Wage Differentials And Structure Research Paper

1. Earnings And Wages

There are several different ways of defining the wage rate, and the appropriateness of any particular definition depends on the questions in which one is interested. Table 1 shows the distribution of average hourly gross (pretax) wage rates of a random sample of 4,300 full-time workers in the United States in 1995. In principle, one would like to include nonwage compensation (employer contributions for fringe benefits and social insurance programs), but such data usually must be collected from employers rather than from individuals. A worker at the 10th percentile of the distribution earned $6.07 per hour, which means that 10 percent of workers earned less than this amount. One way to summarize the wage structure is to compare the average wage in the highest quintile (from the 80th percentile up) to the average wage in the lowest quintile (from the 20th percentile down). These estimated averages for the US in 1995 were, respectively, $33.15 and $5.84, a ratio of 5.7.

An alternative to the use of hourly wage rates is the use of average weekly, monthly, or annual earnings of individuals. Total earnings per year for an individual are by definition equal to the average wage times hours per year. Data on individual incomes rather than hourly wages are often more readily available in most countries. An example of this is reported in Table 2 in which the ratio of the average annual income of workers in the top quintile to that in the lowest quintile is reported for a selection of countries. The study from which this study was taken includes data for 108 different countries. The dispersion of incomes within a country generally is higher than the dispersion of wages becauseannual hours tend to be higher for persons with higher wage rates. Notice that for the US the ratio of the highest to the lowest quintile in terms of income is 8.5 compared to 5.7 for the wage rates of full-time workers.

Table 2 shows that there is substantial variation among countries in the dispersion of incomes—as there would be if the data referred to hourly wages rather than annual incomes. A few facts are illustrated by these data. First, there tends to be substantially less dispersion in individual wages and incomes within industrialized countries than in the developing countries. Second, there is much greater dispersion of incomes in the US than in other industrialized countries. Third, the degree of wage and income dispersion in the formerly socialist countries is relatively low.

2. The Determinants Of Individual Wages

The hourly wage rate (W ) of any individual worker in a particular economy at a moment in time will, in principle, depend on a set of observable variables (such as that person’s level of completed schooling) as well as on a set of factors that generally are not observed (such as the individual’s innate ability, a characteristic that is difficult to measure and usually not obtained in a census-type surveys). Let X represent all the variables determining wages that are observed and U represent all the unobserved determinants. We can then specify what economists call an earnings function in which the value of W for each individual depends on X and U according to

where ε is a random error term (pure ‘luck’) that is unrelated to the values of X and U.

β is a vector of coefficients representing the effects of each of the observable variables on wages and γ a vector of coefficients representing the effects of the unobserved variables. If, for example, we thought that there were only two observable determinants of W, years of completed schooling (S) and age (A), βX might be specified as β₁ S + β₂ A, and the average effect of an additional year of schooling on W would be β₁ .

In practice most estimated earnings functions feature the replacement of the numerical value of W with its natural logarithm. This means that a unit increase in the value of a particular X variable causes W to increase proportionately by the coefficient on that variable. In the case of years of schooling discussed above, with the semilogarithmic form β₁ represents (δW/δ)/W rather than ∆W / ∆S if W is entered in numerical form.

From the earnings function given by Eqn. (1) it is clear that the wage structure, the distribution of W across the working population, depends on both the joint distributions of X and U (as well as the distribution of ε) and the parameters β and γ. Wage differentials, the ceteris paribus difference between the average wages of individuals with one characteristic versus another, are determined by the β parameters.

The various determinants of individual wages, both potentially observable and unobserved, can be classified into five groups. These include:

(a) skill characteristics, (b) job location,

(c) job characteristics,

(d) discrimination/nepotism, and

(e) rents.

We will discuss the variables within these groups in turn.

2.1 Skill Characteristics

Generally, it is presumed that in most societies more skilled workers are more productive and that they earn higher wages. Two fairly crude but readily observable variables relating to skill of each worker are their years of schooling and potential labor market experience. Some microdata sets allow a more detailed specification of education (what subjects were studied, how many resources were allocated to each year of schooling, etc.) and experience (actual vs. potential years, amount of on-the-job-training, etc.), but this sort of detail is rare.

An important source of differences in skill and productivity differences among individuals is variations in unobserved ability and motivation. Although a few surveys have included the scores on simple IQ tests, these factors are either largely or completely ignored, that is, they are considered U rather than X variables. This illustrates a general problem associated with the estimation of earnings functions. The true model is given by Eqn. (1) above, but the researcher is only able to estimate W as a function of observed variables, that is

where e now represents variation in W that cannot be explained by the observed variables. We know that for most empirical earnings functions e is very important, for the fraction of the variation in W jointly explained by the observable variables is usually fairly small— between 25 and 35 percent (see Juhn et al. 1993).

