Some evidence on the relationship between performance-related pay and the shape of the experience-earnings profile.

Sessions, John G.

1. Introduction and Background

In this paper we explore the shape of the experience-earnings profile for individuals employed under different types of contracts: fixed-wage, performance-related pay (PRP), and self-employment contracts. We follow Lazear (1979, 1981) and Lazear and Moore (1984) in hypothesizing that the slope of the experience-earnings profile reflects agency costs, and the reduction thereof, with increased agency costs inducing steeper profiles. If monitoring costs are high, firms may implement compensation schemes designed to encourage employees to self-select behavior in the firms' interests. One such scheme defers a substantial amount of compensation until the later years of tenure. The resulting experience-earnings profile provides a penalty for shirking (Lazear and Moore 1984).

We, therefore, presume that the profiles of workers employed under fixed-wage contracts are steeper than those of self-employees for whom the issue of agency does not arise because of the duality of principal and owner (Lazear and Moore 1984). The interesting case is that of workers employed under PRP. If such contracts represent a hybrid between fixed-wage and self-employment, then one would expect their experience-earnings profiles to lie somewhere between these two extreme cases. Thus, PRP and a steep experience-earnings profile may be regarded as substitute mechanisms for inducing employee effort.

The nature of the experience-earnings profile has important implications for labor market behavior. If agency considerations prevail, then issues arise concerning the credibility of long-term employment contracts--firms may have an incentive to replace expensive "older" workers with cheaper "younger" recruits. The shape of the experience-earnings profile may also influence decisions to quit jobs. Experienced "generally" trained workers may face more options in the labor market than their "firm-specifically" trained counterparts. But both types of worker may have more labor market options than those "older" workers whose earnings reflect agency considerations.

We generalize Lazear and Moore (1984) by allowing for the intermediate category of PRP, which enables us to further explore the agency explanation behind the positive slope of the experience-earnings profile. Our presumption is that all three employment contracts can be nested in the form

[W.sub.j] = (1 - [[lambda].sub.j]) [bar.w] + [[lambda].sub.j]f(e;[theta]), (1)

where j = fw, prp, se denotes "fixed-wage," "PRP," and "self-employment," respectively, [w.sub.j] denotes total remuneration, [bar.w] the component of total remuneration that is "fixed" (i.e., independent of worker performance), and f(e; [theta]) some function mapping the relationship between worker performance (i.e., effort), e, "uncertainty," [theta], and pay. (1) [[lambda].sub.j] represents the degree of equity held by a worker in his/her enterprise, that is, the proportion of total remuneration that is dependent on performance. We assume that for fixed-wage employment, [[lambda].sub.fw] = 0; for PRP contracts, [[lambda].sub.prp] [member of] (0, 1); and for self-employment, [[lambda].sub.se] = 1. Our hypothesis is that the shape of the experience-earnings profile depends critically on [[lambda].sub.j]. To be sure, we predict that agency costs decline monotonically with [[lambda].sub.j] such that the slope of the PRP earnings profile falls between the zero-equity, fixed-wage, and 100% equity, self-employed profiles. (2)

2. Data and Methodology

Our empirical analysis draws on three British data sets: the British Social Attitudes Surveys 1985, 1987, 1993, and 1996, the British Household Panel Surveys 1991-1999, and the British Family Expenditure Surveys 1997/98, 1998/99, and 1999/00.

The British Social Attitudes Surveys (BSAS) are an annual series of cross-section surveys initiated by Social and Community Planning Research in 1983. The sample comprises individuals, aged 18 and over, living in private households whose addresses were on the electoral registrar; 114 out of 650 Parliamentary constituencies in Great Britain were selected. A polling district was randomly selected from each constituency with addresses being randomly chosen from each district.

The surveys conducted in 1985, 1987, 1993, and 1996 contained questions relating to the presence of PRP: whether in the year of interview the respondent had received some component of their total remuneration in the form of: (i) a productivity-linked bonus scheme; (ii) an annual bonus (at the employing organization's discretion); (iii) a share ownership or share option scheme; or (iv) a profit-sharing scheme. Individuals who reported that they had participated in any of the four schemes were labeled as PRP employees. (3) Selecting out all male respondents in fixed-wage, PRP, or self-employment with complete records rendered 1467, 783, and 491 individuals, respectively. (4)

Our second data set is based on the British Household Panel Survey (BHPS). This is a random sample survey conducted by the Institute for Social and Economic Research of each adult member of a nationally representative sample of more than 5000 private households. For wave 1, the interviews took place in 1991. The same individuals are reinterviewed in successive waves.

We explore data from the 1991-1999 surveys. In the first wave (1991), all individuals were asked whether their pay includes bonuses or profit sharing, thereby enabling us to identify PRP employees. For the period 1992-1995, this question was asked only to individuals who changed their jobs. We, therefore, assume that individuals who did not change job remain in their 1991 employment type. We specify an unbalanced panel of data wherein the minimum number of times an individual is in the sample is one and the maximum is nine. Our sample comprises 4594 fixed-wage employees, 2806 PRP employees, and 1153 self-employees.

Our third data set is drawn from the Family Expenditure Survey (FES) for Great Britain. This is a nationally representative survey that has been conducted annually since 1957. Approximately 10,000 households are selected each year to take part in the FES, and the average response rate is approximately 70%.

We use data from the 1997-1998, 1998-1999, and 1999-2000 surveys. (5) Our subsample comprises working males aged between 18 and 65 who are either self-employed or employed under a fixed-wage contract or a contract characterized by PRP. Those individuals classified as being employed under a PRP contract were those in receipt of a productivity-linked bonus, profit-related pay, dividends from employee share ownership, an incentive bonus, or a performance-related bonus. Our sample consists of 8405 male respondents comprising 5965 fixed-wage employees, 1201 PRP employees, and 1239 self-employees.

It is apparent that there is some tension between our theoretical and empirical definitions of PRP. The former defines PRP as any contract in which current pay is related to current performance broadly defined. Some jobs reward performance with a promotion and/or a salary increment rather than with an explicit bonus. Moreover, some bonus schemes may also be used to identify "fast-track" employees, and for these, a revelatory component may be in operation as the employer learns more about a worker's skills and ability in the early years following an initial hire. (6) The BHPS, but not the BSAS or FES, asks specific questions relating to promotions and salary increments in each year, and we, therefore, augment our BHPS empirical analysis by including dummy variables to control for these factors. (7)

The empirical definitions of PRP in the three data sets are quite explicit. The BSAS and FES adopt similarly broad definitions that include productivity-linked and discretionary bonuses, share ownership/options, and profit sharing. The BHPS focuses on just bonuses (type not defined) and profit sharing. (8) Which definition is more appropriate is largely a matter of interpretation, but our results are broadly consistent across the three data sets.

Sample statistics of the key variables for each of the data sets are set out in Table 1. (9) A common finding in all three data sets is the relatively low earnings of self-employed respondents. The problems of accurately measuring pay from self-employment are well documented (see Eardley and Corden 1996; Hamilton 2000). Because our focus is on the slope of the experience-earnings profile, measurement error is not too problematic if the earnings of the self-employed are consistently underreported.

