Click for next page ( 70


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrimination INTRODUCTI ON We have presented evidence indicating that men and women tend to be employed at different jobs and that jobs held mainly by women tend to pay less than jobs held mainly by men, even when account is taken of differences in the complexity and difficulty of jobs and the qualifi- cations and experience of incumbents. Insofar as such pay differences result from the concentration of men and women in different fines, the issue is not one of pa' equity but one of equality of access to firms. When women are concentrated disproportionately in the low-paying jobs within a firm, however, the question anses: can differences In the average pay of men and women be accounted for entirely by differences in their access to or preference for high-paying and low-pay~ng jobs, or are the pay rates influenced by the sex composition of jobs? In the latter case the remedy would be to adjust the pay rates of jobs so as to remo`,e what would be considered the discriminatory component. Such a pro- cedure requires, however, the development of a means for identifying whether and what portion of pay differences in jobs within a finn are .. . . alscnm~natory. One approach to unraveling the components of pay differentials is to measure the worth of jobs directly, using any of a number of job eval- uation procedures. The concept of comparable worth is that jobs of equal worth should be paid equally. The demand and supply of particular skills and similar factors are regarded as legitimate bases of pay differ 69

OCR for page 69
70 WOMEN, WORK, AND WAGES ences among jobs only insofar as such factors are explicitly included in the formulas for specifying job worth.' In this approach, instances of possible discrimination in pay are identified by using the job worth hierarchy resulting from application of the job evaluation plan as a standard against which to assess actual pay rates. Acceptance of a comparable worth approach-the attempt to measure the worth of jobs directly on the basis of their content~oes not require an absolute standard by which the value or worth of all jobs can be measured. In the judgment of the committee, no such standard exists nor, in our society, is likely to exist. The relative worth of jobs reflects value judgments as to what features of jobs ought to be compensated, and such judgments typically vary from industry to industry' even from firm to firm. Paying jobs according to their worth requires only that whatever characteristics of jobs are regarded as worthy of compensation by an employer should be equally so regarded irrespective of the sex, race, or ethnicity of job incumbents. The use of job evaluation procedures without modification may not always be appropriate, however, because some job evaluation plans are designed in such a way that they reproduce whatever biases exist in the pay practices current when the plans were introduced. Moreover, not all employers who use job evaluation plans base pay rates solely on the job worth hierarchy implied by the job evaluation plan in use; in such cases inferences regarding possible discrimination must rest not on spe- c'fic instances of underpayment of jobs relative to their "worth" but on a pattern of underpayment for jobs held mainly by women or minorities. To cope with these complications, we explore two modifications to the conventional job evaluation approach: a multiple regression approach that includes "percent female" among the variables used to predict pay rates, and the use of pay rates of white men in jobs held mainly by white men as a standard of equitable pay. These approaches have potential value in developing bias-free job evaluation plans and in identifying instances of discrimination in pay. In this chapter we first review the conventional job evaluation ap- proach and the two modifications, commenting on their strengths and weaknesses. We then demonstrate the use of these procedures to correct discnminatory pay rates. Throughout we suggest a number of areas in which additional research is necessary before definitive conclusions can be reached or unqualified recommendations offered regarding the de- sirability of adopting any of the procedures renewed. ~ For example, Ocular Wills may to in And because of the se~css of a tcch- Dology; if so, a job factor indicating the accused of ~DoIOgy for each job could be "deaf lo the job c~luadon pled used ('cc Remit, 1978~.

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrimination 71 CONVENTIONAL JOB EVALUATION APPROACHES TO ASSESSING PAY RATES METHODS OF JOB EVALUATION At the present time in the United States many large private oompa- nies, the federal government, and many state governments make use of some form of formal job evaluation as an aid to establishing pay rates for jobs. Although job evaluation systems differ in details of design and implementation, almost all conform to a common methodology and underlying logic: all the jobs in the unit being analyzed (firms, division, or other)2 are described; the descnptions are then rated or evaluated according to one or more "compensable factors" (features defined as legitimate bases for pay differentials); the ratings are added in some way to create a total score, sometimes called a job worth score; and the scores are use~sometimes alone and sometimes with other informa- tio~to assign the jobs to pay classes.3 Particularly in large firms with large internal labor markets, it is not always evident how to set pay rates for different jobs. Even when a firm is committed tO paying the going rate in the local labor market for each job, such information is not always available, and, as we note in Chapter 3, many jobs are filled entirely from within a firm, so that there is no going rate to apply. Moreover, a large firm may be the only or the largest employer in town and hence be in the position of defining the going rate for man, jobs. In these contexts, job evaluation systems prove useful as a was of setting pay rates. One type of job evaluation plan, known generically as the point method or factor point method, is used in many organizations and is therefore used to illustrate this discussion. In this type of job evaluation plan a set of compensable factors is chosen. For each factor a scale is devised representing increasing levels of "worth," and each octet is assigned a given number of points. Each job is rated on each factor separately, then assigned the corresponding number of points for the rated level on each factor. The points are totaled to yield the job worth score. 2 "Other" may be an entire industry; a job c~alustion plan for production jobs in the tool industry is such an example. While for con~cnicocc our delusion usually refers to "firms," it should be understood as applicable to all compensation Items, however they may be organized. 3 TIC properties of cresting job evaluation plans arc described ~ more detail us the oo~ttcc's intenm report to the Equal Employment Oddity ~ on (Tin, 1979~.

