Karen Mumford and Cristina Sechel

Transcription

1 Discussion Papers in Economics No. 17/17 PAY, RANK AND JOB SATISFACTION AMONGST ACADEMIC ECONOMISTS IN THE UK. Karen Mumford and Cristina Sechel Department of Economics and Related Studies University of York Heslington York, YO10 5DD

2

3 PAY, RANK AND JOB SATISFACTION AMONGST ACADEMIC ECONOMISTS IN THE UK. * Karen Mumford 1,2 and Cristina Sechel 3 November We use new data to explore the determinants of pay, rank, and job satisfaction for academic economists in the UK. After allowing for a broad range of characteristics, including measures of individual productivity and workplace features, we find a raw (unconditional) gender salary difference of 15 log percentage points (lpp) and a conditional gender pay gap of 9 lpp. This aggregate pay gap is strongly influenced by the relative concentration of men in higher paid job ranks where there are also within-rank gender pay gaps. Nevertheless, the majority of academic economists (male and female) are satisfied with their job. JEL A1, A11, A2, I3, J01, J31, J7 Key words: economics; gender; pay, satisfaction, gaps, academia. * We would like to thank Heather Antecol, Denise Osborn, Peter Smith, and participants at WEAI 2017 and WPEG 2017, for helpful comments and advice. Corresponding author: Prof. Karen Mumford, karen.mumford@york.ac.uk Department of Economics and Related Studies, University of York. Heslington York YO10 5DD, UK. (phone ). 1

4 The relative position of women in UK academic economics has changed dramatically in the last twenty years. Comparison of balanced samples 1 for 1996 (Mumford, 1997) and 2016 (Tenreyro, 2017) shows the proportion of this workforce that is female increased from less than one-in-six in 1996, to more than one-in-four in 2016, with women improving their relative representation in all of the academic job ranks 2. In 2016 women were 35% of the Lecturers (17% in 1996), 26% of the Readers/Senior Lecturers/Associate Professors (10% in 1996) and 16% of the Professors (4% in 1996). The movement of women into academic economics is not unique to the UK, for example, Canada, America, and Italy have seen similar trends. 3 The growth in female participation in many academic disciplines has been addressed in a series of reports investigating the status of women faculty in high prestige institutions, especially the early MIT study (MIT, 1999) and subsequent studies at the California Institute of Technology (Sargent, 2001); Duke University (Keohane, 2003); and MIT (MIT, 2011). These studies explicitly include discussion of reforms aimed at improving gender equality in academia such as greater awareness of unconscious bias, more equal access to resources, and ensuring female participation in governance. A pertinent empirical measure of relative equality in the labour market is the gender pay gap. Early studies focusing on the gender pay gap for academic economists are rare. Ward (2001; page 1669), in her study of academic pay in Scotland, provides a useful survey. She concludes that evidence of gender differences in salary is typically found although comparisons are difficult due to inconsistent approaches. In an influential study of the gender pay gap amongst US academics in the STEM disciplines between 1972 and 1997, Ginther (2001) finds substantial within job-rank conditional gender salary differences, peaking at 14% amongst full Professors. McDowell et al. (1999 and 2001), using panel data on members of the American Economic Society ( ) and ordered probit analysis, find that women are less likely to 1 There have been considerable institutional changes during this time period, balanced sample analysis is consequently far from perfect. Such analysis has been carried out biennially by the Royal Economic Society (RES) Womens Committee and published in their reports, the regular comparison of these findings provides some support for the balanced sampling (see ). 2 There has also been substantial job rank inflation across the discipline with the numbers of Professors (male and female) amongst all staff doubling over the time period (from 13% of all staff to 26%). In 1996, 6% of the women working full-time were Professors (19% in 2016), and 22% of the males were (35% in 2016). 3 In Canada: 5.4% of Professors were female in 1999, by 2014 this has risen to 14%; 10.9% of Associate Professors were female in 1999 and 27.5% were in 2015 (CWEN, 2015; Table 3). In Italy: 5% of Professors were female in 1999, by 2016 this has risen to 16%; 22% of Associate Professors were female in 1999 and 33% were in 2016 (Corsi et al., 2016 and 2017). In America: 6.6% of Professors were female in 1999, by 2016 this has risen to 13.1%; 13.5% of Associate Professors were female in 1999 and 25.6% were in 2015 (CSWEP, 2017; Table 1). 2