Our analysis is based on a Mincerian earnings equation of the form:

1n [w.sub.ijt] = [X.sub.ijt][B.sub.j] + [[alpha].sub.j][E.sub.ijt] + [[beta].sub.j][E.sup.2.sub.ijt] + [[epsilon].sub.ijt], (2)

where [w.sub.ijt] represents the hourly earnings of an individual i employed in employment type j at time t, with j = s, prp, se representing the three employment categories. (10) [X.sub.ijt] represents a vector of personal and workplace characteristics including education, occupational status, and industrial affiliation, and [E.sub.ijt] denotes labor force experience at time t, proxied by the respondent's age at time t less his/her age when he/she completed full-time education. (11) It is apparent that a first-best estimation of Equation 2 would utilize a panel data source. We adopt such an approach with the BHPS sample, which follows the same individuals across 1991 to 1999. In the case of the BSAS and FES analysis, we are obliged to use pooled cross-section data. This is not too problematic: If we assume that the vector of personal/ workplace characteristics is stable across time then [X.sub.ijt] = [X.sub.ij], [for all] t. (12)

Estimation is further complicated on account of potential sample selection bias. Our earnings data derive from observing a particular employment contract (i.e., fixed-wage, PRP, or self-employment), and there may be variables that affect both the probability of observing such a contract and the return to any factors in the earnings equation. To take account of such considerations, we control for sample selection bias. Probit analysis with three discrete outcomes is used to model the determination of Z, prob([Z.sub.i] = j), that is, the probability of being in one of the three possible types of employment, which is then used to calculate the standard inverse mills ratio term, [[delta].sub.ijt]=[phi]([H.sub.ijt])/[PHI]([H.sub.ijt]), where [H.sub.ijt] = [[PHI].sup.-1]([P.sub.ijt]), [P.sub.ijt] denotes the predicted probability of individual i at time t being employed under contract type j, [phi](.) represents the probability density function of the standard normal distribution, and [PHI](.) represents the cumulative density function of the standard normal distribution.

By incorporating [[delta].sub.ijt] into the wage regression, we control for the possibility that particular types of individuals may be employed under specific types of contract.

To explore the robustness of our findings, we experimented with three alternative measures of labor market experience. First, for all three data sets, we estimated age-earnings profiles. Second, for the BSAS only (such information is not available in the BHPS or FES), we estimated employer tenure-earnings profiles for fixed-wage and PRP employees, making use of the question asked in 1993 and 1996: How long have you been continuously employed by your present employer? And finally, for the BHPS only, we estimated job tenure-earnings profiles, making use of the following question: What was the date you started working in your present position? This question is asked to all employed respondents, whereas all self-employed respondents were asked, On what date did you start doing your present job? By that I mean the beginning of your current spell of doing the work you are doing now on a self-employed basis? Thus, the responses to both questions are used to calculate current job tenure for employees and self-employees.

3. Results

We estimate a standard quadratic Mincerian wage equation as depicted by Equation 2. In the case of the cross-section data, we employ standard ordinary least squares (OLS) techniques, whereas in the case of the BHPS we have employed a random effects estimator. (13) We have experimented with a number of combinations of explanatory variables as well as changes to our set of overidentifying instruments in the selection equation, and we have found the regressions to be generally well specified and highly robust. (14) The selectivity terms suggest that ignoring selection issues would have led to significant bias in the estimated coefficients. In general, our findings accord with the previous literature, suggesting a generally positive relationship between earnings and education as well as a concave relationship between earnings and experience. For brevity, in Tables 2 to 4, only one specification, which includes both educational certificates and years of education, is presented. (15) Tables 2, 3, and 4 present the results relating to the BSAS, the BHPS, and the FES, respectively.

It is apparent that the least robust regressions are those for the self-employed respondents. This is not surprising. Such workers will not be motivated by Mincerian-type arguments to the same extent as their employed counterparts and, under an (extreme) argument whereby the earnings profile is a reflection of agency issues only, would not face the necessity of rewarding themselves with an upward sloping profile to ensure efficient effort over their life cycle (Lazear and Moore 1984).

It is apparent from Tables 2 to 4 that the estimated coefficients on the various labor force experience proxies (i.e., Age, Years in Labor Force, Job Tenure, and Employer Tenure) support our theoretical assumption that the slope of the earnings profile for PRP workers lies between those of their fixed-wage and self-employed counterparts. In terms of the BSAS and FES (Tables 2 and 4), the three profiles are significantly different from one another at the 1% level of significance irrespective of the experience proxy used. In terms of the BHPS (Table 3), the three profiles are significantly different from one another at the 1% level of significance when experience is proxied by age or job tenure, whereas the profiles associated with PRP and self-employment (but not fixed wage and PRP) are significantly different from each other at the 1% level when experience is proxied by years in the labor force. Figures 1 to 5 present the estimated experience-earnings profiles for each set of results, with--for reasons of brevity--the exception of the age--earnings profiles. (16)

Our BHPS results also highlight the positive relationship between the dummy variable indicating whether the respondent had been promoted in the current year (Promotion) and the earnings of fixed-wage and PRP workers, and between the incremental pay scale dummy variable (Annual Increment) and the earnings of PRP workers. We experimented with interacting Annual Increment and Promotion with the various experience proxies. These results confirm the validity of our findings regarding the experience-earnings profiles. (17) Moreover, in the case of the Promotion variable, they also cast additional light on the roles played by human capital and agency. Interacting the various experience proxies with Promotion suggested that the ordering of the three profiles was not affected by whether respondents had enjoyed a promotion in the current year. Indeed, we found that the difference in the three slopes, for example, the agency effect was actually more pronounced for those periods that fell in between promotions. Such findings support the argument that human capital may be more relevant to obtaining promotions, whereas agency effects appear to operate primarily between promotions. (18)

Lambda as a Continuous Variable

The FES and the BHPS (1997-1999) record the total amount of remuneration received in the form of the bonus. In Equation 1, total remuneration is split into a fixed and a variable component, the division depending on the size of [lambda]. For the FES and the BHPS, we can, therefore, proxy [lambda] by calculating the ratio of bonus to total pay:

[??] = Bonus Pay/Total Pay = [lambda]f(e; [theta])/(1 - [lambda][bar.w] + [lambda]f(e; [theta])

Sample statistics for both the FES and BHPS samples relating to [lambda] are set out in Table 5.

Our assumption is that a higher value of [lambda] raises the expected cost of early-term shirking on the part of the worker and thereby leads to a flattening of the experience-earnings profile. To investigate this, we ran three earnings regressions pooling the self-employed, the PRP employees, and the fixed-wage employees together for each of the two data sets (see Tables 6 and 7). In the case of the FES, we conduct OLS analysis, whereas in the case of the BLIPS, we adopt a random effects approach given the panel element to the data. Once again, an unbalanced panel is analyzed in which the minimum (maximum) number of times an individual is in the sample is one (three).

Specification 1 illustrates that higher values of [lambda] are associated with significantly lower earnings--this effect is probably reflecting the presence of self-employed workers within our sample. Specifications 2 and 3 include an interaction between experience and [lambda]. Our results suggest that higher levels of [??] are indeed associated with a flattening of the experience-earnings profile. In the case of the FES, our findings suggest that as the estimated coefficient of [??] moves from zero to unity, the annual rate of return (in terms of log hourly earnings) to an additional year of labor market experience falls from just under 5% to less than 1%. To summarize, these findings provide further support for the hypothesis that the slope of the experience-earnings profile is influenced by both the presence and extent of PRP.

Training

Our results support the hypothesis that experience-earnings profiles reflect agency considerations, with PRP profiles generally lying somewhere between fixed-wage and self-employed profiles. Furthermore, we would argue that our findings are relatively robust given that we have used three different data sets (including both cross-section and panel data) as well as exploring our theoretical assumptions via two different approaches and specifying alternative measures of experience. However, it is apparent that an alternative explanation can be offered. It may be the case that PRP workers undertake relatively less training than their fixed-wage counterparts, implying relatively high (low) starting (future) earnings and thus flatter profiles. It is difficult to rationalize such an argument; it might seem that workers remunerated under PRP are, if anything, more likely to respond to investments in training than their fixed-wage counterparts. We investigated the relative likelihood of fixed-wage and PRP respondents having undertaken some form of training at their place of work. Two of our three data sets (BSAS and BHPS) contained information regarding whether respondents had received training.