OCR for page 69
72 WOMEN. WORK. AND WAGES Many of the factor point job evaluation systems in use today were developed by using a firm's existing pay structure to statistically deter- mine which attributes of jobs best predict their pay rates. In this ap- proach a set of factors that is thought likely to be related to existing pay differences among jobs is identifie~factors representing differences in skill, effort, responsibility. and working conditions. Each job is scored on each of the factors. These factors are then used to predict existing pay rates (usually via the statistical technique of multiple regression analysis), and those factors contributing substantially to the prediction are included in the job evaluation plan q with weights proportional to their contribution. The factors and factor weights can then be used to assign pay rates for new jobs and to adjust the pay rates of existing jobs that are overpaid or underpaid relative to the predictions of the formula. This method provides an empirically derived underlying structure with which the pay rates of all jobs in a firm can be brought into conformity. This is sometimes called a "policy-captur~ngq' approac~the implicit policy underlying the existing pay system is made explicit. Job evaluation plans developed in this way necessarily produce hierarchies of job worth that are closely related to existing pay hierarchies: that is what they are designed to do. Another way to develop a factor point job evaluation system is to define a priori a set of factors and factor weights that expresses what the employer believes are legitimate bases of pay differentials.4 In both methods of developing factor point systems job worth is defined by the factors that measure it, and jobs are assumed to deserve equal pay if they have equal total scores on the job worth scale. In both of these methods, completely different kinds of jobs may have equal total job worth scores and hence be regarded as deserving equal pay. In a set of jobs with equal total scores' one job. for example, may entail great responsibility, while another job may require great skill. The major purpose of job evaluation systems, however designed, is to make the content of different kinds of jobs commensurable for the purpose of determining pay rates. While this is difficult to do, and we discuss a number of the difficulties below, we do not believe it is im- possible in principle. In our discussion of job evaluation we accept the criteria of worth implicit in each pla~"worth~' being determined by those features of jobs that are identified and measure~and focus on "'Employer" as used here refers to Shocker sets policy for a firm, regardless of how policy decisions are made. Decisions regarding c01npcasation and job evaluation plans may wee be the outgrowth of umon-management negotiations or the work of committees that include members drawn from all k~rcis within a firm.

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrimination 73 ways to improve the measurement of jobs according to these cnter~a. Although we considered the potential use of other criten~for example, productivity-we felt it would be most useful to concentrate on the enters that are currently used.5 Because formal job evaluation plans purport to measure the worth of jobs in the precise sense of measuring the factors that are regarded by employers as legitimate bases of pay differentials among job~it has been suggested that such plans can be used to arrive at objectively fair pay rates for jobs and thus to resolve charges of pay discrimination based on sex, race, or ethnicity. Specifically, actual pay rates could be compared with the pay rates that would prevail if jobs were paid ac- cording to their job worth scores, and the difference taken as a measure of possible discrimination. Furthermore, if the difference between actual and predicted pay rates were shown to be correlated tenth the sex, race, or ethnic composition of occupational categories, then a strong suspicion would be created that pay practices are discriminatory, and appropriate action could be taken. That is. if the jobs with actual wage rates lower than indicated by the job worth scores tended to be held dispropor- tionately by women or minorities. and if the jobs with actual wage rates higher than indicated by the job worth scores tended to be held by men or by nonminor~ties then the existence of discriminatory pay practices would be implied. In the judgment of the committee, four aspects of formal job eval- uation procedures are important in assessing their practical application in developing pay plans and in resolving complaints of pay discnmina .. . . . . . . ~ . . . . . . . tlon, particularly In a labor force Highly segregated by sex, race, and cthnicity. (The first three of these features are reviewed in greater detail in the committee's interim report; see Treiman, 1979.) First, the ranking ' jibe committee also concluded that with rare conceptions the difficulties of mcasu~g productivity arc very great. To be useful in job evaluation the relative average productivity (the contribution to output) of each job would have to be measured. For example, to assess the productivity of hospital workers the relative oontabution of DUrSCS, motors, orderlies, xactancs, and maintenance workers to hospital output would bevc to be detennined. At the present time, however, there is no agreement on bow to m~surc hospital output, and little attempt has been made to assist the relative contnbutions of canons hospital workers (Scott, 1979~. While an employer may mat to Jude produc- d~qty measures in job evaluation plans when they become available, the factors Erectly identified in job evaluation plans remeet employers' conceptions of what Would be rc- muncrated. For example, some plans Nymphs slcill more this rcspon~bility; some do the opposite. Presumably these diffcrcaces reflect, at least in part, employers' judgments about what cohanecs the perfoln.ance of their firms. Further research on the mcasuremcst of the productivity of jobs and research OR the ~'poradon of productivity measure job evaluation plans arc advisable.

OCR for page 69
74 WOMEN. WORK. AND WAGES of jobs tends to be highly dependent on which factors are used in the evaluation and how heavily each factor is weighted. And as we have already noted, the principal method for deriving factor weights in most currently used job evaluation plans pegs them to current pay rates, therby reflecting existing pay differences between men and women (and between minority and nonminority) workers. Second like other meth- ods of establishing pay rates, job evaluation involves judgments, making it possible for well-known processes of stereotyping to result in an un- dervaluation of jobs held mainly by women. Third, many employers use more than one job evaluation plan in their fimns (e.g., one for shop jobs and one for office jobs), a policy that makes it difficult to compare the worth of jobs or to determine the likelihood of discrimination in pav- among different sectors of a firm. Fourth' the interpretation of the differences between actual and predicted pay rates as evidence of dis- cnmination requires strong assumptions regarding the adequacy of the prediction model and the measurement of the variables included in the modeWassumptions that may be difficult to satisfy in practice. Each of these features is discussed below. FACTORS AND FACTOR WEIGHTS Although in principle a very large set of compensable factors could be developed, in practice most job evaluation systems use a similar small set of factors. This is due in part to the propensity of designers of job evaluation systems to borrow factors from previously developed systems with only minor modifications. In addition. the choice of factors no doubt reflects the industrial origin of job evaluation plans. Their use was initially In factories, and the job characteristics chosen for evaluation tended to emphasize shop knowledge, responsibility for equipment, strength requirements, work hazards, and so on. Job evaluation plans have subsequently been designed specifically for the evaluation of office jobs as well as executive jobs. It is probably true, however' that the job evaluation systems currently available do not correspond very closely to the character of the contemporary labor force, which is increasingly concentrated in technical and service jobs that did not exist when most plans were developed (for example, jobs involving automated infor- mation processing, such as computer operator, data entry clerk, airline reservations agent, etc.~. It is not clear how important this issue is, but a question must be raised as to how well these new jobs are dealt with by systems designed to evaluate quite different sorts of jobs. Since such technical and service jobs are often held by women, any inadequacy in

