THE impact of unionism on relative wage rates is a topic of long-standing interest in labor economics. Over the years researchers have used a variety of approaches to estimate the wage advantage associated with union membership. Initially, researchers compared the average wages of union and nonunion workers in specific sectors, such as an industry.' This approach has the obvious limitation of not accounting very well for the various quality or productivity differences that exist among workers-differences that command compensating wage payments. More recent studies have used cross-section wage regressions to control for these other determinants of wages and thereby improve estimates of the union wage effect. Using large microeconomic data files containing detailed information on characteristics of individual workers, numerous studies have estimated wage regressions that include controls for worker characteristics (education, experience, etc.) and a dichotomous variable indicating union membership status as explanatory variables.2 The coefficient of the union membership variable in these regressions is taken as an estimate of the average impact that unionism has on wage rates, controlling for measured differences in characteristics among workers. Unfortunately, the wage regression approach does not completely resolve the problem of standardizing the union-nonunion wage comparison for differences in worker quality. Even with detailed microdata on the characteristics of individual workers, it is simply not possible to specify a set of variables that completely captures all worker-specific differences in productive ability. Some aspects of human capital are too subtle to be operationally specified and included in a wage regression, although they are recognized and paid for by employers. To the extent that these unmeasured worker-specific differences are correlated with union membership, the wage regression will incorrectly attribute a relative wage impact to unionism.3 An alternative measurement approach that has several advantages over the wage regression is to compare the wage received by the same worker as a union and nonunion member. In the past, data limitations have prevented researchers from making such a comparison. Since only a small portion of workers change union status in a given period, an unusually large longitudinal file is needed to yield a sufficient number of observed changes for a meaningful analysis.4 Recently, however, large longitudinal data files on individuals participating in the Current Population Survey have become available. This paper uses these data to estimate the impact a change in union status has on the worker' s wage. The

This paper examines the extent of response errors in labor market survey data and explores the implications of such errors for economic analysis. Explicitly examined are responses to questions on industry, occupation, union status, hours worked, and wages. Analyses are based on two sources: (1) a special supplement to the January 1977 Current Population Survey that obtained data from workers and their employers and (2) an exact match of workers and their employers interviewed in the Employment Opportunity Pilot Project Survey. The dual nature of these surveys provides a basis for analyzing the effect of response error on a variety of economic analyses including the trade-off between wages and risk, the wage impact of unionism, and the sensitivity of wage-determination models to alternative responses to earnings questions.