to control for other factors unrelated to Title VII which also influence relative earnings. Additionally, the estimation procedures account for the pervasive censoring in the earnings data.

In Chapter 3, I use a model of unobservable skill to assess the implications of growing wage dispersion on estimated changes in the college premium and black/white relative wages in the 1980s. The key to the analysis is the finding that one can use across-group variation in within-group wage variances from multiple periods to identify and estimate a relatively unrestrictive error-components model of wages which nests competing explanations for observed changes in relative wages. Interestingly, the identification strategy does not require panel data, but rather a series of independent cross-sectional samples is sufficient for implementing the econometric model.

Chapter 1

Over two decades of research on Title VII of the Civil Rights Act of 1964 and Executive Order 11246, which followed it in September 1965, have failed to reach a consensus on the effectiveness of these laws. Two problems confront any analyses of civil rights policies. On one hand, the timing of the legislation (in the mid 1960s) corresponds with the timing of many other significant changes in the U.S. labor market. On the other hand, the nature of these laws, and in particular their nearly universal coverage, makes it difficult to control for changes that would have occurred even in the absence of the legislation.

This paper presents new evidence on the effectiveness of federal antidiscrimination policy, focusing on an important but under-studied amendment of Title VII of the Civil Rights Act. The Equal Employment Opportunity Act (EEOA) of 1972 expanded civil rights coverage of Title VII statutes to employers with 15 24 employees, while leaving unaffected the civil rights protection for employees of larger establishments. In conjunction with already existing state fair employment practice (FEP) laws, which varied in employer coverage, the EEOA set up a useful "natural experiment" for measuring the impact of civil rights law. The effect of the legislation can be estimated by comparing changes in outcomes at newly covered and previously covered (by Title VII or FEP laws) employers with respect to the timing of the amendment. The 1972 amendment should have most directly affected the relative status of blacks employed in the newly covered small establishments in states where small employers were not covered by FEP laws. Many of the problems in the existing literature on the 1964 Civil Rights Act can be avoided by this simple treatment and control group evaluation methodology.

In the absence of a detailed establishment-level data set that would permit precise comparisons of labor market outcomes by employer size, the strategy in this paper is to use individual micro-data aggregated into region and industry cells. Specifically, variation across industries in the fraction of employees in small establishments and across states in the employer coverage of FEP laws is used to define treatment and control groups. Industries are aggregated into three groups with similar fractions of workers in small establishments in each group, and states are aggregated into the South, where FEP laws were virtually non-existent, and the non-South, where most states already had FEP laws covering small employers. Using data from March and October Current Population Surveys (CPS) from 1968-80, movements in three measures of racial inequality (the share of blacks in industry employment, black/white relative annual earnings, and black/white relative occupational status) in each cell are analyzed to measure the impact of the 1972 EEOA on working-age black men employed in the private sector. Comparisons are made between the relative gains experienced by blacks employed in the most affected industry group in the South (the treatment group) and relative gains for blacks in the other five cells (the control groups).

Controlling for a wide set of factors, including permanent differences across regions and industry groups, cyclical effects, the changing relative skills of black workers, and region- and industry-specific trends, I find that black men in the high-impact industries in the South achieved large gains in employment share and relative earnings and more modest gains in relative occupational status after 1972. For blacks employed in the other cells, there are either no improvements or improvements which appear to be the continuation of trends that began some time before 1972.

Most of the gains were concentrated among relatively unskilled black men employed in the construction and service sectors. It appears that the relative demand for less-skilled blacks increased significantly among newly covered employers after the implementation of the 1972 coverage amendment. The location and timing of these relative gains provide evidence that civil rights policies had a positive impact on the labor market status of African Americans.

Chapter 2

Time-series studies assessing the effects of Title VII (and Executive Order 11246) rely on comparisons of pre-policy (before 1965) and post-policy (after 1965) trends in black/white relative earnings calculated from published aggregate tabulations. Because the legislation specifies nearly universal coverage and its timing corresponds with the timing of many other significant changes in the U.S. labor market (e.g., the War on Poverty and changes in the relative skills of black workers), it is difficult to control for changes in relative earnings that would have occurred even in the absence of the legislation. As a result, with aggregate time-series data it is nearly impossible to disentangle the actual effect of the law from other factors unrelated to Title VII which also influence relative earnings.

This paper uses a unique micro data base to reevaluate the impact of federal antidiscrimination policy on black economic progress in the 1960s and early 1970s. A data source is constructed that links the 1973 and 1978 March Current Population Surveys (CPS) to employer-reported longitudinal Social Security Administration earnings records from 1957 to 1975 With disaggregate, detailed longitudinal data, I can use "non-experimental" statistical methods to account for competing explanations for observed changes in relative earnings.

In particular, I introduce and implement a new evaluation strategy for obtaining structural estimates of the impact of Title VII which controls for the effects of both the observed (e.g., education) and unobserved (e.g., school quality and family background) skill gaps between black and white men and changes in the return to these skills on relative earnings. Thus, this study exploits the longitudinal structure of earnings to identify earnings convergence after 1965 attributable to changes in labor market discrimination, presumably the result of the government intervention.

Although the administrative payroll tax records are likely to be an accurate measure of true earnings, many records are censored at zero and at the Social Security tax ceiling. The censoring at zero does not appear to be a serious issue. The censoring at the tax ceiling, however, could be extremely problematic. Because the real value of the taxable maximum changed significantly during the 1960s and early 1970s, estimates of the intervention effects which do not account for the top-coding and changes in it (e.g., least squares estimates) would be seriously biased.

