response to a negative shoclc to labor demand, a firm that views the cost of labor as being given by the contract wage reduces employment a great deal; in terms of the figure, it reduces employment from La to Lb- The marginal product of labor now exceeds the opportunity cost of workers time. Thus when the firm and the workers negotiate a new contract, they will make sure that employment is increased; in terms of the diagram, they will act to raise employment from Lb to Ic Thus if the wage determines employment (and if shocks to labor demand are the main source of employment fluctuations), changes in employment during contracts should be partly reversed when new contracts are signed.

To test between the predictions of these two views, Bils examines employment fluctuations in U.S. manufacturing industries. Specifically, he focuses on twelve industries that are highly unionized and where there are long-term contracts that are signed at virtually the same time for the vast majority of workers in the industry. He estimates a regression of the form

Alnl,f = a, - </.Z,f - e(lnl,f-i - lnI,,f io) + FD,f + f,,f. (10.86)

Here z indexes industries, L is employment, and is a dummy variable equal to 1 in quarters when a new contract goes into effect in industry i. The key variable is Z,,t. If a new contract goes into effect in industry z in quarter t (that is, if D,,f = 1), Z,,f equals the change in log employment in the industry over the life of the previous contract; otherwise, Z,,f is zero. The parameter therefore measures the extent to which employment changes over the life of a contract are reversed when a new contract is signed. Bils includes In I;,t-i -In I,.t-io to control for the possibiUty that employment changes are typically reversed even in the absence of new contracts; he chooses f - 10 because the average contract in his sample lasts ten quarters. Finally, allows for the possibility of unusual employment growth in the first quarter of a new contract.

Bilss estimates are = 0.198 (with a standard error of 0.037), = 0.016 (0.012), and = -0.0077 (0.0045). Thus the resuhs suggest highly significant and quantitatively large movements in employment related to the dates of new contracts: when a new contract is signed, on average 20% of the employment changes over the life of the previous contract are immediately reversed.

There is one puzzling feature of BUss results, however. When a new contract is signed, the most natural way to undo an inefficient employment change during the previous contract is by adjusting the wage. In the case of the fall in labor demand shown in Figure 10.8, for example, the wage should be lowered when the new contract is signed. But Bils finds Uttle relation between how the wage is set in a new contract and the change in employment over the life of the previous contract. In addition, when he looks across industries, he finds essentially no relation between the extent to which employment changes are reversed when a new contract is signed and the extent to which the wage is adjusted.

Interindustry Wage Differences

The basic idea of efficiency-wage models is that firms may pay wages above market-clearing levels. If there are reasons for firms to do this, those reasons are unlikely to be equally important everywhere in the economy. Motivated by this observation, Dickens and Katz (1987a) and Krueger and Summers (1988) investigate whether some industries pay systematically higher wages than others.-*

These authors begin by adding dummy variables for the industries that workers are employed in to conventional wage regressions. A typical specification is

; • . M N

Inw, =a+Y. PjXij + X TfcAk + £i, (10.87)

J=l k=\

where w, is worker zs wage, the X, js are worker characteristics (such as age, education, occupation, and so on), and the D, s are dummy variables for employment in different industries. In a competitive, frictionless labor market, wages depend only on workers characteristics and not on what industry they are employed in. Thus if the Xs adequately capture workers characteristics, the coefficients on the industry dummies will be zero.

Dickens and Katzs and Krueger and Summerss basic finding is that the estimated y,s are large. Katz and Summers (1989), for example, consider wage differences among U.S. workers in 1984 across nvo-dzgzf industries. Since Katz and Summers consider a sample of over 100,000 workers, it is not surprising that they find that most of the ys are highly significant. But they also find that they are quantitatively large. For example, the standard deviation of the estimated ys (weighted by the sizes of the industries) is 0.15, or 15%. Thus wages appear to differ considerably among industries.

Dickens and Katz and Krueger and Summers show that several possible explanations of these wage differences are contradicted by the data. The estimated differences are essentially the same when the sample is restricted

3"See Katz and Summers (1989, pp. 216-247) for a summary of this literature. Groshen (1991) examines wage differences among firms within industries.

Two-digit industries refers to the Standard Industrial Classification (or SIC). One-digit industries are very broad industries, such as durable goods manufacturing, communications, and retail trade. Two- digit industries are narrower classifications within these broad groups; for example, two-digit industries within durable goods manufacturing include furniture and motor vehicles. Three-, four-, and five-digit Industries are even finer distinctions.

Bits suggests two possible explanations of this finding. One is that adjustments in compensation mainly take the form of changes to fringe benefits and other factors that are not captured by his wage measure. The second is that employment determination is more complex than either of the two views we have been considering.

-See, for example, Murphy and Topel (1987b); Hall (1989); and Topel (1989).

to workers not covered by union contracts; tlius they do not appear to be the resuh of union bargaining power. The differences are quite stable over time and across countries; thus they are unUkely to reflect transitory adjustments in the labor market (Krueger and Summers, 1987). When broader measures of compensation are used, the estimated differences typically become larger; thus the results do not appear to arise from differences in the mix of wage and nonwage compensation across industries. Finally, there is no evidence that working conditions are worse in the high-wage industries; thus the differences do not appear to be compensating differentials.

There is also some direct evidence that the differences represent genuine rents. Krueger and Summers (1988) and Akerlof, Rose, and Yellen (1988) find that workers in industries with higher estimated wage premia quit much less often. Krueger and Summers also find that workers who move from one industry to another on average have their wages change by nearly as much as the difference between the estimated wage premia for the two industries. And Gibbons and Katz (1992) consider workers who lose their jobs because the plants they are working at close. They find that the wage cuts the workers take when they accept new jobs are much higher when the jobs they lost were in higher-wage industries.

Two aspects of the results are more problematic for efficiency-wage theory, however. First, although many competitive explanations of the results are not supported at aU by the data, there is one that cannot be readily dismissed. No wage equation can control for all relevant worker characteristics. Thus one possible explanation of the finding of apparent interindustry wage differences is that they reflect unmeasured differences in ability across workers in different industries rather than rents.

To understand this idea, imagine an econometrician studying wage differences among baseball leagues. If the econometrician could only control for the kinds of worker characteristics that studies of interindustry wage differences control for-age, experience, and so on-he or she would find that wages are systematically higher in some leagues than in others: major-league teams pay more than AAA minor-league teams, which pay more than AA minor-league teams, and so on. In addition, quit rates are much lower in the higher-wage leagues, and workers who move from lower-wage to higher-wage leagues experience large wage increases. But there is little doubt that large parts of the wage differences among baseball leagues reflect ability differences rather than rents. Just as an econometrician using Dickens and Katz and Krueger and Summerss methods to study interleague wage differences in basebaU would be led astray, perhaps econometricians studying interindustry wage differences have also been led astray.

Several pieces of evidence support this view. First, if some firms are paying more than the market-clearing wage, they face an excess supply of workers, and so they have some discretion to hire more able workers. Thus it would be surprising if at least some of the estimated wage differences did