the gender gap in early-career wage growth

THE GENDER GAP IN EARLY-CAREER WAGE GROWTH

Alan Manning and Joanna Swaffield

In the UK the gender pay gap on entry to the labour market is approximately zero but ten years afterlabour market entry, there is a gender wage gap of almost 25 log points. This article explores thereason for this gender gap in early-career wage growth, considering three main hypotheses – humancapital, job-shopping and �psychological� theories. Human capital factors can explain about 11 logpoints, job-shopping about 1.5 log points and the psychological theories up to 4.5 log pointsdepending on the specification. But a substantial unexplained gap remains: women who have con-tinuous full-time employment, have had no children and express no desire to have them earn about8 log points less than equivalent men after 10 years in the labour market.

Given the enormous literature on the gender pay gap – see Altonji and Blank (1999)for a relatively recent survey – one might wonder about the justification for yet anotherpaper on the subject. The answer is that this article focuses on the gender gap in wagegrowth in the early years after labour market entry and argues that an understanding ofthis is necessary as it is the primary cause of the overall gender pay gap. The reason whythis is particularly important in the UK can be explained very simply.

Figure 1 presents the data the earnings-experience profile in the UK for men andwomen together with the gender pay gap from one of the main data sets used in thisarticle, the British Household Panel Survey (BHPS). Average log wages are normalised tobe zero for men with zero years of labour market experience. One can see that theearnings of men and women are very similar on entry to the labour market but 10 yearslater a sizeable gender gap in earnings has emerged,1 a gap that reaches its maximumabout 20 years after labour market entry before falling slightly in later years.2 Of course,cross-section profiles like the ones presented in Figure 1 confound true life-cycle effectsand cohort effects. But, while cohort effects can probably explain some of the cross-section profile, there is good reason to think that they are not the most important reasonfor the widening of the gender pay gap after labour market entry. To see this Figure 2presents the gender earnings gap at different ages for different birth cohorts in the NewEarnings Survey (NES).3 The NES contains no information on education (so we cannotcompute experience) and Figure 2 simply presents age profiles. What one can see is thatthe gender pay gap is lower at all ages for younger cohorts but that the contribution ofcohort is small relative to the life-cycle effects. For example, Figure 2 suggests that thegender pay gap that can be expected at age 40 for those born around 1970 is likely to besmaller than for those born around 1960 but extrapolation of the profiles in Figure 2would suggest that a very sizeable gender wage gap will still exist. And for women bornafter 1970 there is some evidence that the progress of women has stalled as the profiles of

1 This might be a finding specific to Britain as, for example, Kunze (2003, 2005) shows that for Germanswith an apprenticeship, there is a sizeable gender gap on labour market entry that does not then change verymuch.

2 This convergence in later years is less apparent in the BHPS than in other British data sets such as theLabour Force Survey. Note also that Kunze (2003, 2005) finds a different pattern for Germany with a pay gapon entry that remains approximately constant through the life-cycle.

3 The BHPS is too short a panel to follow cohorts through time.

The Economic Journal, 118 (July), 983–1024. � The Author(s). Journal compilation � Royal Economic Society 2008. Published by

Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

[ 983 ]

the gender pay gap are very similar for those born around 1970 and those born in the late1970s – if this is true then there will be a gender pay gap for these women of 20 log points intheir early 30s.4 So, young women entering the labour market today can still expect toearn much less than young men after 10 years in the labour market and this genderdifference in early-career wage growth needs explaining.

Most papers on the gender pay gap focus on the level of pay and not the growth in pay.But as the level of pay at any level experience is simply the initial pay on labour market

0

0.2

0.4

0.6

0.8

log

hour

ly w

ages

0 10 20 30 40 50exp

Male FemaleDifference

Fig. 1. The Earnings-Experience Profile for Men and WomenNotes. Data come from the first 12 waves of the BHPS. Log hourly wages are normalised to be zerofor men with zero years of labour market experience.

0

0.2

0.4

0.6

Gen

der

wag

e ga

p (l

og p

oint

s)

20 30 40 50 60Age

Born 1938–42 Born 1948–52Born 1958–62 Born 1968–72Born 1975–79

Fig. 2. The Evolution of the Gender Pay Gap for Different Birth CohortsNotes. Data come from the New Earnings Survey, 1975–2001. Earnings data are hourly earningsexcluding over-time.

4 The flattening in the gender pay gap that one sees in the early 20s in Figure 2 is the result of the entry atthat age into the labour market of college graduates for whom, on labour market entry, the gender pay gap issmall.

984 [ J U L YT H E E C O N O M I C J O U R N A L

� The Author(s). Journal compilation � Royal Economic Society 2008

entry plus cumulated wage growth, any theory of the pay level must be, if only implicitly,also a theory of wage growth (and vice versa). So, theories to explain the level of pay shouldalso be able to explain differences in wage growth. This article focuses on three theories ofwage growth – the human capital hypothesis, job-shopping and �psychological� theoriesand asks how well they can do in explaining the gender gap in early-career wage growth.

The most common approach to explaining the gender pay gap is based on thehuman capital model developed by Becker (1993) and Mincer (1974) and applied tothe gender pay gap in many papers; see Mincer and Polachek (1974), Mincer and Ofek(1982), Light and Ureta (1995) inter alia. This approach seeks to explain the genderpay gap by gender differences in human capital accumulation of which there areseveral dimensions. First, there are gender differences in accumulated work experience(which is assumed to be correlated with levels of human capital) because women aremore likely to have intermittent labour market participation and are more likely towork part-time. Secondly, the anticipation of future intermittence may affect currentinvestments in human capital. The affected investments might be after labour marketentry – see Ben-Porath (1967) for a classic theoretical model of this – or prior to labourmarket entry where it may take the form of gender differences in the quantity or (moreplausible these days for a country like the UK) type of education. Thirdly, Becker(1985) has argued that the greater domestic commitments of women (especiallymarried women with dependent children) mean they put less energy into work andthat translates into lower hourly wages.

The estimates presented below find that the human capital hypothesis can explain anon-trivial amount of the gender gap in early-career wage growth but over 50% of thegender pay gap that emerges by 10 years after labour market entry cannot be explainedin this way. The reason for this conclusion can be explained very simply. First, differ-ences in actual labour market experience between young British men and women after10 years in the labour market are simply not large enough to explain more than 6.5 ofthe 25 log-point gender pay gap at that point in the life-cycle. Secondly, genderdifferences in job-related training do explain a part of the differential (primarilybecause young men are still much more likely than young women to go into appren-ticeships) but, again, the estimated contribution is no more than 4.5 log points. And,thirdly, gender differences in the quantity of education are very small (and, in any case,Mincer’s (1974) famous observation that the experience profiles are similar fordifferent levels of education would suggest that any difference in the quantity ofeducation should show up in a gender pay gap on labour market entry and not agender gap in wage growth) and, while gender differences in the types of educationand associated occupational choices are still large, we do not find that occupationaldifferences contribute much to the gender gap in early-career wage growth. Thebottom line is that the human capital explanations can explain perhaps half of the 25log point gender pay gap observed 10 years after labour market entry.

The article then also investigates the job-shopping hypothesis that an important partof early career wage growth is associated with moving from worse to better-paying jobs,e.g. Topel and Ward (1992), who study only US men, find that something like one-thirdof wage growth in the first 10 years after labour market entry can be put down to jobmobility). Manning (2003a, ch. 7) has documented how women are more constrainedin their opportunities to change jobs than men and are less concerned with money

2008] 985G E N D E R G A P I N E A R L Y - C A R E E R W A G E G R O W T H


when they do. So we might expect to see women making fewer �good� job changes,more �bad� job changes and a lower return to job changes when they do. We investigatethis hypothesis below, and find that it can explain only 1.5 log points of the gender paygap after 10 years in the labour market. This is not because job mobility is unimportantbut because gender differences among young workers in the incidence of and returnsto job mobility are quite small.

The article then considers a more recent but less well-known theory of the genderpay gap – that has its roots in the sizeable psychological differences between men andwomen at the age they enter the labour market, e.g., in attitudes to risk-taking, com-petition, self-esteem and selflessness. These differences are surveyed in Croson andGneezy (forthcoming) and Babcock and Laschever (2003); Gneezy et al. (2003) suggestthat they might help to explain the gender pay gap. We use data on psychologicalattributes of men and women before labour market entry to examine the explanatorypower of these ideas, finding that they can explain a few percentage points of thegender wage gap but that their impact is not enormous.

The plan of the article is as follows. Section 1 describes the data used. Section 2 thenpresents �reduced-form� estimates of wage growth equations that confirm the presenceof a gender gap in early-career wage growth as suggested by Figure 1. Section 3 theninvestigates the extent to which this can be explained by differences in human capitalconsidering the contribution of breaks in employment, of job-related training, of thetypes of education and of occupational choice. Section 4 then considers the contri-bution of job mobility, both good and bad, to wage growth. Section 5 then investigatesthe importance of �personality� for explaining the gender pay gap. Section 6 concludes.The bottom line is that a sizeable part of the gender pay gap after 10 years in the labourmarket remains unexplained: a woman who has done nothing but work full-time eversince leaving education, has had no children and expresses no desire to have them inthe future earns, on average, at least 8 percentage points less than an equivalent man.

1. The Data

The main data used in the first part of this article come from the first 12 waves of theBritish Household Panel Study (BHPS) that cover the period 1991–2002 inclusive,though the Labour Force Survey (LFS) and the New Earnings Survey (NES) are some-times used for supplementary evidence and consistency checks. The BHPS started in 1991with some 5,500 households and 10,300 individuals drawn from 250 different areas ofGreat Britain. It has had a number of booster samples: approximately 1,000 households in1997 that are part of the European Community Household Panel Survey (that includeNorthern Ireland and over-sample low-income households), approximately 3,000Scottish and Welsh households that were added in 1999. It follows all individuals insample households and those who end up in households with them as long as they remainin a household with an original sample member (in many ways, it can be thought of as theBritish PSID).

From the BHPS the sample is constructed in the following way. First, the sample isrestricted to those with no more than 10 years of potential labour market experience asour focus is on what happens to the gender pay gap in the years after labour marketentry. Secondly, only those individuals for whom there is an uninterrupted set of



interviews are included. Thirdly, anyone who has a period in employment when wecannot identify whether they are working for the same employer as at the previousinterview is excluded. Finally, because of the difficulty in measuring earnings for theself-employed, anyone who reported being self-employed at any interview date is ex-cluded: as few individuals move between employment and self-employment (about1.5% of employees become self-employed in any year) this is a relatively minorrestriction. In total about 5% of observations are excluded because of data problemsand the final sample used for analysis is approximately 10,500 observations on wagegrowth for about 3,559 individuals.

Wages are measured as an hourly rate, obtained by dividing usual weekly earnings byusual weekly hours. Those with hourly earnings below £1 or above £100 are excluded.As we do not want the measure of earnings growth on-the-job to include aggregate wagegrowth it is important to deflate wages: as Altonji and Williams (1998) point out,estimates of the return to job tenure can be sensitive to the way in which this is done.5

We experimented with the use of an external earnings index but average earningsgrowth in the BHPS deviates from this aggregate measure. As a result we estimated asimple earnings function of the log of hourly wages on a quartic in experience,regional, race and gender dummies and a set of wave dummies and then deducted thewave dummies to generate our adjusted wage series.

As we are interested in how wages evolve, we need a measure of wage growth. In moststudies of wage growth, the dependent variable is defined as the growth in wagesbetween two fixed points in time, with this measure being undefined for those who arenot in employment at either of the two points. This approach has the disadvantage thatit under-samples those individuals who have intermittent employment, a potentiallyserious problem when analysing the gender pay gap. For example, suppose we measurewages annually but some individuals have a year or more in which they are not inemployment. If one only measures wage growth from one year to the next, one willdrop two observations for such individuals because there will be two years in whichannual wage growth is not defined. To deal with this problem, we define wage growthin the following way. For any year in which an individual is in employment, define thewage growth as the difference in log wages between the next year they are inemployment and the log wage this year. Consequently there may be gaps between thetwo wage observations of any number of years. It is important to control for this �gap� inwhat follows and we describe later how we do this. There remains a potential problemin that the gap between wage observations is censored at the top end by the latest yearof observation being 2002 but this seems to be a small problem6 because long gapsbetween wage observations are relatively rare. The first two columns of Table 1 presentthe distribution of gaps between the wage observations in our data set for men andwomen. One observes, as one would expect, that women are more likely than men tohave intermittent employment. But, what is also striking is the relatively short length ofgaps between periods of employment; see also Myck and Paull (2001) for an analysis of

5 This is less important in the current context where we are primarily interested in the gender gap in wagegrowth than in the literature on the returns to job tenure as incorrect detrending probably has a very similareffect on both men and women.

6 We did experiment with restricting the sample to those in the early waves when censoring is less prob-lematic: this made no difference to the conclusions.



gender differences in accumulated experience. Only 2% of the observations on femalewage growth have a gap in employment of more than 2 years. Because the BHPS is arelatively short panel, there are many censored observations so, to get some idea of theuncensored distribution of the gap, the third and fourth columns of Table 1 report thedistribution of the gap among those in waves 1–6. There are now more young womenwith longer gaps in employment, though it remains remarkable how rare are longbreaks in employment. However these data are still censored: some women may neverreturn to paid employment but some might have very long spells that are not observedin the BHPS. To get some idea of this, the final two columns of Table 1 present, forcomparison, data on gaps between spells of employment from the Labour Force Survey(LFS). In the LFS those not currently in employment are asked when they last workedand we can use this information to work out the gap between employment spells forthose who are employed in the subsequent quarter.7 Again, one sees that women aremore likely to have gaps in employment than men but the average size of gaps and thegender differences are not large – the BHPS suggests an average gap for women that is0.3 years above that for men and the LFS suggests a somewhat smaller gap of 0.25 years.And the very long gaps that contribute quite a lot to this overall gender difference arevery rare among the young workers that are the focus of this article. For example, thework history data of the 1970 British Cohort Study suggests that, among those inemployment at age 30, 23% of men have had a period of non-employment between twoperiods of employment with an average of 0.3 gaps compared to 27% of women with anaverage of 0.35 gaps.