As a result, in this study I use both maximum likelihood and quantile-based semiparametric estimation to implement the evaluation strategy and identify the policy effects while explicitly accounting for the nonlinearity in the panel data censored regression model of earnings. If the distribution of the underlying unobserved components of the regression model are correctly specified (e.g., error terms which are identically normally distributed across individuals), then the maximum likelihood estimator will be consistent and efficient.

However, if the unobserved components are non-normal and/or not identically distributed, only the censored regression quantile estimators will provide consistent estimates of the legislation's impact.

Analyzing data on black and white men in three narrowly defined birth cohorts disaggregated by region (South and non-South), I find that blacks in the two youngest birth cohorts in the South achieved large gains in relative earnings after 1965 even after controlling for black/white skill differences and changes in the return to skill. In addition, there were no post-policy improvements in the economic status of black men in the oldest birth cohort in the South, while in the non-South, only black men in the youngest birth cohort achieved relative earnings gains after 1965.

Although there is evidence that the price of unobserved skill was nonstationary, changes in the skill premium were much too small to have had a significant impact on changes in the black/white earnings gap during the 1960s and early 1970s. I also find no evidence that these results are biased by potentially nonrandom participation in the sector covered by Social Security. The analysis suggests that Title VII legislation led to much of the improvements in the economic status of African Americans from 1965 75.

With respect to estimation methodologies, I find that using the maximum likelihood estimates of the conditional location parameters of the censored regression model provides accurate measures of the impact of the 1964 Act. However, using maximum likelihood estimates of the second moment parameters of the model as an extra source of identification relies heavily on stochastic restrictions on the shape of the error distribution (e.g., joint normality) which do not hold in the data. Quantile estimation of the censored regression model results in estimates of the intervention effects which are very similar to the maximum likelihood estimates, implying that the sources of misspecification in the maximum likelihood approach are fixed over time.

Surprisingly, the quantile-based estimates are more precise than the maximum likelihood estimates due to long tails at the low end of the earnings distribution. I conclude that quantile-based semi-parametric methods provide an extremely attractive approach to estimating censored regression models of the log-earnings process.

Chapter 3

During the 1980s, wage inequality among men grew along several dimensions in the United States. Most notably, after experiencing a decline in the previous decade, the measured college/high school wage differential increased substantially during the 1980s. In addition, wage inequality within narrowly defined demographic groups based on education and experience also rose, continuing a trend that began in the early 1970s. Finally, wage convergence between black and white men stagnated in the 1980s after 15 years of significant black economic progress dating back to the mid 1960s.

Consequently, a great deal of economic research has focused on proposing and evaluating various explanations for these well-documented empirical facts.

Amidst numerous attempts to identify the driving forces behind these observed changes in relative wages, a debate has arisen concerning their connection. In particular, rising within-group residual wage dispersion may reflect an increase in the return to unobservable "skill." As a result, it is an open question whether the rapid growth in the college/high school wage gap in the 1980s represents an increase in the economic returns to a college education or a rise in the payoff to unmeasured factors which are correlated with, but not the result of, educational attainment (e.g., innate ability or family background influences).

Similarly, it is questionable whether the recent slowdown in black/white wage convergence is attributable to an increase in labor market discrimination or a rising premium for such unobserved factors as well as for other difficult-to-measure productivity components (e.g., school "quality").

In this study, we attempt to inform the debate by answering the following questions concerning relative wage changes during the 1980s: how much of the dramatic increase in the college/high school wage differential could be due to a rise in the return to unmeasured "ability" or "skill" rather than to an increase in the true college premium? and, to what extent can the slowdown in black economic progress or the widening black/white wage gap among young workers be explained by a rise in the return to pre-labor market factors correlated with race?

A recent body of empirical work has proposed and used "direct" measures of skill or ability, such as test scores and observable measures of school quality, to control for unobserved heterogeneity biases that may confound estimates of the return to college and the existence of racial wage discrimination. However, due to either the "unspecific" or "too specific" nature of the measure used, the findings of these studies are arguably difficult to interpret.

The approach we adopt to answer the above questions, on the other hand, provides a distinct and more general alternative to using these direct measures. Specifically, in our analysis one component of "skill" is allowed to remain strictly unobservable to the researcher and have an economic payoff which changes over time. More importantly, these changes will have clear implications for the behavior of within-group wage dispersion across groups and over time.

We develop a parsimonious yet general model of the wage process in which it is necessary to identify both 1) the extent of the "unobserved skill" (or omitted-ability) bias at a given point in time and 2) the growth in the unobserved skill premium, in order to identify true changes in the college premium or the residual black/white wage gap. Although one cannot identify the unobserved skill gap from a time-series of conditional means of log-wages, we show that a series of conditional variances of log-wages over time is sufficient to identify changes in the payoff of unmeasured skill.

In addition, based on these estimates of the rise in the value of skill, we are able to generate bounds for changes in the college premium and wage discrimination under various assumptions on the magnitude of unobserved skill differences across education and race groups.

Specifically, our study illustrates that it is possible to use across-group variation in within-group wage variances from multiple periods to identify the change in the return to unobservable skill within a relatively unrestrictive error-components model of wages. Our identification strategy accommodates an unobservable component of skill which differs by education group and race and has a non-stationary return while avoiding full specification of the time-series properties or the functional form of the error components.

