2015 Why do Women Leave Science and Engineering?

http://journals.sagepub.com/doi/full/10.1177/0019793915594597?utm_source=Adestra&utm_medium=email&utm_content=8J0217&utm_campaign=not+tracked&utm_term=& Jennifer Hunt First Published July 23, 2015 Research Article Free Access Abstract The author uses the 2003 and 2010 National Survey of College Graduates to examine the higher exit rate of women compared to men from science and engineering relative to other fields. The author finds that the higher relative exit rate is driven by engineering rather than science, and that half the gap can be explained by the relatively greater exit rate from engineering of women dissatisfied with pay and promotion opportunities. Family-related constraints and dissatisfaction with working conditions are found to be only secondary factors. The relative exit rate by gender from engineering does not differ from that of other fields once women’s relatively high exit rates from male fields generally are taken into account. Keywords exit rate, gender, science and engineering, NSCG An education system and labor market that combine to match workers to the appropriate qualifications and jobs contribute to the static efficiency of an economy. This matching may be especially important for combinations of qualifications and jobs that lead to technological innovation and, hence, economic growth. The low share of women in science and engineering therefore suggests the possibility of both static and dynamic inefficiency, particularly given women’s very low share of patents (7.5% in 2003; Hunt, Garant, Herman, and Munroe 2013). Women’s underrepresentation in engineering may be more significant for innovation and growth than their underrepresentation in science. Hunt (2011) found that most of the gender patenting gap is among holders of science and engineering degrees and that this group’s gap is attributable in large part to women’s underrepresentation in engineering and overrepresentation in the life sciences. Women’s representation in science and engineering is determined by both entry and exit. Women’s low entry to science and engineering has been examined in the context of choice of field of study (recent examples include Stinebrickner and Stinebrickner (2011); Antecol and Cobb–Clark (2013); Zafar (2013)). Preston (1994, 2004, 2006) and Frehill (2012) documented that women leave science and engineering at a higher rate than men. Given that girls receive less encouragement than boys to enter science and engineering, making women in science and engineering likely to be more positively selected for interest and aptitude than their male counterparts, women’s higher exit rate hints at labor market inefficiency. Certain practices in science and engineering firms may prevent women from reaching their full potential, or outright discrimination against women may be present in such firms. The existence of such practices or discrimination could in turn discourage forward-looking women from entering science and engineering. In this article, I introduce a new way of considering female exits from science and engineering, and their link to women’s underrepresentation in those fields. Preston (1994, 2004, 2006) demonstrated that, although female scientists and engineers are more likely than their male counterparts to stop working, they are also more likely to move to a job outside science and engineering; Frehill (2012) provided similar results for engineers. The considerable literature on women leaving science and engineering has highlighted the difficulty of balancing long work hours and family in science and engineering; the isolation of being a minority, and the associated lack of mentoring and networks; the risk-taking environment; the hostile macho culture; and discrimination. Related factors identified were the prevalence of laboratory work and fieldwork (Preston 1994, 2004, 2006; Sonnert and Holton 1995; Stephan and Levin 2005; Hall 2007; Hewlett et al. 2008; Fouad, Singh, Fitzpatrick, and Liu 2012).1Frehill (2007, 2008, 2009, 2012) deemphasized family and highlighted instead changes in career interests. Women’s difficulties were established either using interviews or tabulations of responses to surveys. With the exception of Morgan (2000), however, all the research I am aware of sampled current or former scientists and engineers only, without any comparison to individuals in other skilled occupations.2 Possibly educated women in general simply churn more than men in search of the optimal job, and possibly science and engineering cannot be distinguished from other fields in terms of the gender gap in exit rates. If this is the case, studying exit rates to understand women’s underrepresentation in science and engineering may not be fruitful. If I establish that science and engineering stand out from other fields in terms of excess female exits, further comparisons with other fields are important to understand the reasons for the excess exits and the link with women’s underrepresentation in science and engineering. A common explanation in the literature for women’s quitting science and engineering has been that the work hours are long and, hence, difficult to combine with family. Yet jobs in many other fields have long hours, and educated women may leave all fields at a higher rate than men as they search for a job with optimal work hours. If so, women’s underrepresentation in