There is thus a large potential for empirical estimates of the effects on W of any individual observed variable to be biased upward or downward due to the correlation of that variable with unobserved variables. In the very simple model above in which schooling and age are the only observed variables, we want to estimate the ‘true’ effect of S on W from the model W= β₁ S + β₂ A + γU +ε, but we are forced because of data limitations to estimate W = b₁ S + b₂ A + e. The use of b as an estimate of the true effect β₁ is in general subject to specification error which arises if S is correlated with γU (holding A constant). It seems reasonable to suppose that persons with relatively high levels of intelligence and/or motivation (a) will tend to have higher wage rates (i.e., γU > 0) and (b) be more likely to have relatively high schooling levels. This would suggest that b₁ would be an upward-biased estimate of β₁ . Interestingly, however, the consensus of the mass of research into this question (see the review by Card 1999) is that conventional estimates of the effect of schooling on wages are probably not significantly biased in an upward direction.

Another feature of empirical earnings functions is that their coefficients may change over time. This is especially true for coefficients concerning skill variables. For example, in the US the relative hourly wage of 30-year old male workers with a bachelor’s degree (16 years of schooling) to those with just a high school degree (12 years of schooling) fell from 1.18 in 1973 to 1.11 in 1979, but it then rose during the 1980s to a value 1.41 in 1989 (see Bound and Johnson 1992, Table 1). Why these changes in relative wages occurred (in the US and in other countries with relatively flexible wage determination systems such as Canada and Britain) has received a great deal of attention from economists, but the answer is still a matter of controversy.

2.2 Job Location

A second set of observable variables that are usually included in empirical earnings functions are dummy (one zero) variables describing the location of the worker within the country. Typically the locational variables represent the size of the area in which the person lives (whether or not, for example, it is a metropolitan area with a population in excess of a million) and indicators of residence in particular areas. For example, in a hypothetical study of the determinants of wages in the Great Britain, one would include a large city dummy variable (L_large) and separate dummy variables for persons employed in Greater London (L_london), other parts of Southern England (L_south), and Scotland (L_scot). This specification would be represented as W = b₁L_large + b₂ L_london + b₃ L_south + b₄ L_scot +…with the other observable determinants of wages (the X’s in Eqn. (2)) included. The estimated coefficient b in this earnings function would represent the average ceteris paribus wage differential between a person working in London and another working in a large city in Northern England. If, as is generally the preferred approach, W is entered as a natural logarithm, the estimated coefficient b would reflect the approximate proportionate wage differential associated with employment in London.

What is the economic interpretation of coefficients on location of employment? First, because of higher land prices, it is generally more expensive to live in areas with a highly concentrated population. In order for real wages (nominal wages divided by the price level) to be equal across areas, observed wages must be appropriately higher in expensive areas. This would apply especially to very large cities within individual countries—London, Paris, New York, Tokyo, etc.

A second reason for the existence of wage differentials by job location is the preferences of workers concerning the characteristics of different areas. For example, living in a large urban area involves greater stress and poorer air quality than living in a rural area, so, holding the cost of living constant, the typical worker might have to be paid a compensating wage differential to work in the large urban area. Similarly, some regions have better weather and other amenities (like public services) than others, and one would expect that wages would be higher in the more desirable than in the less desirable areas in order for potential migrants to be indifferent between different areas. For the US, the estimated effect on wages of an area having sunshine 80 percent of the time vs. only 30 percent is, as simple economic theory would predict, a reduction of about 7 percent (Blomquist et al. 1988).

This, of course, presumes that all (or most) workers in the economy have the same preferences for the amenities of different areas. To the extent that preferences differ across the population—some people prefer rural areas (warm weather) to large cities (cold), others the opposite—predictions concerning compensating wage differentials become somewhat less clear.

A final reason for the existence of wage differentials across locations is the effect of changes in the structure of labor demand. Suppose, referring to the example above, that there was for some reason a major boom in Scotland that caused the demand for labor there to increase permanently by 50 percent. It is unlikely that there would be an immediate large migration of workers from England to Scotland, so wages in Scotland would be bid up significantly. The earnings function for Great Britain based on data collected after the demand shift would yield an estimated coefficient b_scot that would now be positive. How rapidly the Scottish/English relative wage would return to one (the value of b_scot to return to approximately zero) depends on how long it takes for a portion of the British population to relocate to Scotland.

