Has Performance Pay Increased Wage Inequality in Britain?

Using data from the British Household Panel Survey (BHPS) we show performance pay (PP) increased earnings dispersion among men and women, and to a lesser extent among full-time working women, in the decade of economic growth which ended with the recession of 2008. PP was also associated with some compression in the lower half of the wage distribution for women. The effects were predominantly associated with a broad measure of PP that included bonuses. However, these effects were modest and there is no indication that PP became increasingly prevalent, as some had predicted, over the decade prior to recession.


Introduction
Income inequality has grown in English-speaking economies in recent decades, largely due to growing wage inequality (Atkinson, Piketty and Saez, 2011). A variety of explanations have been proffered, including increasing returns to skill induced by skills-biased technological change (SBTC) (Autor, Katz and Kearney, 2008), changes in labour market institutions, most notably de-unionisation (Dustmann et al., 2009;Card, Lemieux and Riddell, 2004) and increased trade (Autor et al., 2013). In their seminal paper for the United States Lemieux, MacLeod and Parent (LMP) (2009) show that performance pay (PP) accounted for one-fifth of the growth in wage inequality among men between the late 1970s and early 1990s, and most of the growth in wage inequality among high earners in the top quintile. They show that PP became more widespread between the 1970s and early 1990s, was closely tied to individuals' productive characteristics, and that the returns to these characteristics were rising faster in PP jobs than in fixed wage jobs. Their findings are consistent with a world in which SBTC increases the rewards for more productive workers and induces firms to resort to PP to both attract and incentivise those workers.
LMP's (2009) model, which draws on the work of Lazear (1986; and Prendergast (1999), indicates PP generates higher wage dispersion than fixed rate pay (FP) due to the sorting of high ability workers into PP jobs -a labour market segmentation type argumentand because PP reflects individuals' marginal product more accurately than fixed wage schedules. Growth in PP jobs allows high ability workers to recoup returns to their ability in a way that is not possible with fixed wages, while the higher incidence of PP at the top end of the earnings distribution will also generate higher wage dispersion.
LMP attribute the increased use of PP to SBTC and the declining costs of worker monitoring due to advances in technology. These trends are likely to have continued in the period since the mid-1990s which LMP were studying, both in the United States and in other industrialised countries. For instance, Sommerfeld (2013) documents an almost continuous rise in the share of PP jobs between 1984 and 2009.
And yet LMP's findings have recently been challenged in a series of papers using data for the United States. Using establishment data from the Bureau of Labor Statistics' Employer Costs for Employee Compensation (ECEC) series (which derives from the National Compensation Survey) Gittleman and Pierce (2013) show the proportion of jobs with PP rose in the 1990s, only to fall in the 2000s such that, by 2013, it had declined by about onefifth since LMP's study period, irrespective of how one measures PP. This decline is apparent throughout the wage distribution but is concentrated among low earners.
Furthermore, in a second paper, Gittleman and Pierce (2015) show the contribution of PP to growth in the earnings distribution in the first decade of the 21st Century has been small -in the order of 9 per cent of the growth in variance. Sommerfeld's analysis for Germany also showed that despite the expansion of PP, it did not lead to increased wage inequality because it was associated with higher wages across the board and not just for high earners.
Two more papers find LMP's basic results do not hold for some parts of the working population. Like LMP, Heywood and Parent (2012) analyse the Panel Survey of Income Dynamics (PSID). They find that, during the period 1976-1998, the tendency for PP to be associated with greater wage inequality at the top of the male earnings distribution applies to white workers but not to black workers. In a second paper using the National Longitudinal 4 Survey of Youth (NLSY), Heywood and Parent (2013) find skilled fathers select into PP jobs, whereas skilled mothers select out of PP jobs, a finding which is not consistent with standard assumptions regarding workers sorting into PP jobs on ability. This, in turn, raises questions about the effects of PP on wage inequality.
In Britain wage inequality among full-time workers has been rising since the late 1970s, although the rate of change slowed dramatically in the 2000s, with all the growth being confined to the top part of the wage distribution (Machin, 2011;Lindley and Machin, 2013).
Over the whole period the graduate wage premium rose, despite growth in the graduate share in employment and hours, suggesting demand for highly skilled labour was exceeding its supply (Lindley and Machin, op. cit.). This is consistent with SBTC, and the authors find direct evidence of greater demand for more educated workers in more technologically advanced industries (op. cit.: 175-176). They also point to the introduction of the national minimum wage in 1999 and its subsequent up-rating as a reason for the stability in the 50-10 wage differential in the 2000s.
Although they point to the potential importance of SBTC in the British context, Lindley and Machin do not consider the potential role played by PP in growing wage inequality. There is some evidence that annual bonuses have contributed to an increase in wage inequality at the top of the earnings distribution in the last decade or so, primarily as a result of large bonus receipt by bankers, traders and other well-paid professionals in the Finance sector Van Reenen, 2010, 2013). 1 These employees may be sharing in the substantial rents 5 generated by a lack of competition in the sector. Alternatively, they may be benefiting from productivity "scaling" effects that accrue to highly productive employees when changes such as increased firm size and capital intensification "scale up" worker productivity, increasing returns to their employer. This is the type of effect identified by Gabaix and Landier (2008) and Kaplan (2012)