The BSAS survey for 1987 asked respondents whether, in the two years preceding the survey interview, they had been (i) asked to do anything just for practice in order to learn the work; (ii) given any special talks or lectures about the work; (iii) placed with more experienced people to see how the work should be done; (iv) sent around to different parts of the organization to see how the work is done; (v) asked to read things to help learn about the work; (vi) taught or trained by anyone while actually doing the work; (vii) sent on any courses to introduce new methods of working. We created two variables: a dummy variable (Train) that takes the value of one if the respondent answered "yes" to any of these questions and an index (Trains) that equals the number of these questions to which the respondent had answered "yes" to the various types of training. Thus, Train represents a binary dummy variable indicating whether or not an individual has received any training, and Trains represents an index, which ranges from 0 to 7, indicating the number of types of training undertaken. We are, therefore, able to focus on both the incidence (Train) and the intensity of training (Trains).

Participants in the BHPS were asked whether in the preceding year they had participated in any off-the-job education or training. Relevant summary statistics for the training variables for both data sets are set out in Table 8. Because our focus is on the incidence of training at relatively low levels of experience, we present summary statistics for four levels of experience. In the case of the BSAS, it appears that PRP employees do receive less training than their fixed-wage counterparts. This finding may be taken as support for a training effect. The situation is reversed, however, at the lowest category of experience, which arguably is the focus of our attention. In the case of the BHPS, the support for the training explanation is less clear-cut, with PRP employees, in general, being characterized by a higher incidence of training.

We explore the incidence and intensity of training across PRP and fixed-wage employees by conducting probit analysis (dependent variable Train) and ordered probit analysis (dependent variable Trains). Our findings are set out in Table 9. In the case of the BSAS, the PRP dummy variable has no effect on the probability that the respondent had so engaged in training. The probability of training was, however, significantly negatively related to the respondent's experience in the labor market, perhaps reflecting the fact that most training is undertaken by relatively younger workers. To ascertain whether there was any interaction between such experience and the PRP dummy, we ran a second specification of the dichotomous and ordered probit models, this time including a series of experience-PRP interaction terms such as Experience less than 5 years*PRP; Experience more than 5 years but less than 10 years*PRP; Experience more than 10 years but less than 20 years*PRP; and Experience more than 20 years*PRP. The coefficients on these terms were insignificant in the dichotomous probit, but those on the last three were significantly negative in the ordered probit. Thus, it does appear that more experienced workers receive relatively less training than their non-PRP counterparts, ceteris paribus. It should be noted, however, that we are primarily interested in the incidence of training at the lower level of experience.

We repeated this analysis for the BHPS. In the case of the panel data set, the results from employing a random effects probit estimator suggest that PRP employees are more likely to receive training, which provides evidence contrary to the training explanation. Our results are, therefore, somewhat mixed, and the BSAS findings allude to the possibility that differences in the experience-earnings profile may not be driven solely by agency consideration. But these findings are not reflected in the BHPS.

4. Final Comments

This paper has focused on the relationship between experience--earnings profiles and the degree of worker equity within an enterprise. We further explore Lazear and Moore's (1984) thesis that the nature of the profile reflects agency considerations by focusing not only on those workers with zero or 100% equity (i.e., fixed-wage and self-employed workers, respectively) but also on those with a fractional level of equity, for example, workers remunerated under some form of PRP. Our presumption is that PRP employees face an intermediate level of agency costs and as such require an intermediate profile. Our empirical analysis of three British data sets offers support for this view.

Our results might be interpreted as support for the argument that the shape of the experience-earnings profile reflects agency considerations. As such, they highlight important issues pertaining to the credibility of long-term employment contracts because employers may be tempted to replace experienced workers with less costly, but equally productive, novices. But the latter will not remain "young" forever, and whether they will be inclined to work for a firm that is unable to guarantee them employment in their dotage is an open question.

References

Blinder, Alan S. (ed.). 1990. Paying for productivity: A look at the evidence. Washington, DC: The Brookings Institution.

Booth, Alison L., and Jeff Frank. 1999. Earnings, productivity and performance-related may. Journal of Labor Economics 17:447-63.

Eardley, Tony, and Anne Corden. 1996. Self-employed earnings and income distribution: Problems of measurement. Social Policy Report No. 5, Social Policy Research Unit, University of York.

Hamilton, Barton H. 2000. Does entrepreneurship pay? An empirical analysis of the returns to self-employment. Journal of Political Economy 108:604-31.

Jovanovic, Boyan. 1979. Job matching and the theory of turnover. Journal of Political Economy 87:1972-90.

Lazear, Edward P. 1979. Why is there mandatory retirement? Journal of Political Economy 87:1261-84.

Lazear, Edward P. 1981. Agency, earnings profiles, productivity and hours restrictions. American Economic Review 71:606-20.

Lazear, Edward P., and Robert L. Moore. 1984. Incentives, productivity and labor contracts. Quarterly Journal of Economics 99:275-95.

Murphy, Kevin M., and Finis Welch. 1990. Empirical age-earnings profiles. Journal of Labor Economics 8:202-29.

Sarah Brown, Department of Economics, University of Sheffield, Sheffield S1 4DT, England; E-mail [email protected]; corresponding author. and John G. Sessions, Department of Economics and Intemational Development, University of Bath, Bath BA2 7AY, England; E-mail [email protected].

We are particularly grateful to two anonymous referees for excellent comments. We are grateful to the Data Archive at the University of Essex for supplying the British Social Attitudes Surveys 1985, 1987, 1993, and 1996, the British Household Panel Surveys 1991-1999, and the Family Expenditure Surveys 1997/98, 1998199, and 1999/00. Our analysis has benefited from discussions with Saul Estrin, Gianni De Fraja, and Stephen Pudney. Helpful comments were also received from seminar participants at the Universities of Brunei, Essex, Kent, Lancaster, Leicester, and Sheffield. The normal disclaimer applies.

Received June 2004; accepted June 2005.

(1) Uncertainty may be related, for example, to conditions in the output market.

(2) Equation 1 defines PRP as any contract in which current pay is related to worker performance broadly defined. The practical operation of such schemes is certainly more nuanced; see, for example, Blinder (1990), Booth and Frank (1999).

(3) Clearly, the four schemes are diverse in nature and, as such, create different incentives. Ideally, we would classify individuals according to the type of PRP scheme. Such an approach would be somewhat problematic, however, because of the low number of observations in each scheme.

(4) In order to abstract from issues related to labor market participation, we focus on male employees only.

(5) Before this period, the dataset had a different structure, and some of the variables required for our analysis are not available.

(6) We are grateful to an anonymous referee for highlighting this point.

(7) The specific questions are: Promotion--If you have been promoted or changed grades, please give me the date of that change; Annual Increment--Some people can normally expect their pay to rise every year by moving to the next point on the scale as well as receiving negotiated pay rises. Are you paid on this type of incremental scale? Zero-one dummy variables were created from both of these questions.

(8) There is also some distinction between the BSAS/FES and BHPS definitions of self-employment. In the BSAS and FES, individuals are categorized according to the following question: In your main job, are you an employee or a self-employee? In the BHPS, individuals are asked to specify their current labor force status, with options including "paid employment," "selfemployment," "unemployed," "retired," "on maternity leave," "in family care," "full-time student," "long-term sick/ disabled," "on a government training scheme," and "other."