engineering is unlikely to be attributable to long work hours. In this article, I use the 2003 and 2010 National Survey of College Graduates (NSCG) to investigate whether and why women whose highest degree is in science or engineering cease doing science and engineering–related work at higher rates than do similarly trained men. Unlike earlier authors, I contrast science and engineering, and use non-science and non-engineering fields as a comparison group to establish whether issues identified in the literature are specific to science and engineering. I perform the first quantitative test of whether gender differences in science and engineering exits reflect merely the degree to which the field is male dominated by comparing science and engineering with male-dominated economics and financial management and by controlling for the male share in the field of study. I probe further by exploring what job characteristics the male share may be proxying for. I use more recent data than existing studies and add to the relatively small literature examining the question in a regression context. The results shed light on which interventions are most likely to increase female representation in science and engineering. Data and Descriptive Statistics I use the 2003 and 2010 waves of the NSCG, data collected under the auspices of the National Science Foundation (NSF).3 The 2003 survey is a stratified random sample of respondents to the 2000 U.S. Census long form who reported having a bachelor’s degree or higher; the sampling frame for the similar 2010 survey is the 2008 and 2009 American Community Surveys. I define a worker as having left a field if he or she stated that his or her current work is not related to the field of study of his or her highest degree. The surveys asked, “Thinking about the relationship between your work and your education, to what extent was your work on your principal job held during the week of [. . .] related to your HIGHEST degree field? Was it . . . closely related/somewhat related/not related?” If the respondent answered “Not related,” he or she was asked, “Did these factors influence your decision to work in an area outside of your HIGHEST degree field?” and given a list of possible factors to check: family-related reasons; working conditions; pay, promotion opportunities; job in highest degree field not available; change in career or professional interests; job location; and other reason. The respondent was then asked, “Which factor in [the list] represents your MOST important reason for working in an area outside of your HIGHEST degree field?” I more generally define an individual as having left a field if he or she was either working in an unrelated field or not working. In addition to conventional variables on salary and demographics, I also take advantage of responses to the question asked of all workers: “When thinking about a job, how important is each of the following factors to you . . . ?” followed by a list of job attributes: salary, benefits, job security, job location, opportunities for advancement, intellectual challenge, level of responsibility, degree of independence, and contribution to society. Workers indicated whether each factor was very important, somewhat important, somewhat unimportant, or not important at all. I define the science and engineering fields of study as those coded as such in the data by NSF, with the important exception of social science, which I classify as non-science and -engineering. NSF’s classification excludes the health fields as well as those considered technician preparation, including computer programming and engineering technology (the latter field is tiny). When distinguishing between science and engineering, I include computer science with engineering and include mathematics with science. The advantage of the data is that they allow the identification of who left their field and why, for all fields. The disadvantage is that the date of the exit is not known, which precludes a hazard rate analysis of the type carried out by Preston (1994). Preston (1994, 2004, 2006) used the 1980s and 1990s longitudinal files associated with the NSCG (in addition to her own survey), but these cover only individuals working in science and engineering occupations or who had a degree in science and engineering. The cross-sectional nature of my data also imposes the use of field of study of highest degree to determine the initial field. Unlike in occupation-based samples, exits thus include exits after the completion of studies and before the first job. I discuss the relation between the two types of exit in the Appendix. I use two main samples in my analysis, from which I exclude respondents 65 and older. The main sample used to analyze working in the field of study consists of all workers, except those working part-time because they are students, a sample of 137,206 observations. The second sample, used to analyze the probability of not working, or the probability of not working or doing unrelated work, consists of all respondents except those who said they were working part-time or not working because they were students, yielding a sample of 156,395 observations. The Appendix describes the construction of the samples in more detail. In the sample of all respondents, the weighted shares of workers who were employed are 93% for males and 81% for females; 8.9% of men and 6.0% of women had a