One other peculiarity of the BHPS that is worth mentioning is that its basic measureof job tenure is tenure in a particular job with a particular employer so that within-firm

Table 1

Distribution of Gap Between Successive Wage Observations

Gap between wage obs (yrs)

BHPS (all waves) BHPS (wave < ¼ 6) LFS

Men Women Men Women Men Women

1 95.6 93.0 94.7 91.5 96.6 92.72 3.1 4.8 3.3 5.2 2.4 3.13 0.9 0.9 1.1 1.1 0.4 1.14 0.2 0.4 0.4 0.8 0.2 0.65 0.1 0.2 0.1 0.4 0.1 0.56 0.1 0.3 0.1 0.5 0.1 0.47 0.0 0.1 0.0 0.3 0.0 0.38 0.7 0.1 0.2 0.2 0.0 0.39 0.0 0.0 0.0 0.0 0.0 0.210–14 0.1 0.715–19 0.0 0.0 0.0 0.1 0.0 0.120þ 0.0 0.0No. of obs 5,344 5,756 2,709 2,990 37,384 34,595

Notes. BHPS sample is restricted to those with 10 or fewer years of potential labour market experience.LFS sample is from 1997–2002 and gives the distribution among those currently in employment who are notlabour market entrants and who have less than 20 years of labour market experience.

7 Note that in a steady-state it should not matter whether one looks forward as in the BHPS or backwards asin the LFS.



changes of job reset the tenure clock to zero. This is not the way that the job tenurevariable is usually defined where it refers to job tenure with the same employer.However, the BHPS definition is retained here as job moves within employers (e.g.promotions) are as interesting and important as those between employers in under-standing the emergence of the gender pay gap.8 The data are summarised in theAppendix in Table A1.

2. Reduced Form Estimates

While the evidence presented in Figures 1 and 2 is strongly suggestive of a gender gapin wage growth in the years immediately after labour market entry, it is important toverify this using direct data on wage growth. Hence, this Section presents some�reduced-form� wage growth regressions to get some idea of what the data looks like andto confirm the existence of this gender gap in early-career wage growth.

Accordingly, we estimate the following �reduced-form� model for the wage growthbetween t and (t þ g) of individual i where g is the gap between wage observations:

Dwit ¼ b0 þ b1eit þ b2e2it þ eit ¼ /ðeitÞ þ eit ð1Þ

where e is years since leaving full-time education i.e. potential experience. This model isestimated separately for men and women. The quadratic specification seems adequateto capture the variation in wage growth among the young workers in our sample butone should be wary of extrapolating into other parts of the life-cycle; Murphy andWelch (1990) suggest a quartic for the level implying a cubic for the change if workersof all ages are in the sample.

In specifying this equation we need to be careful about how we deal with individualsfor whom the gap over which wage growth is measured is more than a year. To see whythis matters suppose that the observed wage growth for all individuals (whatever thegap) was 5% but that men are always in employment (so g ¼ 1) but women are only inemployment every two years (so g ¼ 2). After 2 years earnings for men would havegrown by approximately 10% whereas those for women would have only grown by 5%.The gender pay gap would be widening but estimation of (1) that did not take accountof differences in the gap would miss this.

The simplest way of getting a gap-adjusted measure of wage growth would be simplyto divide the observed wage growth by the gap between wage observations to get anestimate of annualised wage growth. However the validity of this procedure is based onthe implicit assumption that wage growth does not vary with experience, something weknow to be untrue. So, a slightly more sophisticated approach is taken here to adjustfor the gap between wage observations. If (1) is the wage growth for those for whomg ¼ 1, then for someone with g ¼ 2 to return to employment at the same level of wagesthey must have expected wage growth between the two observations of:

E Dwitð Þ ¼ /ðeitÞ þ /ðeit þ 1Þ ð2Þ

8 Use of a more conventional job tenure measure does not make much difference to the results but doesresult in a large fall in the number of useable observations because the work history information needed toconstruct it is only available for a fraction of the sample. For this reason, we only report results using the BHPSjob tenure variable.



the right-hand side of which is simply two one-period sets of wage growth. One canreadily extend this formula to any value of g in which case it will be given by:

E Dwitð Þ ¼Xg�1

j¼0/ðeit þ jÞ: ð3Þ

Using the specific functional form in (1), this can be written as:

E Dwitð Þ ¼ b0g þ b1

Xg�1

j¼0ðeit þ jÞ þ b2

Xg�1

j¼0ðeit þ jÞ2: ð4Þ

Thus one can readily estimate wage growth on a consistent basis for individuals withdifferent gaps by computing the �adjusted� levels of experience in (4) and using these asregressors in a wage growth equation. Note that there is no constant in this regression –the constant in (1) gets multiplied by g. If wage growth does not vary with experiencethis approach is equivalent to the simple-minded approach of just dividing wage growthby the interval between wage observations as the gap would be the only remainingregressor and the coefficient on it can be interpreted as annual wage growth.

Table 2 reports estimates of the reduced-form wage growth equations. The first twocolumns of Table 2 report estimates of (4) for men and women separately. We reportthe estimates of earnings growth at 0, 5 and 10 years of experience together with theirstandard errors. Taking the coefficients for men, the estimates suggest that a man canexpect 14.5% annual wage growth on entry into the labour market, falling to 6.8% after5 years and 1.3% after 10 years. For women (the second column) earnings growth onentry is lower than for men (at 12.0% per annum) and still lower after 10 years thoughthe gap in wage growth narrows.9 These estimates are consistent with the finding that

Table 2

�Reduced Form� Estimates of Wage Growth Equation

1 2 3 4 5 6

Men Women Men Women Men Women0 yrs of experience 0.145 0.120 0.161 0.131 0.160 0.160

(0.013) (0.012) (0.013) (0.011) (0.019) (0.016)5 yrs of experience 0.068 0.040 0.070 0.046 0.074 0.041

(0.005) (0.004) (0.004) (0.004) (0.008) (0.006)10 yrs of experience 0.013 0.009 0.030 0.035 0.018 0.020

(0.010) (0.006) (0.007) (0.007) (0.015) (0.011)No. of obs 5,015 5,535 4,833 5,208 5,015 5,535R2 0.076 0.051 0.077 0.058 0.020 0.015Fixed Effects No No No No Yes Yes

P ¼ 1.00 P ¼ 1.00

Sample All All Gap ¼ 1 Gap ¼ 1 All AllImplied Gender Gap at 5 yrs 0.144 0.149 0.096

(0.044) (0.041) (0.062)Implied Gender Gap at 10 yrs 0.248 0.222 0.216

(0.054) (0.044) (0.066)

Notes. These estimates are derived from the estimation of (4) where experience is modelled as a quadratic.Standard errors in parentheses. These are robust standard errors with clustering on the individual.

9 One might wonder whether these gender differences are significantly different from each other. At eachindividual level of experience the answer is often �no� but one can easily reject the joint hypothesis that thereturns to experience for men and women are equal in the first 10 years.



the gender pay gap begins to widen soon after labour market entry, as suggested inFigure 1.10 The bottom two rows of Table 2 report the gender pay gap implied by thesedifferences in estimated wage growth 5 and 10 years after labour market entry. Theseare derived by cumulating the estimates of the gender difference in wage growth fromthe estimation of (4) assuming that there are no breaks in employment and (as is agood approximation) that gender differences in wages on labour market entry are zero.The estimates in columns 1 and 2 of Table 2 imply a gender pay gap of 14.4 log pointsafter 5 years and 24.8 log points after 10 years. These estimates are not far out of linewith those derived from estimates of wage levels.

To check that these estimates are not sensitive to the treatment of those with gaps ofmore than 1 year between wage observations, columns 3 and 4 of Table 2 present the�reduced-form� wage growth equations restricting the sample to those for whom the gapbetween wage observations is only one year. One finds the same pattern of genderdifferences in wage growth. The implied gender pay gap after 10 years is lower for thissample at 22.2 log points suggesting that part of the overall gap can be explained bywomen having more intermittence but the contribution is not that large.

One might also wonder whether the inclusion of other controls on individualcharacteristics alters these conclusions. The bottom line is that they do not have a bigimpact.11 One way to see the irrelevance of personal characteristics is to estimate wagegrowth equations including individual fixed effects. This is done in the final twocolumns of Table 2: the coefficients hardly change and the fixed effects are jointlyinsignificant from zero. We should not be surprised by this finding: if there were verysignificant permanent differences in wage growth across individuals one would expectto see the variance of earnings in a cohort rising explosively over the life-cycle which itdoes not.12 Of course, this insignificance of fixed effects in a wage-growth equationdoes not mean that individual fixed effects would not be very significant in a wage levelsequation – rather, it is a validation of Mincer’s conclusion that earnings profiles areparallel to each other (once one controls for gender).

So, our basic data do contain the feature that there is a noticeable gender gap inwage growth from the start of labour market careers. Although this gender gap in wagegrowth eventually reverses there is a very large gender pay gap by the time it does andwomen never make up the ground lost in the first 10 years after labour market entry.One might be concerned that this conclusion applies only to the BHPS or is an artifactof the quadratic specification used. So, as a useful double-check, Figure 3 presents datafrom the NES that has a large enough sample size to get precise estimates of wagegrowth for men and women at each age. This shows the same pattern as the BHPS. It isthis gender gap in early-career wage growth that this article seeks to explain.

10 Note that, averaging across men and women of all experience levels, there is no significant difference inwage growth – as found by Manning and Robinson (2004). However, that paper makes the (embarrassing)mistake of inferring from this fact that gender differences in wage growth cannot explain the gender pay gap.This is wrong because a male advantage in wage growth in early life off-set by a female advantage in later lifedoes not imply a zero gender pay gap.

11 One proviso to this is that there is some evidence that, in the reduced form, earnings growth is lower inthe early years for the best-educated. But, this can be explained by the larger amount of on-the-job trainingreceived in the early years by the less-educated and this is a variable for which we do control later in the article.

12 See Mincer (1974) and, more recently, Heckman et al. (2003) for a discussion of the variance profile.



3. Gender Differences in Human Capital Accumulation

Human capital theories seek to explain the gender pay gap by gender differences inhuman capital accumulation. There is no doubt that human capital theory is of use inexplaining at least some part of the gender pay gap – some researchers (O’Neill, 2003;Polachek, 2004) claim that it can explain virtually all of the US gender pay gap whileothers (Altonji and Blank, 1999; Blau and Kahn, 2004) find that there remains asizeable unexplained component to the gender pay gap. There are several varieties ofhuman capital explanations for the gender pay gap all of which have their root in �thelesser amount of time and energy that women can commit to labour-market careers as aresult of the division of labour within the family� (O’Neill, 2003, p. 309). First, afterlabour market entry, the weaker labour market attachment of women means they tendto accumulate less labour market experience than men and, given that such experienceis valuable, part of the difference in earnings can be explained as the result of thisdifference in actual labour market experience. Studies (Mincer and Polachek, 1974;Light and Ureta, 1995) sometimes find that the returns to actual experience are verysimilar for men and women though typically other measures of labour market historyare also important in explaining the gender pay gap and a sizeable part of the genderpay gap remains unexplained even after controlling for differences in actual labourmarket experience.13

Secondly, there may be differences in human capital investments after labour marketentry in anticipation of future labour market attachment. So, we might see men doingmore job-related training than women because they expect to spend longer in thelabour market in later years. Thirdly, a similar argument might apply to human capitalinvestments in education prior to labour market entry – men might make more

0

0.05

0.1

0.15

0.2

15 20 25 30 35Age

Male wage growthFemale wage growthGender gap in wage growth

Fig. 3. Gender Differences in Wage Growth: New Earnings Survey DataNotes. The NES does not contain information on education so cannot be used to compute yearsof experience. This is annual wage growth for the sample of workers observed in two adjacentyears.

13 And one should also note that actual labour market experience is endogenous so that part of thecorrelation of wages with actual experience may not be causal.



investments than women in human capital that has a labour market pay-off. Weconsider these three different human capital hypotheses in turn.

3.1. Gender Differences in Labour Market Experience

First, consider whether gender differences in actual labour market experience (theresults of more labour market intermittence and more part-time working by women)can explain the gender gap in early-career wage growth.

In our framework labour market intermittence is captured by the gap between wageobservations as a gap of more than one year must imply a period not in employment. Asshown in Table 1, women have longer gaps between wage observations than men,implying that women do have weaker labour market attachment than men. It may bethat the differences in the gap variable can explain the disadvantage of young womenin wage growth. The specification of the wage-growth equation used so far assumes thatthere is no wage penalty associated with having g > 1 but there are good reasons tothink there might be a penalty. Assume that each year in which one is out ofemployment reduces the level of wages when one returns by b5 log points.14 Then theequation for wage growth (4) will become:

E Dwitð Þ ¼ b5 þ b0g þ b1

Xg�1

j¼0ðeit þ jÞ þ b2

Xg�1

j¼0ðeit þ jÞ2 ð5Þ

so that a simple way to test for a penalty to intermittent employment is to includeconstants in the reduced-form wage-growth equations. This is done in the first andsecond columns of Table 3. These results suggest that each year out of employmentreduces wages by 4.4% for men and 4.7% for women. The estimate for men is slightlylower but broadly consistent with the penalty for spells of unemployment found byArulampalam (2001) and Gregory and Jukes (2001).

But the implied gender pay gap after 10 years (for those without any intermittence)only falls from 24.8 to 22 log points, suggesting a contribution of intermittence of 2.8log points (this is similar to the estimate from columns 3 and 4 of Table 2).

Women accumulate less work experience than men, not just because they have moreintermittence but also because they are more likely to work part-time. For example, inour sample, 3.6% of the men are working part-time (defined as less than 30 hours aweek) and 19.2% of the women. The third and fourth columns of Table 3 include adummy variable for whether the individual is currently working part-time. The few menwho are working part-time have a very large penalty of 19.3 log points while the part-time women have a pay penalty of 6.0%. The implied gender gap after 10 years forthose who work continuously full-time now falls to 18.3% suggesting a contribution ofthe gender difference in part-time working of 3.7%. This is probably an over-estimate ashourly earnings are obtained by dividing weekly earnings by reported hours, leading toa division bias problem that overstates the initital wages of those in part-time employ-ment.

14 In some of the literature, notably Mincer and Ofek (1982), there is a debate about whether it is theincidence or the length of the break in employment that is the more important. But, as Table 2 makes clearwe do not really have sufficient variation in the employment gap in our data to explore this further.



Tab

le3

The

Impa

ctof

Hu

man

Cap

ital

Var

iabl

eson

the

Gen

der

Gap

inW

age

Gro

wth

12

34

56

78

910

1112

Men

Wo

men

Men

Wo

men

Men

Wo

men

Men

Wo

men

Men

Wo

men

Men

Wo

men

0yr

so

fex

per

ien

ce0.

151

0.13

00.

172

0.14

10.

173

0.14

10.

153

0.13

80.

118

0.08

80.

159

0.15

0(0

.013

)(0

.011

)(0

.013

)(0

.011

)(0

.013

)(0

.011

)(0

.013

)(0

.011

)(0

.021

)(0

.013

)(0

.018

)(0

.015

)5

yrs

of

exp

erie

nce

0.07

40.