(9) A small number of individuals with more than one job, individuals employed by the armed forces, and agricultural workers were excluded from the analysis of each data set.

(10) The definition of hourly earnings differs across our three data sets according to the survey questions asked of respondents. For the BHPS, hourly earnings are defined as labor income in the previous month divided by the number of hours normally worked per month. For the BSAS, they are defined as the respondent's gross annual earnings divided by the number of hours the respondent works per week multiplied by 52. For the FES, hourly employed earnings are defined as the normal gross weekly wage divided by usual weekly hours, whereas hourly self-employed earnings are defined as normal gross income from self-employment divided by usual weekly hours worked.

(11) Following Murphy and Welch (1990), we also experimented with cubic and quartic experience terms; the results are available on request.

(12) However, where a good match has been made between the employer and the employee, earnings will be relatively high, and tenure will be relatively long (Jovanovic 1979). Cross-sectional data, which contain no information about the quality of a job match, may bias estimates of the returns to job tenure.

(13) The fixed-effects estimation results are available from the authors by request.

(14) The overidentifying instruments for the underlying sample selection model included region of residence and a variety of demographic controls (ethnicity, marital status, number of children, previous spells of unemployment, and private education). The sample selection models, which were estimated using probit analysis with three discrete outcomes, are generally well specified. For reasons of brevity, the sample selection results are not presented here but are available from the authors on request. The results reported in Tables 2 to 4 have all been corrected for sample selection bias (uncorrected results are also available on request).

(15) As pointed out by an anonymous referee, the shape of the earnings profiles may be influenced by trade union membership, unionization of the workplace, and employment in the public sector. Our results are robust to the inclusion of such controls.

(16) Because we are primarily interested in the relative slopes of the earnings profiles, the intercept terms in Figures 1 to 5 have been set to zero.

(17) The results are available on request.

(18) We are grateful to an anonymous referee for highlighting such issues.

Table 1. Summary Statistics: Key Variables

 Fixed-Wage
 (N = 1467)

 Standard
Variable Mean Deviation

British Social Attitudes Surveys 1985, 1987, 1993, 1996
 Log hourly earnings 1.716 0.619
 Years in labor force (YILF) 21.722 12.694
 Age 38.644 11.902
 Employer tenure (a) 9.227 11.505
 Years of education 11.941 2.191
 Degree 0.192 0.394
 Further education 0.179 0.384
 A level 0.141 0.348
 GCSE grades A to C 0.197 0.398
 GCSE grades below C 0.065 0.247
 Other qualification 0.013 0.113

 Fixed-Wage
 (N = 14,284)

 Standard
Variable Mean Deviation

British Household Panel Survey 1991-1999
 Log hourly earnings 1.926 0.517
 Years in labor force (YILF) 24.820 12.435
 Years of education 13.224 4.018
 Job tenure 4.944 6.451
 Age 38.088 11.568
 Annual increment 0.294 0.455
 Promotion 0.053 0.224
 Degree 0.177 0.381
 Further education 0.241 0.428
 A level 0.144 0.351
 GCSE grades A to C 0.195 0.396
 GCSE grades below C 0.195 0.396
 Other qualification 0.035 0.185

 Fixed-Wage
 (N = 5965)

 Standard
Variable Mean Deviation

Family Expenditure Survey 1997/98, 1998/99, and 1999/00
 Log hourly earnings 2.120 0.564
 Years in labor force (YILF) 22.283 12.413
 Age 39.644 11.618
 Years of education 12.361 2.777
 Degree 0.224 0.417
 Further education/A level 0.192 0.394
 GCSE 0.365 0.482
 Less than GCSE 0.218 0.413

 PRP (N = 783)

 Standard
Variable Mean Deviation

British Social Attitudes Surveys 1985, 1987, 1993, 1996
 Log hourly earnings 1.771 0.610
 Years in labor force (YILF) 22.253 12.325
 Age 38.789 11.643
 Employer tenure (a) 10.757 10.713
 Years of education 11.540 1.885
 Degree 0.102 0.303
 Further education 0.171 0.377
 A level 0.160 0.367
 GCSE grades A to C 0.223 0.417
 GCSE grades below C 0.089 0.286
 Other qualification 0.010 0.101

 PRP (N = 6212)

 Standard
Variable Mean Deviation

British Household Panel Survey 1991-1999
 Log hourly earnings 2.029 0.515
 Years in labor force (YILF) 23.588 11.754
 Years of education 13.214 3.485
 Job tenure 4.468 6.106
 Age 36.830 11.005
 Annual increment 0.418 0.493
 Promotion 0.113 0.316
 Degree 0.159 0.366
 Further education 0.278 0.448
 A level 0.160 0.366
 GCSE grades A to C 0.206 0.404
 GCSE grades below C 0.054 0.225
 Other qualification 0.032 0.176

 PRP (N = 1201)

 Standard
Variable Mean Deviation

Family Expenditure Survey 1997/98, 1998/99, and 1999/00
 Log hourly earnings 2.492 0.574
 Years in labor force (YILF) 21.767 11.247
 Age 39.493 10.376
 Years of education 12.726 2.735
 Degree 0.277 0.448
 Further education/A level 0.216 0.412
 GCSE 0.357 0.479
 Less than GCSE 0.149 0.356

 Self-Employed
 (N = 491)

 Standard
Variable Mean Deviation

British Social Attitudes Surveys 1985, 1987, 1993, 1996
 Log hourly earnings 1.613 0.802
 Years in labor force (YILF) 25.193 12.398
 Age 41.433 12.258
 Employer tenure (a) -- --
 Years of education 11.566 1.968
 Degree 0.112 0.316
 Further education 0.147 0.354
 A level 0.175 0.380
 GCSE grades A to C 0.220 0.415
 GCSE grades below C 0.104 0.305
 Other qualification 0.016 0.127

 Self-Employed
 (N = 3716)

 Standard
Variable Mean Deviation

British Household Panel Survey 1991-1999
 Log hourly earnings 1.855 0.842
 Years in labor force (YILF) 29.939 11.872
 Years of education 12.925 3.933
 Job tenure 8.409 8.544
 Age 42.903 11.078
 Annual increment -- --
 Promotion -- --
 Degree 0.120 0.325
 Further education 0.249 0.432
 A level 0.126 0.332
 GCSE grades A to C 0.222 0.416
 GCSE grades below C 0.044 0.205
 Other qualification 0.047 0.212

 Self-Employed
 (N = 1239)

 Standard
Variable Mean Deviation

Family Expenditure Survey 1997/98, 1998/99, and 1999/00
 Log hourly earnings 1.601 1.085
 Years in labor force (YILF) 27.195 11.104
 Age 44.320 10.535
 Years of education 12.125 2.810
 Degree 0.194 0.395
 Further education/A level 0.157 0.364
 GCSE 0.362 0.481
 Less than GCSE 0.287 0.453

(a) Figures relate to 1993 and 1996 only because this information
was provided only in these years.