5 be promoted than men at every job rank. Ginther and Hayes (2003) consider US academics working in the humanities ( ) and also find substantial differences in the promotion paths for men and women. This is particularly so for academic economists in the US relative to other disciplines between 1975 and 2000 (Ginther and Kahn, 2004). An early withininstitution study of gender salary differences is provided for Princeton where an unconditional gender pay gap at the mean of 18% is found across all faculty in 2002, with a conditional gap of 8% once measures of experience and accomplishment were included (Tilghman, 2003; page 22). The conditional gap for different broad discipline groups (including Social Sciences) were found to be across, rather than within, job ranks (Girgus, 2005). The Royal Economic Society (RES) commissioned an early cross-sectional survey of pay and possible discrimination amongst academic economists in the UK in The survey was carried out by the RES Working Party on the Representation of Ethnic and Other Minorities in the Economic Profession (see Blackaby and Frank, 2000). The data are subsequently used in Blackaby et al. (2005) to explore the gender pay gap; they find an unconditional gender salary difference of 17.7 log percentage points (lpp) and a conditional gender pay gap of 9.4 lpp, with a substantial component of this conditional gap occurring within job ranks. As with many developed countries (Blau and Kahn, 2017), the UK has seen a substantial decrease in gender pay differences across its national workforce in recent years: the mean UK full-time unconditional gender pay gap has fallen from 21% in 2004 to 18.7% in 2011; and further to 16% in 2015 (Butcher et al., 2016; page 36). We might expect to see a fall in the gender pay gap amongst academic economists in the UK as well. A rare recent study of salaries and the gender pay gap, at the London School of Economics (LSE) in 2015, finds an unconditional gender pay difference for academics of 16.5 log percentage points (lpp) and a gap of 10.5 lpp after controlling for age, tenure and research productivity (Bandiera et al., 2016). The within-job rank gender pay differential accounted for some 25% of the conditional gap and the within-job rank pay gap increased with seniority. Bandiera et al. (2016) also finds that, when using a balanced sample, the gender pay gap conditional on age and experience has actually risen at the LSE from 3 lpp in 1998 to 9 lpp in Contemporary studies of the relative position of women in academia tend to provide detailed analysis of what may be considered components in the determination of salary. For 3

6 example, gender differences in the production and recognition of quality (Sarsons, 2017) and/or quantity of research publications (Hammermesh, 2013; Aiston, 2014; Joeks et al., 2014; Eagan and Garvey, 2015; Krapf et al., 2017); marriage and promotion (Mason et al., 2013); applying for, and being awarded, research grants (Marsh et al., 2011); and mentoring and career progression (Blau et al., 2010). However, these studies do not include direct information on salaries and the gender pay gap. In this paper, we return to consider gender pay differentials for academic economists across institutions. We employ a particularly rich source of new data generated by the authors from surveying individual academic economists in These data are combined with institutional information collected from the Royal Economic Society Women s Committee Surveys to explore the current determinants of pay, job rank, and job satisfaction for academic economists in the UK. The data are described in section 1 of the paper; section 2 discusses the estimation of the earnings function; decomposition analysis is presented in section 3; the probability of being in different job ranks is considered in section 4; changes in the gender pay gap over time are explored in section 5; job satisfaction is addressed in section 6; and section 7 presents conclusions. 1. Data Information was gathered by the authors from an online survey ed to individual academic staff members via their Heads of Department or similar department contact between February 26 and March 28, The individual staff member s responses were collated automatically via the survey software (Qualtrics) in an anonymised manner. Hard copies of the survey were also circulated at the 2016 Royal Economic Society Conference (March 21-23, 2016), resulting in a further 24 usable responses. In total, there were 668 responses, however, many had little or no information and may have been accessed to simply look at the questionnaire rather than to participate in the survey. There were 543 responses providing information on job rank. Given missing information on other variables of interest this allowed for the estimation of ordered probits for job rank with 526 observations. There was a substantial reduction in the number of respondents who provided salary information; 383 did, allowing for the estimation of wage 4 Contact details for the Heads of Departments were obtained from CHUDE (the Conference of Heads of University Departments of Economics). CHUDE was established by the Royal Economic Society (RES) in 1987 in collaboration with the Association of University Teachers of Economics. Additional information on the survey distribution and invitation is provided in the Appendix, Section A1. 4

7 regressions with 367 observations. Respondents were reticent when it came to revealing their job satisfaction in combination with other potentially identifying information; 443 respondents reported levels of job satisfaction but given missing information on the other explanatory variables, the probability of being satisfied with job estimation covers a sample of 306 individuals. Supplementary institutional information is collected from the Royal Economics Society Women s Committee Survey (Mitka et al., 2015; Tenreyro 2017). This data series was started by Mumford and Osborn in 1996, Mumford returned to collect the data (with others) 5 from 2008 to 2012, and with Sechel in 2013 and The Women s Committee Survey harvests information from CHUDE listed university department webpages on the individual academic staff by grade of employment and gender. These survey entries are then ed biennially to respective institutions for verification. Their overall verified survey response rate was high in 2015 with some 84% of the institutions responding, it was 57% in The Women s Committee Survey 2016 (Tenreyro, 2017) suggests there were 2077 workers across the entire UK academic economics workforce in This would imply a total response rate for our survey of 32.7% (668/2077) with a useable response rate of 26.1% (543/2077), falling to 18.4% for those providing salary information. Summary statistics are provided in Appendix Table A1 (column 1 for the pooled sample, column 2 for men, and column 3 for women). 6 There are concerns that the sample does not fully reflect the population. This concern is obvious in two main places. First, females make up some 43% of our total sample; however, Tenreyro (2017) found the proportion of the UK academic economics workforce that is female is only 28%. Second, a little over a third of our sample are Professors (Table A1), 28.7% of the women and 42.3% of the men. Tenreyro (2017; Table 1) found 25.5% of the workforce were Professors; 29.9% of the men and only 14.2% of the women. There is clearly an overrepresentation of Professors, especially female, in our sample. The sample accordingly has some underrepresentation of Lecturers amongst the 5 The RES Womens Committee reports, including authorship details, can be found at 6 Comparable summary statistics for the samples used in the job ranking analysis are provided in the Appendix Tables A2, and Table A3 for the job satisfaction analysis. 5