046

0.07

70.

056

0.06

30.

043

0.05

90.

040

0.04

90.

026

0.05

80.

047

(0.0

05)

(0.0

04)

(0.0

05)

(0.0

04)

(0.0

08)

(0.0

06)

(0.0

08)

(0.0

07)

(0.0

10)

(0.0

04)

(0.0

12)

(0.0

08)

10yr

so

fex

per

ien

ce0.

023

0.03

20.

028

0.04

6�

0.00

30.

014

�0.

004

0.01

0�

0.00

10.

025

�0.

010

0.02

5(0

.009

)(0

.008

)(0

.009

)(0

.008

)(0

.014

)(0

.012

)(0

.014

)(0

.012

)(0

.019

)(0

.004

)(0

.018

)(0

.014

)C

ost

of

gap>

10.

044

0.04

70.

046

0.05

90.

054

0.06

60.

054

0.06

70.

119

0.09

30.

072

0.04

9(0

.024

)(0

.015

)(0

.024

)(0

.015

)(0

.024

)(0

.016

)(0

.025

)(0

.017

)(0

.029

)(0

.021

)(0

.033

)(0

.017

)P

art-

Tim

e�

0.19

3�

0.06

0�

0.19

4�

0.06

2�

0.18

9�

0.06

0�

0.15

3�

0.06

6�

0.17

3�

0.05

8(0

.041

)(0

.011

)(0

.041

)(0

.012

)(0

.042

)(0

.012

)(0

.045

)(0

.014

)(0

.049

)(0

.016

)D

ecad

eso

fjo

bte

nu

re�

0.04

0�

0.04

1�

0.03

7�

0.04

5�

0.04

4�

0.06

8�

0.04

1�

0.03

2(0

.017

)(0

.015

)(0

.017

)(0

.015

)(0

.023

)(0

.021

)(0

.019

)(0

.016

)T

rain

ing

inP

ast

Year

(yrs

)0.

087

0.02

80.

119

0.07

30.

066

0.02

1(0

.058

)(0

.038

)(0

.091

)(0

.031

)(0

.064

)(0

.051

)T

rain

ing

bet

wee

nw

age

ob

s(y

rs)

0.10

7�

0.00

50.

077

�0.

039

0.11

4�

0.03

6(0

.063

)(0

.035

)(0

.081

)(0

.046

)(0

.062

)(0

.036

)H

ou

sew

ork

(wee

kly

hrs

-10)

/10

�0.

001

�0.

003

(0.0

11)

(0.0

06)

Lab

ou

rM

arke

tM

oti

vati

on

0.00

50.

004

(0.0

06)

(0.0

05)

No

.o

fo

bs

5,01

55,

535

5,01

55,

535

5,01

55,

535

4,83

05,

306

2,50

53,

127

3,62

23,

923

R2

0.07

80.

055

0.09

00.

061

0.09

10.

062

0.08

80.

060

0.06

40.

042

0.08

50.

070

Co

ntr

ols

for

Init

ial

Occ

up

atio

nN

oN

oN

oN

oN

oN

oN

oN

oYe

sYe

sN

oN

oP¼

0.07

P¼

0.14

Imp

lied

Gen

der

Gap

at5

yrs

0.13

60.

147

0.14

60.

096

0.06

20.

073

(0.0

42)

(0.0

42)

(0.0

46)

(0.0

49)

(0.0

95)

(0.0

74)

Imp

lied

Gen

der

Gap

at10

yrs

0.22

00.

183

0.18

30.

138

0.12

40.

057

(0.0

45)

(0.0

48)

(0.0

86)

(0.0

91)

(0.1

65)

(0.1

32)

Not

es.T

hes

ees

tim

ates

are

der

ived

fro

mth

ees

tim

atio

no

f(5

)w

her

eex

per

ien

ceis

mo

del

edas

aq

uad

rati

can

d,w

her

eap

pro

pri

ate,

ad

um

my

vari

able

for

hav

ing

inte

rmit

ten

tem

plo

ymen

t,a

par

t-ti

me

du

mm

y,ye

ars

of

job

ten

ure

and

year

so

ftr

ain

ing

are

incl

ud

ed.

Stan

dar

der

rors

inp

aren

thes

es.

Th

ese

are

rob

ust

stan

dar

der

rors

wit

hcl

ust

erin

go

nth

ein

div

idu

al.

Inco

lum

ns

5to

10th

ees

tim

ates

of

the

gen

der

gap

afte

r5

and

10ye

ars

are

bas

edo

nas

sum

ing

no

chan

gein

job

sojo

bte

nu

reri

ses

wit

hex

per

ien

ce.I

nco

lum

ns

7to

10th

ees

tim

ates

of

the

gen

der

gap

afte

r5

and

10ye

ars

are

bas

edo

nas

sum

ing

no

rece

ipt

of

on

-th

e-jo

btr

ain

ing.

Fu

ll-t

ime

wo

rkis

alw

ays

assu

med

.In

colu

mn

s9

and

10w

her

ein

itia

locc

up

atio

nis

incl

ud

edas

aco

ntr

ol,

the

esti

mat

eso

fw

age

gro

wth

and

the

gen

der

gap

are

bas

edo

nth

eav

erag

eu

sin

gth

ed

istr

ibu

tio

no

fin

itia

locc

up

atio

nac

ross

bo

thm

enan

dw

om

enw

ith

less

than

10ye

ars

of

exp

erie

nce

.Usi

ng

the

mal

eo

rth

efe

mal

ed

istr

ibu

tio

nm

akes

no

dif

fere

nce

toth

ere

sult

s.T

he

p-v

alu

efo

ra

test

of

the

hyp

oth

esis

that

occ

up

atio

nal

vari

able

sar

ejo

intl

yin

sign

ifica

nt

fro

mze

rois

also

rep

ort

ed.



The estimates so far suggest that differences in actual work experience when youngcan explain at most 30% of the gender gap in early-career wage growth. The explana-tion for this is simple: the gender gap in wage growth is largest when the gender gap inaccumulated actual labour market experience is rather small. To illustrate this furtherFigure 4 presents data from the 1970 British Cohort Study, a longitudinal study of allchildren born in Britain in one week in 1970 and which collects monthly informationon labour market status. Figure 4 shows the gender differences in actual experience(measured as months in work) over the first 10 years after labour market entry.Ten years after labour market entry the men have accumulated about 12 months morework experience than the women. The gap in full-time experience is larger at20 months. However, for the purpose of understanding the gender pay gap for those inwork (as is the focus of this article) these differences are misleading as the pay dif-ferences relate to those in work and the gender differences in actual experience aremuch smaller for those in work. These differences are also plotted on Figure 4: after10 years male workers have about 3 months more work experience than female workersand about 12 months more full-time experience. These differences are simply not largeenough to explain the magnitude of the gender pay gap after 10 years: for example, ifone thinks of the male profile as being the �true� one and women are one year behindmen after 10 years this can only account for wage differentials of the order of 2 logpoints. The contribution is a bit larger if one recognises that the earnings profile isconcave so it is not just the mean but also the variance that is of importance and womenhave slightly higher variance than men. After 10 years the gender difference in thevariance of full-time work experience among workers is about 1 year which adds at mostan extra one log point to the predicted gender difference. These back-of-the-envelopecalculations are smaller than the regression estimates of Table 3 though, as we havealready noted, the contribution of part-time working is probably over-estimated.

The discussion so far has focused on the accumulation of general human capital butgender differences in the accumulation of specific human capital might also beimportant. Job-specific human capital is often assumed to be measured by job tenure;

0

5

10

15

20

Mon

ths

1 2 3 4 5 6 7 8 9 10Potential experience (years)

Gender gap in actual experienceGender gap in full-time actual experienceGender gap in actual experience for workersGender gap in full-time actual experience for workers

Fig. 4. Gender Differences in Accumulated Work ExperienceNotes. These figures come from the 1970 British Cohort Study. We thank Fernando Galindo-Rueda for allowing us to use the work-history data he had constructed.



see Mincer and Jovanovic (1981) for this theoretical argument and Jones and Make-peace (1996) for a British study in which this variable is important for explaining agender gap. The fifth and sixth columns of Table 3 include job tenure in the currentjob as an extra control in the wage-growth equations. We find that higher job tenure isassociated with significantly lower wage growth. But, this does little to narrow thegender gap in wage growth on labour market entry and has no impact on the impliedgender pay gap after 10 years.

3.2. Gender Differences in On-the-Job Training

Of course, it may be differences in expected future labour market attachment thataffect current human capital accumulation either before or after labour marketentry. Let us start with the training after labour market entry. In the BHPS,respondents are asked about the total amount of job-related training they have doneover the previous year. We simply include a measure of the total years of trainingreceived as a control variable in our wage growth regression. We might expect thatcurrent job-related training depresses current wages but raises future wages andwage growth. However, the strongest empirical effect of on-the-job training seems tocome not from the measure of the training received between the two wage obser-vations but from the measure of training received over the year prior to the initialwage observation. So, measures of training in both years are included in the wagegrowth regressions: the results are presented in columns 7 and 8 of Table 3. Thereturn to a year of job-related training for men is higher (at approximately 8–10%)than for women (approximately 3% for training received in the past year and zerofor training between wage observations).

The difference in the return to training is compounded by the fact that young menreceive more on-the-job training than young women – for example among workers withless than 5 years of potential experience, the men receive on average 0.039 years oftraining per year and the women 0.032. This seems to be the result of the fact that men(especially those with little education) are much more likely than women to be inapprenticeships (which have a very large training component). Among men who leaveschool at 16 but are aged 20 or under, 25% are in apprenticeships compared to 10% ofwomen (figures from the 2004 LFS). Figure 5 shows that the gender gap in training isentirely driven by the behaviour of the low-educated: among college graduates youngwomen receive more training than young men. When one cumulates this one can seethat the predicted gender wage gap after 10 years falls to 0.138 from 0.183 when onecontrols for training, so something like 4.5 log points of the gap after 10 years can beexplained by gender differences in the receipt and the return to on-the-job training.

3.3. Gender Differences in Occupation and Education

Now let us consider gender differences in human capital accumulation beforelabour market entry. In terms of the quantity of education there are few genderdifferences (see the summary statistics in Table A1). But the type of education maybe different, e.g., Brown and Corcoran (1997), Black et al. (2003) for the US andMachin and Puhani (2003) for the UK and Germany document that differences in



subject of degree can explain a portion of the gender pay gap among collegegraduates. Unfortunately the BHPS does not contain detailed information on thesubjects studied by college graduates or others. But it does contain information onthe occupation of the first job that might be expected to capture much of genderdifferences in career choices.

Columns 9 and 10 of Table 3 include dummy variables for the occupation in the firstjob so the assumption is that initial choice of occupation has a permanent effect onwage growth.15 Sample sizes preclude using more than 1-digit occupation controls butthere are sizeable gender differences even at this level in the type of job first done.Details are in Table A1 in the data Appendix but the big differences among youngworkers (with less than 10 years of potential experience) are that 38% of women firstenter clerical and secretarial occupations compared to 16% of men, 20% of womenenter personal and protective service occupations compared to 8% of men, 2% ofwomen enter craft occupations compared to 23% of men and 7% of women enterelementary occupations compared to 15% of men.

In producing the summary statistics on wage growth at different levels of experienceand the gender gap we assume a distribution of first jobs equal to the average acrossmen and women with less than 10 years of experience – evaluating at male or femalecharacteristics makes no difference. For both men and women the coefficients oninitial occupation are jointly not significantly different from zero though the rejectionis marginal for men (the rather curious mix of managerial, professional, clerical andelementary occupations seem to have higher wage growth than the others). But, whenit comes to explaining gender differences in wage growth, initial occupation has littleexplanatory power with the predicted gender pay gap after 10 years falling from 0.138to 0.124 after controlling for initial occupation.16

–0.04

–0.02

0

0.02

0.04

0.06

Gen

der

gap

in jo

b-re

late

dtr

aini

ng (

Yea

rs)

0 5 10 15 20Years of potential experienceAllSchool leaving age ≤17School leaving age ≥18 and ≤20School leaving age ≥21

Fig. 5. Gender Differences in On-the-Job TrainingNotes. This data comes from the Labour Force Survey for the years 2000–3.

15 We did experiment with interacting first job with years of experience so the impact of initial job falls overtime but the coefficients on these interactions were always insignificantly different from zero.

16 One might be concerned that the sample sizes in columns 9 and 10 are considerably smaller than incolumns 7 and 8 because initial job is missing for many observations. But, estimating the model of columns 7and 8 using the restricted sample of columns 9 and 10 leads to a predicted wage gap after 10 years of 0.13 sothe contribution of occupation is probably smaller than it appears from a simple inspection of Table 3.



However this evidence has only used 1-digit controls for initial occupation and onemight be concerned that more marked differences would emerge if a finer measure ofoccupation was used. That is not possible because of small sample sizes in the BHPS butwe can get some idea of whether this is likely using the much larger sample sizesavailable in the NES where we can use 3-digit occupation of which there are approxi-mately 370. Once one goes to this level of disaggregation there is a problem that someoccupations are so overwhelmingly dominated by one gender that one has very fewobservations on the other. To deal with this problem of matching men and women atthe 3-digit occupation level the sample is restricted to occupations that have at least 2%of the minority gender. This excludes from the sample 17.5% of men in 74 occupations(broadly, production managers in manufacturing and construction, senior police andfire officers, drivers of trains, boats, and planes, almost all of construction, almostanything that involves metal-working, and refuse collection). 9.6% of women areexcluded in 9 occupations (broadly, midwives, secretaries, dental nurses, childcareworkers and beauticians).

We take two approaches to controlling for occupation. In the first we simply in-clude a dummy variable for each occupation in the gender-specific wage growthequations effectively assuming that occupation only affects the level and not the life-cycle profile of wage growth. The results are summarised in Panel (a) of Figure 6.The line marked �whole sample� reproduces the raw gender gap in wage growthalready shown in Figure 4. The line �matched sample� reports the raw gap in wagegrowth for those in occupations that are not excluded from the analysis – one noticesthat the gender gap in wage growth is now much smaller for those aged 16–22. Themost likely explanation for this finding is that many of the men who are excludedfrom the matched sample are in construction and/or metal-working trades and theyoung men in particular are doing apprenticeships that are associated with fast wagegrowth. This is consistent with our earlier finding of gender differences in trainingreceipt among this group.

The remaining two lines in Figure 6(a) show the impact of adjusting the matchedsample for differences in the gender distribution of occupation. We show two esti-mates – one evaluated at the male occupation distribution and one at the femaleoccupation – though they are very similar and similar to the line that does notcontrol for occupational segregation at all. Figure 6(a) suggests a sizeable gender paygap in early-career wage growth remains even after controlling for differences inoccupation.