Furthermore, our approach does not require panel data on individuals. Instead, a series of independent cross-sectional samples is sufficient for implementing our econometric model of unobservable skill and assessing empirically the implications of growing wage dispersion for conventional estimates of changes in the college premium and racial discrimination.

Earnings data for men from the Current Population Survey (CPS) show that there is useful variation in within-group wage variances across narrowly defined demographic groups. This variation across groups and over time allows us to estimate a growth in the return to unobservable skill of about 10 20 percent during the course of the 1980s. In addition, our model provides a relatively accurate description of changes in within-group wage inequality over time.

Even given our largest estimate of the change in the value of unobservable skill and under the assumption that the entire initial education differential is attributable to nonrandom sorting, we find that college-educated workers still gain substantially relative to high school-educated workers in the 1980s after controlling for the effects of the rising skill premium. In particular, the rise in the payoff to unobserved skill can account for at most 30 to 40 percent of the observed rise in the college premium for relatively young workers, leaving a 0.10 to 0.17 log point growth in the true college premium as the lower bound estimate.

In addition, we find that an increase in the return to unmeasured skill cannot account for the stagnation of black economic progress during the 1980s, even under the assumption that all of the initial racial difference in earnings results from unmeasured productivity differences. Specifically, young, well-educated black men still experience at least a 0.13 log point decline in wages relative to their white counterparts in the 1980s.

1997 Honorable Mentions
Sandra E. Black, Harvard University
Lawrence Katz, Dissertation Advisor

Human capital accumulation has long been a point of interest for economists, particularly in recent years when rising wage inequality has been associated with increasing premiums on education and skills. As a result, national attention has focused more closely on the availability of educational opportunities and training programs in order to assist low-skilled individuals who are hurt the most by recent increases in inequality. In many cases, it is difficult to measure the value of these opportunities. What is the value of a better education? What is the value of firm-provided training?

A number of studies have focused on the "value" of education by relating the wages individuals earn when they are older to the quality of their education, measured either by inputs or outputs, when they were younger. However, a key limitation of this work is that there may be other factors influencing wages that are also related to school quality.

For example, if parents who are more concerned about school quality are more likely to send their children to better schools, and the children will do better because their parents care more, regardless of the school quality, then we may see a positive relationship between school quality and wages even if none were there. In addition, the choice of using inputs (such as school spending) or outputs (such as test scores) to measure school quality affects the conclusions made. As a result, the literature has been controversial.

A very different way to approach the issue is to focus on the value that parents place on better schools. One way to do this is to look at the housing market and see how much more parents are willing to pay to live in an area with better schools. The first chapter in the dissertation looks at houses in three counties in Massachusetts and calculates the value, in terms of increased housing prices, of better schools as measured by higher test scores.

While understanding the value of better schools is essential when evaluating a number of educational policies, other programs, such as the implementation of school vouchers, require an understanding of both the value of increased access to private schools and the effect of increased private school access on public school quality. There is a very limited understanding of the implications of voucher programs and the effects of increased private school access, and a first step in understanding the potential effects is to look at the value of private school access in the absence of school vouchers.

We can then infer what the effects of increased access will be. The second chapter in the dissertation focuses on the valuation of private school access in terms of the value as capitalized in housing prices and the effect on public school quality, shedding light onto an area that has had few previous empirical findings.

Finally, schools are not the only avenue by which we can improve the skills of low-skilled individuals; another method is firm-provided training. If schools are not providing the necessary skills, it may be the responsibility of the firms to do so. However, firms may be unwilling to invest in workers if training is not profitable. In order to determine the value of firm-provided training and other workplace practices, we first must understand their relationship to firm productivity.

The third chapter in the dissertation looks at different workplace practices and relates them to firm productivity, informing an area that has previously been lacking due to data limitations.

Do Better Schools Matter? Parental Valuation of Elementary Education

From a policy perspective, understanding the value of better schools is integral to the improvement of our educational system. Relevant policies include busing programs, school choice, school voucher programs, and school finance laws. Of particular interest to policymakers are whether parents are informed enough as consumers of education to be sensitive to differences in the performance of local schools and how they value these differences.

The first part of this dissertation focuses on estimating the value parents place on education, and it does so by looking at the housing market. The first chapter looks at the valuation of public elementary school quality as capitalized into housing prices. Previous literature that used the housing market to look at the relationship between housing prices and school quality has been limited by the fact that better schools are located in better neighborhoods, so omitted neighborhood characteristics may be biasing estimates of the valuation of school quality upwards.

It is therefore difficult to isolate the effect of school quality on house prices. Ideally, we would look at similar quality houses on opposite sides of the same street, where the children on one side of the street attend one school and those on the other attend another school. There would be virtually no neighborhood variation, so differences in house prices would reflect only differences in school quality and school peers.

Using Massachusetts data on individual house sales with exact geographic location information, I am able to observe something very close to this ideal, thereby minimizing the confounding effects of unmeasured neighborhood characteristics.

I do so by looking at houses on attendance district boundaries, the geographic boundaries that determine which school a child goes to within a district. By comparing houses within the same city, I eliminate all variation in property tax rates. By comparing houses within the same school district, I am controlling for differences in per-pupil spending. By comparing houses on opposite sides of the attendance district boundaries, I am comparing houses that are within a very small geographic distance of each other and therefore I am controlling for variations in neighborhoods.