8 women (Tenreyro found 31.8% of women were Lecturers, we have 24.2%) and an underrepresentation of Researchers amongst the men (12.2% in Tenreyro, 8.6% in our sample). One compensation of the over representation of female Professors in the sample is the inclusion of a reasonable number of observations in the analysis as there are 45 female Professors in the sample (54% of the potential population, Tenreyro 2017). In contrast, Blackaby et al. (2005) could include only 7 female Professors in their analysis. Variable definitions are provided in Appendix Table A1, most of these are selfexplanatory but some are worthy of discussion here. The measure of annual salary is the current gross full-time equivalent wage for the main job. On average, the UK academic economists in our sample earn a full-time equivalent salary of 55,389 at the median in their main job: males average 60,000 and females average 52,000. This implies a raw gender pay difference of 15.4% at the median. 7 Only 2% of the men believe they have ever suffered from gender discrimination, in striking contrast to 24% of the women 8. Furthermore, the men perceive the average Professorial gender pay difference to be 9%, with the women predicting it to be 26%. In our sample the Professorial gender pay difference is 14.3% at the mean. It would appear that these academics do not have a good understanding of their wage relative to their fellow employees, male or female, unlike most employees (Hampton and Heywood, 1993). If we consider only the predictions made by the Professors, the predicted gap is less biased but there is still a substantial gender difference. The male Professors predict this salary difference to be 11% and the female Professors predict it to be 18%; the women are clearly unduly pessimistic and the men unduly optimistic. Most authors adopt the human capital model as the theoretical basis for the earnings function (Becker, 1974; Mincer, 1975). This approach will also be used here. It is assumed that wages increase with measures related to individual productivity: own education; research output and funding; and self-reported teaching excellence. The earnings function is augmented with the addition of further categories of explanatory variables including: demographic 7 Unsurprisingly, the gender pay difference at the mean is higher; the mean male wage is 73,109 and the mean female wage is 60,418, implying a raw gender pay gap at the mean of 21%. 8 Various specifications considering the probability of a relationship between perceived gender discrimination and the unexplained component of earnings were considered, no significant relationship was established for either gender. Analogously, the measure was not pertinent when considering the determinants of job satisfaction. Results are available from the authors upon request. 6

9 variables which may constrain an individual s choice of jobs (having children, marital status, ethnic identification, and age); workplace characteristics (working in a stronger research department, regional location, the percentage of women in the department, workplace network available, mentoring, if the workplace is perceived to be cooperative, or if the workplace is perceived to be competitive); and a range of variables loosely reflecting the individuals response to the labour market (being an external appointment, taking a career break, working part-time, engagement with the promotion process, and attracting outside job offers). Beginning with the demographic variables, the great majority of this workforce classify themselves as white (86%), they are relatively young with close to half aged between 35 and 49, three quarters are married and more than half have children (see Table A1). The women are less likely to be married and less likely to have children. It may be argued that if women believe they will be primarily responsible for childcare after marriage they will be less willing to incur the necessary investment expenditures for entering this occupation (Becker 1985; Summers 2005). For those who have chosen to enter, however, the relationship between academic salary and being married and/or having children is not clear. Empirical evidence is mixed. For example, Mason et al. (2013) find having children has a negative association with female academic career progression but is positive for males. Ginther and Hayes (1999 and 2003) find no marriage effect but establish a positive and significant promotion effect from having children for men; they find married women and mothers are less likely to be promoted. In contrast, Wolfinger et al. (2008) argues being a parent increases the likelihood of tenure, regardless of the gender of the parent. There is a noticeable difference in the age distribution for men and women, women are considerably more likely to be aged below 50 (75% of the women compared to 63% of the men). Whilst there is clearly not a one-to-one relationship between age and academic job-rank, there are incremental salary steps within many of the job ranks in different institutions and it is important to control for age accordingly. Considering the measures expected to be positively associated with individual productivity and earnings, a little over one in five of the total sample has a UK first class undergraduate degree (slightly more males than females). Males are more likely to have a PhD (92% compared to 82% of the women), and to consider themselves to be better teachers (19% of men consider themselves to be excellent teachers, whilst 16% of the women do). Females are, however, more likely to have been awarded more than 100,000 in research grants in the previous 5 years (35% relative to 24% for the men). Individuals were asked to provide a REF 7