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrimination 75 the ability of existing job evaluation plans to measure properly their compensable features may undercut the usefulness of these plans in resolving pay discrimination disputes. In our judgment, consideration should be given to redesigning job evaluation systems currently in use to take account of changes in the content of jobs in the American economy in the 40 years since most of them were initially developed. Despite the fact that most job evaluation plans appear to tap the same basic features of jobs skill, effort, responsibility, and working condi- tions the particular operational indicators of the factors may vary more widely, a possibility with important consequences. For example, "skill" Is sometimes measured operationally by the amount of experience re- quired to become fully trained in a job and sometimes by the amount of formal education required to qualify for a job. Clerical jobs, for example, tend tO entail considerable formal education but little actual on~the-job training, while the reverse tends to be true of crah jobs. The choice of which operational indicator of skill is used in a job evaluation system could have substantial impact on the job worth scores of these two categories of jobs. Furthermore' since clerical jobs tend to be held mainly by women and craft jobs to be held mainly by men, the choice of indicators could effectively determine the outcome of any analysis of whether pay differences between these job categories involve discrim- ination on the basis of sex. Proper use of job evaluation techniques to resolve disputes involving pay discnmination depends on clear under- standing of precisely what factors are used in the evaluation formula. We relative weight accorded the different compensable factors used in a job evaluation plan can have substantial impact on the resulting hierarchy of job worth. That is, different weightings of factors can sub- stantially alter the ordering of jobs. If the content of the jobs held by men and women or by minorities and nonminonties differs substantially, then different weightings of the factors can result in different outcomes as to the average worth of jobs held by men and women or by minorities and nonminonties, and hence in different judgments regarding the pres- ence or extent of pay discrimination on the basis of sex or minority status. (For a more detailed discussion of this point see Truman, 1979:5~54 ) We noted above that a frequently used method of determining the relative weight to be accorded each factor in developing a job evaluation plan is to predict existing market wage rates from a set of potential job evaluation factors. This method, although it has advantages, has two major drawbacks. One of these drawbacks is technical. The resulting weights may vary

OCR for page 69
76 WoMEN, WORK, AND WAGES substantially, depending on the set of jobs chosen for the prediction and the data base used in determining the average wages of those jobs.6 In the typical application a subset of all jobs in a firm is used to determine factor weights: these are often called "key" or "benchmark" jobs. If, for example, only the highest-paid blue-collar jobs and the lowest-paid white-collar jobs in a firm are used as benchmark jobs and blue-collar and white-colIar jobs typically have different characteristics, then the weights for the white-collar job characteristics derived by this procedure would be unusually low and those for the blue-collar job characteristics unusually high, compared with their actual compensation in the firm. Hence the choice of benchmark jobs can affect the weights derived. Second, even if the benchmark jobs are themselves representative of all jobs in a firm, a wide range of wage rates typically exists for each job, and the choice of wage used to represent the job (minimum, mid- point, mean, etc.) can affect the weights resulting from the procedure. These technical difficulties mean that the derived weights may not ad- equately reflect compensation. This aspect of the implementation of job evaluation-procedures is an extremely important one, and care must be taken that even a well-designed system is not poorly implemented at this stage. The second major drawback of using existing wages to derive factor weights is that the weights wall then necessarily resect in turn any biases that exist in market wages. To the extent that existing wages incorporate the effects of discnnunatory practices (and we argue in Chapter 3 that those effects can be substantial in some cases), the weights derived from those wages as well as the resulting job worth scores also incorporate those effects. It is hardly optimal to use job evaluation scores as a standard against which to assess the legitimacy of existing pay differences among jobs if the job evaluation scores themselves are designed to replicate as closely as possible existing pay differences. On the other hand, using existing wages to derive factor weights has the advantage, from the point of view of the employer and many em ' In a typical prediction operose, a set of jobs has a~rcrage ~gc rates Y and preliminary, arbitranly specified scores on job worth factors J' (i = 1, . . ., n). An equation of the following type is estimated, ~ ~ a + ~ b' JJ 9 d-1 where Y represents the estimated wage rates of the jobs, "d b, represent the estimated weights for the factors J.. l~c constant term, a, Ed the weights, b', arc derived, Mug multiple regression technigocs, such that the "ready," Y - ),, arc as small as possible (~ifi~y, ply _ ~2 ~ minimized for the ~Ic). Tbc weights c" theD be used to rescalc Tic furors for use m denying pb worth scores.