2.3 Job Characteristics

In addition to the location of employment, an important set of determinants of wage rates is the attributes of the job held by each individual worker. We can arbitrarily define each job attribute (the set of which we will call A) as a bad—the degree to which it is physically onerous, monotonous, a threat to the worker’s health and safety, has low social status, etc. It would be expected that higher values of the elements of A would require a compensating differential. Put differently, holding all other (observed and unobserved) determinants of wages constant, the market equilibrium wage of each job would be higher for ‘bad’ jobs (those with high values of A) than for ‘good’ jobs (those with low values of A). Thus, for the specification W = b₁ A₁ + b₂ A₂ + … , where the A_is are numerical representations of bad characteristics (like the injury rate of the job), we would expect to estimate positive b_is.

There are numerous examples of the operation of compensating differentials. (For an excellent introduction to the topic, see Chapter 8 of Ehrenberg and Smith 2000.) On the other hand, there are many instances in which the estimated effects of certain job attributes tend to be zero or have the ‘wrong’ sign (from the point of view of the theory) (see Brown 1980 for a refreshingly honest discussion of these problems).

An important set of applications of estimates of job amenities on wages is to the question of the economic value of life. If one of the As (say A₁ ) is the rate of accidental death associated with the occupation of the worker, then one can infer from the estimated value of δW/δA₁ how much individuals value—apart from the opportunity to earn and consume—place on the continuation of their lives. This information can, in principle, be used to do cost benefit analyses of programs to reduce risks to life and health.

An example of wage differentials associated with differences in job characteristics is given in Table 3. These results illustrate some of the difficulties in the interpretation of the effect of job attributes. The sample refers to individuals working full-time in the US in 1995 who had received a Ph.D. in science or humanities. The regression equation relates the annual salary of each respondent to years since receipt of the doctorate, gender, race, and region as well as the field of receipt of the Ph.D. and the current type of employment.

The range of field effects (with economics equal to 100) is from 73 for humanities to 105 for specialists in computer science. This means that, holding gender, race, years of experience, type of employment, and region constant, the computer /humanities relative salary is 1.44. There are also substantial compensation differences across employer types. For example, Ph.D.s employed in the for-profit business sector earn, ceteris paribus, 29 percent more than those in tenure track academic positions.

There are various explanations of why relative compensation differentials of this magnitude exist—in this case among sets of individuals who have the same quantity of education. First, these differentials may reflect compensating differentials in the sense that some fields (economics, for example) are intrinsically boring relative to other fields (like humanities) and require higher monetary rewards in order to induce persons to study them. Some of the relative compensation levels by employer type—for example, the business/academia comparison—most likely reflect premiums that are necessary to get Ph.D.s to work in jobs that are perceived to be less rewarding in a nonfinancial sense.

A second potential explanation of differentials of this sort is that they reflect, in part, average ability differences of the individuals who are in each field or employment type. This is especially relevant for comparisons of employment type in which there is partial sorting on the basis of ability. For example, the compensation difference between academics in tenure track positions and those in both nontenure track positions and lower education (secondary schools and 2-year colleges) is very likely to reflect practices by which the ‘better’ Ph.D.s are granted tenure.

A third explanation of these compensation differentials is that some of them reflect temporary market factors, and there is a presumption that the high relative compensation of the computer science field in Table 3 is an example of this. This phenomenon is illustrated theoretically in Fig. 1. The demand for a particular type of labor is assumed to increase such that the demand curve shifts from D` to D“. In the short run the supply curve, S<sub>SR</sub>, is quite inelastic, reflecting the fact that the pool of persons qualified to work in this market is fixed. (In the case of computer scientists, some additional labor can be obtained from people not on the labor force or from closely related fields, but these possibilities are limited.) Thus, the short-run effect of an increase in the demand for this type of labor is to increase its relative compensation level. In the long run the supply of labor to this market is very or completely elastic, for new entrants to the labor force can choose to train for this type of work. (In the case of computer specialists, young persons are induced to get Ph.D.s in that field rather than in mathematics or physics.) Over time, therefore, the relative compensation level in this market would be expected to return to its initial equilibrium value, C`, rather than stay at its temporary value, C“ in Fig. 1.

Analysis of wage differentials of various sorts can proceed along the above lines. A major example of this is the analysis of average compensation across industries. An example of this is Table 4 in which average monthly salaries across broad industrial groups are reported for six rather diverse countries. Wages in agriculture, trade, hotels and restaurants, and personal services tend to be relatively low; wages in mining, utilities, and finance tend to be relatively high. Some of these differentials reflect quality and compositional differences (that are not adjusted for in the table). But some of the differentials reflect differences in the amenities associated with work in the particular industries.