10 (Research Excellence Framework) 9 style publication score for each of their career best three publications (ranging from zero to four), these were averaged into a single mean value. Women report a slightly lower average REF style score (self-reported over their three publications) than men, but not significantly so. It is important to explicitly include productivity measures in the empirical analysis not least because there is a mixed literature on the relationship between gender and research productivity of academics (Hammermesh, 2013). Many recent studies find no gender differences (European Commission, 2011; Marsh et al., 2011; Aiston, 2014; Eagan and Garvey, 2015). In contrast, Krapf et al. (2017) find no relationship between research productivity and fatherhood, but a loss of between 2 to 4 years of research output for mothers (of two or three children). Of the workplace characteristics, a similar proportion of both genders work in a top 6 (on 2014 REF ranking) department, although women are more likely to work in the old universities. 10 We might expect the top 6 research departments and the old institutions to provide more facilities conducive to academic performance and for earnings to be higher on this basis. Women are also more likely to work with other women (or, alternatively, in a more feminised work-place) than are men. The percentage of females in the workplace is taken from the RES Women s Committee survey data for 2014 (Mitka et al., 2015), this avoids potential difficulties extrapolating from our sample when calculating this measure. Using this institutional measure, the men in our sample are typically working in a workforce which is 23% female whilst for women this value is 28%. Working in a more feminised workplace is commonly associated with lower salaries (Groshen, 1991; Solnick, 2001; and Bayard, 2004) and is often argued to be linked to over-crowding and a decrease in bargaining power (Babcock and Laschever, 2003; Leibbrandt and List, 2015). Most (79%) of the UK academic economists work in England (60.5% excluding London). Women are more likely to work in London and in Scotland than are men. A positive relationship between working in London and earnings is expected as universities provide a London weighting (an additional salary component to partially compensate for the higher costs associated with living in London). 9 The Research Excellence Framework (REF) is an ongoing exercise judging the research quality of academic institutions in the UK. REF incorporates a range of measures but focusses largely on the quality of publications (subject to submission rules) of individual staff members. There is not an official list of journal quality used across economics departments, nor is there a commonly accepted list. 10 Old universities are those that were awarded their charter prior to the substantial movement of former Poytechnic and Central Institutions into the university sector in 1992 (with the Further and Higher Education Act, 1992). 8

11 Men are more likely to report that there are networks in their workplace they can use for advice concerning professional advancement (62% of the men compared to 55% of the women). However, some one in five of either gender have never had a formal mentor they could turn to for work related advice. The relationship between mentoring and salary is not clear in the literature (Quinn, 2012) although we might expect a positive relationship in the long run (Blau et al., 2010). It would seem that there are unofficial support processes in these workplaces that, whilst relatively commonly available, are operating more effectively for men than women. It may also be that women are less willing to seek out these networks if they lack confidence in their relative abilities and/or fear being judged. Women in our sample are considerably more likely to report that they feel their workplace is competitive or very competitive (49% relative to 39% for the men). Neiderle and Versterlund (2007) argue that, even with equal ability and productivity, women are more likely to shy away from competition than men. They argue that this gender difference is due to men being overconfident in their own abilities and women preferring non-competitive work environments. Interestingly, both genders in our sample report a similar average for feeling their workplace is cooperative or very cooperative (40% of the men and 38% of the women). Cooperation, or active recognition of mutual advantage, has long been associated positively with productivity in the labour economics literature (Mas and Moretti, 2009) and increasingly so amongst behavioural economists (Bruni and Sugden, 2013). Turning to the remaining variables loosely grouped together as labour market related, 46% of men and 44% of women have received an outside job offer in the previous five years and men are more likely to have been appointed from an external position (50% versus 41%). Men are less likely to have: applied for promotion in the previous 10 years or been rejected in this promotion process. These outcomes may be due to many factors (Leibrandt and List, 2015), including the males being on average older and in more senior ranks (Artz et al., 2016). We will return to consider these issues more fully in the analysis below. One in ten of the workforce currently works part-time, and this is much the same for men and women. Whilst we do not have strong priors on the relationship between part-time employment and full-time equivalent earnings in this sector, previous studies suggest a bimodal relationship across the economy with high and low skill employees choosing to work part-time (Mumford and Smith, 2009). Finally, the majority report that they are satisfied with their job (see Appendix Table A3); 67% of the men and 62% of the women, although this difference is not significant at standard confidence levels. The summary statistics are very similar for the samples used to 9

12 analyse the determinants of salary (Table A1), job rankings (Appendix Table A2) and job satisfaction (Appendix Table A3). 2. Estimating the Earnings Function There is an enormous literature examining the gender wage differential in the context of the human capital model developed by Becker (1975) and Mincer (1974). Following in this literature, using semi-logarithmic wage equations, we estimate the earnings equation as: = + X G + W i i i i where Wi is the natural log of the wage for individual i; α is an intercept term; Xi is a vector of regressors capturing the individual characteristics expected to impact on wages; and residual term. An indicator variable G identifies males in the dataset. In the pooled model, the parameters are common parameters for men and women on individual characteristics and δ measures the relationship between gender and the constant component. We begin our analysis with pooled wage equations for men and women (Neumark, 1988), estimating the earnings function using ordinary least squares, the results are presented in Table (1) i is a [TABLE 1 AROUND HERE] Column 1 of Table 1 shows the unconditional gender wage gap to be log percentage points (lpp), there are 367 observations in the regression, and the goodness of fit (in this case, the adjusted R-squared) measure is low at 2.4%. Additional categories of explanatory variables are added to the model from column 1 to column 4 of Table 1. Model 2, presented in column 2, includes individual productivity measures. The results are consistent with prior expectations, salary is positively and significantly related to the variables measuring individual productivity. In particular, having: a first class UK undergraduate degree; a PhD; higher REF style publication score; and more than 100,000 of research income in the previous 5 years are all associated with higher earnings. The exception is being an excellent teacher, which is not 11 Separate estimates for the two genders, and semi-pooled estimates including some gender interactive terms are also considered: = + X G G X + W i i i i i i The coefficients measure where the coefficients on any of the variables X i differ between men and women. The fully pooled model sets all =0 whereas separate models for male and female earnings would allow all to be non-zero and set δ=0 10