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrimination 77 ployees, of by and large preserving customary wage relationships among jobs and of formalizing an evaluation of the relative worth of different attributes of jobs that already exists implicitly. The rationalization of a system already largely in place is likely to be far less disruptive than imposition of a substantially different hierarchy and criteria of job worth. Moreover. when as is sometimes done area wage rates rather than a fimn s own wage rates are used. employers can be reasonably certain that their internal pas structure is consonant with external market re- wards and that their pay scale will enable them to recruit and retain the necessary workers. THE ROLE OF JUDGMENT It is important to recognize that job evaluation ultimately rests on judgments. Jobs are described in terms of their tasks, duties, and re- sponsibilities. and these descriptions are rated or ranked with respect to some set of factors. The factor ratings are seldom based on objective information; rather. they represent judgments about such amorphous features of jobs as the responsibility entailed or the experience required. The nature of job evaluation makes it possible for bias to enter at two points: in the writing of the job descriptions and in the evaluation of the descriptions with respect to a set of factors. One can question to what extent traditional stereotypes regarding the complexity and responsibility of different types of work are reflected in job descriptions. Unfortunately. there is no evidence on this issue: to our knowledge there have been no studies of the accuracy or validity of job descriptions. Such studies would be extremely useful. Moreover, methods of systematic job analysis, such as structured job analysis and task analysis, ought to be explored for their applicability to job evalu- atio~in particular the job component method of job evaluation (McCormick and llgen 1980:Ch. 18~. which uses structured job analysis. One can also question to what extent sex stereotypes may affect the evaluation process. That is' are jobs held by women evaluated differently from jobs held by men, even when their content is virtually identical? The e~dence in answer to this question is very sparse. Almost no evi- dence is available that pertains directly to sex stereotyping in the c~al- uation of jobs; there is, however, very strong e~dence that female workers are evaluated as less worthy than male workers with identical charactenstics. This evidence denves from a genre of studies in e%per- ~mcntal social psychology in which subjects are presented with a set of vignettes describing the performance or qualifications of individuals and asked to rate them on one or several dimensions. Various aspects of

OCR for page 69
78 WOMEN' WORK, AND WAGES the vignettes are systematically vaned, and the effect of the variation on the ratings is studied. For our purposes, the vanable of interest is the sex of the individual described in the vignette. These studies are reviewed in detail in our interim report (Treiman, 1979:43~5~. Here we report our conclusion (p. 4S): In a variety of contexts the mere fact of identifying a performance as done by a woman results In a lower evaluation and a lower likelihood of reward hinng, promotion, etc. than when the identical performance is attributed to a man. The only exception is when the performer is certified as competent on ~nde- pendent grounds; in such cases there is no significant tendency to evaluate women more poorly than men. While most of the studies ated . . . refer to the evaluation of people rather than jobs, the evidence for sex stcreo~ping ire jo~related contexts is certainly strong enough to suggest the likelihood that sex stereotyping will pervade the evaluation of jobs strongly identified with one se% or the other. That is, it is likely that predominantly female jobs will be undervalued relative to predom- inantly male jobs in the same way that women arc undervalued relative to men. Given the paucity of the essence, much additional work needs to be done to cianfy to what extent and under what circumstances sex ster- cotyping is likely to be operating in the evaluation of jobs. For example, experimental studies of the same type as those cited above, in which job descriptions rather than job incumbents are evaluated, would be very useful. If the same description received a higher rating when iden- tified with a "male" title than with a "female" title (e.g., waiter versus waitress, orderly versus nurse's aide), the process of sex stereotyping the job evaluation process could be inferred. Moreover, ways should be explored to niinim~ze the impact of stereotyped perceptions of jobs. One possibility would be to carry out job evaluations on the basis of job descriptions only, omitting job titles from the description. MULTIPLE JOB EVALUATION PLANS IN A SINGLE FIRM Another source of difficulty in using job evaluation plans to assess complaints of pay discrimination is the tendency of firms to use different job evaluation plans for diffcrcut categories of workers. When more than one plan is used to evaluate the jobs in a firm, there is no way of directly comparing all jobs ~ that firm (unless a formula exists for translating the scores from one plan into the scores of the others, which Is rarely the case). Since the job categories Covered by different plans (typically shop, office, and czecutive jobs) tend to be highly segregated by sex, race, and ethnicity (i.e., women arc overrepresented in office

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrimination 79 jobs and underrepresented in both shop and executive jobs; minorities are underrepresented in executive jobs), the use of job evaluation ratings to assess the existence of pay discrimination would appear to require the ability to make comparisons among as well as within the different plans. The development of different job evaluation plans for different types of jobs appears to have two explanations: first, plans were onginally developed for blue-collar jobs and only later for white-collar office and professional and executive jobs; second, different types of jobs are widely believed tO have vent different job characteristics. For example, manual dexterity, a factor thought to be unponant for most blue collar jobs and some office jobs, is nonnally not thought to be important for professional and executive jobs. However, it is quite possible that, as the economy has changed, so has the nature of jobs. As both clerical and factory jobs are becoming more automated, they perhaps are coming to require skills that are more similar than they were previously. It has been suggested that job evaluation plans for technical jobs, which cur- rently form a kind of bridge between plans for shop jobs and office jobs, could be used to evaluate all jobs. Moreover, some large fines and some state governments do in fact use a single plan for all workers, and the U.S. Department of Labor, in its Dictionary of Occupational Titles, descnbes all occupations in the economy using one set of factors. Me committee is divided on the question of whether the jobs usually found within a single firm can be adequately evaluated by a single job evaluation plan or whether several plans would be required to measure job characteristics adequately. Two issues are involved. First, should a given score on a given factor be worth the same amount for all jobs in a firm? Many on the committee would say ye~that is the meaning of comparable worth. If, for example, degree of responsibility differen- tiates supervisory from line jobs on the plant floor, it should do so in the office as well. Quite possibly, of course, an entire group of jobs could have the same rating on a particular factor. For example, a factor for undesirable working conditions might be scaled in such a way that all office jobs but only some plant jobs are scored in the lowest category, so in effect the factor would be relevant only in differentiating among plant jobs; this would not vitiate the applicability of a single job eval- uation plan to an entire finn. The second, more difficult problem is whether indicators can be devised to measure accurately a given factor for all jobs in a firm. Is it possible, for example, to specify the meanung of responsibility in such a way as to differentiate between managerial jobs of greater and lesser responsibility and also to differentiate between