7
In the next section we outline the theoretical links between wage dispersion and PP. Section Three then introduces the data, followed by section Four which presents results relating to the incidence and correlates of PP followed by its links to wages and wage dispersion in Britain. Finally section Five discusses the implications of the findings and draws some conclusions.

Wage Dispersion and Performance Pay
In perfectly competitive labour markets in which firms and workers have perfect information employees would be paid their marginal product, that is, they would be paid for their performance. However, employers and employees often prefer fixed wage contracts based on time rather than effort or output. Employers may find fixed wages less costly to administer, especially if labour inputs or outputs are costly to monitor: it can be costly for firms to identify the contribution of individual employees to output, while factors beyond the control of the employee, and even the firm, mean output is affected by factors other than employees' talent and effort. In standard economic theory wage dispersion rises when employees are paid for their performance, compared to a counterfactual scenario in which they are paid a fixed wage. Under fixed wage schedules employees are paid for time worked, whereas under PP they are paid for output. Heterogeneity in individuals' ability to increase output, either by virtue of talent or effort, is ignored in fixed wage schedules, but it does have a bearing on earnings when pay is linked to performance.
There are three channels that may lead to higher earnings dispersion in the presence of PP.
The first is a mechanical effect: PP reveals underlying differences in individuals' productivity that were previously ignored. Second, PP may have the effect of incentivising effort: employees can raise (lower) their earnings through higher (lower) effort such that variance in effort induces variance in earnings, whereas employees' earnings are not a function of effort in fixed wage jobs. Third, employees will sort into (out of) PP jobs according to talent and other traits (such as their tastes for effort and risk) that may affect their earnings. If more able workers sort into PP jobs where they can command higher earnings, while less able workers prefer the guarantee of a fixed wage, the market will segment into high and low earners along PP lines. Thus via all three of these channels, the introduction of PP should lead to greater wage dispersion than might obtain if all workers were paid a fixed wage. 2 Of course this is an over-simplistic picture because job retention and job progression are often performance-related, even when workers are paid a fixed wage, because wage levels and earnings progression reflect workers' efforts and talent, while career concerns can incentivise effort (Prendergast, 1999;Papps et al., 2011). But the link between performance and pay is usually more explicit and more direct in the presence of PP schemes.
While all these considerations suggest that PP will be associated with greater wage dispersion in cross section, the impact of PP on changes in wage dispersion are less clear. LMP (2009: 3-4) discuss some reasons why PP may induce growth in earnings dispersion. If demands for more skilled and more able workers are rising due to SBTC or globalisation, this will raise the market value of more talented workers such that firms may bid up their price relative to less talented workers as they try to influence the job matching process. This, in turn, may induce greater worker sorting between PP and FP jobs, contributing to growth in the dispersion of earnings between PP and fixed pay jobs. If there is an increase in the prevalence of PP, particularly at the top end of the earnings distribution, this will also contribute to a growth in earnings dispersion.