13 significantly related to wages. The conditional gender pay gap is 11.1 lpp in column 2 and the overall goodness of fit increases to 30.9%. Model 3 adds demographic and workplace characteristics. No significant relationship is found between earnings and ethnicity, being married or having children. Age is found to be important, however, a strong relationship is found between older age bands and the omitted year olds. Job rank is not included in these regressions due to obvious endogeneity concerns, so this relationship between age and earnings might be predicted. We return to discuss these issues further when we explicitly consider job rank in section 4 of the paper. Amongst the workplace variables, working in a top 6 REF ranked department and working in an old university are both positively associated with higher earnings as expected. The strong regional effect associated with working in London relative to the omitted other England, is also not surprising as these universities provide a London weighting. The percentage of the departmental workforce that is female is negatively associated with salary. This is a finding consistent with the traditional industrial relations, and more recent gender, literature and is often argued to be associated with gender segregation, over-crowding and a decrease in bargaining power (Groshen, 1991; Bayard, 2004; Bayer and Rouse, 2016). A negative and significant relationship is found between having a professional network available in the workplace and earnings and no significant relationship is found between having had a mentor and earnings. The gender pay gap is estimated to be 9.09 lpp and the measure of fit has increased substantially to 50.9% in Model 3. Model 4 (column 4) presents the full model including variables related to the labour market as reflected in the work environment. The gender pay gap is estimated to be 9.07 lpp and the measure of fit is 52.5%. Having received an outside job offer is positively and significantly related to earnings. We do not find a part-time pay penalty, however, this may be partly due to the dichotomous nature of the job ranks using part-time employment amongst academic economists; part-time employment is more common amongst Researchers (21%) but also amongst the Professors (13.4%). Gender specific estimation of Model 4 reveals very little difference (results are presented in columns 1 and 2 of Appendix Table A4). There are some qualitative differences, for example, higher salary is associated with marriage for men but not for women. However, 11

14 the standard errors are so high (especially in the estimation of female earnings) that these differences are not significantly different at standard confidence levels. It is particularly interesting that no significant gender difference in the relationship between salary and receiving an outside offer is found 12. These results do not suggest women face lower earnings because employers do not respond to them having an outside job offer as fully as they do with men. Instead, a strong positive relationship between salary and having received an outside job offer is found for both men and women. Similarly, we do not find a gender difference in the impact of applying for promotion in the last 10 years and salary Decomposing the Earnings Gap An unconditional earning gap between men and women in these data of lpp and a conditional gap of 9.07 lpp is found in section 2 above. Further insight into these gaps may be provided by decomposition analysis (Oaxaca, 1973). The approach we adopt to apportion the gap in the mean earnings of men and women here is that proposed by Neumark (1988) and discussed further in Oaxaca and Ransom (1994) where the reference set of parameters is given by the pooled estimates presented in Table 1. The decomposition of the mean earnings gap is calculated as: W W ( X X ) ˆ X m f m f f ˆ (2) Where ( X m X f ) ˆ captures the difference in the productive characteristics across the genders; and X ˆ f is the remaining gender gap which is usually regarded as unexplained. The decomposition results present calculations of the total unexplained gap as well as those parts of the explained gap that can be attributed to individual or groups of variables. As discussed by Jann (2008) amongst others, there may be a problem in interpreting these results for categorical variables where the results may depend on the choice of omitted category. The solution proposed by Jann, and implemented in Stata, is adopted here; taking the standard coefficient estimates and computing elasticities for all categories, including the omitted category, by reweighting. 12 Including a male-interacted-with-outside-job-offer term in Model 4 is found to have a low coefficient of with a standard error of The coefficient on the male-interacted-with-promotion-application term if added to Model 4 is with a standard error of