However, the estimates that lie behind Figure 6(a) assume an impact of occupationon wage growth that can be modelled as a shifting intercept. This is quite restrictive e.g.it might be the case that women, anticipating future labour market withdrawal, go intooccupations that offer high initial wage growth and slower wage growth later on. Toallow for this possibility we also present – in Figure 6(b) – a reweighting estimator thatuses the methodology of diNardo et al. (1996) to reweight the female (male) data toreproduce the male (female) occupational distribution. We then report weighted wagegrowth by age. The resulting estimates of the gender gap in wage growth are muchnoisier than the regression-based estimates (unsurprisingly because less structure is puton the data) but one can still see that virtually all of the observations of the gender gapin wage growth up to the age of 35 are positive. So, differences in occupation do not



seem to be able to explain much of the gender-gap in early-career wage growth onceone has controlled for gender differences in training.17

This conclusion that occupational differences cannot explain much of the gendergap in early-career wage growth can be reconciled with the findings of Machin andPuhani (2003) that field of study can explain a sizeable part of the gender wage gapamong British college graduates. First, there is a small gender wage gap on entry to thelabour market among graduates that can be explained almost entirely by differences in

(a) Regression Model

(b) Reweighting Model

0

0.01

0.02

0.03

0.04

Gen

der

gap

in w

age

grow

th

15 20 25 30 35Age

Matched sampleMale occupation distribution – reweighted modelFemale occupation distribution – reweighted model

0

0.01

0.02

0.03

0.04

0.05

Gen

der

gap

in w

age

grow

th

15 20 25 30 35Age

Whole sampleMatched sampleMale occupation distribution – regression modelFemale occupation distribution – regression model

Fig. 6. The Impact of Occupational Job Segregation on Gender Differences in Wage Growth.Notes. These data come from the New Earnings Survey for 1996–2001. Occupation is measured atthe 3-digit level. The matched sample excludes those men and women in occupations with verylow representation of the opposite sex, as discussed in the text. The regression model simplyincludes dummy variables for each occupation in the wage growth equations. The reweightingmodel reweights observations to obtain the desired occupational distribution and computes thegender gap in wage growth for this distribution.

17 Though it may be that even the 3-digit occupation is not fine enough – see Wood et al. (1993) forevidence that among lawyers, women are over-represented in the lower paid fields.



field of study but field of study seems to have little impact on wage growth. Secondly,gender differences in curriculum are larger for graduates than non-graduates, as isfound for the US by Brown and Corcoran (1997).

3.4. Gender Differences in Housework and Labour Market Motivation

Becker (1985) suggested that part of the gender pay gap can be explained by the factthat women put less effort into paid work even when working the same number ofhours because of greater domestic commitments. To investigate this we include as acontrol in our wage growth equation the number of hours of housework done per week(as expected women in our sample do 10 hours more on average than the men) and acomposite measure of labour market motivation; as also used by Vella (1994), Swaffield(2000) and Chevalier (2004). The labour market motivation variable is constructed as acomposite measure (standardised to have mean zero and variance one) of how muchthe respondent agrees or disagrees with 6 statements on the role of women in thelabour market.18 To allay fears about how the responses might themselves be influ-enced by labour market outcomes, we construct this measure using only the earliestrecorded response to each of these questions. As might be expected women have, onaverage, a more positive view about the role of women in the labour market than menand working women a more positive view than non-working women.

Columns 11 and 12 of Table 3 include both these variables in the wage growthregression. The sample size is lower because the question on housework is not askedevery wave. Both of these variables have coefficients that are of the expected sign butare insignificantly different from zero for both men and women. The net effect of thesevariables on the predicted gender gap in wages is small after 5 years but appears largeafter 10 – however, this is more the result of the smaller sample than the variablesthemselves and the standard error on the estimate is very large. Hence, we find, withthe variables available to us, little evidence for the Becker hypothesis being important.

3.5. Conclusions on the Human Capital Hypothesis

The evidence presented here suggests that human capital factors can explain perhapshalf of the gender gap in early-career wage growth. Of the overall 25 log point gapimplied after 10 years, 2.8 log points can be explained by differences in labour marketintermittence, 3.7% by differences in part-time working, 4.5% by differences in training(primarily because young men who leave school early are still more likely to doapprenticeships) and perhaps 1.5% by differences in occupational choice. Labourmarket intermittence and part-time working is more important between 10 and20 years after labour market entry for the reason that this is when most women havebreaks in employment associated with childcare. The bottom line is that a woman whoworks continuously full-time can expect to be 12% behind an equivalent man after10 years in the labour market.

18 The six statements are �A pre-school child suffers if mother works�, �A family suffers if mother works full-time�, �A woman and family are happier if she works, �A husband and wife should both contribute�, �A full timejob makes a woman independent�, �A husband should earn, wife stay at home�



4. Job-Shopping Models

Part of the rapid wage growth in the years immediately after labour market entry may bethe result of job mobility – finding a good match for one’s skills or simply finding the goodjobs (if, as evidence suggests, there is a sizeable amount of equilibrium wage dispersion).Theoretical analysis of the implications of job search models for the returns to experienceand job tenure can be found in Burdett (1978) and Manning (1997, 2000). Topel andWard (1992) found that for US men something like a third of wage growth in the first10 years after labour market entry can be ascribed to job mobility.

In this process of job-shopping it seems plausible that there are gender differences.For example Manning (2003a, ch. 7) documents, using the BHPS, that women aremore constrained in their job choices (primarily by family commitments), they travelless far to work, again, suggesting a narrower choice of jobs – see Manning (2003b), andtheir job changes are less motivated by money and are less likely to be �for a better job�.All of these factors might be expected to be potential explanations of the gender gap inearly-career wage growth.

First, consider gender differences in job mobility. For current purposes we define jobmobility rates as the percentage of those currently in employment for whom the nextobservation is non-employment or a change in job. Note that a change in job need notbe associated with a change in employer so a promotion will count as a job change; seeBooth et al. (2003) for an analysis of promotions in the BHPS. Figure 7 presents theoverall job mobility rates by experience for men and women. One sees the well-knownpattern that job mobility rates start off very high, then decline before flattening out andrising slightly for the older workers. Women have noticeably higher mobility rates thanmen between 10 and 25 years of experience i.e. in the prime child-bearing years.

It is likely that there are gender differences in the types of job mobility as well asthe level. Although there is an argument (based on a frictionless competitivelabour market) that there is no meaningful distinction between quits and lay-offs(McLaughlin, 1991) most labour economists think there is a useful distinction betweenquits that are voluntary on the part of workers and lay-offs that are not. Consistent withthis, quits tend to be associated with wage gains while lay-offs are associated with wagelosses (Ruhm, 1991; Jacobson et al., 1993; Kletzer, 1998). The job history files of theBHPS record reasons for why jobs ended and Table 4 tabulates them. 47% of youngmen leave jobs because they are promoted or get a better job compared to 43% ofwomen.19 15% of male jobs end in redundancy or dismissal compared with only 9% ofwomen. 7% of female jobs end to have a baby or look after other family with,unsurprisingly, this proportion being close to zero for men. To summarise, male jobsare slightly more likely to end with a move to a better job, but men are more at risk ofjobs ending because of redundancy. On the other hand, women’s jobs are more likelyto end to have a child.

Not all job changes have reasons associated with them (approximately 12% do not).But, we can get an idea of the relative importance of different mobility rates for

19 There is also a question about why the new job is better, the reasons for which can be crudely dividedinto pecuniary and non-pecuniary reasons. We did experiment with disaggregating the �left for better job�category but there are many missing values and ambiguous answers (e.g. many saying they left for a better jobbecause the new job is better) so this was not pursued.



different sorts of job change by multiplying the overall mobility rate by the fraction ofjob changes due to particular reasons (this implicitly assumes that the job changes withno given reason are drawn at random). To prevent information overload we dividereasons for jobs ending into two categories: those that end with a move to a better joband those that do not. Figure 8 presents the experience profiles for changes to a betterjob for men and women: there are no large gender differences. Figure 9 presents thejob mobility rates not to a better job: this is higher for women in the early years of thelabour market, primarily because of the role played by family responsibilities. Thesefigures suggest that over the first 10 years in the labour market men can expect 3 goodjob moves (compared to 2.9 for women) and 2 bad job moves (compared to 2.4 forwomen).

0.2

0.3

0.4

0.5

0.6

0.7

Mob

ility

rat

e

0 5 10 15 20Years of potential experience

Men Women

Fig. 7. Job Mobility Rates: TotalNotes. These data come from the BHPS. The mobility rate is measured as the fraction of workerswho change jobs or leave a job for non-employment in the coming year.

Table 4

Reasons Given for Jobs Ending

Reason given Men (%) Women (%)

Promoted 19.1 16.7Left For Better Job 28.0 26.9Made Redundant 12.7 7.8Dismissed or Sacked 3.1 1.4Temporary Job Ended 10.5 9.8Took retirement 1.9 2.9Health Reasons 3.6 4.2To Have Baby 0.0 3.7Children/Home care 0.5 3.9Moved Area 3.6 5.0Other 17.1 17.1No. of obs 3,611 3,749

Notes. Sample is restricted to BHPS from waves 5 on as extra categories are then added. Sample is those withbetween 0 and 10 years of labour market experience inclusive. Data refer to reasons given for job ending.



Now let us consider the extent to which wage changes are related to different sorts ofmobility. There is an existing literature on the impact of different sorts of moves onwage changes that starts with a series of papers in the early 1980s (Bartel, 1980, 1982;Borjas, 1981; Bartel and Borjas, 1981). More recent papers are Topel and Ward (1992)and, with a specific focus on gender differences, Loprest (1992), Crossley et al. (1994),Keith and McWilliams (1997, 1999) and Cobb-Clark (2001).

We examine the impact of job mobility by simply including dummy variables fordifferent sorts of move.20 There are a large number of reasons for moves given in

0.1

0.15

0.2

0.25

0.3

0.35

Mob

ility

rat

e


Men Women

Fig. 8. Job Mobility Rates: for Better JobNotes. As for Figure 7 but with mobility defined as being for a better job.

0.1

0.15

0.2

0.25

0.3

Mob

ility

rat

e


Men Women

Fig. 9. Job Mobility Rates: not for Better JobNotes. As for Figure 7 but with mobility not being defined as being for a better job.

20 There are reasons to doubt whether this specification is adequate. For example, theory predicts somevariation in the returns to job mobility both in observables (like experience, job tenure and the current levelof wages) and unobservables (because individuals are less likely to leave jobs with high wage growth). How-ever, experimentation with the specification did not lead to any substantive change in the results and we onlyreport the simplest specification here.



Table 4 and we aggregate these into four categories: moves for a better job (called�good moves�), moves because jobs ended in redundancy, dismissal or end of contract(called �bad� moves), moves that end for family reasons (called �kid� moves) and movesfor other reasons (called �other� moves). This is relatively crude: for example Gibbonsand Katz (1991) provide evidence that those dismissed suffer a greater wage penaltythan those made redundant through plant closure. But, we do not have enough datahere to do a very fine disaggregation.

The first two columns of Table 5 provide estimates of wage growth equationsaugmented by job mobility variables. For men a �good� move is associated with awage gain of 8.6% while for women it only produces a gain of 6.3%. This result thatthe returns to job mobility are lower for women is consistent with the evidencepresented in Manning (2003a, ch. 7). �Bad� moves are associated with significantwage losses; 3.9% for men but not for women. For kid moves, there is a very largepay penalty for the small number of men who make such a move but the mostinteresting finding is that there is no wage penalty associated with such moves forwomen (the coefficient is actually positive though insignificantly different fromzero). This is different from the findings in Waldfogel (1998) or Joshi and Paci(1998) but they use data from a much earlier period and find that the wage penaltyis smaller for women with maternity leave entitlements who return to the sameemployer. As these entitlements have been extended it is possible that now there is arelatively small penalty associated with having children if the woman returns to paidwork quickly (there is a penalty to intermittence of employment as captured by thegap variable). Finally the �other� move category has insignificant effects on wagegrowth.

The inclusion of the job mobility variables essentially leaves the implied gender payafter 10 years of continuous employment slightly higher than it was without the jobmobility variables (compare columns 1 and 2 of Table 5 with columns 7 and 8 of Table 3).But, the specification used here assumes that there is no heterogeneity in the returns tojob mobility. In particular, the theory of job-shopping would suggest that very youngworkers will not suffer much a of a wage loss when making a bad move because they havelittle to lose but older workers will face larger wage losses because they are more likely tohave worked themselves into a good job.21 To test this hypothesis we also included aninteraction of experience with a bad move – the results are reported in columns 3 and 4 ofTable 5. For both men and women this interaction variable is negative though thecoefficient is much larger for men (probably because older men have more to lose thanolder women). As predicted by the theory, the coefficient on the bad move variable itself isinsignificantly different from zero (this is the estimate of the cost of a bad move for aworker just entering the labour market). Accordingly the fifth and sixth columns ofTable 5 include only the interaction of bad moves with experience. The implied genderpay gap after 10 years for those who remain in the same job now falls to 0.122 suggesting acontribution of job mobility of 1.5 percentage points (comparing with columns 7 and 8 ofTable 3). This modest contribution is not because job mobility has no effect on wagegrowth, just that the differences between young men and women are not that large.

21 The theory might also be used to predict other interactions but, while we experimented with otherinteractions, we never found any to be significant.



Hence, job-shopping models do not appear to be able to explain more than a small part ofthe gender gap in early-career wage growth.