OCR for page 69
80 WOMEN, WORK, AND WAGES shop jobs of greater and lesser responsibility? Lee answer to this ques- tion is uncertain. Not enough is yet known about the measurement of job characteristics to be able to assess the validity of job evaluation plans for different categories of jobs. More research on the nature of job characteristics and on the properties of job evaluation plans used throughout a firm is needed before the usefulness of such plans can be established. MODELING AND MEASUREMENT Two bases for specifying factors and factor weights in job evaluation formulas are reviewed above: (1) an a priori approach, in which the choice of compensable factors and their relative weights are decided as a matter of policy, without regard for existing pay practices, and (2) a policy-captunng approach, in which existing pay rates are predicted from a set of compensable factors Ma multiple regression procedures and weights are denved from the regression model. In the latter case, the adequacy of the predicted scores as the basis for assigning jobs to pay grades or for assessing whether jobs are "underpaid" or "overpaid" depends on the adequacy of the statistical mode! used to predict existing pay rates. There are a number of potential difficulties involved in regres- sion procedures of this kind. First, if variables that in fact affect pay rates are omitted from the model, two distortions result: the predictions wall be less accurate than otherwise and may be biased, and the relative weights of the vanables included in the mode] will be distorted. Second, if variables in the mode] are measured imperfectly, they wall appear to have less importance than they truly have. Third, if the functional form of the mode] is not correct, the predictions wall be less accurate as well as distorted. There are now wel]-developed procedures for estimating regression models with nonlineanties and interactions of venous kinds and for testing the relative adequacy of alternative functional fonns, but these do not seem to have been widely used in fob evaluation systems. Techniques used in job evaluation have not kept pace with devel- opments in econometncs, psychometncs, and sociological measurement. Serious attention should be given to the selection and measurement of compensable factors, the functional form of regression models, and assumptions about error structures, each of which can seriously affect the factor weights and the pay rates predicted by these models. Regres- sion-based models of the type discussed here require considerable sen- siti~rity and ingenuity on the part of the analyst; mechanical applications of the technique can easily produce seriously misleading results.

OCR for page 69
Wage-Adjustmen' Approaches to Overcoming Discrimination B] SUMMARY A number of features of existing job evaluation systems make them less than optimal for use in the resolution of pay discrimination disputes. First, formal job evaluation systems order jobs by reference to a set of compensable factor~that is, factors thought to be legitimate bases of pay differentials. As we have shown, the factors and their relative weights are ohen chosen in such a way as to closely replicate costing wage hierarchies. For that reason, they can hardly serve as an mde- pcndent standard against which to assess the possibility of bias in c~ust~g pay rates. Second, it is possible that the process of describing and eval- usting jobs reflects pervasive cultural stereotypes regarding the relative worth of work traditionally done by men and work traditionaDy done by women. These features of job evaluation systems make it probable that their use as a standard of job worth understates the extent of differences in pay based on sex and perhaps on race or ethnicity. Third, most firms currently use more than one job evaluation plan, a pumice that restricts comparisons between jobs to those wit n sectors of a fi~n~e.g., shop jobs, office jobs, or executive jobs. Finally, there are potentially serious technical shortcomings in the way regression pro- cedures are used to create job evaluation formulas. Nonetheless, it would be unwise to reject the use of job evaluation plans altogether. Despite their limitations, they do pronde a systematic method of comparing jobs to determine whether they arc fairly com- pensated. Because job evaluation plans as ardently implemented are likely to understate the extent of differences In pay based on the sex, race, or ethnic composition of occupational categories, estimates of discrimination derived from the application of current job evaluation plans are probably low. Still, using job evaluation scores to determine pay rates wall generally go some way toward reducing discriminatory differences in pay when they exist. It may be possible to improve fob evaluation glans. The oommiKee urges fob evaluation practitioners and users to scrutinize exists plans for fairness in light of the considerations reviewed here and in its Interim report (Treiman, 1979~. In addition, we urge further research into the many unresolved technical issues regarding job cvaluabon pnoc~ples and practices. The techniques used in job evaluation plans have been for the most part designed and Implemented by practitioners hobo have not been well grounded in advances in measurement in the social sci- ences. These procedures have not to date been subject to the kind of rigorous analytic scrutiny necessary to put them on a technicaDy sound

OCR for page 69
82 WOMEN. WORK, AND WAGES footing. Considering their potential usefulness in resolving wage dis- crimination disputes and the growing interest in that use, further re- search on techniques of job evaluation would be very valuable. STATISTICAL APPROACHES TO ASSESSING PAY RATES The comparable worth approach attempts to use job evaluation pro- cedures to determine "fair" rates of pay for jobs, but as we note above, that approach involves the difficulties inherent in using the job evalu- ation plans that are currently available. One of the main difficulties is that the factor weights used in job evaluation systems frequently are denved from a regression procedure designed to make job worth scores replicate existing wage hierarchies as closely as possible. When this is the case, the job worth scores themselves may reflect any existing bias. The committee therefore explored two modifications of the statistical approach conventionally used in developing job evaluation plans. Tl2ese procedues may be used both to modify existing job evaluation plans to reduce bias and to develop new bias-free job evaluation plans. They may also procure helpful in identifying specific instances of pay discnm- ination. The first is a multiple regression approach that, although anal- ogous to the standard job evaluation procedure of using existing wages to derive factor weights, includes "percent female" among the predictor ~ranables. The second is the use of wage rates of jobs held mainly by white men as a standard of "fair" wages. These are not, of course, the only statistical approaches available; indeed, we encourage the devel- opment and testing of alternative statistical approaches. We wish to make clear at the outset that we discuss these approaches because of their potential and not because of their proven value. Hey are at present completely untried, and their application would entail the solution of many theoretical and practical problems of measurement. Were is also a serious question as to whether the quality of the data generated by job evaluation plans In current use Is adequate to sustain the kinds of statistical adjustments we describe. (The discussion in the previous section details some of the problems.) Moreover, there is con- siderable debate regarding the interpretation of the statistics generated by these adjustments (specifically, regression cocffiacats), especially given Imperfect measurement (see B~rnbaum, 1979~. In order to en- courage the kind of practical research and development that should be earned out in order to create technically sound procedures for identifying and correcting pay discrimination, we present these approaches as ex- amples of two that might be considered.

