Gender differences in research performance and its impact on careers: a longitudinal case study

Gender bias in science

That bias may play a role was very clearly put on the agenda when Wennerås and Wold (1997) published their results of a study on grant decision-making. They showed that not meritocracy was the standard, but cronyism: having friends in the relevant committees proved to help considerably in getting a grant. On top of that, they also showed the role of gender bias: women needed a substantially (160 per cent) higher performance than male researchers to be successful in biomedical grant applications. Replicating that study some 10 years later, Sandström and Hällsten (2008) found again nepotism, but no sexism anymore: female researchers even had a slightly better chance than male researchers. Obviously, the council studied in both papers changed its gender policy in the meantime.¹ After the Wennerås and Wold study, quite a few other studies found that that gender matters, but some evidence suggests differently in different disciplines. For example, a Dutch study showed that in science fields female researchers received positive evaluations and high success rates, even higher than could be expected from past performance. In contrast, the life sciences were characterized by negative gender bias, as no substantial differences between track records of unsuccessful women and successful men were found (Brouns 2000).

Apart from the issue of gender bias in grant reviewing, there is also an issue about gender bias in academic recruitment and selection, leading to lower success rates of female applicants (Van den Brink et al. 2006; Van den Brink 2009). Female researchers have a slower career, and on average end at lower positions; women are still underrepresented in the higher academic positions and men outnumber women in positions of formal power, authority and high income (Xie and Shauman 1998; Timmers et al. 2010). Here again the question is at stake as whether these career differences relate to performance differences, or are based on biased decision-making.

In a recent review, Ceci and Williams (2011) discuss the evidence about gender bias in science, in journal reviewing, grant funding, and in hiring. They claim not to find evidence supporting the existence of discrimination against women in science. The authors conclude that the unequal position of women in science would be based on quality differences between male and female researchers that may partly be based on own career related choices, and partly on discriminatory arrangements not in science but in society at large — e.g., inequalities related to division of domestic work and child care. If this analysis is correct, we are back from gender bias to performance differences.

The main problem with most studies arguing that no gender bias exists is that they do not take performance into account at all. This is also the case with recent meta-reviews (Mutz et al. 2014; Marsh et al. 2009), claiming that gender bias no longer exists in peer review. These studies do not refer to performance, but look only at the success rates.

This is remarkable, as bias can only be measured against performance differences. More directly, if the slower career and the glass ceiling would be determined by performance differences alone, one might be willing to accept this—and focus policy on improving female performance. But with this omission, the studies mentioned actually are not very instructive and fail to show what they claim. They have no information about whether the applicants are statistically representative of their respective groups (male–female), and they also lack information about any self-selection processes that correlate with performance. For example, it might be that self-selection of female applicants applying for a job is stronger than of male researchers in the sense that female applicants have a higher average performance than male applicants. If that is the case we would expect better average bibliometric scores for the female group and, on average, less good results for the male group. These performance data then could be correlated with the committee decisions. We assume that bibliometric data are more or less unbiased for each applicant group and that full bibliometric data with relative citations scores will produce relevant correlates to the grading and ranking procedures of standing and ad hoc committees.

Productivity differences

Several explanations of the productivity puzzle (Cole and Zuckerman 1984) have been proposed: scientific ability, self-selection, social selection, and accumulated disadvantage (Zuckerman 2001).² According to the scientific ability explanation, male and female academics differ in biological and psychological characteristics, influencing research output. However, more recent research did not find a direct gender ability effect as was established in earlier research (e.g. Xie and Shauman 1998). Above that, one can observe at all levels of education that girls tend to outperform boys, which also contradicts this assumption (Buchmann et al. 2008; Pekkarinen 2008).

The second explanation is more widespread. In the early career, researchers get children and start a family. In fact, this may affect women more than men, as women still to the larger part of domestic work and child-care. This may translate in less time for research and therefore a lower scientific performance in the early career (Long 1992; Symonds et al. 2006). Only later in their career women more or less catch up with male researchers (Long 1992; Symonds et al. 2006), but this lower early productivity has a negative effect on careers (Prozesky 2008; Fuchs et al. 2001; Hunter and Leahey 2010; Karamessini 2004).³ However, other studies indicate that the academic career of female scientists does not suffer from parenthood (Fox & Faver 1985; Cole & Zuckerman 1991; Astin & Bayer 1979; Dryler 2011).

One may argue that getting kids is choice, but the fact that the bulk of domestic labor and of child-care is done by women, is no ‘self-selection’ but a social selection process (c.f. Fox 2005, Fox et al. 2011). This brings us to a variety of differences between male and female researchers, which are partly choice (self-selection) and partly based on social selection. Female researchers have a lower degree of specialization (Leahey 2006), tend to work in other disciplines, tend to focus more on teaching, tend to work at universities and departments with lower reputation, and have a less developed international collaboration and co-authoring network (Allison and Long 1990; McNamee et al. 1990; Dundar and Lewis 1998; Prpic 2002; Lee and Bozeman 2005; Bland et al. 2006; Carayol and Matt 2006; Leahey 2006; Taylor et al., 2006; Puuska 2010), which all affects performance and career in a negative way. However, Badar et al., (2013) report the opposite, but in the context of a developing country. This suggests that a specific study on gender and performance focusing on the latter context may be useful. Furthermore, women receive less academic support and mentoring than men (Landino and Owen 1988; Fuchs et al. 2001). This may be a disadvantage for women too, as academic careers depend on support by academic mentors (Van Balen et al. 2012). A slower career progress recursively may lead to less research and more teaching (Taylor et al. 2006; Snell et al. 2009), and consequently to less scholarly productivity. The accumulation of all these self and social selections over time leads to cumulative disadvantage (Zuckerman 2001).