Table 2. British Social Attitudes Survey 1985, 1987, 1993, 1996:
Dependent Variable, Log Hourly Earnings

 Fixed-Wage

 Spec. A

 Coeff T Stat

Age 0.0675 9.93 *
[Age.sup.2] -0.0007 -8.47 *
YILF -- --
[YILF.sup.2] -- --
Employer tenure -- --
Employer [tenure.sup.2] -- --
Years of education 0.0317 3.47 *
Selectivity term 0.0088 0.13
Constant -1.2737 -6.40 *

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 48.2
[R.sup.2] 0.5797
F statistic [87.80.sub.(24, 1442)]
Number of obs. 1467

 Fixed-Wage

 Spec. B

 Coeff T Stat

Age -- --
[Age.sup.2] -- --
YILF 0.0483 13.38 *
[YILF.sup.2] -0.0008 10.57 *
Employer tenure -- --
Employer [tenure.sup.2] -- --
Years of education 0.0439 4.64 *
Selectivity term -0.2068 -2.97 *
Constant -0.3485 -2.08

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 30.2
[R.sup.2] 0.5675
F statistic [87.50.sub.(24, 1442)]
Number of obs. 1467

 Fixed-Wage

 Spec. C

 Coeff T Stat

Age -- --
[Age.sup.2] -- --
YILF -- --
[YILF.sup.2] -- --
Employer tenure 0.0234 8.32 *
Employer [tenure.sup.2] -0.0003 -4.85 *
Years of education 0.0104 0.90
Selectivity term 0.0740 0.93
Constant 1.0783 5.64 *

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 39.0
[R.sup.2] 0.4541
F statistic [30.76.sub.(22, 801)]
Number of obs. 1467

 PRP

 Spec. A

 Coeff T Stat

Age 0.0551 5.97*
[Age.sup.2] -0.0006 -4.86*
YILF -- --
[YILF.sup.2] -- --
Employer tenure -- --
Employer [tenure.sup.2] -- --
Years of education 0.0301 2.54
Selectivity term -0.4043 -3.97 *
Constant -0.2468 -0.74

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 45.9
[R.sup.2] 0.6499
F statistic [64.80.sub.(24, 758)]
Number of obs. 783

 PRP

 Spec. B

 Coeff T Stat

Age -- --
[Age.sup.2] -- --
YILF 0.0360 7.36*
[YILF.sup.2] -0.0005 -5.25*
Employer tenure -- --
Employer [tenure.sup.2] -- --
Years of education 0.0421 3.38 *
Selectivity term -0.3600 -4.50 *
Constant 0.3199 1.28

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 36.0
[R.sup.2] 0.651
F statistic [71.66.sub.(24, 758)]
Number of obs. 783

 PRP

 Spec. C

 Coeff T Stat

Age -- --
[Age.sup.2] -- --
YILF -- --
[YILF.sup.2] -- --
Employer tenure 0.0159 3.73 *
Employer [tenure.sup.2] -0.0002 -3.67 *
Years of education 0.0315 1.66
Selectivity term -0.2590 -2.48
Constant 1.5331 3.93 *

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 39.8
[R.sup.2] 0.434
F statistic [17.22.sub.(22, 335)]
Number of obs. 783

 Self-
 Employed

 Spec. A

 Coeff T Stat

Age 0.0171 0.80
[Age.sup.2] -0.0001 -0.51
YILF -- --
[YILF.sup.2] -- --
Employer tenure -- --
Employer [tenure.sup.2] -- --
Years of education 0.0647 1.52
Selectivity term 0.8313 10.90 *
Constant 0.4793 0.54

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 85.5
[R.sup.2] 0.6882
F statistic [120.16.sub.(24, 466)]
Number of obs. 491

 Self-
 Employed

 Spec. B

 Coeff T Stat

Age -- --
[Age.sup.2] -- --
YILF 0.0175 1.73
[YILF.sup.2] -0.0002 -1.08
Employer tenure -- --
Employer [tenure.sup.2] -- --
Years of education 0.0241 0.73
Selectivity term 0.2151 2.12
Constant 0.7479 1.07

Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
Turning point (years) 74
[R.sup.2] 0.303
F statistic [16.99.sub.(24, 466)]
Number of obs. 491

Highest education certificates: Degree, Further education, A level,
GCSE grades A-C, GCSE grades below C, Other qualification.

Occupation dummies: Professional, Other nonmanual, Skilled manual,
Semiskilled, Unskilled manual.

Industry dummies: Energy, Metal extraction, Metal goods, Other
manufacturing, Construction, Distribution, Transport and
communications, Banking, Other services.

Robust standard errors are reported.

Tests of equality of the estimated coefficients of age and age
squared across the PRP (self-employed) and fixed-wage (PRP) wage
equations led to a test statistic of 7.23 (8.67).

Tests of equality of the estimated coefficients of YILF and
YILF squared across the PRP (self-employed) and fixed-wage (PRP)
wage equations led to a test statistic of 7.10 (8.46).

Tests of equality of the estimated coefficients of Employer tenure
and Employer tenure squared across the PRP and fixed-wage wage
equations led to a test statistic of 6.25.

* Statistically significant at the 1 % level for a two-tailed test.

Table 3. British Household Panel Survey: 1991-1999:
Dependent Variable, Log Hourly Earnings

 Fixed-Wage

 Spec. A Spec. B

 Coeff T Stat Coeff T Stat

Age 0.0895 35.29 * -- --
[Age.sup.2] -0.0010 -30.83 * -- --
YILF -- -- 0.0578 37.00 *
[YILF.sup.2] -- -- -0.0008 -30.27 *
Job tenure -- -- -- --
Job [tenure.sup.2] -- -- -- --
Years of
 education -0.0007 -0.44 0.0155 9.13 *
Selectivity term -0.1025 -10.58 * -0.1059 -10.92 *
Promotion 0.0284 2.61 * 0.0292 2.68 *
Annual increment 0.0076 1.19 0.0078 1.22
Constant -0.2252 -4.19 * 0.6561 18.33 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation
 dummies Yes Yes
Industry dummies Yes Yes
Turning point (yr) 44.8 36.3
[R.sub.2] within 0.0780 0.0749
[R.sub.2] between 0.4186 0.4196
[R.sub.2] overall 0.3926 0.3918
Wald chi-square [4145.sub.30] [4124.sub.30]
Number of obs. 14,284 14,284
Nos. of groups 4594 4594

 Fixed-Wage PRP

 Spec. C Spec. A

 Coeff T Stat Coeff T Stat

Age -- -- 0.0821 21.99 *
[Age.sup.2] -- -- -0.0009 -18.90 *
YILF -- -- -- --
[YILF.sup.2] -- -- -- --
Job tenure 0.0192 15.00 * -- --
Job [tenure.sup.2] -0.0005 -9.73 * -- --
Years of
 education 0.0016 0.93 0.0001 0.04
Selectivity term -0.0676 -6.79 * 0.0457 2.85 *
Promotion 0.0661 5.68 * 0.4917 4.17 *
Annual increment 0.0011 0.17 0.0191 2.24
Constant 1.5417 53.44 * -0.0901 -1.05

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation
 dummies Yes Yes
Industry dummies Yes Yes
Turning point (yr) 19.2 45.6
[R.sub.2] within 0.0356 0.1153
[R.sub.2] between 0.348 0.4284
[R.sub.2] overall 0.3269 0.4206
Wald chi-square [2507.sub.30] [2538.sub.30]
Number of obs. 14,284 6212
Nos. of groups 4594 2806