OCR for page 69
Wage.Adjustmer't Approaches to Overcoming Discrimination 83 INCLUDING PERCENT FEMALE'' IN THE ESTIMATION OF PAY RATE S As we discuss above, the fact that men tend on the average to be paid more than women guarantees that any job evaluation factor that is correlated with the sex composition of jobs will receive some weight, and those factors that are more strongly correlated with sex composition w'll on the whole receive greater weight. Thus, the use of job evaluation plans perpetuates existing pay differences among jobs, which-as we have argued in Chapter May reflect discriminatory as well as leoi~- unate components of pay. ~. . . . . ~ _ a One potential method of removing the bias built into factor weights denved in the conventional way is to estimate the factor weights from an expanded equation that includes measures of the sex (or race or ethnic) composition of occupational categones. As we showed above (see note 6), the conventional approach to denying factor weights for job evaluation plans is to estimate a regression equation of the forth y=a+~:bili (~) over a set of occupations in a Fin., where Y is the average pay rate of each occupation, the Ji are potential compensable features of jobs (meas- ures of skill, effort, responsibility, working conditions, etc.), and the hi are the estimated weights for the factors hi.' Now consider a modification of eq. ~ of the form Y= a' + I:b,']i + cF, (2 ' While these equations arc in linear form, as required by the regression estimating cdurc, it is possible to represent quite complex relations, including interactions and nonlincarities of various kinds, by appropriate transformations of the vanables. For e%- amplc, an exponential equation of the foully Y= e.~"`x cut be written in a mathematically equivalent way as a linear additive equation in(Y) = a + ]:b~Y,. Similarly, an equation of the form can be rewritten as Y = Q + b,X, ~ b2X2 + b3X,X2 Y = a + b,X, ~ b2X2 + b3X3 simply by defining X3 = X,X2. See Goldberger (1968), MostcDcr "d Tubby (l977), and Stolzenberg (1919) for a discussion of these and similar t~sfo~s.

OCR for page 69
84 WOMEN, WORK, AND WAGES where Y and the li are as before, F is the percentage of incumbents of each job who are women, the b,' are the estimated weights for the factor li (in general, these will not be the same as the corresponding coeffi- cients, hi, estimated by eq. 1), and c is the weight for the percent female. The coefficients, hi', associated path each of the compensable factors, Ji, can be interpreted as indicating the contribution of each factor in determining the average pay rate of workers in these occupations, hold- ~ng constant each other factor, plus percent female. Specifically, each hi' indicates the number of dollars a one-point change in the factor score is worth, on the average, for occupations that have identical scores on the other factors and the same percent female. For example, suppose the first factor, ],, indicates the number of months of training necessary to become fully qualified. Estimating the equation yields a value for the coefficient bit' that indicates how many dollars each month of training is worth. That is, for 2 jobs equal in all other respects except that one requires 2 months of training and the other requires 12 months of train- ing, a difference in their average pay rates of job,' dollars would be predicted, since they differ by 10 units on the variable I,. These coefficients differ from the corresponding coefficients in eq. I, the conventional job evaluation equation, in that they are adjusted for the propensity of characteristics that distinguish between jobs held mainly by women and jobs held mainly by men to be heavily weighted as a consequence of the strong negative association in most firms be- tween the percent female and the average earnings of jobs; the explicit inclusion of percent female in the equation is what adjusts the weights. Given the usual association between sex composition and earnings, job evaluation scales built by weighting the factors in proportion to the coefficients b,' from eq. 2 will ordinarily be less sex-biased than scales built by weighting the factors in proportion to the corresponding coef- ficients of eq. 1. Me coefficient c indicates the effect of sex composition on pay rates for occupations that are identical with respect to all of the other vari- ables. Specifically, c indicates the difference in pay rates that would be expected on the average between two occupations that differ by one percentage point in their sex composition but are identical with respect to all other measured variables. As such, c can be taken as a direct measure of potential discrimination. Whenever c is significantly different from zero, the sex composition of occupations would have to be inter- preted as a compensable factor-but to pay jobs on the basis of the sex of their incumbents would be regarded as discnnunatory. An objection could be raised that variables excluded from consider- ation in estimating job worth may cubist that arc both vatid indicators