However, over time gender roles and responsibilities in family life are changing (Xie and Shauman 1998; Taylor et al., 2006; Prozesky 2008), which may also influence work and career orientation of women. One may expect that the gradually changing gender roles in the last decades in many countries may also result into changed behavior. Indeed we have witnessed increased performance of girls in the educational system, and at all levels, girls are outperforming boys (Buchmann et al. 2008; Pekkarinen 2008). Recent data suggest that the performance gap has been closed for PhD students (Miller & Wai 2015; Ceci et al. 2014). If this is indeed a slow and generational process, one would wonder whether this gender performance turn has also reached the research system, first of all the generation of early career researchers.⁴

An earlier study (Van Arensbergen et al. 2012) indeed confirmed (i) the traditional gendered performance pattern for the older generations of researchers, but (ii) when isolating the group of early career researchers, gender differences have become smaller (in e.g., economics) or are disappearing (in psychology). The question remains as whether women now at a later moment face the performance drop.

Early career phase

Mid career phase

First for the whole sample: As far as (fractionalized) publications are concerned, the median performance and the distribution of performance of women are lower than the median and distribution of the male researchers. However, in all (field normalized) citations-based impact indicators, the differences are small, non-significant, and sometimes women score higher, sometimes men—as Table 4 (part: All) shows. For the three fields separately, the pattern is about the same—although in the case of psychology, male researchers have also in the citation-based indicators a somewhat (but non-significant) higher score.

In other words, the null hypothesis that men and women perform equal impact-wise cannot be rejected for most of the indicators. But, we have to reject the hypotheses that productivity is equal among the sexes. Indeed, the productivity gap has increased over the 10 years as becomes visible when comparing Table 1 and Table 3. The Male/Female productivity ratio was 1.09 (mean) and 1 (median), around 2004, and 9 years later the figures were 1.17 (mean) and 1.31 (median). Male researchers now publish on average about 17 % more publications than female researchers, and the former have a 31 % higher median production. For fractional counted publication the increase of the difference is even substantially higher, as is the case for the individual disciplines.⁸

Performance and careers

After analyzing the development of male and female research performance, the second question can be addressed: does performance determine academic careers? We collected data about the current position of the researchers in our dataset. Figure 1 shows that gender differences in career are still obvious.

Open image in new window

In terms of careers, male researchers score in average 13.7—almost associate professor. Female researchers score in average 12.6—slightly above assistant professor. Associate professor is also the median position for male applicants, whereas assistant professor is the median for women; so the gender difference is substantial. For the disciplines separately, we find about the same differences, with economics slightly higher and behavioral & education research slightly lower than psychology. Figure 2 shows that the unequal pattern is also visible at the level of the various disciplines separately.

Open image in new window

Taking only one variable into account (gender) gives an incomplete picture. In a meritocratic system one would expect that research performance has a strong influence on careers. As the gender differences are in the same (small) order of magnitude for all citation-based indicators, we selected only one impact variable (field normalized citations, with a citation window) for the model, plus productivity. We use fractional counted productivity as we expect panel and committee members to account in some imprecise way for the number of coauthors.

Apart from the performance variables, a few other variables should be taken into account. As we have three academic age groups and relative young researchers, the year of PhD may be a factor: the younger, the less chance to promote to higher positions. The average differences are small—only 1 year—between the groups (2003, 2004, 2005), but as we only consider a 10-years period, a difference of 1 or 2 years is substantial. Another factor may be mobility. If mobility influences the career, and men are more mobile than women, this together may lead to gender differences in careers. We distinguish three levels of mobility: (i) no mobility, (ii) national mobility, and (iii) international mobility. Finally, labor academic markets may de different for the different fields under study. Therefore we also use discipline of the researcher as variable.

First of all, the labor market seems better and careers seem faster in economics than in psychology, and in psychology better than in education & behavior.

Secondly, the year group variable has an effect on the achieved career level: if one belongs to the older cohorts 2003 and 2004, the average achieved career level is higher than for those in the cohort 2005. This is of course expected: the longer the career, the more time one has had for moving up.

Thirdly, mobility has only a small (and non-significant) effect. Note that the small effect does follow what would be expected: international mobility is slightly better than national mobility, which is slightly better than no mobility.

Fourthly, scholarly performance has a statistically significant positive effect on the career.

Finally, the analysis shows that gender has an effect on achieved career level as male researchers have achieved a higher average career level than female researchers, when controlling for all other variables.