 PRP PRP

 Spec. B Spec. C

 Coeff T Stat Coeff T Stat

Age -- -- -- --
[Age.sup.2] -- -- -- --
YILF 0.0542 23.72 * -- --
[YILF.sup.2] -0.0008 -19.02 * -- --
Job tenure -- -- 0.0151 7.18 *
Job [tenure.sup.2] -- -- -0.0003 -3.88 *
Years of
 education 0.0147 6.17 * 0.0017 0.72
Selectivity term 0.0476 2.96 * -0.0099 -0.59
Promotion 0.0500 4.24 * 0.0777 5.97 *
Annual increment 0.0195 2.29 0.0030 0.34
Constant 0.7021 11.69 * 1.6222 33.70 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation
 dummies Yes Yes
Industry dummies Yes Yes
Turning point (yr) 33.9 25.2
[R.sub.2] within 0.1152 0.057
[R.sub.2] between 0.4304 0.3618
[R.sub.2] overall 0.4195 0.3536
Wald chi-square [2553.sub.30] [1668.sub.30]
Number of obs. 6212 6212
Nos. of groups 2806 2806

 Self-Employed Self-Employed

 Spec. A Spec. B

 Coeff T Stat Coeff T Stat

Age 0.0607 2.26 -- --
[Age.sup.2] -0.0007 -2.33 -- --
YILF -- -- 0.0336 1.96
[YILF.sup.2] -- -- -0.0006 -2.11
Job tenure -- -- -- --
Job [tenure.sup.2] -- -- -- --
Years of
 education -0.0086 -0.58 -0.0064 -0.41
Selectivity term 0.0019 0.03 0.0046 0.06
Promotion -- -- -- --
Annual increment -- -- -- --
Constant -0.1138 -0.18 0.6329 1.45

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation
 dummies Yes Yes
Industry dummies Yes Yes
Turning point (yr) 43.4 28
[R.sub.2] within 0.0099 0.0097
[R.sub.2] between 0.0353 0.0345
[R.sub.2] overall 0.0381 0.0389
Wald chi-square [68.45.sub.28] [67.49.sub.28]
Number of obs. 3716 3716
Nos. of groups 1153 1153

 Self-Employed

 Spec. C

 Coeff T Stat

Age -- --
[Age.sup.2] -- --
YILF -- --
[YILF.sup.2] -- --
Job tenure 0.0022 0.22
Job [tenure.sup.2] 0.0002 0.80
Years of
 education -0.0059 -0.40
Selectivity term 0.0362 0.51
Promotion -- --
Annual increment -- --
Constant 0.9139 3.08 *

Year dummies Yes
Highest ed. cert. Yes
Occupation
 dummies Yes
Industry dummies Yes
Turning point (yr) 5.5
[R.sub.2] within 0.0101
[R.sub.2] between 0.0337
[R.sub.2] overall 0.0383
Wald chi-square [67.82.sub.28]
Number of obs. 3716
Nos. of groups 1153

Highest education certificates: Degree, Further education,
A level, GCSE grades A to C, GCSE grades below C, Other qualification.

Occupation dummies: Managerial, Professional, Intermediate nonmanual,
Sales, Clerical, Personal services, Skilled manual, Semiskilled manual,
Unskilled manual.

Industry dummies: Energy, Extraction, Engineering, Manufacturing,
Construction, Distribution, Transport, Storage and communication,
Finance, Other nonmanufacturing.

Robust standard errors are reported.

Tests of equality of the estimated coefficients of age and age
squared across the PRP (self-employed) and fixed-wage (PRP) wage
equations led to a test statistic of 18.37 (126.76).

Tests of equality of the estimated coefficients of YILF and YILF
squared across the PRP (self-employed) and fixed-wage (PRP) wage
equations led to a test statistic of 1.82 (18.08).

Tests of equality of the estimated coefficients of job tenure
and job tenure squared across the PRP (self-employed) and
fixed wage (PRP) wage equations led to a test statistic of
12.30 (40.98).

* Statistically significant at the 1% level for a two-tailed test.

Table 4. Family Expenditure Survey 1997/98, 1998/99, and 1999/00:
Dependent Variable, Log Hourly Earnings

 Fixed-Wage

 Spec. A Spec. B

 Coeff T Stat Coeff T Stat

Age 0.0677 15.59 * -- --
[Age.sup.2] -0.0007 -14.28 * -- --
YILF -- -- 0.0422 15.44 *
[YILF.sup.2] -- -- -0.0007 -14.36 *
Years of education 0.0160 2.10 0.0311 4.33 *
Selectivity term 0.1679 2.99 * 0.1956 3.04 *
Constant 0.7400 6.54 * 1.5724 17.09 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation dummies Yes Yes
Industry dummies Yes Yes
Turning point (years) 47.9 30.1
[R.sup.2] 0.3306 0.3304
F statistic [141.13.sub. [143.52.sub.
 (25, 5939)] (25, 5939)]
Number of obs. 5965 5965

 PRP

 Spec. A Spec. B

 Coeff T Stat Coeff T Stat

Age 0.0553 5.27 * -- --
[Age.sup.2] -0.0005 -3.77 * -- --
YILF -- -- 0.0380 7.21 *
[YILF.sup.2] -- -- -0.0005 -4.34 *
Years of education 0.0412 3.15 * 0.0610 4.66 *
Selectivity term -0.4378 -4.15 * -0.4896 -4.47 *
Constant 1.3877 4.02 * 2.0280 8.20 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation dummies Yes Yes
Industry dummies Yes Yes
Turning point (years) 55.3 38.0
[R.sup.2] 0.4356 0.4430
F statistic [33.48.sub. [35.88.sub.
 (25, 1201)] (25, 1201)]
Number of obs. 1201 1201

 Self-Employed

 Spec. A Spec. B

 Coeff T Stat Coeff T Stat

Age -0.0019 -0.06 -- --
[Age.sup.2] -0.0001 -0.40 -- --
YILF -- -- -0.024 -1.31
[YILF.sup.2] -- -- 0.0001 0.47
Years of education -0.0162 -0.68 -0.0388 -1.53
Selectivity term -0.403 -1.57 -0.6098 -2.44
Constant 3.4795 2.61 * 4.4272 4.30 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation dummies Yes Yes
Industry dummies Yes Yes
Turning point (years) 9.5 120.0
[R.sup.2] 0.3306 0.1522
F statistic [141.13.sub. [10.26.sub.
 (24, 1214)] (24, 1214)]
Number of obs. 1239 1239

Highest education certificates: Degree, Further education, A level,
GCSE grades A-C, GCSE grades below C, Other qualification.

Occupation dummies: Professional, Other nonmanual, Skilled manual,
Semiskilled, Unskilled manual.

Industry dummies: Energy, Metal extraction, Metal goods, Other
manufacturing, Construction, Distribution, Transport and
communications, Banking, Other services.

Robust standard errors are reported.

Tests of equality of the estimated coefficients of age and age
squared across the PRP (self-employed) and fixed-wage (PRP) wage
equations led to a test statistic of 56.02 (163.98).

Tests of equality of the estimated coefficients of YILF and YILF
squared across the PRP (self-employed) and fixed-wage (PRP) wage
equations led to a test statistic of 68.71 (345.83).

* Statistically significant at the 1% level for a two-tailed test.