The only difference in the houses I am comparing is in the elementary school the child attends, so any difference in house price can be attributed to differences in elementary school quality.

Using a standardized test score administered by the state (the Massachusetts Educational Assessment Program [MEAP]) as my measure of school quality, I find that parents are willing to pay for better schools. Chapter One shows that parents are willing to pay about 2.1% more for houses associated with elementary schools having 5% higher test scores at the mean. This estimate is $3,948 at the mean house price ($188,000) and is approximately half the estimate I get when I run the more typical hedonic housing price regression with census block group controls for neighborhood characteristics.

This suggests that there was a significant amount of bias due to omitted neighborhood characteristics. The estimate provides information about the value of programs that would involve busing or sending students from poorer neighborhoods into better neighborhoods.

The Effects of Private School Access: A Study of the Housing Market

The second chapter focuses on the valuation of private school access, again as capitalized in housing prices, through its direct effect and its indirect effect (through changes in public school quality, which is also capitalized into housing prices). When considering policies such as the implementation of a school voucher program, which would essentially increase private school access for low-income families, understanding the implications of private school access is essential.

There is a large theoretical literature focusing on the relationship between private school access and public school quality, and the results are ambiguous depending on the assumptions made. There is, however, very little empirical evidence resolving the uncertainty. Some evidence suggests that private school access increases public school efficiency, but tempering this are theories that private school access may lead to decreases in public school spending or reduced peer quality in public schools. The second chapter sheds light on this literature and provides evidence on this relationship.

By looking at the effects of private school access on the housing market, I am able to look at both the direct effect of private school access on house prices and the indirect effect through public school quality. I use data on the housing market in Middlesex County in Massachusetts, where approximately 16% of elementary school children in the county attended one of the 64 available private schools in 1993.

I find that, controlling for neighborhood characteristics, there is a significant positive effect of private school access, measured in a number of ways, on house prices. However, this effect is reduced somewhat by the fact that private school access seems to have a negative effect on public school quality as measured by elementary school test scores. Although I am only able to measure a partial equilibrium result (because my study is limited to one county in Massachusetts and is not able to pick up such things as beneficial increases in public school efficiency), these results still suggest that private school access does somewhat reduce public school quality and still on net raising house prices.

How to Compete: The Impact of Workplace Practices on Firm Productivity

The third chapter of the dissertation shifts focus from education to firm-provided training and workplace practices. In joint work with Lisa Lynch, we look not at the dollar value of additional training, but its value in terms of increased firm productivity. By looking at the relationship between training and workplace practices and firm productivity, we can better understand the skills that will be needed in the workplace. We use a new, nationally representative survey of U.S. establishments and look at the association between establishment workplace practices and productivity.

In earlier work, we examined the relationship between different types of firm training and the depth of firm training and workplace practices. Our results, which showed that types of training and training intensity vary significantly across manufacturing and nonmanufacturing sectors, across industries, and even across firms within industries, suggested the need for an understanding of how these practices are related to firm productivity.

There is much discussion today of moving away from more traditional assembly-line production technology to technologies where workers have more input into the production process. We see examples of this type of workplace almost every day, even on television commercials advertising that products are superior because the employees have a voice in how it is produced, and therefore they "care" more.

In the third chapter, we evaluate this new type of workplace technology by assessing its relationship to firm productivity. We estimate a standard Cobb-Douglas production function with cross-section data that is augmented by measures of workplace practices and human capital investments, along with panel data estimation using data from the Longitudinal Research Database (LRD). Using a two-step procedure, we first estimate our production function and calculate firm fixed effects (the firm average residual), which are then regressed on our measures of workplace practices and other employee and employer characteristics to determine their association with productivity.

We find that workplace practices do matter regardless of how the production function is specified. We also find that the key is not whether or not an employer adopts a particular workplace practice, but how the practice is implemented. That is to say, simply adopting a Total Quality Management (TQM) system has an insignificant effect on productivity, but raising the proportion of workers involved in decision making within the plant, either through regular meetings (an important component of TQM systems) or through unionization, does have a significant positive association with establishment productivity.

Our findings suggest that establishment practices that encourage workers to participate in the production process decisions are associated with higher firm productivity.

Conclusion

While my dissertation presents evidence on the value of education and training, it is only a first step. In terms of education, a study of what elements of better schools parents are willing to pay for is a natural next step. Do they value better test scores because of the better students at the school, or is it the better teachers and administrators that are important? With respect to private school access, we still have many unanswered questions. Not only do we need to understand what are the more general equilibrium effects of private school access on public school quality, we also must determine the value of different types of private schools instead of looking at them as a homogenous group.

Finally, my results suggest that workplace practices are related to firm productivity, but further research and more data are necessary in order to convincingly attribute causality.

Carole Roan Gresenz, Brown University
Robert Moffitt, Dissertation Advisor

Recent welfare reform legislation replaced the Aid to Families with Dependent Children (AFDC) program with a largely defederalized welfare program, Temporary Assistance to Needy Families (TANF). Benefit generosity, enforcement of program rules, availability of training, and other aspects of TANF are expected to vary widely across states as a result of the defederalization. That has led to concern that some welfare recipients will migrate to states that have more generous programs.

The concern is not a new one. Before the legislation was passed, many questioned whether large inter-state differences in the AFDC benefit paid to families induced migration. This thesis examines whether that was the case: did AFDC benefit differentials affect individuals' location choices? Among other things, the answer will help us better understand how individuals may respond to variation in welfare programs under TANF.

