Despite progress made in the conceptualization of labor market discrimination, the sources of discrimination continue to elude economists in their empirical investigations (D'Amico ). The primary reason for this failure is inadequate data, since studying the sources of discrimination demands information on workers' institutional settings. In particular, the examination of employee-based discrimination requires information on coworkers' attributes, in addition to the characteristics of the worker himself. Because of this, direct tests of employee-based discrimination in the modern economy are nearly impossible to make. Most tests must be rather indirect.(1)
Here we attempt to examine the sources of discrimination by focusing on discrimination by fellow employees in an era in which intergroup tensions were important, as they are now. We link two themes in labor economics by analyzing employee-based discrimination as a special example of a compensating wage differential. The data, from the 1889 Michigan Bureau of Labor Annual Report on the furniture industry, are uniquely suited to the analysis of employee-based discrimination because they include information on the individual worker, his firm, and his coworkers for over 5000 workers and because the workers surveyed come from several ethnic groups. Using the data, we measure the magnitude and character of employee-based discrimination as we examine specific aspects of ethnic interaction in the labor market in the late 1800s.
Empirically we find that employee-based discrimination was important in this labor market. Workers of one ethnic group received a higher wage when working with members of other groups. Wages rose by about 0.1 percent when the proportion of coworkers in the worker's own ethnic group fell by one percentage point. This response was greater in small cities, small firms, and among certain ethnic groups including the American-born, Dutch, and Poles. Protestants were generally paid more when working with Catholics. Thus, the wage of the furniture worker was determined not only by his ability to "lathe" but also by "fear" and other feelings toward his coworkers.
We close by attempting to measure the additional labor costs generated by employee-based discrimination within a multi-ethnic work force, concluding that they were probably offset by several benefits which rendered complete segregation unnecessary.
II. MODEL AND DATA
Our model of the labor market unites Becker's  model of employee-based discrimination with Rosen's  hedonic wage model. We assume that employers sell goods in a competitive output market and hire mobile workers in a competitive input market. Following the hedonic wage model, firms make job offers to workers that include both wage and nonwage benefits. A worker chooses the wage-nonwage benefit bundle to maximize utility. The nonwage benefits bundle to which we restrict our attention is the ethnic composition of the firm.(2) As the work force of his firm includes more workers whom he likes (dislikes) the worker's wage should fall (rise).
The data from the Seventh Annual Report of the Michigan Bureau of Labor  include the worker's wage, birth place, father's birth place, and several human capital measures: age, occupation, years of experience in occupation, years with present employer, a literacy proxy, place of residence, and marital status for each individual.(3) (Table I shows the characteristics of the American-born, foreign-born, and sons of immigrants in the sample.) The report is organized by firm, with the name and location of the firm followed by a list of the employees in the firm containing thirty-three pieces of information about each individual.(4) Therefore it is possible to measure the ethnic characteristics of each worker's fellow employees, and the effect of work force composition on the worker's wage.
III. EXPECTATIONS ABOUT EMPLOYEE AND EMPLOYER DISCRIMINATION, 1889
Nativism reached a peak in the United States around 1890, and Michigan was a center of nativist feeling. …