5. Personality and the Gender Pay Gap

There is an enormous literature in psychology on the size and nature of gender dif-ferences in personality, a literature that spills out into the popular imagination withbooks of the �men are from mars, women from venus� variety. This literature – seeCroson and Gneezy (forthcoming) for an excellent survey for economists – shows thatby labour market entry, there are quite significant gender differences in personalities22

Table 5

The Impact of Job Mobility on Wage Growth

Men Women Men Women Men Women

0 yrs of experience 0.135 0.120 0.130 0.112 0.131 0.107(0.014) (0.012) (0.014) (0.012) (0.014) (0.012)

5 yrs of experience 0.051 0.029 0.047 0.028 0.047 0.022(0.008) (0.007) (0.008) (0.006) (0.008) (0.006)

10 yrs of experience �0.005 0.013 �0.001 0.013 �0.001 0.005(0.015) (0.012) (0.014) (0.012) (0.014) (0.012)

Cost of gap > 1 0.031 0.041 �0.002 0.033 �0.001 0.036(0.037) (0.017) (0.037) (0.019) (0.036) (0.019)

Part-Time �0.193 �0.060 �0.194 �0.061 �0.194 �0.060(0.043) (0.012) (0.043) (0.012) (0.043) (0.012)

Decades of tenure �0.017 �0.031 �0.022 �0.032 �0.023 �0.033(0.018) (0.015) (0.018) (0.015) (0.018) (0.015)

Training in Past Year (yrs) 0.082 0.011 0.081 0.011 0.080 0.011(0.058) (0.038) (0.059) (0.038) (0.059) (0.038)

Training between wage obs (yrs) 0.085 �0.001 0.088 �0.002 0.087 �0.003(0.066) (0.035) (0.066) (0.035) (0.065) (0.035)

Good Move 0.086 0.063 0.087 0.063 0.087 0.063(0.011) (0.010) (0.011) (0.010) (0.011) (0.010)

Bad Move �0.039 0.013 0.008 0.029(0.022) (0.023) (0.031) (0.027)

Bad Move � Decades of experience �0.890 �0.277 �0.813 �0.143(0.452) (0.240) (0.323) (0.202)

Kid Move �0.143 0.041 �0.189 0.033 �0.186 0.035(0.176) (0.052) (0.182) (0.053) (0.181) (0.053)

Other Move �0.032 0.006 �0.037 0.004 �0.037 0.005(0.021) (0.019) (0.021) (0.019) (0.021) (0.019)

No. of Obs 4,762 5,142 4,762 5,142 4,762 5,142R2 0.105 0.072 0.106 0.073 0.106 0.073Gender Gap at 5 yrs 0.108 0.087 0.081

(0.052) (0.052) (0.052)Gender Gap at 10 yrs 0.155 0.127 0.122

(0.090) (0.091) (0.091)

Notes. These estimates are derived from the estimation of (5) where experience is modelled as a quadraticand, where appropriate, years of job tenure and a dummy variable for having intermittent employment isincluded.Standard errors in parentheses. These are robust standard errors with clustering on the individual.The estimates of the gender gap after 5 and 10 years are based on assuming no change in job so job tenurerises with experience and the job mobility variables are zero.

22 There is a heated debate about whether this is nature or nurture, a question we will avoid here.



(something we confirm with our data below). Recently, economists have becomeinterested in the possibility that these differences can help to understand differences ineconomic outcomes. There have been applications in a number of areas e.g. Barberand Odean (2001) suggest that the propensity of men to be more over-confident leadsthem to be more active traders in the stock market, but the main areas of interest havebeen the labour market (Eckel and Grossman, 1998; Gneezy and Rustichini, 2004;Gneezy et al., 2003; Babcock and Laschever, 2003; Niederle and Westerlund, 2005).Most of these studies have been in experimental settings – while these can demonstratethe existence of significant differences in behaviour it is hard to translate the resultsinto a portion of the gender pay gap that can be explained in this way. That is what weattempt here.

One might hope that psychological differences could help to explain the gender paygap because a number of recent studies have demonstrated intriguing correlationsbetween earnings and �personality� variables23 (Goldsmith et al., 1997, Feinstein, 2000;Bowles et al., 2001; Nyhus and Pons, 2004). A natural extension of this work is toinvestigate the hypothesis that psychological differences between men and women onentry into the labour market can explain part of the gender pay gap. The only paper tohave tried to do this directly is Mueller and Plug (2004) who use data from 50-year oldhigh school graduates from Wisconsin. They investigate whether the �five-factor� modelof personality structure can help to explain the gender pay gap in their data, findingonly modest effects. But the personality data are contemporaneous with the earningsdata (raising issues of reverse causation) and the personality traits used might not bethe most important ones in the labour market, making this unlikely to be the last wordon the subject. In our study we try to remedy both of these weaknesses.

There are a very large number of potential personality traits that one might use toexplain the gender pay gap. To put some structure on the issue, we use as a frameworkthe survey by Croson and Gneezy (forthcoming) and the book by Babcock and Las-chever (2003). Croson and Gneezy (forthcoming) identify a number of main areas inwhich men and women seem to have different preferences:

• Risk-taking – men seem inclined to take more risks than women possibly be-cause they are less risk-averse and possibly because their assessment of risks isaffected by the fact that men seem to have more self-esteem

• Self-esteem – men seem to have a higher opinion of their own abilities than dowomen

• Attitudes to competition – possibly because of the two previous factors butperhaps also because of some other factors, women seem more reluctant thanmen to enter competitive situations and their performance is affected differ-ently when they do.

• Other-regarding behaviour – women seem to be less selfish than men in anumber of dimensions and care more about what others think about them.

The specific argument put forward by Babcock and Laschever (2003) is that women areless effective than men in negotiation. This gender divide is related to more funda-

23 This research can be seen as one part of the renaissance in behavioural economics that seeks insightsfrom psychology for application in economics.



mental psychological differences between men and women, some of which are in theCroson and Gneezy list above (e.g. self-esteem – see Babcock and Laschever, p. 72) butsome that are slightly different e.g.

• Locus of Control (Babcock and Laschever, p. 23) – men are more likely to feelthat their fate can be influenced by their actions.

• Ambition – men seem to be more career-oriented than women (Babcock andLaschever, p. 30).

In addition we add another set of factors that are sometimes argued to be important(Fisher, 1999).

• Social skills – women have better �soft� skills then men.

The addition of the last set of factors is potentially important because our exercisewould have a bias to it if we only sought to model the aspects of personality that putwomen at a disadvantage in the labour market.

The BHPS has collected data on youths living within sample households since wave4 and asks them, among other things, the �psychological� questions described inTable 6. The answers to these questions show that a higher incidence of lower self-image and greater self-doubt in young women as compared to young men. Forexample, only 13.9% of male youths agreed with the statement �I don’t have much tobe proud of� compared to 17.5% of females. The lower self-image for females isshown even more starkly in the responses to the statements �I certainly feel useless attimes� – 29.1% of male youths agreed compared to 44.8% of females. In addition,22.6% of male youths compared to 35.3% of females agreed with the statement �Attimes I feel I am no good at all�. These differences might be expected to producewage differentials, e.g., Babcock and Laschever (2003) argue that low self-esteemleads women not to negotiate wage rises for themselves. Unfortunately the BHPSpanel is not long enough to provide a large enough sample of previously surveyedyouths who are then observed earning in the labour market so that we cannot, as yet,use the BHPS for this purpose.

Table 6

Gender Differences in Personality Among Teenagers: BHPS Data

Percentages agreeing with the statement Men Women Sample

I feel I have a number of good qualities 93.0 91.0 1,883 (1,882)I feel that I do not have much to be proud of 13.9 17.5 1,883 (1,882)I certainly feel useless at times 29.1 44.8 3,948 (3,753)I am able to do things as well as most other people 91.0 90.5 1,883 (1,882)I am a likeable person 93.0 93.6 3,948 (3,753)I can usually solve my own problems 90.2 86.0 1,883 (1,882)All in all, I am inclined to feel I am a failure 8.9 12.1 3,948 (3,753)At times I feel I am no good at all 22.6 35.3 3,948 (3,753)I feel left out of things when I am with friends 15.6 25.0 2,752 (2,679)

Notes. In the final column female sample sizes are shown in the parentheses.Rows 1, 2, 4 and 6 includes BHPS waves 9–11. Rows 3, 5, and 7 includes BHPS waves 4–11. Row 9 includesBHPS waves 7–11. �Agree� includes strongly agree and agree.



For this reason we turn to another data set, the British Cohort Study (BCS) that alsocontains the type of data required to investigate the impact of personality on thegender wage gap. The BCS is a cohort of all individuals born in a single week in 1970;see Joshi and Paci (1998) for more description of this data. Information has beencollected on the cohort at a variety of ages: here we use the latest information collectedin 2000, for 30-year-old individuals that have been (roughly) 10 years in the labourmarket. In addition we use the detailed information collected before they entered thelabour market – during their childhood and adolescence – on their personality. Thesepersonality controls are primarily constructed using the cohort members� question-naire responses at age 16.

This survey contains hundreds of questions that one might conceivably use as apotential explanation of the gender pay gap. This embarrassment of riches presentssome problems in deciding which variables to include and exclude. There are so manythat one cannot include them all so selection is necessary. If a large number of variablesare used there is an obvious danger that the exercise becomes a �fishing expedition� inwhich one seeks those variables which, if only by chance, happen to be capable of�explaining� the gender pay gap in our data. At the same time, restricting attention to asmall set of variables which we have a priori reason to think might be important runs therisk of missing the variables that do matter, a particular danger when, as is the casehere, our a priori views are not very strong and also of only estimating attenuatedcoefficients if there is measurement error in those variables.

Our strategy tries to steer a middle course between these two dangers. We start byusing a fairly long list of questions that seem most closely related to the factors iden-tified in the existing literature as the main psychological differences between men andwomen – we then investigate which of these variables are significant in explaining theearnings of men and women separately and finally evaluate the contribution of thesevariables to the gender pay gap. By only considering the ability of these variables toexplain the gender pay gap at the end we hope to minimise the fishing trip aspect ofthe exercise though variables will be retained based on their significance in explainingearnings. However, in the Appendix we report results from an alternative strategy inwhich we passively use a very large number of questions combined into modules (as wasdone in the original questionnaire).

Although the primary aim of this Section is to investigate the impact of the per-sonality variables on the gender wage gap, it is important to establish that our earlierfindings about the gender pay gap in the BHPS are also found in the BCS. As the BCSonly has information on earnings at a few points in time we focus on the gender paygap in 2000 when the cohort members were aged 30 and had, on average, been in thelabour market for approximately 10 years. But, we are estimating the level of thegender pay gap (which can be thought of as cumulated wage growth) rather than directestimates of the gender gap in wage growth.

Table 7 presents some estimates of the gender pay gap using the BCS 2000, We startwith the estimation of equations similar to those we have used for the BHPS. Thestructure of the Table is as follows: in the second column we present the raw gender paygap for the sample under consideration. The third column then reports the coefficienton a dummy variable for being male when other controls are introduced. The third(fourth) column then reports the results when separate earnings equations for men



and women are estimated and the gender pay gap evaluated for a full-time worker with10 years of actual experience whose other characteristics match those of the averagewoman (man).24

The raw gender pay gap is approximately 18 log points for the wage levels as shown inthe second column of the first row, somewhat lower than estimates from the BHPS.25

When controls for whether there are children in the household, marital status, race,tenure and qualifications are introduced (but not fertility intentions or actual labourmarket experience) the estimated gender wage gap falls to approximately 16 log points(row 1, column 3). The second row adds in the expected future fertility controls –a dummy for whether the individual is planning (further) children in the future butthe impact of this control is negligible.

In the third row a quadratic for total labour market experience is included. This hasa relatively small impact on the wage gap estimates, reducing slightly the wage gap withcontrols and the wage gap evaluated with male characteristics. By comparison, theinclusion of full-time and part-time actual labour market experience in the row below(row 4) has a much greater impact on the wage gap estimates. The estimated gender

Table 7

The Gender Gap at Age 30 (BCS wage data at age 30)

Raw LogWage gap

Wage Gapwith controls

Wage Gapevaluated at

10 yrs of actualexp with: Sample size

Femalechars

Malechars

Women(Men)

1. Basic equation withoutfertility or experience controls

0.181 0.163 0.163 0.163 3,281(0.016) (0.015) (0.015) (0.015) (3,681)

2. Row 1 plus expected fertility controls 0.181 0.163 0.163 0.163 3,281(0.016) (0.015) (0.015) (0.015) (3,681)

3. Row 2 plus actual experience 0.181 0.159 0.169 0.152 3,281(0.016) (0.015) (0.002) (0.002) (3,681)

4. Row 2 plus actual full-timeand part-time experience

0.181 0.119 0.127 0.115 3,281(0.016) (0.016) (0.002) (0.003) (3,681)

5. Row 4 plus 1-digit occupation 0.181 0.125 0.121 0.142 3,281(0.016) (0.017) (0.002) (0.002) (3,681)

6. Basic equation with sample restrictedto �always in FT employment� with �no kids�

0.081 0.110 0.121 0.115 1,310(0.025) (0.023) (0.004) (0.004) (1,589)

7. Row 5 plus 1-digit occupation 0.081 0.095 0.086 0.107 1,310(0.025) (0.025) (0.005) (0.005) (1,589)

Notes: The basic equation for each row includes controls for whether there are any children in the household,quadratic in actual full-time and part-time labour market experience, age left full-time education, quadraticfor current tenure, qualifications, marital status, ethnic, establishment size, whether a supervisor and futureplans for (further) children.Average wage gaps are evaluated at the means of the full-time only workers.

24 Our reason for showing this particular gender gap is that this is closest to the cumulated gap that hasbeen the focus of attention in the BHPS data.

25 It is worth noting that throughout this Section we use wage levels rather than wage growth because theyare the only data available to us. But, as the gender gap is approximately zero on labour market entry thisshould not make a huge difference.



wage gap with controls (column 3) falls by roughly 4 log points. The unexplained paygap also falls by roughly 4 log points. The fourth row includes 1-digit occupationcontrols – the unexplained gender gap rises when evaluated for the average man byalmost 3 log points. The figures based on the female evaluated wage gaps are similar tothe wage gap estimates (with controls) in the previous (third) column. All of theseestimates suggest a pay gap in the region of 11–16% between British men and womenaged 30 in 2000 who have spent all their careers in full-time employment, an estimatesimilar to that derived from the BHPS earlier in the article.

To illustrate this even more starkly, the fifth row restricts the sample to those indi-viduals who at age 30 have not had any children and who have had every month in full-time employment since leaving full-time education. These are individuals whosecommitment to the labour force cannot be questioned. The raw gender pay gap for thisgroup is 8.1 log points but the adjusted gender pay gaps are larger at 11%, primarilybecause women in this group are better educated than the men. It should be noted thatthere is no correction for sample selection bias here - such a correction is likely to makethe adjusted gender pay gap even bigger as the �career� women are likely to be muchmore positively selected than the �career� men. The final row shows that the inclusion of1-digit occupation makes little difference to the conclusion for this group. Theseestimates are also in line with the unexplained gender pay gap reported in Dolton et al.(2002). Having established that the BHPS and BCS data lead to similar conclusions, weturn to the impact of personality variables on the gender pay gap – first, consider thepro-active strategy.

Table 8 provides the list of variables that we use, divided into six categories – riskaversion, competitiveness, self-esteem, other-regarding behaviour, career orientationand an �other� category. It is very likely that some of these variables could arguably beplaced in other categories from the one we have chosen but as this categorisation ismainly for ease of exposition this does not matter greatly. Table 8 also shows thegender differences in the responses to these questions. For the most part these dif-ferences are in line with what a reading of the literature on gender differences wouldsuggest.