The results of the study also have implications for academic research. Welfare analysts have used the variation in AFDC benefits across states to identify its effect on individuals' marital status, childbearing, and labor supply choices, for example. If AFDC benefits play a role in individuals' location choices, then the variation in benefits across states is endogenous and cannot be used to identify the effects of AFDC on other decisions.

There is a substantial literature on AFDC and migration, but unresolved theoretical and methodological issues have led to mixed results in studies to date. This thesis builds on the existing literature by developing a unique theoretical framework of decision making and by improving on two types of estimation strategies. In a departure from the previous literature, an individual's choice of location and choice of whether to work and/or receive welfare are modeled as sequential. The sequential model allows welfare benefits to act as insurance against a bad wage outcome in a particular location. The model yields hypotheses about the effect of migration in response to AFDC benefit differentials on employment.

The theoretical model guides two estimation strategies. The first strategy uses a "location choice" model, in which individuals' location choices at a point in time are analyzed, to ascertain whether in equilibrium and holding all else constant, individuals are distributed more heavily in higher-benefit states.

The second strategy uses a "mobility" model, in which individuals' decisions to move from or stay in a location are analyzed to determine whether changes in benefits over time bring about changes in location choices over time.

In the next section, the sequential model of decision-making is developed and its theoretical implications are discussed. In Section III, two empirical strategies are explained. Estimation results are summarized in Section IV, and Section V is the conclusion.

Theoretical Model of Location and Work/Welfare Choice

The theoretical model encompasses two decisions. Individuals choose which state to reside in and they choose whether to work and/or receive welfare. An individualÕs location decision is made by comparing the utility of the initial location to the utility of each alternative location in the choice set. An individual chooses a particular location if the expected utility from living there, less the cost of moving there, is greater than the expected utility, net of moving costs, of each of the other locations.

An individual's work/welfare choice is one of four alternatives: individuals can receive welfare, neither work nor receive welfare, only work, or do some combination of work and welfare. Individuals also decide upon work and welfare by maximizing utility.

At the end of an initial time period, individuals must choose a location and work/welfare decision for the subsequent period. Both the choice of location and the work/welfare choice depend in part on what wages are in each location. Wages are determined by individual characteristics and regional economic characteristics, but they are also determined by a random component which reflects a premium or discount that certain individuals get when working for a particular employer, which can be thought of as "luck."

As a result, when individuals are considering different locations, they have information about the mean wage and variance around the wage for each location, but they do not know the actual wage in each location. Wages are only revealed when individuals carry out job search activities in the chosen location. Because of the uncertainty in wages, individuals calculate an expected utility and expected probability of each work/welfare alternative at the time when the location decision must be made, but make no work/welfare decision. Thus, each individual first chooses a location, based on an expected work/welfare choice in each location, and then chooses a work/welfare alternative, once the location decision is made and carried out.

The sequential framework allows for a direct effect of benefits on individuals who later receive welfare and an indirect effect of benefits on individuals who later choose not to receive welfare. The latter may be thought of as an insurance effect: individuals who move to a state and later find employment may have been affected by the insurance AFDC benefits provided against the possibility of a poor wage draw. The empirical implications are twofold. First, the initial sample of individuals includes both individuals initially observed receiving welfare and those not receiving welfare.

Both types of individuals are assumed to have some uncertainty about future work and welfare choices. Second, wages and benefits are allowed to affect the location choices of both those who receive welfare in the new location and those who work in the new location.

The advantages of the sequential framework can be clarified by comparing it to alternative models of decision making. In some studies, the work/welfare decision is not modeled at all, but rather "receipt of welfare" is included as a right-hand-side variable. The problem with that specification is that welfare receipt is endogenous. Other empirical analyses have included only those initially observed receiving welfare in the sample population.

The exclusion of nonrecipients may underestimate the true effect of AFDC benefits on location choice. Simultaneous models have also been implemented. These models assume individuals decide whether or not to receive welfare at the time the location decision is made despite uncertainty about what wages will be in different locations. The empirical result is that benefits are allowed to affect only those sample members who receive welfare in the chosen location. The simultaneous framework misses the effect of benefits on the location choices of individuals who consider welfare but who later choose not to receive it.

One hypothesis about the effect of migration on employment levels is that higher AFDC benefits induce some individuals to leave work in lower-benefit states to join the welfare rolls in higher-benefit states, decreasing overall employment levels. The implication from the sequential model of location and work/welfare choices that benefits may act as insurance against a poor wage draw in a location suggests certain modifications to that hypothesis.

Consider an individual who is deciding whether to move from a location A where wages are relatively low to a location B where wages are relatively high. Individuals are likely to have better information about the distribution of wages in the initial state A than in B because of information-sharing networks among neighbors, family, and friends, and because individuals may have investigated job opportunities in the local area in the past.

Thus, the variance of wages in B is greater than the variance of wages in A. The higher variance in wages increases the individual's risk of a poor wage draw in B and decreases the probability that the individual will move to B. If benefits are relatively high in the high-wage state B, then benefits may facilitate migration to B by offsetting some of the risk of a bad wage draw.

On the other hand, if benefits are relatively high in the low-wage state A, then they may mitigate labor-market-equilibrating migration to B. Thus, migration in response to benefits has the potential for increasing employment levels by facilitating movement to higher-wage states, as well as the potential for decreasing employment levels by impeding migration out of low-wage states. The effect of migration in response to benefits on employment is in part determined by the relationship between benefits and wages across states.