15 [TABLE 2 AROUND HERE] Aggregate decompositions for the earnings function of Model 4 (the full model) are presented in the first panel of Table 2. Comparing male with female academic economists, the total earnings gap is log percentage points in favor of the males. Of this gap, 5.98 log percentage points is due to this group of females having more of those characteristics associated with lower earnings than do the corresponding males. This explained portion makes up over a third (39.7%) of the total raw gap (of lpp) and is weakly significant at an 85% confidence level. The explained component can be further decomposed, see panel 2. The demographic variables make up some log percentage points of the lpp explained gap, clearly the largest component of the explained pay gap. The demographic component can be further decomposed into the portion associated with age (-4.04 lpp) and other (ethnicity, having children, and being married) which is only lpp. Thus, of the explained portion of the gender pay gap, the great majority (71.1%) is associated with demographic characteristics; gender age differences are associated with 95.1% of the total demographic component and 67.6% of the total explained pay gap. The age component can be further decomposed into the portions associated with each of the age bands included in the earnings function (also reported in panel 2 of Table 1). Each age band is associated with a negative component of the total negative age association; in accordance to the summary statistics presented in Table A1, women are more likely to be in the younger age bands associated with lower pay and less likely to be in the older age bands that are linked to higher pay. The youngest age band (20 to 34 year olds) is associated with lpp or 47.8% of the total age component; or 30.9% of the total explained gender pay gap. The relationships with the other age bands, the total age component, 14 and thereby the explained gender pay gap, are more modest and are not significant at standard confidence levels. The component of the decomposition related to feminisation at the workplace is some 14 The 35 to 50 age band is associated with 16.6% of the total age component, the 51 to 64 is associated with 20.1% and the above 65 age band is related to 17.3%. 13

16 -1.97 lpp or 32.9% of the explained pay gap. 15 The other workplace characteristics aggregate in favor of higher salaries for women (in total making up 2.52 lpp or 42.1% of the aggregate explained gap). The remaining components of the decomposition are not significant at standard confidence levels: women have on average lower mean levels of those characteristics associated with productivity (-1.83 lpp or 30.6% of the total explained pay gap); and are less likely to have labour market response characteristics (-0.45 lpp or 7.5%) that are related to higher pay. The remaining 9.07 log percentage points (clearly the major component of the gap at 60.3% and significant at the 99% confidence level) is unexplained and is due to the characteristics (as estimated in the full model) being in aggregate rewarded at a lower rate for females than for males (differences in the coefficients). To reiterate, the model does not explain why they are being rewarded differently (hence the term unexplained ). 4. Gender Differences in Job Ranks It is reasonable to expect there may be within-job rank gender differences that are also part of the picture of the relative status of academic economists. Table 3 presents ordered probit rank estimates for the full model, Model 4 above, for the pooled job rank sample of 526 individuals. The overall goodness of fit measure is not high (pseudo R 2 of 22.5%), nevertheless this is consistent with estimations of this type and the model presented in Table 3 is both well specified and consistent with prior expectations. [TABLE 3 AROUND HERE] 15 It constitutes lpp (standard error ) of the 5.98 lpp explained gap and lpp (standard error ) of the unexplained gap. The relevant decomposition of the mean earnings gap is calculated as: ^ ˆ k k ( ) ˆ m f m f f (% m % f ) ˆ W W X X X F F, where ( X m X f ) captures the difference in the productive characteristics across the genders (male m or female f); k is the workplace; %F is the percent proportion of that gender in the workplace and ^ k k [ X ˆ (% F % F )] f m f is the remaining gender gap. 14

17 The estimation results reported in Table 3, row 1, show that gender is significantly and strongly related to job rank for academic economists in the UK. Males are less likely to be employed as Lecturers, Researchers or Teaching Fellows; and are more likely to be Readers/Senior Lecturers 16 and, especially, Professors. The remaining explanatory variables are also consistently associated with higher job ranks. Being older 17, married, having a first class degree, having a PhD, higher publications score, receiving more than 100,000 in research income, working full-time and having an outside offer in the last 5 years are all significant and negatively associated with lower job ranks, and positively associated with higher job ranks. The impact of these trends can be most clearly seen by comparing Lecturers (column 3) with Professors (column 5): males are 6.6 percentage points less likely to be employed as Lecturers and they are 10 percentage points more likely to be Professors. Whilst there is clearly a gender difference in the probability of being employed at a higher rank, there may not necessarily be a within job rank gender pay gap. Re-estimating Model 4 (the full model) earnings equation in section 2 for each job rank does not suggest a gender pay gap in any of the job ranks at standard confidence levels, 18 as is seen by the conditional estimated relationship between being male and earnings presented in Table 4. It is worth noting, however, that the 7.3 lpp gender pay gap found amongst Senior Lecturer/Readers is significant at the 85% confidence level and the 6.9 lpp gap amongst the Professors is significant with 75% confidence. These pay gaps are not very precisely measured but may be considered as evidence of within rank gender pays gaps at the higher paid senior job ranks. [TABLE 4 AROUND HERE] 5. Changes Over Time Our study focuses on new data for a single cross section of UK academic economists in 2016 and, as such, does not allow for an intertemporal analysis. However, the results may be 16 Whilst many departments have moved away from having Readerships, there are still some institutions (such as York) that have two separate (but partially overlapping) pay scales for Senior Lecturers and Readers, implying that combining the job ranks may not allow for a true within rank estimate. 17 The great majority of this workforce is on a seniority based incremental pay structure within job ranks, however, these pay ladders are not uniform across institutions. 18 The Teaching Fellow rank could not be included in this exercise as the number of observations were too few to allow for the estimation procedure. 15