Table 5. Lambda ([??]) Summary Statistics

 Family Expenditure Survey 1997/98,
 1998/99, 1999/00

[??] All Workers PRP Workers

Mean 0.152 0.054
Standard deviation 0.349 0.069
Minimum 0.000 9.28 x [10.sup.-6]
Maximum 1.000 0.476
Observations 8405 1201

 British Household Panel
 Survey 1997-1999

[??] All Workers PRP Workers

Mean 0.234 0.302
Standard deviation 0.364 0.223
Minimum 0.000 0.001
Maximum 1.000 0.981
Observations 9187 2840

Table 6. Continuous Lambda, FES 1997/98, 1998/99, 1999/00:
Dependent Variable, Log Hourly Earnings

 Spec. 1 Spec. 2

 Coeff T Stat Coeff T Stat

Age 0.0755 18.08 * 0.0753 19.06 *
[Age.sup.2] -0.0008 -15.75 * -0.0008 -16.71 *
YILF -- -- -- --
[YILF.sup.2] -- -- -- --
Lambda -0.5354 -16.81 * -- --
Lambda * age -- -- -0.0193 -5.69 *
Lambda * [age.sup.2] -- -- 0.0002 2.21
Lambda * YILF -- -- -- --
Lambda * [YILF.sup.2] -- -- -- --
Years of education 0.0109 1.56 0.0112 1.60
Constant 0.8587 7.21 * 0.8444 7.23 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation dummies Yes Yes
Industry dummies Yes Yes
[R.sup.2] 0.3071 0.3088
F statistic [154.92.sub. [152.53.sub.
 (25, 8379)] (26, 8378)]
Nos. of obs. 8405 8405

 Spec. 3 Spec. 4

 Coeff T Stat Coeff T Stat

Age 0.0783 21.06 * -- --
[Age.sup.2] -0.0008 -18.45 * -- --
YILF -- -- 0.0457 20.48 *
[YILF.sup.2] -- -- -0.0008 -17.03 *
Lambda 0.6085 1.37 -0.5566 -17.51 *
Lambda * age -0.0478 -2.22 -- --
Lambda * [age.sup.2] 0.0005 1.88 -- --
Lambda * YILF -- -- -- --
Lambda * [YILF.sup.2] -- -- -- --
Years of education 0.0111 1.58 0.0257 3.78 *
Constant 0.7882 6.94 * 1.826 19.36 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation dummies Yes Yes
Industry dummies Yes Yes
[R.sup.2] 0.3088 0.3077
F statistic [152.53.sub. [[156.21.sub.
 (26, 8378)] (25, 8379)]
Nos. of obs. 8405 8405

 Spec. 5 Spec. 6

 Coeff T Stat Coeff T Stat

Age -- -- -- --
[Age.sup.2] -- -- -- --
YILF 0.0488 22.77 * 0.0489 23.19 *
[YILF.sup.2] -0.0008 -19.29 * -0.0008 -19.57 *
Lambda -- -- 0.0259 0.19
Lambda * age -- -- -- --
Lambda * [age.sup.2] -- -- -- --
Lambda * YILF -0.0413 -9.91 * -0.0432 -3.72 *
Lambda * [YILF.sup.2] 0.0007 5.59 * 0.0007 3.15 *
Years of education 0.0256 3.76 * 0.0256 3.76 *
Constant 1.7869 19.04 * 1.7850 19.03 *

Year dummies Yes Yes
Highest ed. cert. Yes Yes
Occupation dummies Yes Yes
Industry dummies Yes Yes
[R.sup.2] 0.3113 0.3113
F statistic [156.61.sub. [155.47.sub.
 (26, 8378)] (27, 8377)]
Nos. of obs. 8405 8405

Highest education certificates, occupation dummies, and industry
dummies are as in Table 4.

Standard errors adjusted according to the White-Huber approach
because of the presence of heteroskedasticity.

Lambda = PRP/total pay.

* Statistically significant at the 1% level for a two-tailed test.

Table 7. Continuous Lambda, BHPS 1997-1999:
Dependent Variable, Log Hourly Earnings

 Spec. 1

 Coeff T Stat

Age 0.0872 10.81 *
Age (2) -0.0010 -10.04 *
YILF -- --
YILF (2) -- --
Job tenure -- --
Job tenure (2) -- --
Lambda -0.2620 -7.51 *
Lambda*age -- --
Lambda*age (2) -- --
Lambda*YILF -- --
Lambda*YILF (2) -- --
Lambda*job tenure -- --
Lambda*job tenure (2) -- --
Years of education -0.0071 -1.74
Constant -0.1690 -1.01
Year dummies
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0020
[R.sub.2] between 0.1647
[R.sub.2] overall 0.1473
Wald chi-squared [808.sub.26]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 2

 Coeff T Stat

Age 0.0791 9.58 *
Age (2) -0.0009 -8.17 *
YILF -- --
YILF (2) -- --
Job tenure -- --
Job tenure (2) -- --
Lambda -- --
Lambda*age 0.0063 1.74
Lambda*age (2) -0.0003 -3.83 *
Lambda*YILF -- --
Lambda*YILF (2) -- --
Lambda*job tenure -- --
Lambda*job tenure (2) -- --
Years of education -0.0069 -1.69
Constant -0.1071 -0.63
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0034
[R.sub.2] between 0.1677
[R.sub.2] overall 0.1498
Wald chi-squared [848.sub.27]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 3

 Coeff T Stat

Age 0.0774 8.56 *
Age (2) -0.0008 -7.34 *
YILF -- --
YILF (2) -- --
Job tenure -- --
Job tenure (2) -- --
Lambda -0.1945 -0.46
Lambda*age 0.0159 0.75
Lambda*age (2) -0.0004 -1.59
Lambda*YILF -- --
Lambda*YILF (2) -- --
Lambda*job tenure -- --
Lambda*job tenure (2) -- --
Years of education -0.0069 -1.69
Constant -0.0748 -0.41
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0034
[R.sub.2] between 0.1676
[R.sub.2] overall 0.1496
Wald chi-squared [847.sub.28]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 4

 Coeff T Stat

Age -- --
Age (2) -- --
YILF 0.0510 10.55 *
YILF (2) -0.0008 -9.39 *
Job tenure -- --
Job tenure (2) -- --
Lambda -0.2637 -7.55 *
Lambda*age -- --
Lambda*age (2) -- --
Lambda*YILF -- --
Lambda*YILF (2) -- --
Lambda*job tenure -- --
Lambda*job tenure (2) -- --
Years of education 0.0021 0.48
Constant 0.8103 7.82 *
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0021
[R.sub.2] between 0.1620
[R.sub.2] overall 0.1453
Wald chi-squared [792.sub.26]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 5

 Coeff T Stat

Age -- --
Age (2) -- --
YILF 0.0483 9.63 *
YILF (2) -0.0007 -7.51 *
Job tenure -- --
Job tenure (2) -- --
Lambda -- --
Lambda*age -- --
Lambda*age (2) -- --
Lambda*YILF -0.0023 -0.56
Lambda*YILF (2) -0.0002 -2.28
Lambda*job tenure -- --
Lambda*job tenure (2) -- --
Years of education 0.0023 0.54
Constant 0.7823 7.56 *
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0037
[R.sub.2] between 0.1657
[R.sub.2] overall 0.1486
Wald chi-squared [839.sub.27]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 6

 Coeff T Stat

Age -- --
Age (2) -- --
YILF 0.0525 9.74 *
YILF (2) -0.0008 -7.79 *
Job tenure -- --
Job tenure (2) -- --
Lambda 0.3631 2.13
Lambda*age -- --
Lambda*age (2) -- --
Lambda*YILF -0.0280 -2.20
Lambda*YILF (2) 0.0002 0.73
Lambda*job tenure -- --
Lambda*job tenure (2) -- --
Years of education 0.0025 0.58
Constant 0.7265 6.81 *
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.004
[R.sub.2] between 0.1659
[R.sub.2] overall 0.1492
Wald chi-squared [843.sub.28]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 7

 Coeff T Star

Age -- --
Age (2) -- --
YILF -- --
YILF (2) -- --
Job tenure 0.0201 4.90 *
Job tenure (2) -0.0005 -3.35 *
Lambda -0.2412 -6.87 *
Lambda*age -- --
Lambda*age (2) -- --
Lambda*YILF -- --
Lambda*YILF (2) -- --
Lambda*job tenure -- --
Lambda*job tenure (2) -- --
Years of education -0.0050 -1.20
Constant 1.4589 18.66 *
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0025
[R.sub.2] between 0.1485
[R.sub.2] overall 0.1315
Wald chi-squared [689.sub.26]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 8