Data
We analyse data from the British Household Panel Survey ( The analyses focus on employees aged 18-64 years, excluding those reporting total weekly hours of 100 or more or 5 or less. This restriction, which affects fewer than 2% of observations, reduces possible measurement error in hourly wages arising from extreme reports of hours worked. It also eliminates very small jobs. To account for the possibility of different wage determination processes across gender, we perform separate analyses for men and women. We also analyse a sample of women in full-time jobs only to make sure that any gender differences do not reflect the much larger proportion of women in part-time jobs (where PP is less common). All our estimates are weighted using the cross-sectional weights 10 provided with the survey, which account for survey design and the likelihood that a respondent appears in a particular wave.
As is standard in the literature our wages measure is hourly wages, which we compute as (usual gross pay/(usual basic hours+1.5 X usual paid overtime). The usual gross pay variable includes regular bonuses, commission and tips, so the hourly wage measure will take account of these components of PP. We can also construct a second hourly pay measure including more irregular bonuses (such as seasonal bonuses), derived from a separate question in the survey. While in principal this second measure better reflects total bonus payments received, it carries a risk of double counting if respondents report some bonus payments in answer to both questions (the second question does not explicitly exclude all regular bonuses). As a result we use the first wage measure as our baseline dependent variable, but as a robustness check we also run all analyses with the wage measure including irregular bonuses (the results are almost identical).
BHPS contains two measures of PP. The first, relating to bonuses, is derived from the question: "In the last 12 months have you received any bonuses such as a Christmas or quarterly bonus, profit-related pay or profit-sharing bonus, or an occasional commission?". The second measure relates specifically to performance-related pay (PRP).
Respondents are asked: "Does your pay include performance related pay?" The bonus question was asked in Waves 6-18 and the PRP question in Waves 8-18. As we wish to combine information from the two measures we focus on Waves 8-18 covering the period 1998-2008. 4 Gittleman and Pierce (2013) emphasise the importance of recognising that PP measures often capture different types of PP, some more closely related to individual productivity than others. In our data, the PRP question arguably captures pay linked to individual performance, while the bonus question captures payments like Christmas bonuses and rewards, such as profit related pay, that are probably linked to team or firm performance.
Across the pooled sample PRP is roughly half as prevalent as bonus receipt (15% compared to 32%, see Table 1).  (Table 2). So to some extent PRP and bonus receipt are distinct types of compensation. As we show later it is also the case that employees in PRP and bonus jobs differ somewhat in their characteristics.  Table 2 shows that 62% of employees did not get either PRP or bonus, thus the prevalence of PP broadly measured is 38%, whereas 15% receive PP narrowly defined.
Tables 1 and 2 above relate to the receipt of PP. However, throughout the analysis presented in Section Four we follow other papers in the literature by focusing on PP jobs, not receipt.
A job is a period of employment with the same employer in the same "grade", i.e. if they get promoted it is a new job. A job is classified as a PP job where the respondent has been in receipt of PP on at least one occasion. This adjustment is made in recognition of that fact that some jobs are PP jobs but that, for whatever reason (poor performance on the part of the firm or individual, for instance) there has been no receipt of PP in a particular year -that is to say, the respondent may be in a job that pays for performance but, in a given year, the PP due is £0, thus making it hard to distinguish from a fixed pay job.
year, and also combine the bonus and PRP measures for some of the analysis, we do not use the earlier bonus question.

13
Since the probability of a PP job paying out for performance is partly a function of the number of times that job is observed in the data (which is lower for jobs near the ends of the data window) it is necessary to make an "endpoint adjustment" which accounts for the presence of jobs of different durations. Following LMP's approach we construct an adjusted measure of the prevalence of PP jobs by estimating probit models for the probability of appearing in each wave of the data based on the number of times a job is observed. The resulting predicted probabilities are used to construct a weight which then effectively holds the distribution of the number of times a job is observed to that observed in the middle of our sample. 14 Figures 1a shows the incidence of PP jobs using the broad measure (PRP plus bonuses). It is apparent that men are more likely to be in PP jobs than women and that, among women, PP jobs are more common in full-time jobs. Throughout the period a little under two-thirds of jobs undertaken by men were PP jobs. Among women, the figure is closer to one-half. The incidence of PP jobs declines a little over the decade before the recession began, the drop being particularly notable among full-time women. There is no evidence at all that PP jobs became more common except perhaps a small rebound among women towards the end of the period. is based solely on the performance-related pay question referred to in Section Three. The male-female gap is smaller on this narrow measure, and it narrows over the period since 1998 because the percentage of jobs undertaken by men that are PP jobs has been falling.