18 compared with previous studies. As discussed briefly above, a directly comparable survey was carried out by the Royal Economic Society Working Party on the Representation of Ethnic and Other Minorities in the Economic Profession in 1999 (see Blackaby and Frank, 2000). CHUDE-listed Heads of Departments were asked to distribute a hard copy survey to full-time academic staff in their departments; 516 individuals completed the survey from 1600 distributed, a very similar response rate to our survey of 32%. Blackaby and Frank also encountered difficulties with missing data, their job rank analysis is for 452 individuals (implying a response rate of 28.3%), and their earnings estimations are for 405 (25.3%). The data were subsequently used in Blackaby et al. (2005) to explore the gender pay gap more fully; they included information on 351 individuals (or 21.9%) in their earnings estimations. As discussed in section 1 above, there is an overrepresentation of Professors, especially female, in our sample. Blackaby et al. (2005) encountered a similar overrepresentation of Professors, especially female Professors, in their sample. They could include only 7 female Professors in their analysis; however, this was also some 54% of the potential population at that time (Booth et al., 2000). The Blackaby and Frank (2000) data are no longer available and so a direct comparison cannot be performed with the 2016 data. Nevertheless, Models 1 and 3 (reported in columns 1 and 3 of Table 1 above) are similar to those presented by Blackaby et al. (2005), where they find an unconditional gender pay gap of 17.7 lpp and a conditional gender pay gap of 9.4 lpp (in contrast to our unconditional pay gap of lpp and a conditional pay gap of 9.1 lpp in model 3). Whilst there are differences in the variable specification, and Model 3 includes additional workplace variables, there is general consistency in the qualitative and quantitative results for the inclusion of (what we call) the demographic, productivity and workplace characteristics between our results and those in Blackaby et al. (2005). Results for a closer approximation to the Blackaby et al. (2005) specification are provided in Table A5 of the Appendix. There are still substantial differences in the specifications of the models, for example, the 2016 estimation uses mid-points of the age bands and does not include a measure of non-academic years of work experience. With this simpler specification, unsurprisingly, the residual gender pay gap in 2016 is found to be larger than that presented in Table 2 (10.3 lpp compared to 9.1 lpp). It is worth stressing how striking this finding is; between 1999 and 2016 there is no notable fall in the unexplained gender pay gap for UK academic economists. As discussed 16

19 above, a study of 2015 salaries at the London School of Economics (LSE) found a very similar unconditional gender pay gap for academics of 16.5 lpp and a gap of 10.5 lpp after controlling for age, tenure and research productivity. (Bandiera et al., 2016). Bandiera et al. (2016) also found, using balanced sample analysis, that the gender pay gap conditional on age and experience had actually risen at the LSE from 3 lpp in 1998 to 9 lpp in Our results would suggest that the gender pay gap at the LSE has moved into line with the broader body of UK academic economic workplaces over the time period. Considering job rank, compared to Blackaby et al. (2005) we find a weaker negative male probability of being a Lecturer in 2016 (-6.6% in 2016, Table 3, relative to -13.8% found by Blackaby et al. (2005, Table 2)) and a stronger male probability of being a Professor (10% in 2016 relative to 7.4% in 1999), all significant at the 99% confidence level. Men are relatively less likely to occur amongst the Lecturers and are more likely to be Professors in 2016 than they were in Blackaby et al. (2005) also found evidence of a within rank gender pay gap amongst Lecturers of 6.3 lpp (with 99% confidence) but a clearly insignificant gap of 3.4 lpp amongst Professors in In contrast, the results reported in Table 4 indicate a clearly insignificant gender gap of 3.1 lpp amongst Lecturers in 2016, and a weakly significant (at 75% confidence) gender pay gap of 6.9 lpp amongst Professors. The within rank gender pay gap appears to have changed in nature over time: falling away for Lecturers but increasing amongst Professors. This conclusion is broadly consistent with Bandiera et al. (2016) who find: at the Assistant Professor (Lecturer) level the within rank gender gap is small at around 2% and insignificantly different from zero; with a significant gap of 7% for Associate Professors (Senior Lecturer/Reader) and 11% for Professors at the LSE. In summary, the gender pay gap for UK academic economists in 2016 is strongly influenced by the relative concentration of men in the higher paid job ranks where the unexplained gender pay differential is considerable. 6. Job Satisfaction There is a small, but influential, literature that seeks to address the common finding that whilst women typically earn less than men, they usually report higher levels of job satisfaction (Clark 17