Estimation
Location Choice Model

Location choice studies analyze individualsÕ choice of residence at a single point in time to ascertain whether in equilibrium and holding all else constant welfare recipients are distributed more heavily in higher-benefit states. Blank (1988) employs this type of approach. She estimates a model of individuals' choice of one of twelve regions controlling for the characteristics of each region, including the expected wage rate, tax rate and unemployment level, in addition to the AFDC benefit level. Blank finds a positive effect of benefits on location choice.

Two important methodological issues in her location choice study are the aggregation of location choice to the regional level and the validity of the implicit equilibrium assumption. As a result of aggregation, the effect of AFDC benefits on location choice is underestimated to the extent that benefit levels affect intraregional decisions. In addition, variation in distances between regions, variation in wages within regions, and variation in AFDC benefit levels within regions are not captured. The second methodological issue concerns the implicit assumption in the model that individuals are in equilibrium at the point in time in which they are observed.

In Blank's analysis, there is no empirical support for the assumption. By using cross-sectional data, Blank observes individuals at different points in their welfare-eligible years. Those who have received welfare for many years are more likely to be in equilibrium than those who have only recently become eligible for welfare, as they may not have had a chance to react, in terms of choosing where to live, to the circumstance that brought about eligibility.

The first estimation strategy improves upon Blank's location choice analysis. One improvement is straightforward: I model an individual's choice of state rather than the aggregated choice of region, thereby avoiding problems of measurement error. Another improvement stems from innovations of the theoretical model. I allow for an insurance effect of benefits by including benefits as regressors for all sample members, rather than only for those who receive welfare in the chosen location. A third improvement is that I employ longitudinal data to empirically support the assumption in the location choice model that individuals are in equilibrium with respect to location.

I assume that individuals are in disequilibrium during their first year of sample eligibility because they are not likely to have had a chance to change location in response to their eligibility. I then allow individuals five years to reach equilibrium. Thus, the sample consists of individuals' observations five years after their initial eligible observation.

Mobility Model

Studies of mobility analyze changes in individuals' location choices between two time periods, or their migration behavior. Clark (1990) estimates a model of individualsÕ decisions to move or stay and includes as a determinant of that choice the AFDC benefit level in the state in which an individual is initially observed.

Zimmerman and Levine (1995) compare the migration decisions of individuals eligible for AFDC to the migration decisions of individuals ineligible for AFDC. The foremost problem in these and other mobility studies is the endogeneity of the AFDC benefit level in the initial locationÑthe benefit level is endogenous because individuals have chosen the location in which they are observed in time periods before observation. Thus, these studies may underestimate the effect of benefits on migration to the extent that migration in response to benefits occurred before observation.

In a second estimation strategy, I improve upon previous mobility studies. Two contributions are noteworthy. First, I deal with the problem of endogenous benefits by analyzing changes in location as a function of changes in covariates, which are exogenous, instead of levels, which are endogenous. Second, I parameterize individuals' choices with a dependent variable indicating which state the individual chose and a lagged independent variable indicating location in the initial time period.

Previous mobility studies have parameterized individualsÕ choices as the dichotomous "move" or "stay" decision. The move/stay decision must be estimated with either probit or logit models, which are not able to incorporate controls for characteristics of all alternative locations. The model I employ can be estimated with a conditional logit model that readily incorporates characteristics of all locations.

The specification of right-hand-side variables in changes requires that individuals be in equilibrium with respect to location in the initial observation and in the terminal observation. If individuals are in equilibrium in their initial location, they will only react to shifts in variables that affect the utility of each location, not the levels of these variables. The shifts induce disequilibrium in location, and individuals respond by choosing a location that is optimal under the new conditions. As in the location choice model, I support the equilibrium assumptions with a specific sample selection methodology involving longitudinal data.

In this model, I assume that individuals reach equilibrium quickly, by the year following their first eligible year. For the terminal equilibrium observation, I use each individual's observation five years later.

Results and Discussion

In the location choice model, I find a positive but insignificant effect of AFDC benefits on location choice. Even were the hypothesis that the AFDC effect is different from zero accepted, the practical effect of AFDC is negligible. The magnitude of the AFDC coefficient can be discussed in terms of its "retentive" and "attractive" powers, the former referring to the effect of an increase in benefits in the initial state on staying in the initial state and the latter to the effect of an increase in benefits in an alternative state on moving to the alternative state.

The retentive effect is smaller than the attractive effect, but neither effect is large. A $100 per month increase in AFDC benefits on average increases the probability of staying in the same location less than 1%. The retentive effect is small because the probability of staying is large to begin with the mean probability of staying was 91%. The probability of choosing a location other than the original location rises on average 3.8% from a $100 change in AFDC, while for half of the sample the average effect was more than 12.9%.

But even an attractive effect of 13% is rarely enough of an effect to evoke a move, given the large probability of staying in the original location and low probabilities of choosing alternative locations. Thus, the effect of inertia is significantly stronger than the attractive effect of AFDC. In the mobility model, I find that changes in AFDC have a positive but insignificant effect on location choice, and the magnitude of the AFDC effect is comparable to that in the location choice. A $100 increase in AFDC in a state other than the initial state increases the probability of choosing that location 2.6% on average.