The Prevalence of PP and Its Correlates
The percentage of PP jobs among women was broadly stable, though as for the broad measure, there is some indication that it may have increased slightly after 2006. Overall, though, we see no clear evidence that PP jobs expanded over this period. Table 3 shows a bivariate probit estimated for the pooled sample which establishes the correlates of bonus-paying jobs and PRP jobs, having accounted for the positive and statistically significant correlation in unobservables between the two. There are a number of points worth noting. First, consistent with the graphical evidence, the incidence of both bonus jobs and PRP jobs has declined significantly since the turn of the century having conditioned on employees' demographic, job and workplace characteristics. For PRP jobs there was an abrupt decline in 2000 followed by further decline after 2006. Bonus jobs also fell sharply in 2000 but then declined more steadily. Second, those in receipt of both types of PP have higher gross hourly earnings than those in fixed pay jobs: even after accounting for occupation, tenure, and other characteristics that influence wages (such as firm size) those in PP jobs have gross hourly wages that are around one-third higher than those among observationally equivalent fixed pay employees.
Third, the male-female differential in PP jobs, apparent in the figures above, is not significant having accounted for other factors. PP jobs are more likely to be full-time, permanent, and in managerial, clerical and sales occupations. PRP jobs are more likely to be unionised than fixed wage jobs, but this is not the case for bonus jobs. The quadratic in years of job tenure turns at about 18 years for bonus jobs and 15 years for PRP jobs, both of which are above the 90th percentile of the job tenure distribution, so the probability of bonuses increases in tenure for most employees. Unsurprisingly both types of PP job are more likely in larger organizations and the industry patterns are as found in the literature.

Is there a Performance Pay Premium?
Before looking at the growth in wage dispersion in Britain and the role PP may have played we run log hourly wage regressions to establish whether there is a PP premium at the mean and, if so, how much of it can be explained by the selection of workers into jobs. The results are presented in Table 4. The first column shows the raw wage gap between those in PP jobs and those in fixed wage jobs. The second column is the regression-adjusted gap. The third column introduces person fixed effects identified using workers who switch between PP and FP jobs. The top half of the table presents results for the broad PP measure, while the bottom half focuses on the narrow PP measure.
Among men, the raw differential is about 25% (= exp(0.221)-1), but falls by around half when conditioning on observable differences between PP and FP employees, and to around 4% controlling for fixed unobservable differences across employees. The pattern is similar whether one uses the broad or narrow measure of PP. The fact that the premium falls markedly when adjusting for person fixed effects is a clear indication that there is positive selection into PP jobs among men. Turning to women, a similar pattern emerges: there is a sizeable raw wage differential which falls when regression-adjusted and falls still further with the introduction of person fixed effects. So, once again, there is clear evidence of positive sorting into PP jobs among women. This is the case among all women and women in full-time jobs. The premium when accounting for fixed unobservable differences across employees is around 6% in most cases, so a little higher than for men. 5 To see how the PP premium may have changed over time, But overall there appears to be evidence of an increase in the returns to PP over the period. Whether this increase also leads to rising wage inequality will depend on where PP workers are in the wage distribution and also on how the decline in the prevalence of PP played out across the distribution. In the next section we turn to the net effect of all these factors. . Regression-adjusted estimates also control for quadratics in age and job tenure, and dummies for marital status, part-time work, temporary and fixed terms jobs, trade union coverage, public sector status, occupation (9 categories), industry (11 categories), establishment size (3 categories), region (13 categories), and wave. Standard errors in parentheses. * significant at 10%; ** significant at 5%.

Does Performance Pay Affect Wage Dispersion?
In this section we look at changes in log hourly wage dispersion between 1998-2008 in BHPS for men, women, and full-time women. First we graph dispersion in both tails of the Finally, we compare the actual wage distribution with a counterfactual wage distribution to recover the effect of PP on wages in different parts of the wage distribution. We will explain the methodology behind this below. Women's log hourly wages (lower tail)