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrinunation 85 of worth and correlated with percent female. Insofar as such vanables exist, sex composition would stand as a surrogate for differences in job worth, in which case the coefficient c could not be regarded as a valid measure of potential discnmination. For example, training in mathe- matics could be a job-related factor correlated with sex; its exclusion would cause c to overstate the extent of potential discrimination. In the judgment of the committee, however, the burden should rest on the designer of the job evaluation system to identify and explicitly incor- porate all factors regarded as legitimate components of pay differences between women and men, not merely to assert the possibility that in- cluding unspecified and unmeasured factors or improving the measure- ment of existing factors could reduce the `~disc~imination" ooeffic~ent c.8 One variable in particular that might usefully be added to pay pre- diction equations is the average experience of job incumbents. It is well known that men tend to have more occupational experience than women' and occupational experience is generally regarded as a com- pensable factor for individual workers. By not including a measure of the average work experience of incumbents in each occupation, an equation such as eq. 2 overstates the amount of sex-based discrimination to the extent that the average work experience of incumbents Is nega- tively correlated with percent female. Other worker characteristics, such as educational qualifications, may also be important determinants of wages. Equation 2 could be expanded to include those characteristics of workers that are regarded as legitimate bases of pay differentials. An equation of the form Y = a + I:bi ~' ~ IcjXj ~ OF, (~3) ~ Ibc approach outlined here is an example of a class of models that treat discrimination as a residual factor. He characteristic feature of such models, which have been unduly applied in economics and sociology, is that they attempt to explain observed diffcrenocs between groups with respect to some attribute (e.g., income) by predicting that attribute from a small number of other charactenstics, then relating the residual diffcrcocc between observed and predicted values to group membership, intcrpretmg group diffcrcaces In the evcra~c size and sign of the residual as evidence of disnmination. From a technical point of view, howe~cr, the residuals estimated from such equations indicate the effect of all factors not explicitly measured pus any error m mcasuremcat; discrimination per sc may or may not account for a large fraction of the residual ~rananCc. Interpreting the residual as indicating damnation, then, requires Other the monk Ed clearly untenable assumption that all relevant factors have been measured, and measured without error, or a determination that discrimination ~ lively. ~ our jUd - Cut, the proper interpretation in light of the c~idencc reduced in Chapters 2 and 3 is to treat the unexplained differences in average pay rates between men "d women "d between mDontics and nonrrunontics as indicating the probability of disenmizutory protest unless the contrary can be shown.

OCR for page 69
86 WOMEN, WORK. AND WAGES where the X, represent average characteristics of incumbents and the other vanables are as defined above, can be interpreted in ~ manner identical to eq. 2. The substantive problem of determining which char- actenstics of workers should be regarded as legitimate bases of pay differentials corresponds to the problem of determining which aspects of job content should be regarded as legitimate bases of pay differentials. Ultimately, both determinations are matters of values. We should point out that job evaluation plans currently in use do not ordinarily include worker characteristics because these plans typically attempt to measure only the required elements of jobs. Experience that incumbents happen to have that is not required by the job is ordinarily considered irrelevant, while required experience is usually included as a job element, one of the lit If job evaluation procedures are to.be used for the resolution of claims of pay discrimination, however, their usefulness would be en- hanced by including those worker characteristics regarded as legitimate compensable factors. JOBS HELD BY WHITE MEbt AS A STANDARD We now turn to a different approach, the use of features of jobs held mainly by white men as a standard for assessing the fairness of the pay rates of other jobs. The key assumption of this approach is that white men in jobs held mainly by white men are not subject to discrimination.9 With this assumption, the average pay rates of white men in jobs held mainly by white men can be used as an indication of the relative worth of these jobs; for this subset of jobs, then, there is an objective, market- based criterion of job worth. In this approach one determines what features of those jobs contribute to differences in their level of compensation, using the familiar technique of regressing average pay rates on job characteristics. That is, an equa- tion analogous to eq. 1, };,,,,, = a ~ I;biJi, (4) can be estimated for the subset of jobs held mainly by white men, where Y.,,,,, Is the average pay rate for white male incumbents. The coefficients of such an equation can then be used to assign a job worth score to ' Discnniinatory treatment of women or minorities would have the effect of driving down the pay of white men in jobs with substantial proportions of women and minontics. We mow empirically that pay rates of men arc negatively correlated with the percent female among incumbents of aD occupation (see Chapter 2) and would expect a similar Cation tenth the proportion in an occupation who are minority workers.

OCR for page 69
Wage-Adjustmen~ Approaches to Overcoming Discruninaiion 87 every job in a firm if it can be assumed that whatever factors differ- entiate the actual earnings of white men in jobs held mainly by white men also, and in equal proportion, differentiate the actual earnings of women and minorities and of white men in other jobs within a finn. The weakness of this approach is that if jobs held mainly by white men have characteristics substantially different from those held mainly by m~nonties or women, the coefficients from the earnings equation for jobs held mainly by white men wall not be good estimates for jobs held mainly by women or minorities. This is an empirical issue that should be resolved in each specific case. The strength of this approach is that it entirely avoids the question of what ought to be the bases of compensation question we regard as being a matter of values and admitting of no technical answer and takes the marketplace as the arbiter of pay rates for that subset of jobs for which there is no suspicion that discriminatory processes affect the rates. To use the results of such an analysis as the standard of equitable compensation, one would apply the estimated coefficients to jobs held mainly by other workers and estimate the average wage or salary level that would be expected if such workers are compensated in the same way as white men in jobs held mainly by white men. The average earn- ings levels estimated in this way could then be compared tenth the actual earnings of those workers and the difference taken as a measure of possible discrimination. If the expected value is greater than the actual value, one could conclude that those workers are underpaid relative to the worth of their jobs; if the expected value is smaller than the actual value, one could conclude that they are overpaid. It would then be possible to adjust the pay rates of those jobs or workers who are sub- stantially underpaid (or overpaid). USING STATISTICAL PROCEDURES TO CORRECT DISCRIMINATORY PAY RATES It may be possible to use the sorts of models outlined ~ the previous Actions to make adjustments in discriminatory pay rates, although me- chanical application of these approaches without careful consideration of the measurement issues noted In this chapter would be iB-adv~sed. Many other procedures for adjusting discriminatory aspects of pay dif- ferences could be devised; we have selected several as illustrations. The first procedure is to pay each job according to its worth as de- termined by the job evaluation plan In use. This ensures that aD jobs In a firm are compensated on the basis of the same cntena. If a firm Is willing to use job evaluation ratings as the sole basis for establishing