What explains the absence of a strong positive and significant AFDC effect? Factors that limit individuals' ability to migrate and factors that limit migration given an ability to migrate are at work. One of the factors that limits mobility is liquidity constraints. Money to finance a move may not be readily available. Estimation results indicate that this may be the case: the effects on the probability of choosing a location other than the initial location of distance and number of children, which increase the monetary cost of a move, are negative and significant.

Another monetary cost not captured in the model is that of switching from receiving welfare in one state to receiving welfare in another. Bureaucratic delays may result in a lag between receipt of the last check from the initial state and receipt of the first check in the new state.

Liquidity-constrained individuals may not be able to afford the time without income. Another factor that limits mobility is the network of familial and other support that low-income parents develop in their initial state. The presence of non-family members in the household and the proxy for familial ties in a state, "state in which one grew up," both have large and significant negative effects on the probability of a move. Others in the household and family members may provide services which would be lost in the event of a move.

Low levels of education are also a migration deterrent. Most AFDC eligible or nearly eligible individuals have less than a high school education. Even without controlling for income, mobility is limited among less educated individuals. Education likely increases the ability of workers to get placed in jobs, either because they are more employable or because they have better information about opportunities, leaving less educated individuals at a disadvantage in new markets.

Why might AFDC differentials not affect migration, given an ability to migrate? One consideration is information. Individuals may not know the differing benefit levels across states.

Within-community information networks are strong, but it is not clear that information about benefits across the nation is easily and readily accessible. A second issue involves multiperiod optimization and thus is not captured explicitly in estimation of the two-period model used in this paper. In a multiperiod model, individuals have a longer time horizon to consider, and decisions are made in the first time period considering the effect of that decision on all future time periods. Thus, an individual thinking about whether or not to move to a new location considers the one time cost of moving compared to the benefits of moving that extend over all future years in which the individuals lives in the new location.

The shorter the time over which the benefits accrue, the less likely a move will be. (Indeed, the effect of age on the probability of migration is negative.) Consider the fact that for many individuals the duration of AFDC recipiency is relatively short: Blank (1989) finds a median spell length of between 19 and 22 months, while Ellwood (1986) notes that about 47% of new spells last less than two years.

The value of moving to a state with higher AFDC payments will be smaller the shorter is the time period over which individuals expect to receive payments. Multiperiod optimization also involves individuals' expectations about future values of wages, benefits and the like. If individuals are aware that AFDC benefits in higher-payment states have been falling more rapidly over time than in lower-payment states and they expect that trend to continue, then their perceived long run value of moving for higher benefits decreases.

Conclusion

Estimation of individuals' choices of location and individuals' decisions to change location yields consistent results. Benefit differentials across states do not appear to shift the equilibrium distribution of individuals toward higher-benefit states and changes in AFDC over time do not significantly affect individuals' decisions to change locations.

What are the implications of the findings? Recall that the issue is a concern in welfare research where the variation in benefits is used to identify other incentive effects of AFDC. The results are heartening for academic research as they indicate that the variation in AFDC benefits over states can be considered exogenous at least with regard to individuals' location decisions.

The results also help in our understanding of how individuals are likely to react to variation in welfare across states under TANF. Given the lack of response in terms of migration to large AFDC benefit differentials, we expect that migration in response to TANF differentials will be limited, with the caveat that individuals may react differently to variation in time limits or work requirements than they do to differences monthly monetary payments. Continued research is necessary to more fully address the issue. The models presented here provide a useful framework for future research on differences in TANF programs and migration.

Carolyn Heinrich's dissertation studies the administration and delivery of services by a local JTPA facility. This research evolved from her role as an advisor to a local JTPA service delivery area (SDA) that was implementing a demonstration project to increase the participation of the economically disadvantaged. Ms. Heinrich assisted in the design, implementation, and evaluation of the program. In the process, she observed first-hand the daily operations of an SDA, including the process by which service providers and training professionals select clients and determine their assignments to training programs. This close relationship with a local SDA allowed her to access information and gain insights into the operations of the organization that would not have been possible from a more distant vantage point or with secondary data.

Her case study addressed two basic questions. First, how are participants selected into the program and assigned to services? Second, how effective was the demonstration program in attracting and serving economically disadvantaged job-training eligibles?

The first question confronts the long-standing debate on whether JTPA providers select or "cream" the more highly qualified clients who are eligible for the program. Ms. Heinrich's results suggest that current performance standards encourage "creaming." Creaming is particularly prevalent when budget reductions force providers to cut back on lengthy and expensive training programs and thus incline them to recruit more qualified clients who would do as well finding jobs without the more intensive assistance.

With respect to the success of the demonstration project, Ms. Heinrich found that targeting program funds to a spatially concentrated area increased awareness of and participation in the program, increased involvement by community organizations, and raised earnings gains relative to other JTPA programs. She concludes from these findings that performance standards may be effective management tools, but in the case of JTPA programs, the performance standards should be based on changes in earnings before and after participation in the program and not on the current system of gross, placement-oriented outcomes.

1996 First prize co-winner
Jeffrey Smith , University of Chicago
James J. Heckman, Dissertation Advisor

Jeffrey Smith also focuses on the JTPA system, but the questions he considers extend to the evaluation of social programs in general. In the first two of three essays, he examines the extent to which program participation affects the performance of nonexperimental estimators of program outcomes. Researchers, starting with Ashenfelter, have observed that mean earnings of participants in employment and training programs often decline prior to receiving reemployment services.