Figure 2c -Dispersion in Full-time Women's Log Hourly Wages
. At the bottom we see no real change except in the 50-1 differential which fell until 2001-2 then increased sharply. In their analysis L&M find that the 90-50 differential increased over the period while the 50-10 differential reduced slightly (they do not consider the extreme tails). For women, we find increasing dispersion at the top and reducing dispersion at the bottom over 1998-2008, which is similar to the trends reported by L&M in the 90-50 and 50-10 differentials.
We therefore see some evidence of a growth in wage dispersion over the period, as do L&M, though for men there are some differences as to where precisely in the distribution this widening occurred. These differences could relate to sample differences, such as the incomplete ASHE coverage of low paid workers or the lack of coverage of new immigrants in BHPS, or the fact that very high and very low earners are more difficult to reach with household surveys (Bollinger et al., 2014).
How is PP related to wage dispersion over the period? Table 6 shows the mean and variance of log hourly wages in the PP and FP sectors for the pooled years. It is apparent that mean wages are higher in the PP sector, in keeping with the wage premium analysis above, but there is no evidence to suggest that the variance in wages is greater in the PP sector than the FP sector, whether one is looking at men, women or full-time women. In fact, the variance for women is slightly lower in PP than FP jobs (especially by the narrow measure). One possible reason for this is that there is greater homogeneity among PP employees than there is among FP employees in terms of traits that affect their earnings. In spite of this finding, PP has the potential to affect the earnings distribution owing to the fact that employees receive a PP wage premium and, as also seen in Table 3, tend to lie higher in the earnings distribution even conditioning on other job characteristics. Our estimates of the relationship between PP and the wage distribution are based on a reweighting estimator originally deployed by DiNardo, Fortin and Lemieux (1996) and then applied in a modified form by LMP (2009). The method constructs a counterfactual wage distribution which proxies the wage distribution that would obtain in the absence of PP in the economy. This is achieved by reweighting those sample members who are not in receipt of PP such that their observable characteristics closely resemble the overall population of workers. This in turn is achieved by running a probit estimate for the probability of being in a PP job and then using the predicted probabilities to reweight the FP employees in such a way as to give additional weight to those with high estimated probabilities of being in a PP job (because these employees are underrepresented in the FP sample). One can then recover the "effect" of PP at different parts of the wage distribution by comparing the actual distribution of wages among all workers to the counterfactual distribution observed among the reweighted set of employees not in PP jobs. Table 7 summarises the results of the counterfactual reweighting exercise for men. We consider two points in time, namely early in the period we study (1998)(1999)(2000) and 8 We do not report measures involving the 1 st and 99 th percentiles because the estimates were too noisy to be included. Noise in the percentile estimates is worsened by the fact that most estimates in the table involve comparing distributions (actual vs counterfactual and/or changes over time), and so the figures are differences not levels (unlike in Figure 2). Following LMP (2009)  10 Close to two-fifths of male employees received bonuses during this period. BHPS also asks "What was the total amount of bonus you received over the last twelve months?" For those receiving a bonus they were the equivalent of around about 3.5% of base pay in the late 1990s, rising to 4.5-5% towards the end of our period of investigation.  Table 7 presents estimates of the size of PP effects on wage dispersion among men at specific points of the wage distribution, we can also illustrate the distributional effects using graphs. Figures 3a and 3b present the effects of broad PP over the wage distribution for men. The solid line in Figure 3a represents the difference broad PP made to log hourly wage dispersion in the period 2006-08 by comparing the actual log hourly wages of all male employees -who are a mixture of PP and FP workers -with counterfactual wages based on a scenario in which nobody receives PP (the corresponding summary measures are in the top panel of Table 7 in column 3). The counterfactual gap is fairly flat in the bottom half of the wage distribution, but then it begins to rise such that the log wage differential is around .15 log points towards the top of the wage distribution.
The dotted line presents the same information but for the period 2006-2008 (corresponding to column 6 in the top panel of Table 7). The effect of PP is more pronounced in the later period, rising much more steeply in the top half of the wage distribution. Consistent with the summary measures reported above, the graphs indicate that PP has a disequalising effect on wages which increased in the later period.  Table 7), while the dotted line shows how the male wage distribution would have changed in the absence of PP (that is, the difference in the counterfactual scenarios for 1998-2000 and 2006-08). Thus the gap between the two lines gives the effect of PP on changes in the distribution (corresponding to column 8 in the upper panel of Table 7). PP makes little difference to the change in the wage 30 distribution in the lower half of the wage distribution: wage dispersion grew in the lowest quartile of the distribution, and would have done in a similar fashion in the absence of PP.
We noted above that PP widened the 50-5 gap (relative to the counterfactual without PP jobs) but it is clear from the graph that most of the change was due to a small rise in the median (the solid line is higher than the dotted one) and not a fall in the 5th percentile. This illustrates how the graph can provide a more complete picture than just comparing two points alone.
In contrast, the graph confirms the figures in Table 7  increased to a small extent, but actual dispersion, reflecting the effect of PP jobs, increased much more. Table 7, the picture looks rather different for men if we focus on the narrow measure of PP. PP measured in this way does result in a wider wage dispersion than would be the case in its absence. Although this was the case both at the beginning and the end of our period of investigation, the effect was attenuated in the second period ( Figure 4a).