As measures of attitudes towards risk, we use variables relating to whether therespondent normally wears a seat belt, whether they have taken a first aid course, theirinterest in safety at home, in traffic and in the water, and whether they smoke and drinkalcohol.26 These variables are similar to others that have been used to proxy riskaversion in other studies and there does seem some link between them and more directmeasures of risk aversion (Dohmen et al., 2005). Women are significantly more likelythan men to wear a seatbelt, be interested in safety and drink less, all suggesting morerisk aversion. However, young women smoke significantly more than young men.

The next panel of Table 8 looks at variables to proxy for competitiveness. Here weuse variables related to thinking that fighting (a form of conflict) can be exciting, thenumber of team and individual sports in which the respondent participates, the fre-quency of participation in sport and how often they play cards and electronic games.Again, we see significant gender differences of the type we would expect if women are

26 We also experimented with measures of illegal drug use but these variables were never significant andlow response rates meant their inclusion substantially reduced the sample size.



Table 8

Gender Differences in Psychological Variables: BCS Data at Age 16

Exact Question Responses

Gender Differencein Response

[standard error]

Risk AversionHow often fasten seat belt in front seat? 0 ¼ every time

1 ¼ not every time�0.039 [0.010]**

Done first-aid course in past two years? 0 ¼ yes1 ¼ no

�0.023 [0.012]

How interested are you in safety at home? �1 ¼ very interested0 ¼ somewhat interested1 ¼ not interested

�0.185 [0.019]**

How interested are you in safety in traffic? �1 ¼ very interested0 ¼ somewhat interested1 ¼ not interested

�0.058 [0.019]**

How interested are you in water safety? �1 ¼ very interested0 ¼ somewhat interested1 ¼ not interested

�0.03 [0.020]

When did you last smoke? �1 ¼ never0 ¼ 3 months ago or more1 ¼ more recently

0.101 [0.022]**

How often drink alcohol in past year? 0 ¼ never/special occasions1 ¼ no more than once a week2 ¼ more than once a week

�0.14 [0.019]**

CompetitivenessThink fighting is wrong 0 ¼ agree

1 ¼ don’t know2 ¼ disagree

�0.302 [0.020]**

No. of team sports participate in Count �0.101 [0.011]**No. of individual sports participate in Count 0.019 [0.008]*I am keen on sports 0 ¼ does not apply

1 ¼ applies somewhat2 ¼ applies very much

�0.384 [0.020]**

How often do you play card/board games? 0 ¼ rarely/never1 ¼ sometimes

�0.151 [0.013]**

How often do you play electronic games? 0 ¼ rarely/never1 ¼ sometimes

�0.228 [0.012]**

How often do you play sports? 0 ¼ rarely/never1 ¼ sometimes

�0.216 [0.013]**

Self-EsteemI am clever 0 ¼ does not apply

1 ¼ applies somewhat2 ¼ applies very much

�0.182 [0.015]**

In comparison with others my job prospects are �2 ¼ much less. . .2 ¼ much more

�0.111 [0.026]**

In comparison with others I get worried �2 ¼ much more. . .2 ¼ much less

�0.337 [0.027]**

Do you think of yourself as a worthless person 0 ¼ not at all1 ¼ sometimes

0.086 [0.013]**

Have you been feeling unhappy or depressed 0 ¼ not at all1 ¼ sometimes

0.107 [0.014]**

Have you been losing confidence in yourself 0 ¼ not at all1 ¼ sometimes

0.115 [0.013]**

Do you feel miserable or depressed 0 ¼ sometimes1 ¼ rarely or never

0.199 [0.013]**



Table 8

Continued


GenderDifference in

Response[standard error]

Other-regarding BehaviourDo you care what your mother thinks about you? 0 ¼ cares a lot

1 ¼ cares a little/not at all�0.139 [0.011]**

Do you go out of your way to help others in trouble 0 ¼ yes1 ¼ no

�0.111 [0.013]**

In your job how much does it matterto you to be able to help others

�1 ¼ very much, 0 ¼ somewhat1 ¼ not at all

�0.241 [0.016]**

Are you interested in old people’s needs? �1 ¼ very interested. . ...2 ¼ not at all interested

�0.459 [0.023]**

Are you interested in handicappedpeople’s needs?

�1 ¼ very interested ..2 ¼ not at all interested

�0.52 [0.024]**

Are you always nice to people 0 ¼ no1 ¼ yes

0.031 [0.012]**

I am helpful �1 ¼ does not apply0 ¼ applies somewhat1 ¼ applies very much

0.103 [0.014]**

Do you do volunteer/community work 0 ¼ rarely/never1 ¼ sometimes/often

0.053 [0.011]**

It is up to Africans to grow food for themselves 0 ¼ disagree1 ¼ agree

�0.172 [0.014]**

I am always willing to help the teacher �1 ¼ not true at all0 ¼ partly true1 ¼ very true

0.104 [0.016]**

Career OrientationIn your future job, how important is itto have high earnings?

�1 ¼ doesn’t matter0 ¼ matters somewhat1 ¼ matters very much

�0.157 [0.016]**

In your future job, how importantis it to get ahead?

�1 ¼ doesn’t matter0 ¼ matters somewhat1 ¼ matters very much

�0.069 [0.018]**

Does getting married matter to you? �1 ¼ matters very much0 ¼ matters somewhat1 ¼ doesn’t matter

0.019 [0.021]

Does having children of your own matter to you? �1 ¼ matters very much0 ¼ matters somewhat1 ¼ doesn’t matter

�0.085 [0.020]**

Women can do the same jobs as men 0 ¼ agree fully1 ¼ agree partly2 ¼ disagree

�0.412 [0.017]**

Women’s lib is a good thing �1 ¼ agree fully0 ¼ agree partly1 ¼ disagree

�0.272 [0.017]**

Mother’s attitudes to women’s place in society �0.015 [0.014]

Locus of ControlDo you feel that wishing can makegood things happen?

0 ¼ yes1 ¼ don’t know2 ¼ no

�0.069 [0.029]**

Is a high mark just a matter of luck for you? 0 ¼ yes1 ¼ don’t know2 ¼ no

�0.146 [0.027]**

Are tests just a lot of guess work for you? 0 ¼ yes1 ¼ don’t know2 ¼ no

�0.113 [0.025]**



less enthusiastic about competition – they play fewer sports and other games, enjoythem less and are less likely to think fighting can be exciting.

The third panel of Table 8 looks at variables to proxy for self-esteem. Here we usevariables about whether the respondent has confidence in themselves, whether theythink they are clever,27 whether they think they are worthless, whether they have beenfeeling unhappy or miserable and how they rate their job prospects. In all thesedimensions women seem significantly less self-confident than men and more prone tofeeling unhappy or depressed.

The fourth panel of Table 8 considers other-regarding behaviour – whether therespondent cares about what their mother thinks about them, various measuresof whether they want to help others, whether they do volunteer work and whether

Table 8

Continued


Gender Difference inResponse [standard

error]

Are you a person who believes thatplanning ahead makes thingsturn out better?


0.173 [0.031]**

When bad things happen to youis it usually someone else’s fault?


0.171 [0.027]**

When nice things happen toyou is it only good luck?


0.052 [0.026]**

Do you think studying fortests is a waste of time?


0.079 [0.021]**

Other variablesI am violent 0 ¼ does not apply

1 ¼ applies somewhat/very much�0.12 [0.011]**

I am reliable 0 ¼ applies very much1 ¼ applies somewhat/does not apply

�0.054 [0.013]**

I am popular �1 ¼ does not apply0 ¼ applies somewhat1 ¼ applies very much

�0.029 [0.015]*

I am friendly �1 ¼ does not apply0 ¼ applies somewhat1 ¼ applies very much

0.099 [0.014]**

I am lazy �1 ¼ does not apply0 ¼ applies somewhat1 ¼ applies very much

�0.057 [0.017]**

I am punctual �1 ¼ does not apply0 ¼ applies somewhat1 ¼ applies very much

0.021 [0.018]

I am obedient �1 ¼ does not apply0 ¼ applies somewhat1 ¼ applies very much

0.043 [0.015]**

Notes. The sample size varies with the question but is generally in the region of 2500.

27 This variable is, unsurprisingly, correlated with school attainment but even controlling for this, womenare less likely to think they are clever. The regressions reported below contain detailed controls for educationso we hope we are picking up the confidence not the ability part of this variable.



Africans should grow food for themselves. In all of these measures, women are sig-nificantly more likely than men to report caring about others.

The fifth panel of Table 8 reports the measures we use to capture career orientation.We use questions about the importance of full-time work, high wages, �getting ahead�and also marriage and having children. We also include in this Section variables thatcapture attitudes to women’s role in society generally – whether �women’s lib� is a goodthing and also a composite measure of the mother’s attitudes to work/home from thequestionnaire at age 5. Among these variables we find women caring significantly lessabout work, more about children but also having more positive views about genderequality than men.

The sixth panel of Table 8 reports the variables used to capture the �locus of control�idea that women are less likely than men to think their destiny is in their hands. We usequestions about whether things that happen are a matter of luck, the fault of others orwhether outcomes can be influenced by one’s own actions. Women are more likely tothink that high marks are a matter of luck, that tests are guesswork, not to believe inplanning ahead. But, at the same time women are less likely to think studying for tests isa waste of time, suggesting they are not always fatalistic about outcomes. Indeed the factthat girls are, on average, more diligent in their school work than boys, might bethought something of a problem for the locus of control idea.

The final panel of Table 8 considers a collection of other traits that might be thoughtto affect employability – whether the individual is violent, reliable, popular, friendly,lazy, punctual and obedient. On most of these criteria women are significantly morelikely than men to have the �desirable� characteristic.28

The results in Table 8 show that it is very easy to find significant gender differencesin personality traits along the lines suggested by the existing literature. But how muchof the gender pay gap can be explained in this way? – this depends on whether thesetraits have a significant effect on wages. This is the question addressed in Table 9 wherewe report results from estimating separate male and female wage equations anddecomposing the difference using the standard Oaxaca approach. We could include allthe variables (and we do report a specification that does this) but requiring responsesto all variables drastically reduces sample sizes. So, Table 9 first reports the resultsincluding the psychological variables one category at a time. Because the sample sizevaries according to the block considered we report the raw log wage gap in the sampleconsidered, the portion of the wage gap that can be explained at male and femalecoefficients (using the standard Oaxaca approach), and the portion that can beexplained by gender differences in the personality traits. Finally we also report a p-valueof an F-test of the hypothesis that the psychological variables are jointly zero in bothfemale and male wage equations.

The first row of Table 9 reports the results when the �risk attitudes� are included. Forboth men and women these variables are jointly insignificant (and none individuallysignificant) and the portion of the gender pay gap that can be explained using them issmall and insignificantly different from zero. We have been unable to find any variablesrepresenting attitudes to risk that are of much use in explaining the gender pay gap.

28 Some authors (e.g. Fisher, 1999) have gone so far as to argue that women’s advantage over men in skillslike these that become more important over time mean the gender gap will fall over time and even reverse.



The second row of Table 9 reports the results when the �competitiveness� variablesare included. These are jointly significantly different from zero for women but not formen with the variable for liking sport being significant for women and estimated toraise wages by 0.089. This finding that different variables are significant in male andfemale earnings functions will be a recurrent theme, indicating that (as is oftenclaimed) different traits are regarded differently when displayed by a man and awomen. Overall the gender differences in competitiveness are estimated to be capableof explaining 0.017 to 0.037 of the gender pay gap depending on whether male orfemale coefficients are used.

The third row of Table 9 includes the self-esteem variables. These are jointly sig-nificant for men but marginally so for women. The individually significant variables forwomen is whether they rated themselves as clever and for men it is those who reportnot getting worried about things who have significantly higher earnings. Overall thegender differences in competitiveness are estimated to be capable of explaining 0.018to 0.034 of the gender pay gap depending on whether male or female coefficients areused.

So far gender differences in personality traits have all been capable of explainingthe gender pay gap. But the fourth row shows that the �other-regarding� traits may workin the opposite direction. These are jointly significant for men but not women, with

Table 9

The Impact of Psychological Variables On Wages at Age 30

Raw LogWage gap(standard

error)

ExplainedWage Gap

Contribution ofpsychological variables

p-value forpsych

variablesSample

size

Femaleco-efficients

Maleco-efficients

Femaleco-efficients

Maleco-efficients

Female(Male)

Female(Male)

1. Risk 0.168 0.071 0.067 0.014 0.005 0.48 1184(0.030) (0.008) (0.009) (0.74) (952)

2. Competitiveness 0.166 0.075 0.098 0.017 0.037 0.022 1203(0.030) (0.019) (0.019) (0.253) (963)

3. Self-Esteem 0.179 0.085 0.087 0.018 0.034 0.108 1421(0.030) (0.010) (0.010) (0.040) (1091)

4. Other-regarding 0.195 0.095 0.031 0.009 �0.039 0.215 1382(0.028) (0.014) (0.015) (0.001) (1041)

5. CareerOrientation

0.178 0.109 0.055 0.043 0.002 0.030 1320(0.029) (0.017) (0.015) (0.459) (1013)

6. Locus of Control 0.176 0.072 0.069 0.008 0.006 0.34 1507(0.027) (0.008) (0.008) (0.897) (1183)

7. Other 0.182 0.061 0.046 �0.007 �0.011 0.103 1511(0.027) (0.007) (0.006) (0.387) (1209)

8. All Variables 0.203 0.123 0.091 0.056 0.005 0.087 724(0.041) (0.045) (0.040) (0.016) (536)

9. Selected Variables 0.177 0.114 0.076 0.043 0.009 0.001 1459(0.027) (0.012) (0.013) (0.001) (1123)

Notes. The basic equation for each row includes controls for whether there are any children in the household,quadratic in actual full-time and part-time labour market experience, age left full-time education, quadraticfor current tenure, qualifications including detailed controls for the number of GCE and CSEs obtained,marital status, ethnic, establishment size, whether a supervisor and future plans for (further) children.Standard errors in parentheses.



the individually significant variables for men being that earnings are 0.11 higher formen who care a lot about their mothers and 0.12 lower for men who report they arealways nice to others.29 The net effect is that, when evaluated at male coefficients,these traits are predicted to be to the advantage of women as they care more abouttheir mothers.

The fifth row of Table 9 investigates the role of the �career orientation� variableswhich are jointly significant for women but not men. For women those who reporthaving kids in later life as mattering very much to them have significantly lowerearnings. This can explain 0.043 of the gender pay gap when evaluated at femalecoefficients but nothing when evaluated at male coefficients.