2.4 Gender And Ethnicity

A set of observable variables that are generally included in earnings functions are the demographic characteristics of the workers—his/her gender, race, or ethnicity. For example, the inclusion of a one/zero dummy variable for women as one of the Xs in Eqns. (1) and (2) is very common. For the theoretically complete model given by Eqn. (1), which includes all observed and unobserved influences on wage rates, the interpretation of the (presumably negative) coefficient on women must be in terms of some sort of labor market discrimination against women. Why else would otherwise identical workers receive lower compensation? The same conclusion applies to conclusions concerning potential dummy variables on race and ethnicity for earnings functions in probably every country in the world (blacks in the US, French speakers in Quebec, Arabs in France, natives of Greenland in Copenhagen, etc.).

The problem with attributing the negative effects of gender and certain racial and ethnic distinctions entirely to labor market discrimination is that we are only able to estimate the model for which we have data, Eqn. (2) rather than the theoretically complete Eqn. (1) above. Thus, there is the possibility that the negative coefficients on gender and ethnicity reflect, wholly or in part, differences in the average value of omitted variables rather than some form of direct labor market discrimination. Further, it is necessary to know exactly what is the cause of an observed wage differential if a society decides it wants to eliminate this differential (see Altonji and Blank 1999 for an extensive discussion of these issues).

2.5 Rents

Most of the above discussion proceeded as if wages are determined in markets that are similar in process to the markets for melons and equities. The fact that the commodities involved in labor markets are human beings requires that this ‘model’ be modified to some extent. Indeed, a set of institutions has emerged to provide modification of the outcomes that would follow from the unfettered equilibrium of the labor market. Among these institutions is trade unionism, which has, over time and across societies, taken several forms. Government policy—both directly through legislation determining wages and indirectly through its role as employer—also affects the wage structure.

At one level the existence of institutions such as trade unions creates another wage differential in which to be interested. In terms of Eqns. (1) and (2), being a union member or working in which wage rates are determined by a collective bargaining contract means that a worker could earn a higher wage than another worker who is not so situated. In this case, W = β₁ U + … is the complete wage equation, including all observed and unobserved variables, where U equals one with union representation and equals zero otherwise. We do not, of course, observe everything about the workers and his job, so we estimate the equivalent of Eqn. (2), W = b₁ U + …

How good the estimator b₁ is of the ‘true’ effect of unionism β₁ depends, as in the estimation of other wage differentials, on the correlation of the omitted determinants of wages with U. In the US, the consensus estimate of the average effect of unionism on wages is about 15 percent (see Lewis 1986).

A more important effect of unionism and government policy is their influence on other wage differentials. Much of the above discussion of the structure of wages and of wage differentials reflected the implicit assumption that relative wages are, at least in the long run, free to adjust to their market-clearing levels. This assumption is all right for the analysis of labor markets in the US, Japan, and (during the past 20 years) the United Kingdom, but there are serious problems associated with its application to most other industrialized countries in which bargaining coverage is over 75 percent (vs. 18 percent in the US and 22 percent in Japan in the early 1990s). To the extent that union and/or government policy attempts to prevent increases in the dispersion of earnings, wage differentials will not behave as neatly as the elementary theory suggests. Instead, wage structures will be subject to a great deal of inertia, and they will adjust at best very slowly to shocks in demand.

That the wage structure is slow to change is not per se a bad thing. Indeed, I would join most economists in agreeing that stability of the distribution of income is a good thing. The problem with such policies, however, is that they often lead to heavy unemployment of those groups whose relative wages are protected.

Bibliography:

1. Altonji J G, Blank R M 1999 Race and gender in the labor market. In: Ashenfelter O, Card D (eds.) Handbook of Labor Economics, Vol. 3C. Elsevier Science, Amsterdam, pp. 3143– 3260
2. Blomquist G, Berger M, Hoehn J 1988 New estimates of the quality of life in urban areas. American Economic Review 78(1): 89–107
3. Bound J, Johnson G 1992 Changes in the structure of wages in the 1980s: An evaluation of alternative explanations. American Economic Review 82(3): 371–92
4. Brown C 1980 Equalizing differences in the labor market. Quarterly Journal of Economics 94: 113–34
5. Card D 1999 The causal effect of education on earnings. In: Ashenfelter O, Card D (eds.) Handbook of Labor Economics, Vol. 3A. Elsevier Science, Amsterdam, pp. 1801–63
6. Deininger K, Squire L 1996 A new data set measuring income inequality. The World Bank Economic Review 10(1): 565–91
7. Ehrenberg R, Oaxaca R 2000 Modern Labor Economics, 7th edn. Addison-Wesley, Reading, MA
8. International Labour Office 1999 Yearbook of Labour Statistics 1999. ILO, Geneva
9. Juhn C, Murphy K M, Pierce B 1993 Wage inequality and the rise in returns to skill. Journal of Political Economy 101(1): 410–42
10. Lewis H G 1986 Union Relative Wage Effects: A Survey. University of Chicago Press, Chicago