We examine inference in panel data when the number of groups is small, as is typically the case for difference-in-differences estimation and when some variables are fixed within groups. In this case, standard asymptotics based on the number of groups going to infinity provide a poor approximation to the finite sample distribution. We show that in some cases the t-statistic is distributed as t and propose simple two-step estimators for these cases. We apply our analysis to two well-known papers. We confirm our theoretical analysis with Monte Carlo simulations.

Studies of the earnings of union workers have consistently shown that they earn considerably more than nonunion workers.This paper considers whether part of this observed union/nonunion differential is due to unions organizing high paying primary sector jobs.We extend our earlier work on the dual labor market in which we used an unknown regime switching regression to identify two labor market sectors --a high wage primary sector and a low wage secondary sector.Here we estimate a model where worker's wages are determined by one of three wage equations: a union wage equation, a nonunion primary equation or a nonunion secondary equation.If individuals are in the union sector their sector is treated as known.If they are not then their sector is treated as unknown.Parameter estimates for this model suggest that union/nonunion differences are very large for average workers even when comparing union and nonunion primary workers.We continue to find distinct primary and secondary sectors with wage equations similar to those that would be expected from the dual market perspective.Since it appears that union workers may be receiving large wage premiums it seems likely that there is non-price rationing of union jobs.If there is, our finding in previous papers of non-price rationing of primary sector jobs may have been due only to the rationing of union jobs.We test for the existence of non-price rationing of nonunion primary sector employment in this three sector model and continue to find evidence that at least black workers find it difficult to secure primary sector employment.

A simple econometric model of investment in schooling is developed and estimated. The measure of individual discount rates implicit in their educational investment decisions suggests no difference between individuals from different socioeconomic backgrounds. Differences in individual speeds of educational attainment, which do vary with background, explain most of the variation in levels of attainment that is attributable to family background.

Using a model of statistical discrimination and educational sorting, we explain why blacks get more education than whites of similar cognitive ability, and we explore how the Armed Forces Qualification Test (AFQT), wages, and education are related. The model suggests that one should control for both AFQT and education when comparing the earnings of blacks and whites, in which case a substantial black-white wage differential emerges. We reject the hypothesis that differences in school quality between blacks and whites explain the wage and education differentials. Our findings support the view that some of the black-white wage differential reflects the operation of the labor market.

Much nonmanagerial work is routine, with all workers having similar output most of the time. However, failure to address occasional challenges can be very costly, and consequently easily detected, while challenges handled well pass unnoticed. We analyze job assignment and worker monitoring for such “guardian” jobs. If monitoring costs are positive but small, monitoring is nonmonotonic in the firm’s belief about the probability that a worker is good. The model explains several empirical regularities regarding nonmanagerial internal labor markets: low use of performance pay, seniority pay, rare demotions, wage ceilings within grade, and wage jumps at promotion.

Using Italian Social Security records for male workers from a sample of firms in Turin from 1981 to 1983, we show that conditional on the worker's own wage the average wage in the establishment for similar workers is negatively related to quits. We also find that this variable predicts future wage growth. This is consistent with an economic model in which workers compare the longrun value of employment opportunities when making quit decisions.

We provide microfoundations for dynamic skill formation with a model of investment in multiple skills, when jobs place different weights on skills. We show that credit constraints may affect investment even when workers do not exhaust their credit. Firms may invest in their workers’ skills even when there are many similar competitors. Firm and worker incentives can lead to overinvestment. Optimal skill accumulation resembles—but is not—learning by doing. An example shows that shocks to skill productivity benefiting new workers but lowering one skill’s value may adversely affect even relatively young workers, and adjustment may be discontinuous in age.

The Metropolitan Council for Educational Opportunity (Metco) is a desegregation program that sends students from Boston schools to more affluent suburbs. Metco increases the number of blacks and reduces test scores in receiving districts. School-level data for Massachusetts and micro data from a large district show no impact of Metco on the scores of white non-Metco students. But the micro estimates show some evidence of an effect on minority third graders, especially girls. Instrumental variables estimates for third graders are imprecise but generally in line with ordinary least squares estimates. Given the localized nature of these results, we conclude that peer effects from Metco are modest and short lived.

According to dual labor market theory, the labor market can be usefully described as consisting of two sectors: a high-wage (primary) sector with good working conditions, stable employment, and substantial returns to human capital variables such as education and experience, and a low-wage (secondary) sector with the opposite characteristics. Moreover, primary jobs are rationed, that is, not all workers who are qualified for primary sector jobs and desire one can obtain one. Finally, the sector of the labor market in which an individual is employed directly influences his or her tastes, behavior patterns, and cognitive abilities. Thus the dual labor market model or, more generally, segmented labor market models, is simultaneously a description of the income distribution, a claim about the absence of market clearing, and a radical departure from the standard neoclassical assumption of fully rational actors and exogenously determined preferences. While this last element is potentially the most interesting, even its proponents fail to give it the attention it deserves, and related work has not been incorporated into the segmented labor market model. In this paper, we therefore concentrate on the first two elements of the model. Segmented labor market theory was sufficiently popular in the late 1960's and early 1970's to be taken seriously by prominent mainstream labor economists. However, two influential and largely negative reviews (Glenn Cain, 1976; Michael Wachter, 1974) portrayed the segmented labor market hypothesis as largely atheoretical and based, at best, on questionable statistical analysis. It seems fair to say that even sympathetic mainstream critics felt that key insights from the segmented labor model could be incorporated into neoclassical analysis and that the remaining elements of the model did not form a sufficiently coherent theory to pose a challenge to the neoclassical model. Whatever the merits of this perception, it is clear that advocates of the segmented labor market approach did not develop a formal theory which conformed to the standards of mainstream economists. With some notable exceptions (Michael Piore, 1975; David Gordon, 1972), the work was atheoretical. Moreover, the empirical methods used tended to fall outside the norm (for example; interviews, observational studies, and historical and institutional analysis). Advocates of the segmented labor market perspective, mostly radical political economists, chose instead to develop their own research program outside the mainstream. The reemergence of segmented labor market theory is linked with the reversal of these two tendencies. The theory has been pursued by economists using modern tools of imperfect information theory and state-of-the-art econometrics. As a result, the approach has again attracted the attention of the mainstream. Even a few years ago, it would have been a clairvoyant observer who predicted that Lawrence Summers would be working on a theoretical model of labor market duality (with Jeremy Bulow, 1986), that Robert Solow would count among his recent work a dual market model (with Ian McDonald, 1985) and that James Heckman would publish an article in which he undertook an empirical test of a dual market model, failed to reject the model, and then devoted much of the rest of the article to attacking his and other tests of the dual labor market view (see his article with V. Joseph Hotz, 1986). Since the theoretical developments are largely associated with efficiency wage and *Departments of Economics, University of California, Berkeley, CA 94720 and NBER, and Boston University, 270 Bay State Road, Boston MA 02215 and NBER, respectively. This study was supported in part by NSF grant no. SES-8606139. Lang acknowledges support from a Sloan Faculty Research Fellowship; Dickens acknowledges support from the Institute of Industrial Relations at Berkeley.

We develop an estimator that allows us to calculate an upper bound to the fraction of unrejected null hypotheses tested in economics journal articles that are in fact true. Our point estimate is that none of the unrejected nulls in our sample is true. We reject the hypothesis that more than one-third are true. We consider three explanations for this finding: that all null hypotheses are mere approximations, that data-mining biases reported standard errors downward, and that journals tend to publish papers that fail to reject their null hypotheses only when the null hypotheses are likely to be false. While all these explanations are important, the last seems best able to explain our findings.