20 and Oswald, 1996; Clark, 1997; Clark et al., 2009; Clark and Senik, 2010). Clark (1997) argued that this gender difference would diminish over time as younger female workers enter the labour market with expectations more closely matched to their relative labour market characteristics. Stevenson and Wolfers (2009) provide international evidence showing women s happiness (as measured by subjective measures of well-being linked to life satisfaction) has fallen absolutely, and relatively, to men to such an extent than men are now recording higher levels than women. Interestingly, in our job satisfaction sample (see Table A3 of the Appendix for summary statistics) the males are more likely to report being satisfied with their job (67%) than the women (62%) although the difference is not significant at standard confidence levels. The inclusion of a relative wage measure is also argued to be important in the recent literature on job satisfaction, as discussed in Mumford and Smith (2015). Card et al. (2012) highlight alternative explanations where: an employee values their own relative utility and may be dissatisfied if their wage is lower than other workers (see also Zizzo and Oswald, 2001); or a model of co-worker wage as a signal of their own future wages (see also Clark et al., 2009). To consider fully the alternatives, it is important to find the true comparator wage when setting the average wage of the reference group. For example, we could consider the within job rank, within institution, average wage as comparator averages. Unfortunately, our sample is not rich enough to generate such a measure. Comparator group averages are instead set at the average wage in either the old or new universities. Following Clark and Senik (2010) we model the probability of reporting job satisfaction S of worker i in institution k (Sik): S ( W, W W, X ) ik ik ik i ik where Wik is the wage of that academic; W i (3) is the average wage of his/her reference (or comparator) group and Xik is a vector of observable individual and institutional characteristics correlated with job satisfaction We do not claim to establish causality; as Ferrer-i-Carbonell and Frijters (2004; pages ) discuss in detail, such a claim would require all unobserved factors to be either orthogonal to the observables or suitably controlled for (e.g. via an assumed structure) and would require a cardinality assumption for the ranking of measured job satisfaction requiring exceptionally rich data and simultaneous estimation with suitable instruments. 18

21 A series of probit regressions is estimated with the unobserved latent dependent variable Sik set equal to 1 if the individual lecturer i reported above 5 to the survey question Overall how satisfied are you with your job these days? ; and zero otherwise. The scale of potential answers ranges from 1 (completely unsatisfied) to 10 (completely satisfied). The estimation results are presented in Table 5, marginal effects at the mean of the explanatory variables are reported with differential effects for binary variables (estimated coefficients are available from the authors upon request). In general, the models are not particularly well defined. The overall fit is also not high in absolute terms: the pseudo R-squared measures suggest the models are explaining less than a third of the variation in the reported job satisfaction amongst academic economists. [TABLE 5 AROUND HERE] Considering the results in more detail, column 1 of Table 5 reports the unconditional relationship between relative wage and job satisfaction. Academic economists are more satisfied when they earn more than their comparator group although the size of the effect is not large. This is consistent with a model of relative utility (Card et al., 2012). Column 2 adds the demographic variables, none of these are found to be significantly related to satisfaction (except that year olds are less likely to report job satisfaction relative to the omitted year old age group). It is particularly noteworthy that whilst the relationship between being male and reporting job satisfaction is positive, it is not significant at standard confidence levels. Results for the estimation of the full model, consistent with Model 4 in section 2, are reported in column 3 of Table 5. Of the demographic characteristics, or the individual characteristics associated with productivity, only having a UK first class undergraduate degree is found to be significantly related with job satisfaction. Amongst the workplace and labour market characteristics significant results are found for: regional differences; working in an old university (negatively related to job satisfaction); working in more feminine departments (negatively related); never having had a mentor (negatively related); being rejected from promotion in the last 10 years (negatively related); working in a cooperative environment (positively related); and being a Researcher (positively related). The relationship between relative wage and job satisfaction is smaller (and negative) than the relationship established in columns 1 or 2 and is no longer significantly different to zero. 19

22 The finding of an insignificant relationship between relative wage and job satisfaction may be due to a poor comparator group measure. Column 4 presents results for the estimation of the full model but including own-wage rather than relative wage, none of the results change significantly. A significant relationship between job satisfaction and own wage is also not found. We also estimated this model separately by gender (results are presented in columns 3 and 4 of Appendix Table A4). The model is typically not well specified and the sample sizes are small. Whilst there may appear to be some qualitative gender differences comparing columns 3 and 4, these were not found to be significantly different. For example, comparing the marginal effects on satisfaction of being male for married individuals (all else being held at mean values), and the marginal effect of being male for non-married individuals reveals no significant difference. Considering the pooled results, the conditional probability of being satisfied for males in model 4 (column 3) of Table 5 is and for females it is very similar at The predicted probability of being satisfied is also similar across different levels of relative salary and does not differ by gender at any of these levels (this is also true for own salary). The predicted probability of being satisfied is, however, significantly lower for departments with higher share of female staff and this pattern is observed both for males and females. Of particular note, and as discussed in section 1 above, it has been argued that women prefer less competitive work environments (Neiderle and Vesterlund, 2007; Flory et al., 2015), although this effect may be smaller for younger women (Garratt et al., 2013). We find no significant gender differences in the relationship between reported job satisfaction and workplace competitiveness. 20 To reiterate (see Table 5), job satisfaction is related to workplace characteristics for academic economists in the UK. With the exception of having a UK first class degree, no demographic or productivity related characteristic is found to be significantly related to job satisfaction. It is particularly notable that we find no gender differences in reported job satisfaction. Furthermore, the relationship between comparator wage (or own wage) and job 20 The results for job dis-satisfaction (i.e. reporting job satisfaction 4 or lower) also do not support this premise: the estimated coefficient for reporting a competitive environment in Model 4 is (standard error ) in female only estimates of job dis-satisfaction; and (0.3123) in male only estimates. Results are available from the authors upon request. 20