highest degree in science, and 15.5% of men and only 3.5% of women had a highest degree in engineering. I also consider the fields of economics and financial management; 5.1% of male respondents and 2.0% of female respondents had a highest degree in one of these two fields of study. Turning to the sample of workers, as shown in Table 1, panel A, about 20% of both men and women reported that the work in their current job was unrelated to the field of study of their highest degree. Men and women differed in their distribution between the other two categories (closely related and somewhat related), with more women than men saying their work was closely related to their field of study. For those reporting that their work was unrelated to their field of study, I show in Table 1, panel B, the distribution of the main reason given for this. The main gender differences are the higher share of women who were working in a different field for family-related reasons (3.8% of women compared to only 1.2% of men) and the smaller share of women who were working in a different field because of pay and promotion opportunities (4.4% of women compared to 6.6% of men). Women were somewhat more likely than men to be working in a different field because of dissatisfaction with working conditions. The second part of Table 1, panel B, shows a similar pattern for the shares of workers in a different field who cited the seven factors as either the main reason or a lesser reason for this. Table A.1 (in the Appendix) shows the same distribution separately for science, engineering, and economics and finance. Table Table 1. Relation between Job and Highest Degree (%) Table 1. Relation between Job and Highest Degree (%) View larger version In Table 2, I show the occupations of those men and women whose highest degree is in science and engineering but who responded that they were working in a job unrelated to their highest degree field. Of these respondents, 10.3% of men and 6.1% of women were working as computer or information scientists, 4.1% of men and 1.5% of women were working in another science or engineering field, and 2.8% of men and 1.7% of women were working in technical occupations that the NSF considers to be below the level of science or engineering. Table Table 2. Occupations of Workers with Highest Degrees in Science and Engineering Doing Work Unrelated to Highest Degree (%) Table 2. Occupations of Workers with Highest Degrees in Science and Engineering Doing Work Unrelated to Highest Degree (%) View larger version Also in Table 2, I identify other occupations in which 5% or more of this sample of exiters was working. Management occupations not considered by NSF to be related to science and engineering and management-related occupations, such as accountant, are common destination occupations, as are sales and marketing occupations.4 Fully 18.8% of women were working as a secretary or receptionist or in another administrative occupation, and 6.4% of men worked in transportation (presumably as taxi drivers, although the occupations are not so finely distinguished).5 More detail on the field of study by gender and on the exit rates from fields of study is given in Table A.2; Tables A.3 and A.4 show the means of the other covariates used in the regressions. Do Science and Engineering Have Excess Female Exits? I begin by establishing whether the fields of science and engineering have excess female exits compared to other fields of study for workers with a college education or higher. Figures 1–3 present graphical evidence on the rates at which men and women leave science and engineering compared to other fields. Figure 1, based on all respondents, displays the male and female exit rates from science, engineering, and non-science and non-engineering, including non-employment as an exit, by years since earning the highest degree (which I aggregate to smooth the graph). The gap between men and women is small to start with but then expands for 20 years, with the gap largest for engineering. figure Figure 1. Exit Rates, Including Non-employment Rates, from Fields of Study of Highest Degree by Gender Note: The exit rate from a field is the share of respondents reporting that their work was unrelated to the field of study of highest degree or that they were not working. figure Figure 2. Exit Rates from Fields of Study of Highest Degree by Gender Note: The exit rate from a field is the share of working respondents reporting their work was unrelated to the field of study of highest degree. figure Figure 3. Non-employment Rates by Gender The next two figures split this broad exit rate into its components of exits to other fields and exits to non-employment. Figure 2, based on all workers, shows that in terms of exits to other fields a gender differential exists only for engineering graduates; women with a highest degree in engineering have slightly higher exit rates than men soon after earning the highest degree, with some further divergence between 10 and 20 years after earning the highest degree.6Figure 3, based on all respondents (employed and non-employed), shows that women are less likely to be employed than men in all three fields and that this drives the patterns in Figure 1. To assess excess female exits more quantitatively, I turn to difference-in-differences analysis. Here I compare science and engineering not only to all other fields of study but also to