 Coeff T Stat

Age -- --
Age (2) -- --
YILF -- --
YILF (2) -- --
Job tenure 0.0286 5.85*
Job tenure (2) -0.0006 -3.33 *
Lambda -- --
Lambda*age -- --
Lambda*age (2) -- --
Lambda*YILF -- --
Lambda*YILF (2) -- --
Lambda*job tenure -0.0376 -4.85 *
Lambda*job tenure (2) 0.0007 2.41
Years of education -0.0051 -1.22
Constant 1.4065 18.04 *
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0019
[R.sub.2] between 0.1475
[R.sub.2] overall 0.1313
Wald chi-squared [683.sub.27]
Nos. of observations 9187
Nos. of groups 4464

 Spec. 9

 Coeff T Stat

Age -- --
Age (2) -- --
YILF -- --
YILF (2) -- --
Job tenure 0.0242 4.75 *
Job tenure (2) -0.0005 -2.64 *
Lambda -0.1600 -3.41 *
Lambda*age -- --
Lambda*age (2) -- --
Lambda*YILF -- --
Lambda*YILF (2) -- --
Lambda*job tenure -0.0177 -1.82
Lambda*job tenure (2) 0.0002 0.69
Years of education -0.0050 -1.21
Constant 1.4409 18.36 *
Year dummies Yes
Highest ed. cert. Yes
Occupation dummies Yes
Industry dummies Yes
[R.sub.2] within 0.0019
[R.sub.2] between 0.1520
[R.sub.2] overall 0.1336
Wald chi-squared [698.sub.28]
Nos. of observations 9187
Nos. of groups 4464

Highest education certificates,
occupation dummies, and industry
dummies are as in Table 3.

Standard errors adjusted according
to the White-Huber approach because
of the presence of heteroskedasticity.

Lambda = PRP/total pay.

* Statistically significant at the
1% level for a two-tailed test.

Table 8. Training Frequencies

 PRP Employees

 Standard
 Obs. Mean Deviation

BSAS 1987
 Train
 Exp. < 5 yr 28 0.9286 0.2623
 5 < Exp. < 10 yr 31 0.7742 0.4250
 10 < Exp. < 20 yr 61 0.7541 0.4342
 Exp. > 20 yr 153 0.5882 0.4938
 All employees 273 0.6813 0.4668
 Trains
 Exp. < 5 yr 28 3.8929 2.0788
 5 < Exp. < 10 yr 31 2.6129 2.0278
 10 < Exp. < 20 yr 61 2.4590 2.1952
 Exp. > 20 yr 153 1.7190 1.9717
 All employees 273 2.2088 2.1395
BHPS 1991-1999
 Train 6212 0.1718 0.3772
 Exp. < 5 yr 68 0.3088 0.4654
 5 < Exp. < 10 yr 765 0.2065 0.4051
 10 < Exp. < 20 yr 2015 0.1856 0.3889
 Exp. > 20 yr 3364 0.1528 0.3598
 All employees 6212 0.1718 0.3772

 Fixed-Wage Employees

 Standard
 Obs. Mean Deviation

BSAS 1987
 Train
 Exp. < 5 yr 50 0.9000 0.3030
 5 < Exp. < 10 yr 52 0.7885 0.4124
 10 < Exp. < 20 yr 101 0.6931 0.4635
 Exp. > 20 yr 203 0.6847 0.4658
 All employees 406 0.7266 0.4463
 Trains
 Exp. < 5 yr 50 3.7000 2.0923
 5 < Exp. < 10 yr 52 2.6923 2.2798
 10 < Exp. < 20 yr 101 2.4158 2.2058
 Exp. > 20 yr 203 1.9803 1.9447
 All employees 406 2.3916 2.1400
BHPS 1991-1999
 Train 14,284 0.1246 0.3303
 Exp. < 5 yr 198 0.2475 0.4326
 5 < Exp. < 10 yr 1792 0.1735 0.3788
 10 < Exp. < 20 yr 4000 0.1465 0.3537
 Exp. > 20 yr 8294 0.1006 0.3008
 All employees 14,284 0.1246 0.3303

 All Employees

 Standard
 Obs. Mean Deviation

BSAS 1987
 Train
 Exp. < 5 yr 78 0.9103 0.2877
 5 < Exp. < 10 yr 83 0.7831 0.4146
 10 < Exp. < 20 yr 162 0.7160 0.4523
 Exp. > 20 yr 356 0.6433 0.4797
 All employees 679 0.7084 0.4548
 Trains
 Exp. < 5 yr 78 3.7692 2.0758
 5 < Exp. < 10 yr 83 2.6627 2.1768
 10 < Exp. < 20 yr 162 2.4321 2.1951
 Exp. > 20 yr 356 1.8680 1.9579
 All employees 679 2.3181 2.1401
BHPS 1991-1999
 Train 20,496 0.1389 0.3458
 Exp. < 5 yr 266 0.2632 0.4412
 5 < Exp. < 10 yr 2557 0.1834 0.3871
 10 < Exp. < 20 yr 6015 0.1596 0.3663
 Exp. > 20 yr 11,658 0.1156 0.3198
 All employees 20,496 0.1389 0.3458

Table 9. PRP and Training Incidence (Summary of Results)

 BSAS 1987

 Probit (1) Ordered Probit (1)

 Dep Var = Train Dep Var = Trains

 Coeff T Stat Cceff T Stat

Experience -0.0335 -5.95 * -0.0324 -7.47 *
PRP dummy -0.0700 -0.59 -0.0071 -0.08
(Exp. < 5) * PRP -- -- -- --
(5 < Exp. < 10) * PRP -- -- -- --
(10 < Exp. < 20) * PRP -- -- -- --
(Exp. > 20) * PRP -- -- -- --
Log likelihood -339.3235 -1219.0619
Nos. of observations 679 679

 BSAS 1987

 Probit (2) Ordered Probit (2)

 Dep Var = Train Dep Var = Trains

 Coeff T Stat Coeff T Stat

Experience -0.0311 -4.72 * -0.0292 -5.81 *
PRP dummy 0.2736 0.64 0.6316 2.04
(Exp. < 5) * PRP 0.0130 0.10 -0.0878 -0.99
(5 < Exp. < 10) * PRP -0.0525 -0.87 -0.0899 -2.07
(10 < Exp. < 20) * PRP -0.0211 -0.76 -0.0440 -2.15
(Exp. > 20) * PRP -0.0116 -0.88 -0.0211 -2.12
Log likelihood -338.37818 -1215.8068
Nos. of observations 679 679

 BHPS 1991-1999

 Probit (1) Probit (2)

 Dep Var = Train Dep Var = Train

 Coeff T Stat Coeff T Stat

Experience -0.183 10.49 * -0.020 10.15 *
PRP dummy 0.106 3.30 * 0.015 0.12
(Exp. < 5) * PRP -- -- 0.099 2.27
(5 < Exp. < 10) * PRP -- -- -0.002 0.14
(10 < Exp. < 20) * PRP -- -- 0.003 0.40
(Exp. > 20) * PRP -- -- 0.005 1.15
Log likelihood -6738.750 -6733.579
Nos. of observations 20,496 20,496

Controls were also included for occupation,
region, industry, education, firm size,
marital status, and ethnicity.

* Statistically significant at the 1% level
for a two-tailed test.