OCR for page 69
88 WOMEN, WORK, AND WAGES pay rates for jobs-as, for example, most steel manufacturers do for shop jobs and as the federal government does for white-collar jobs (Treiman, 1979 - and if the job evaluation system is free of bias, this procedure may be satisfactory. A major disadvantage of this procedure, however, which we have noted several times above, is that, since the factor weights are usually denved in such a way as to ma~cim~ze the prediction of existing pay rates and since pay rates may be strongly correlated with the sex composition of jobs, those factors most strongly correlated with sex composition may receive the heaviest weights. To the extent that this is the case, any sex bias in pay rates Will tend to be preserved. To overcome this disadvan- tage, we suggest two modifications to the procedure. The first modification is to use factor weights derived for white men in jobs held mainly by white men as the standard applied to all jobs. This technique has the advantage of adjusting all pay rates to a level commensurate with the highest returns currently offered, which is prob- ably psychologically preferable but has the corresponding disadvantage of increasing the cost of the total pay package (unless, of course, the pay rates for all jobs are reduced by some constant). This technique has another disadvantage: factor weights chosen to differentiate among jobs held mainly by white men may not differentiate well among other jobs. The second modification is to adjust the weights to remove sex com- position as a compensable factor and to use the adjusted weights to develop a '`bias-free" job evaluation plan. Formally this involves com- puting an expected income for each occupation by substituting the mean percent female offer all jobs into eq. 2 (or eq. 3) and evaluating the equation. That is, for each job, I, a fair pay rate, Y', can be estimated by fj = a + I:bi'Jij ~ c(F), (5) where l,j is the score on the ~ factor for the jth job, and F is the mean percent female for aB jobs in a firm. This technique has the advantage of of adjusting the coefficients b,' to estimate the contribution of each factor to total earnings among jobs that have the same percent female. The conventional job evaluation procedure and its modifications are similar In that they determine earnings entirely on the basis of a weighted sum of compensable factors. In this sense they arc all versions of a comparable worth approach. An employer may feel, however, that such an approach is overly deterministic. After all, a variety of idiosyncratic factors may legitimately create pay differences among jobs and, it could be argued, these ought not to be ignored or arbitrarily omitted. Hence,

OCR for page 69
Wage-Adjustment Approaches to Overcoming Discrimination 89 we suggest another procedure, which preserves differences in the pay rates of jobs insofar as the unmeasured idiosyncratic components of pay differentials are uncorrelated with sex composition. This second procedure makes use of the coefficient of percent female in eq. 2 (or eq. 3) as an adjustment factor, adding to the existing pay rate of each job a constant equal to -c(F') or, to keep the total wage cost unchanged, - c(Fj - F), where Fj is the percent female in the jth job, F is the mean percent female for all jobs In a firm, and c Is the associated net regression coefficient. That is, -c dollars are added to the pay rate of each job for each additional percent female among the incumbents. Me result of such a procedure is to reduce the net effect of sex composition to zero. Of course, since the legitimate compensable factors will typically be correlated with sex composition, the resulting zero-order correlation between sex composition and earnings ordmaniy will not be exactly zero, although it usually will be reduced. Me basic difference between the first procedure and its modifications and the second is that the former adjust all pay rates to the policy line defined by a job evaluation formula, while the latter permits pay rates to vary around the policy line, with a sole constraint: the Sedations from the policy line must be uncorrelated with the sex composition of occupational categories. With the exception of the use of unadjusted job evaluation scores as the basis for setting pay rates, these procedures have seldom been im- plemented, so it Is difficult to anticipate what practical difficulties may be involved. Hence it would seem prudent to exercise considerable caution in applying them, attending carefully both to the statistical issues discussed above and to substantive concern~the possibility that some workers may perceive new inequities as replacing old ones, that to avoid such perceptions may require a substantial increase in the wage bill for an enterprise, and that statistical adjustment procedures often generate tension between the need to eliminate discrimination for groups in the aggregate and the need to protect the rights of individuals. Despite these caveats, we urge the exploration of these and similar procedures as a means of eliminating discrimination in pay rates for aB workers. CONCLUSION Starting from the evidence that erasing occupational pay hierarchies sometimes embody discnnunatory elements, this chapter proposes and reviews several approaches to Feting and conechug discrimination in pay rates. These approaches are of two kinds: one involves improvements m Me design and impiementai~on of job evaluation pears ourrendy In use

OCR for page 69
go WOMEN, WORK. AND WAGES and the other involves statistical adjustments to pay rates to estimate and remove the effect of the sex, race, and ethnic composition of job categories on their pay rates. Both kinds of approaches depend on two assumptions: Mat the basis on which jobs are paid at different rates can be made largely explicit and measurable, and that whatever cannot be measured does not favor any sex, race, or ethnic group. These assumptions seem to us to represent useful foundations for the design of methods to assess the fairness of existing pay rates within a firm. Me approaches reviewed here and the procedures we illustrate are at present, however, not very well developed and are almost completely untested. Hence, it is not possible to recommend any of them unequivocally at this time. We prefer to encourage experimentation with and exploration of the properties of these approaches in order to determine their usefulness in eliminating discrimination in pay rates. In particular, efforts should be made to improve job evaluation techniques. Research on the pos- sibility that stereotypes are operating in the evaluation of jobs as well as research on the actual characteristics of jobs held by different groups is extremely important in ~mprowng job evaluation systems. In addition, further research on the discriminatory components of pay rates is needed. It is important to note, however, that we do not recommend requiring the installation of a job evaluation plan in a fimn not using one in an attempt to ensure that the firm's pay system is nondiscr~mi- natory. At present we know of no method that would guarantee a "fair" pay system.