The sixth row of Table 9 reports the results when the �locus of control� variables areincluded. For both men and women these variables are jointly insignificant (and noneindividually significant) and the portion of the gender pay gap that can be explainedusing them is small and insignificantly different from zero. We have been unable tofind any variables representing these �locus of control� variables that are of much use inexplaining the gender pay gap.

The seventh row of Table 9 investigates the other personality traits, these not beingsignificant for either men or women and the only individually significant variablebeing the finding that, for women, those who are punctual seem (oddly) to havelower earnings. The net effect of these variables on the gender pay gap is small butnegative.

In row 7 we include all the variables together except the �risk� and �locus of control�variables.30 These are jointly significant for men but for women only at the 10% level.Taken together these variables can explain 0.056 of the gender pay gap (about 25% ofthe total) at female coefficients but 0.005 at male coefficients. This result can be bestunderstood in the next row where we restrict the included psychological variables tothose found to be most significant in male or female wage equations. Because thenumber of variables is now manageable we report the estimates in Table 10. Looking atall the rows in Table 9 one can see that, generally, the psychological variables do helpto explain the gender pay gap. There is a large range of estimates about their import-ance and these estimates are sensitive to the variables included. One can find specifi-cations in which the psychological variables can explain about 5% of the gender paygap but that is the upper end of what we have found.

The approach we have taken so far can be criticised either as a fishing expedition inwhich we have sought out variables that are significant and reported specifications thatmaximise the importance of the psychological variables or as a very partial exercise inwhich many potentially relevant variables have not been considered. To allay some ofthese fears, the Appendix reports the results from a more passive approach in which wesimply take batteries of questions as originally put into the age 16 questionnaire. Forthis purpose we use the segment on attitudes, divided into 21 modules, spanning a vastrange of subjects. To this we add another module from another part of the question-

29 These variables are only weakly positively correlated, something that might be found surprising. But oneshould remember that the Kray twins, notorious East End gangsters in 1960s London, were famously lovingtowards their mother but definitely not nice in their behaviour towards (some) others.

30 This is just to keep the sample size up and because the risk and locus of control variables are alwaysinsignificant. Results are not very different if these variables are included.



naire designed to measure self-esteem as this has been argued to be important byBabcock and Laschever (2003) and others. In total there are 458 questions covering avery wide range of issues from those that have been deemed important in the literatureof gender differences (e.g., self-esteem, locus of control, attitudes to work), to issuesthat may be thought very tangential, e.g., the teenager’s taste in soft drinks. The gen-eral description of the modules is listed in Table A2.31 The analysis of these questions isin the Appendix but the conclusion is similar, namely that gender differences in�personality� variables can help to explain at most a few percentage points of the genderpay gap. A sizeable part of the gender pay gap remains unexplained.

6. Conclusions

This article has argued that an understanding of the gender pay gap in Britain needs tofocus on the explanation of the gender gap in early-career wage growth that causeswomen who entered the labour market with the same average level of pay as men to beapproximately 25 log points behind 10 years later. Human capital theory can explain atmost half of this (primarily because of gender differences in the receipt of on-the-jobtraining, and because of modest differences in accumulated labour marketexperience). Job-shopping theories seem to be able to explain at most 1.5 log pointsand the �psychological� differences appear capable of explaining up to 4.5 log points,though that is probably an upper bound. There remains a sizeable unexplained

Table 10

Impact of Selected Variables on Male and Female Wages at Age 30

Female Male

I am keen on sports 0.076 0.026[0.022] [0.025]

I am clever 0.074 0.037[0.034] [0.038]

In comparison with others I get worried �0.005 0.035[0.018] [0.020]

Are you always nice to people �0.065 �0.088[0.037] [0.043]

Do you care what your mother thinks about you 0.009 �0.148[0.042] [0.040]

In your future job, how important is it to get ahead? 0.027 0.045[0.025] [0.029]

Does having children of your own matter to you? 0.035 0.007[0.022] [0.026]

Observations 1,459 1,123R-squared 0.24 0.24

Notes. The basic equation for each row includes controls for whether there are any children in the household,quadratic in actual full-time and part-time labour market experience, age left full-time education, quadraticfor current tenure, detailed qualifications, marital status, ethnicity, establishment size, whether a supervisorand future plans for (further) children.Standard errors in parentheses.Details of the questions can be found in Table 8.

31 For reasons of space we cannot list all the individual questions – they can be found in the on-linecodebook at http://www.cls.ioe.ac.uk/studies.asp?section¼00010002000200070001 Document C.



component. This gender gap in wage growth means that, although men and womenhave similar earnings when entering the labour market, the women will be somethinglike 8% behind the men ten years later even if they have been in continuous full-timeemployment, have had no children, do not want any and have the same personality as aman.

Appendix

An Alternative Approach to Investigating the Importance of �Personality� Variables

In the alternative approach we use the battery of questions contained in the age 16 questionnairesegment on attitudes plus one other battery of questions, designed to measure self-esteem. Intotal there are 458 questions divided into 22 modules. These questions cover a very wide range ofdifferences between men and women along many different dimensions though as the ques-tionnaire was not designed with a main purpose of investigating gender differences thesequestions are not tied so closely to the factors discussed above that have been thought to be themost important gender differences in personality – though there are sizeable gender differencesin the response to most questions as we report below.

The gender differences in the responses to these questions are summarised in Table A2.Table A2 details the different modules and the number of questions for which the coefficient issignificantly different from zero at the 5% level. In 383 of the 458 questions in total, thereare gender differences in responses that are significantly different from zero.

There are too many individual questions to include all in an earnings function so weproceed by creating composite indices for each of the 22 modules. For each module we createa weighted average of the responses to create a single measure for each of them.32 For this tobe a sensible procedure requires that the responses to questions within a module reallymeasure the same thing. As a check on this we report Cronbach’s alpha for each module inthe fourth column of Table A2 – a value above 0.7 is conventionally taken to mean that thereis a valid underlying psychometric variable and most of the modules do show this level ofcongruence.

Table A3 reports the coefficients on these modules in earnings functions for male and femaleworkers using the BCS data. The first point to note is that most of the modules have estimatedeffects on wages that are insignificantly different from zero – out of 22 modules 2 are significantlydifferent from zero for women and 1 for men. A joint hypothesis test finds one cannot reject thenull hypothesis that all of the modules have no effect on wages. In spite of this the right-handpanel of Table A3 shows that the total contributions of the modules to the gender pay gap is non-trivial, being a significant 6.9 log points when evaluated at female coefficients and an insignificant4.3 when evaluated at the male. One might conclude that a best estimate of the contribution ofthe personality variables to earnings is about 5 log points, though with a large standard error. It isthe �my interests� and �my point of view� modules that seem most important in explaining thegender pay gap. The �my point of view� module contains questions that seem most closely relatedto other-regarding attitudes – women seem �softer� and more tolerant in their attitudes than menand the labour market seems to punish such attitudes. The �my interests� module is harder tointerpret – the questions are mostly about interest in health issues with women being moreinterested than men and this being punished in the labour market for some reason. The resultsalso suggest that some factors others have suggested are important, e.g., self-esteem and locus ofcontrol, are not. It is also important to note that there remains a sizeable unexplained gender pay

32 This aggregation might also be thought to reduce the problem caused by measurement error in theresponses to individual questions.



gap – the predicted wage gap between men and women who have worked continuously full-timefor 10 years is 9.3 log points when evaluated at average female characteristics and 7.4 log pointswhen evaluated at the average male characteristics.

Columns 3 and 4 of Table A3 show the results when we include only the modules that appearsignificant. The results are qualitatively very similar.

Table A1Summary Statistics for BHPS Data

Young Workers (� 10 yrs of potentialexperience)

Men Women

Log Wage Growth 0.084 0.065(0.306) (0.278)

Potential experience (decades) 0.48 0.47(0.28) (0.28)

Job Tenure (decades) 0.15 0.13(0.22) (0.19)

Gap between wage obs 1.02 1.02(0.18) (0.18)

Part-Time 0.036 0.196Training in Past Year (yrs) 0.035 0.030

(0.106) (0.095)Training between wage obs (yrs) 0.030 0.030

(0.090) (0.099)Good Move 0.254 0.231Bad Move 0.072 0.056Kid Move 0.0005 0.010Other Move 0.062 0.072Manager 0.035 0.036Professional occupations 0.060 0.049Associate professional 0.104 0.104Administrative and secretarial 0.156 0.378Skilled trades 0.239 0.022Personal services 0.083 0.200Sales/customer services 0.081 0.109Machine operatives 0.094 0.037Elementary 0.147 0.068Graduates 0.283 0.290�A� Level 0.396 0.430GSCEs 0.286 0.267No qualifications 0.036 0.012Black 0.011 0.010Asian 0.025 0.010Married 0.442 0.543Child in Household 0.253 0.259No. of kids 0–2 0.098 0.070No. of kids 3–4 0.051 0.066No. of kids 12–15 0.107 0.099No. of kids 16–18 0.028 0.031Housework (weekly hrs-10)/10 �0.442 0.594Labour Market Motivation 0.005 0.119

Notes. Sample is those with observations on wage growth aged 16–65 with less than 45 years of potentialexperience. Sample numbers for wage growth are 5,015 for men and 5,535 for women – some variables havesmaller sample sizes because of non-response.Means are reported and standard errors in parentheses. Standard errors not reported for binary variables.



Tab

leA

2G

ende

rD

iffe

ren

ces

inP

sych

olog

ical

Var

iabl

es–

Alt

ern

ativ

eA

ppro

ach

Tit

lein

Qu

esti

on

nai

reT

ype

of

qu

esti

on

Nu

mb

ero

fq

ues

tio

ns

Gen

der

dif

fere

nce

inco

mp

osi

tesc

ore

Inm

od

ule

Wit

hsi

gnif

gen

der

dif

fC

ron

bac

h’s

alp

ha

Co

effi

cien

tSt

and

ard

erro

r

Wh

atab

ou

tw

ork

?D

eter

min

ants

of

succ

ess

inla

bo

ur

mar

ket

98

0.56

70.

027

[0.0

13]

Rig

ht

and

mig

ht

Att

itu

des

tola

w-b

reak

ing

and

hel

pin

go

ther

s9

70.

533

�0.

152

[0.0

12]

Hav

ea

dri

nk

Alc

oh

ol

1717

0.72

50.

197

[0.0

12]

Wh

at’s

ina

job

Imp

ort

ance

of

dif

fere

nt

job

char

acte

rist

ics

1612

0.60

70.

096

[0.0

10]

Lo

oki

ng

ahea

dW

hat

wil

lb

eim

po

rtan

tas

adu

lt15

130.

661

0.02

[0.0

11]

Up

insm

oke

Smo

kin

g17

140.

746

�0.

026

[0.0

12]

Co

mp

ared

wit

ho

ther

sP

hys

ical

/em

oti

on

alat

trib

ute

s28

210.

822

�0.

122

[0.0

11]

Kn

ow

ing

mys

elf

Per

son

alch

arac

teri

stic

s27

230.

724

�0.

038

[0.0

10]

Ho

wI

feel

Wo

rrie

san

dan

xiet

ies

1210

0.83

30.

147

[0.0

16]

At

leis

ure

Lei

sure

acti

viti

es47

400.

781

�0.

036

[0.0

08]

My

inte

rest

sIn

tere

sts

inp

hys

ical

/m

enta

lh

ealt

hto

pic

s49

460.

923

�0.

289

[0.0

12]

Fat

ean

dfo

rtu

ne

Lo

cus

of

con

tro

l–

bel

ief

inab

ilit

yto

infl

uen

ceo

utc

om

es15

60.

717

�0.

025

[0.0

12]

Wh

atI

read

Typ

eo

fre

adin

g25

240.

817

0.22

4[0

.011

]M

ean

dth

eb

ox

Typ

eso

fT

Vp

rogr

amm

esw

atch

ed22

220.

721

�0.

101

[0.0

11]

Fee

lin

gh

ealt

hy

Ph

ysic

al/

men

tal

hea

lth

2218

0.83

9�

0.18

2[0

.013

]M

yp

oin

to

fvi

ewO

pin

ion

so

nw

om

en’s

/ga

y/m

ino

rity

righ

ts,

dru

gs,

dea

thp

enal

tyet

c21

170.

597

0.21

5[0

.009

]

Wo

talo

tigo

tM

ater

ial

po

sses

sio

ns,

actu

alan

dd

esir

ed30

270.

649

0.16

9[0

.008

]M

ean

dm

yfa

mil

yA

ctiv

itie

sd

on

ew

ith

fam

ily

1513

0.75

7�

0.06

[0.0

13]

Soft

dri

nk

spec

ial

Typ

eo

fSo

ftd

rin

ksco

nsu

med

106

0.63

2�

0.08

5[0

.013

]H

om

eru

leP

aren

tsex

pec

tati

on

so

fb

ehav

iou

rw

ith

inth

eh

om

e23

190.

743

�0.

039

[0.0

11]

Wh

atI

eat

Ho

wo

ften

eat

dif

fere

nt

foo

dit

ems

1814

0.71

8�

0.15

3[0

.011

]Se

lf-e

stee

m11

60.

702

�0.