As indicated in
Consequently, the effect of narrow PP jobs on changes in the wage distribution over the period was actually to reduce that dispersion, though not by very much (Figure 4b). 11 As noted earlier, close to two-fifths of male employees received bonuses during this period. BHPS also asks "What was the total amount of bonus you received over the last twelve months?" For those receiving a bonus they were the equivalent of around about 3.5% of base pay in the late 1990s, rising to 4.5-5% towards the end of our period of investigation. Now we turn to wage dispersion among women. It is apparent from column 7 in Table 8 that overall wage dispersion among women grew over the period, but only very marginally (variance increased by 0.011). This is partly because trends went in The overall effects of these counterveiling effects of the broad measure of PP on women's wages is best illustrated graphically. Figure 5a indicates that the broad measure of PP resulted in a wage distribution for women that was more U-shaped than it would have been in its absence. However, the U was flatter further up the wage distribution in the second period relative to the first. This is why PP contributed to a growth in wage dispersion in the top half of the wage distribution compared with a counterfactual world without PP (compare the solid line with the dotted line in Figure 5b). This pattern of results is similar for women when using the narrow measure of PP jobs (Figures 6a and 6b).  Turning to the effects of PP on the dispersion of full-time women's earnings and focusing first on the broad PP measure it is apparent that PP is associated with greater wage dispersion at the top of the distribution but lower dispersion at the bottom of the distribution. This is the case in both 1998-2000 and 2006-2008 (upper panel Table 9, columns 3 and 6). These effects change very little over the whole period, with the exception of the compressing effect of broad PP on the 50-5 differential, which almost doubles.
Looking at the whole distribution graphically using the broad PP measure the PP effect on full-time womens' earnings relative to a counterfactual world without PP is highest at the top and bottom of the earnings distribution, forming the U-shape referred to above for all women. The size of this effect is larger in the second period (2006)(2007)(2008) relative to the early period (1998)(1999)(2000) but it is similar across all the distribution, except at the very top ( Figure   7a). For this reason PP (broadly defined) resulted in higher earnings among full-time working women, but it had little effect on changing inequality (which decreased below the 38 10 th percentile and increased above the median), except perhaps to mitigate the increase at the very top ( Figure 7b). If we turn to the narrow PP measure and consider its effects on the log hourly earnings of women working full-time this is a shallow U-shape, in both periods, but the size of the effect is greater in 2006-08 (Figure 8a) such that PP increases wage dispersion over the period, as indicated by the rising solid line in Figure 8b relative to the dotted line for the counterfactual "no PP" world, once one moves beyond the lowest quartile of the earnings distribution.

Conclusions
There has been much speculation about the various causes of growing wage dispersion in Britain, the United States and elsewhere. The seminal paper by LMP (2009) showed PP contributed significantly to the growth in earnings dispersion in the United States through to the early 1990s. Using data from the British Household Panel Survey (BHPS) we adopt a similar estimation approach to LMP but applied to Britain during the decade of economic growth that ended abruptly with the recession of 2008. In contrast to LMP, we find that rather than increasing, the prevalence of PP declined (or at most stayed flat) over the decade to 2008. This applies to both broad and narrow measures of PP (although there is some evidence that bonus payments increased), and the trends appear comparable with more recent declines in the US identified by Gittleman. and Pierce (2013).
We confirm others' findings that wage inequality grew overall during the decade to 2008, largely due to growing earnings dispersion in the top half of the wage distribution, but there was also some reduction of inequality among women in the bottom of the distribution. The contribution of PP to these changes depends on how the incidence and returns to PP changed and where workers sit in the wage distribution. While the incidence of PP fell, there was still a substantial wage return to PP and indeed it appears to have increased over time.
Comparing the actual wage distribution with a counterfactual world without PP, we find the net effect of these changes to be that PP is associated with greater wage dispersion towards the top and, particularly for broad PP, that this disequalising effect increased over the period, possibly because of increased bonuses. This was accompanied by some counterveiling effects at the bottom, in particular broad PP is associated with more compressed wages in the lower half of the distribution for women.
Overall PP contributed to earnings dispersion for both men and women in the upper half of the distribution although most of the growth was attributable to a particular type of PP namely bonuses. PP also explains some of the reduction in inequality among women in the