Using experimental data from the National JTPA Study, Mr. Smith shows that the earnings dip is transitory, in that earnings of those who did not participate in the program returned to their pre-dip level shortly after being randomly assigned to the study's control group. Consequently, he concludes that estimators of program outcomes based on difference between pre-program and post-program experiences are biased upward, which may call into question the estimates of large positive impacts obtained in early evaluations of federal employment and training programs.

To understand why the characteristics of program participants differ from nonparticipants, Mr. Smith models the participation process in JTPA. He finds that the recently unemployed are more likely to be aware of the JTPA program and seek out its services than those who have been out of the labor force for a time. He concludes, therefore, that the allegation that JTPA case workers "cream" the eligible individuals may simply be the fact that more job-ready applicants are aware of the program and seek to enroll in it.

1996 Honorable Mentions
John Pepper , University of Wisconsin-Madison
Charles F. Manski, Dissertation Advisor

John Pepper tackles a vexing problem that has been at the center of much of the debate regarding welfare reform -- does welfare dependency breed further dependency across generations? Despite the rhetoric and research, definitive answers have not been found for the fundamental reason that it is not possible to observe how children who actually grew up under welfare would have behaved if they had grown up in families not dependent upon welfare.

Mr Pepper explores different scenarios for welfare dependency, suggested by research and the policy debate, to see how they affect the likelihood and length of time that daughters of welfare mothers will themselves be dependent on welfare programs. His approach is informative in that it narrows the range of assumptions that are consistent with the transmission of welfare dependency.

Mark Turner , University of Maryland at College Park
William N. Evans, Dissertation Advisor

Mark Turner also addresses a controversial and politically charged topic regarding the effects of raising the minimum wage on low-wage workers. His research focuses on educational outcomes, by examining the impact of employment on academic achievement among high school students and investigating the effects of the minimum wage on school enrollment.

Mr. Turner's treatment of the latter relationship had direct bearing on the current policy debate. He demonstrates that recent research showing that a minimum wage hike would increase the high school dropout rate rests on a faulty measure of school enrollment. Substituting a more accurate enrollment measure, Mr. Turner finds no impact of the minimum wage on dropout rates. He goes on the show that students who work more than 30 hours per week have lower test scores, lower grade point averages, are more likely to drop out of school, and less likely to enter college.

David Jaeger, now at Hunter College/CUNY Graduate School, wrote three essays that examine the role of educational and skill attainment in the labor market. One of his essays, "Skill Differences and the Effect of Immigrants on the Wages of Natives," addresses whether the influx of immigrant labor during the 1980s has had a significant effect on the wage structure of natives. The increase in immigration is one of several explanations put forth for the decline in real wages among low-skilled workers and for the increase in the wage gap between workers of different education levels during the last two decades. However, previous studies have yielded mixed conclusions. Mr. Jaeger approaches this issue in two steps. He first finds that immigrants and natives within narrow skill levels are nearly perfect substitutes in production. Based on this finding, he then estimates a production function to examine the effects of changes in the supply of immigrants on the wage structure of natives. He finds that immigration accounts for nearly half the decline in real wages among high school dropouts, but contributes little to the growing wage gap between high school and college educated natives. These impacts differ markedly across cities, with immigration into Miami, Los Angeles, and San Francisco having the largest effect on wages.

Mr. Jaeger's second essay, "Degrees Matter: New Evidence on the Sheepskin Effects in the Returns to Education," addresses a problem that has plagued labor economists when estimating the advantage of education on earnings. Most data sets record only years of education, but the number of years of schooling does not necessarily indicate when and if a degree has been obtained. Using a unique data set that has actual records of degrees obtained along with years of education, Mr. Jaeger finds that the estimated diploma effect is significant. For example, a bachelor's degree earns 33 percent more than simply four years of "some college."

1995 Honorable Mentions
Anne Morrison Piehl , Princeton University
David Card, Dissertation Advisor

Anne Piehl, who is now at Harvard University's Kennedy School, confronts the mounting problem of what can be done to curb the increase in criminal behavior, particularly among the lower-educated. As she states in her dissertation entitled Economic Issues in Crime Policy, expanding prison populations and the lingering skepticism about the effectiveness of rehabilitation have led to the development of less-expensive modes of supervision. Professor Piehl's dissertation provides strong evidence that education does reduce criminal behavior. She shows that inmates who have completed an adult basic or high school education program while in prison are significantly less likely to be reincarcerated four years after release. Also, recidivism is lower for those inmates with higher levels of education before entering prison. She also finds that youths with more schooling are less likely to commit crimes and be convicted. In addition, her results suggest that public perception that immigration is associated with higher crime rates is unfounded.

Matthew Slaughter , Massachusetts Institute of Technology
Paul Krugman, Dissertation Advisor

Matthew Slaughter, now at Dartmouth University, investigates whether international trade has contributed to sluggish real wage growth and increasing wage inequality during the last two decades. The title of his first chapter, "International Trade and American Wages in the 1980s: Giant Sucking Sound or Small Hiccup?" reflects the rhetoric and perceptions expressed during the last presidential election. Professor Slaughter finds no support for these concerns. Rather, slow wage growth and rising wage inequality result from sluggish labor-productivity growth in the service sector and skill-biased technological change, domestic factors not foreign ones. His second chapter, "International Trade, Multinational Corporations, and American Wage Divergence in the 1980s," finds that outsourcing by multinationals has little to do with America's wage divergence.