042

[0.0

14]

To

tal

458

383



Table A3The Impact of Psychological Variables on Wages at Age 30 – Alternative Approach

Regression Coefficients Contribution to Gender Pay Gap

1 2 3 4 1 2 3 4

Female Male Female MaleFemalecoeffs

Malecoeffs

Femalecoeffs

Malecoeffs

What about work? 0.007 0.129 0.012 0.13 0.000 �0.003 0.000 �0.003[0.044] [0.047] [0.043] [0.044] [0.006] [0.001] [0.001] [0.001]

Right and might 0.037 0.001 0.006 0.000[0.040] [0.041] [0.006] [0.006]

Have a drink 0.019 �0.031 �0.004 0.007[0.049] [0.048] [0.010] [0.010]

What’s in a job �0.035 �0.062 0.003 0.006[0.052] [0.052] [0.005] [0.005]

Looking ahead 0.017 0.004 0.000 0.000[0.047] [0.051] [0.001] [0.001]

Up in smoke 0.041 0.001 0.002 0.000[0.045] [0.044] [0.002] [0.002]

Compared with others 0.016 0.018 0.001 0.002[0.052] [0.051] [0.005] [0.005]

Knowing myself �0.074 �0.091 �0.003 �0.003[0.063] [0.068] [0.002] [0.003]

How I feel 0.039 �0.001 �0.005 0.000[0.032] [0.040] [0.004] [0.005]

At leisure 0.106 �0.038 0.003 �0.001[0.073] [0.069] [0.002] [0.002]

My interests 0.09 0.048 0.077 0.019 0.027 0.014 0.023 0.006[0.047] [0.045] [0.043] [0.040] [0.014] [0.013] [0.013] [0.012]

Fate and fortune 0.122 �0.023 0.135 0.001 0.001 0.000 0.000 0.000[0.048] [0.053] [0.045] [0.049] [0.000] [0.000] [0.000] [0.000]

What I read �0.069 0.038 �0.042 0.051 0.014 �0.008 0.009 �0.011[0.050] [0.050] [0.049] [0.049] [0.010] [0.010] [0.010] [0.010]

Me and the box 0.087 0.081 0.068 0.065 0.008 0.008 0.007 0.006[0.055] [0.054] [0.053] [0.051] [0.005] [0.005] [0.005] [0.005]

Feeling healthy 0.026 0.074 0.018 0.106 0.004 0.011 0.003 0.016[0.048] [0.050] [0.039] [0.041] [0.007] [0.007] [0.006] [0.006]

My point of view �0.099 �0.037 �0.113 �0.058 0.022 0.008 0.025 0.013[0.064] [0.055] [0.061] [0.052] [0.014] [0.005] [0.014] [0.012]

Wotalotigot �0.009 �0.071 0.001 0.010[0.063] [0.065] [0.009] [0.010]

Me and my family �0.044 0.021 �0.002 0.001[0.044] [0.040] [0.002] [0.002]

Soft drink special �0.006 �0.066 �0.001 �0.006[0.043] [0.043] [0.004] [0.004]

Home rule �0.107 �0.001 �0.006 �0.000[0.048] [0.048] [0.003] [0.003]

What I eat �0.024 �0.024 �0.004 �0.004[0.044] [0.047] [0.007] [0.007]

Self-esteem 0.005 0.035 0.000 0.001[0.034] [0.037] [0.001] [0.001]

Observations 1,582 1,220 1,618 1,2650.23 0.24 0.22 0.23

R-Squared



Table A3Continued

Regression Coefficients Contribution to Gender Pay Gap

1 2 3 4 1 2 3 4

Female Male Female MaleFemaleCoeffs

MaleCoeffs

FemaleCoeffs

MaleCoeffs

Total Explained 0.069 0.043 0.067 0.027[0.027] [0.025] [0.023] [0.021]

Notes. The basic equation for each row includes controls for whether there are any children in the household,quadratic in actual full-time and part-time labour market experience, age left full-time education, quadraticfor current tenure, detailed qualifications, marital status, ethnic, establishment size, whether a supervisor andfuture plans for (further) children.Standard errors in parentheses.Details of the questions can be found in Table A2.

London School of EconomicsUniversity of York

Submitted: 7 June 2005Accepted: 29 March 2007

ReferencesAltonji, Joseph G. and Blank, Rebecca M. (1999). �Race and gender in the labor market�, in (Orley Ashenfelter

and David Card, eds.), Handbook of Labor Economics. Volume 3C, pp. 3143–259, Amsterdam: North-Holland.Altonji, Joseph G. and Williams, Nicolas (1998). �The effects of labor market experience, job seniority, and

job mobility on wage growth�, in (Solomon W. Polachek and John Robst, eds), Research in Labor Economics,vol. 17, pp. 233–76, Stamford, CN and London: JAI Press.

Arulampalam, Wiji (2001). �Is unemployment really scarring? Effects of unemployment experience on wages�,Economic Journal, vol. 111, pp. F585–606.

Babcock, Linda and Laschever, Sara (2003). Women Don’t Ask: Negotiation and the Gender Divide, Princeton:Princeton University Press.

Barber, Brad M. and Odean, Terrance (2001). �Boys will be boys: gender, overconfidence, and common stockinvestment�, Quarterly Journal of Economics, vol. 116, pp. 261–92.

Bartel, Ann P. (1980). �Earnings growth on the job and between jobs�, Economic Inquiry, vol. 18, pp. 123–37.Bartel, Ann P. (1982). �Wages, nonwage job characteristics and labor mobility�, Industrial and Labor Relations

Review, vol. 35, pp. 578–89.Bartel, Ann P. and Borjas, George J. (1981). �Wage growth and job turnover: an empirical analysis�,

in (Sherwin Rosen, ed.), Studies in Labor Markets, pp. 65–90, Chicago: Chicago University Press and NBER.Becker, Gary S. (1985). �Human capital, effort and the sexual division of labor�, Journal of Labor Economics,

vol. 3, pp. S33–58.Becker, Gary S. (1993). Human Capital: a Theoretical and Empirical Analysis, with Special Reference to Education,

3rd ed, Chicago: University of Chicago Press.Ben-Porath, Yoram (1967). �The production of human capital and the life-cycle of earnings�, Journal of Political

Economy, vol. 75: pp. 352–65.Black, Dan, Haviland, Amelia, Sanders, Seth and Taylor, Lowell (2003). �Gender wage disparities among the

highly educated�, unpublished manuscript, Syracuse University.Blau, Francine D. and Kahn, Lawrence M. (2004). �The US gender pay gap in the 1990s: slowing convergence�,

NBER Working Paper No. w10853, available at SSRN.Booth, Alison, Francesconi, Marco and Frank, Jeff (2003). �A sticky floors model of promotion, pay and

gender�, European Economic Review, vol. 47, pp. 295–322.Borjas, George J. (1981). �Job mobility and earnings over the life cycle�, Industrial and Labor Relations Review,

vol. 34, pp. 365–76.



Bowles, Samuel, Gintis, Herbert and Osborne, Melissa (2001). �The determinants of earnings: a behavioralapproach�, Journal of Economic Literature, vol. 39, pp. 1137–76.

Brown, Charles and Corcoran, Mary (1997). �Sex-based differences in school content and the male-femalewage gap�, Journal of Labor Economics, vol. 15, pp. 431–65.

Burdett, Kenneth (1978). �A theory of employee job search and quit rates�, American Economic Review, vol. 68,pp. 212–20.

Chevalier, Arnaud (2004). �Motivation, expectations and the gender pay gap for UK graduates�, IZADiscussion Paper No. 1101.

Cobb-Clark, Deborah (2001). �Getting ahead: the determinants of and payoffs to internal promotion foryoung men and women�, Research in Labor Economics, vol. 20, pp. 339–72.

Croson, Rachel and Gneezy, Uri (2004). �Gender differences in preferences�, available at http://www.hks.harvard.edu/wappp/research/rachelcrosonandurigneez.pdf.

Crossley, Thomas F., Jones, Stephen R.G. and Kuhn, Peter (1994). �Gender differences in displacement cost�,Journal of Human Resources, vol. 29, pp. 461–80.

DiNardo John, Fortin, Nicole M. and Lemieux, Thomas (1996). �Labor market institutions and the distri-bution of wages 1973-1992: a semiparametric approach�, Econometrica; vol. 64, pp. 1001–44.

Dohmen, Thomas, Falk, Armin, Huffman, David, Sunde, Uwe, Schupp, Jurgen and Wagner, Gert (2005).�Individual risk attitudes: new evidence from a large, representative, experimentally-validated survey�, IZADiscussion Paper No. 1730.

Dolton, Peter, Joshi, Heather and Makepeace, Gerry (2002). �Unpacking unequal pay between men andwomen across the lifecycle�, Working Paper No. 2., Centre for Longitudinal Studies, Institute of Edu-cation, University of London.

Eckel, Catherine C. and Grossman, Philip J. (1998). �Are women less selfish than men? Evidence from dictatorexperiments�, Economic Journal, vol. 108, pp. 726–35.

Feinstein, Leon (2000). �The relative economic importance of academic, psychological and behaviouralattributes developed in childhood�, unpublished, LSE CEP Discussion Paper and University of Sussex,Discussion Paper in Economics, No. 62.

Fisher, Helen (1999). The First Sex: The Natural Talents of Women and How They are Changing the World, New York:Random House.

Gibbons, Robert and Katz, Lawrence (1991). �Layoffs and lemons�, Journal of Labor Economics, vol. 9, pp. 351–80.Goldsmith, Arthur H., Veum, Jonathan R. and Darity, William Jr. (1997). �The impact of psychological and

human capital on wages�, Economic Inquiry, vol. 35, pp. 815–29.Gneezy, Uri and Rustichini, Aldo (2004). �Gender and competition at a young age�, American Economic

Review, Paper and Proceedings, May, pp. 377–81.Gneezy, Uri, Niederle, Muriel and Rustichini, Aldo (2003). �Performance in competitive environments:

gender differences�, Quarterly Journal of Economics, vol. 118, pp. 1049–74.Gregory, Mary and Jukes, Robert (2001). �Unemployment and subsequent earnings: estimating scarring

among British men, 1984-94�, Economic Journal, vol. 111, pp. F607–25.Heckman, James J., Lochner, Lance J. and Todd, Petra E. (2003). �Fifty years of Mincer earnings regressions�,

NBER Working Paper No. 9732.Jacobson, Louis S., LaLonde, Robert J. and Sullivan, Daniel (1993). �Earnings losses of displaced workers�,

American Economic Review; vol. 83, pp. 685–709.Jones, David R. and Makepeace, Gerald H. (1996). �Equal worth, equal opportunities: pay and promotion in

an internal labour market�, Economic Journal, vol. 106, pp. 401–9.Joshi, Heather and Paci, Pierella (1998). Unequal Pay for Women and Men: Evidence from the British Birth Cohort

Studies, Cambridge, MA: MIT Press.Keith, Kirsten and McWilliams, Abigail (1997). �Job mobility and gender-based wage growth differentials�,

Economic Inquiry, vol. 35, pp. 320-33.Keith, Kirsten and McWilliams, Abigail (1999). �The returns to mobility and job search by gender�, Industrial

and Labor Relations Review, vol. 52, pp. 460–77.Kletzer, Lori G. (1998). �Job displacement�, Journal of Economic Perspectives, vol. 12, pp. 115–36.Kunze, Astrid (2003). �Gender differences in entry wages and early career wages�, Annales d’Economie et Stat-

istique, vol. 71/72, pp. 245–66.Kunze, Astrid (2005). �The evolution of the gender wage gap�, Labor Economics, vol. 12, pp. 73–97.Light, Audrey and Ureta, Manuelita (1995). �Early-career work experience and gender wage differentials�,

Journal of Labor Economics, vol. 13, pp. 121–54.Loprest, Pamela J. (1992). �Gender differences in wage growth and job mobility�, American Economic Review,

vol. 82, pp. 526–32.Machin, Stephen and Puhani, Patrick (2003). �Subject of degree and the gender wage differential: evidence

from the UK and Germany�, Economics Letters, vol. 79, pp. 393–400.Manning, Alan (1997). �Mighty good thing: the returns to tenure in a search model�, LSE CEP Discussion

Paper, No. 383, available at http://cep.lse.ac.uk/pubs/download/DP0383.pdf.



Manning, Alan (2000). �Movin� on up: interpreting the earnings-experience profile�, Bulletin of EconomicResearch, vol. 52, pp. 261–95.

Manning, Alan (2003a). Monopsony in Motion: Imperfect Competition in Labor Markets, Princeton: PrincetonUniversity Press.

Manning, Alan (2003b). �The real thin theory: monopsony in modern labour markets�, Labour Economics, vol.10, pp. 105–31.

Manning, Alan and Robinson, Helen (2004). �Something in the way she moves: a fresh look at an old gap�,Oxford Economic Papers, vol. 56, pp. 169–88.

McLaughlin, Kenneth J. (1991). �A theory of quits and layoffs with efficient turnover�, Journal of PoliticalEconomy, vol. 99, pp. 1–29.

Mincer, Jacob (1974). Schooling, Experience and Earnings, New York: National Bureau of Economic Research.Mincer, Jacob and Jovanovic, Boyan (1981). �Labor mobility and wages�, in (Sherwin Rosen, ed.), Studies in

Labor Markets, pp. 21–64, Chicago: Chicago University Press and NBER.Mincer, Jacob and Ofek, Haim (1982). �Interrupted work careers: depreciation and restoration of human

capital�, Journal of Human Resources, vol. 17, pp. 3–24.Mincer, Jacob and Polachek, Solomon (1974). �Family investments in human capital: earnings of women�,

Journal of Political Economy, vol. 82, pp. S76–108.Mueller, Gerrit and Plug, Erik (2004). �Estimating the effect of personality on male-female earnings�, IZA

Discussion Paper No. 1254.Murphy, Kevin M. and Welch, Finis (1990). �Empirical age–earnings profiles�, Journal of Labor Economics; vol. 8,

pp. 202–29.Myck, Michal and Paull, Gillian (2001). �The role of employment experience in explaining the gender wage

gap�, IFS Working Paper, No. 01/18.Niederle, Muriel and Westerlund, Lise (2007). �Do women shy away from competition?, Quarterly Journal of

Economics, vol. 122, pp. 1067–101.Nyhus, Ellen K. and Pons, Empar (2005). �The effects of personality on earnings�, Journal of Economic Psy-

chology, vol. 26, pp. 363–84.O’Neill, June (2003). �The gender gap in wages, circa 2000�, American Economic Review Papers and Proceedings,

vol. 93, pp. 309–14.Oaxaca, Ronald (1973). �Male–Female Wage Differentials in Urban Labor Markets�, International Economic

Review, vol. 14, pp. 693–709.Polachek, Solomon W. (2004). �How the human capital model explains why the gender wage gap narrowed�,

IZA Discussion Paper, No. 1102.Ruhm, Christopher J. (1991). �Are workers permanently scarred by job displacements?�, American Economic

Review; vol. 81, pp. 319–24.Swaffield, Joanna (2000). �Gender, motivation, experience and wages�, CEP Discussion Paper, No. 457,

available at http://cep.lse.ac.uk/pubs/download/DP0457.pdf.Topel, Robert H. and Ward, Michael P. (1992). �Job mobility and the careers of young men�, Quarterly Journal

of Economics, vol. 107, pp. 439–79.Vella, F (1994). �Gender roles and human capital investment: the relationship between traditional attitudes

and female labour market performance�, Economica, vol. 61 (May), pp. 191–211.Waldfogel, Jane (1998). �The family gap for young women in the United States and Britain: can maternity

leave make a difference?�, Journal of Labor Economics, vol. 16, pp. 505–45.Wood, Robert G., Corcoran, Mary E. and Courant, Paul N. (1993). �Pay differences among the highly-paid: the

male-female earnings gap in lawyers� salaries�, Journal of Labor Economics, vol. 11, pp. 417–41.

1024 [ J U L Y 2008]T H E E C O N O M I C J O U R N A L


the gender gap in early-career wage growth

Documents