economics and financial management. Some previous authors proposed that women leave science and engineering more than men because their male colleagues create an atmosphere of competition and risk-taking and a macho culture that women do not like (Preston 1994, 2004, 2006; Sonnert and Holton 1995; Xie and Shauman 2003; Stephan and Levin 2005; Hall 2007; Hewlett et al. 2008; Rosser and Taylor 2009; Fouad, Singh, Fitzpatrick, and Liu 2012). If these are features of male-dominated occupations, they should also be features of economics and finance, which are each 73% male, compared to 60% male for science (and 68% male if biological sciences are excluded) and 83% male for engineering. These occupations are also similar in requiring mathematical skills. Economics and finance are also very different, however, in that they do not deal with the natural world or make use of equipment (other than computers) or laboratories. The comparison could thus shed light on the reason for any excess female exits from science and engineering. Economics and finance are the only sizable fields, other than religion, that are more than 70% male and unrelated to science and engineering (except for the use of mathematics); other large, male-dominated fields of study, such as architecture (72% male) and medicine (68% male), also deal with the natural world and often make use of equipment and laboratories. In Table 3, I use the sample of all respondents to examine the shares by field of study of individuals who either worked in an unrelated field or who were not employed; I consider science, engineering, all non-science and non-engineering, and economics and finance. In all these fields, women are more likely to be working outside the field of study of their highest degree or not working at all. What is potentially interesting is how the gender gap differs across the fields. Science has the lowest excess female exits, at 6.5 percentage points, whereas economics and finance have the highest excess female exits, at 13.3 percentage points. The difference-in-differences section of Table 3 shows that the gender gap for science is not significantly smaller than for non-science and non-engineering (1.2 percentage points lower) but that it is a statistically significantly 6.8 percentage points lower than for economics and finance. More consistent with conventional wisdom, engineering has a statistically significantly higher gender gap than non-science and non-engineering (5.2 percentage points higher), although its gap is indistinguishable from that of economics and finance (only 0.4 percentage point lower). Table Table 3. Individuals in Job Unrelated to Highest Degree or Not Employed, by Gender and Field of Highest Degree (%) Table 3. Individuals in Job Unrelated to Highest Degree or Not Employed, by Gender and Field of Highest Degree (%) View larger version In Tables 4 and 5, I split this broad definition of exits into its components. In Table 4, using the sample of workers, I consider the share of workers whose job was not related to their field of study of highest degree. Engineers are more likely to work in their field of study than any of the other groups—only 10.1% of men trained as engineers were doing unrelated work, and the exit rate for women is also low, at 15.5%; however, consistent with conventional wisdom, engineering has the highest excess female exits, at 5.4 percentage points. The next largest gender gap is for economics and finance (3.7 percentage points), whereas neither science nor non-science and non-engineering has excess female exits. The lack of excess exits from science is not necessarily at odds with the existing literature, which typically does not distinguish between science and engineering. Table Table 4. Workers in Job Unrelated to Highest Degree, by Gender and Field of Highest Degree (%) Table 4. Workers in Job Unrelated to Highest Degree, by Gender and Field of Highest Degree (%) View larger version Table Table 5. Non-employment Rate by Gender and Field of Highest Degree (%) Table 5. Non-employment Rate by Gender and Field of Highest Degree (%) View larger version As shown in the difference-in-differences part of Table 3, science has a gender gap in exit rates similar to that of non-science and non-engineering (0.5 percentage point lower), whereas science’s lower female excess exit rate compared to that of economics and finance (4.8 percentage points lower) is significant at the 10% level. In addition, compared to non-science and non-engineering, engineering has a sizable and statistically significant (6.0 percentage points higher) excess female exit rate, but no statistically significant difference exists between engineering and economics and finance. In Table 5, I consider exits to non-employment, using the sample of all respondents. As expected, the first three rows indicate that, for all field-of-study groups, the share of non-employed women is higher than the share of non-employed men. The gender gap ranges in size from 9.4 percentage points for engineering to 12.7 percentage points for economics and finance. Nevertheless, the difference-in-differences part of Table 5 indicates that these differences are not statistically distinguishable. The results for the probability of working part-time rather than full-time (not shown) are similar; the gender gap in science is the same as in the non-science and non-engineering fields, whereas the gender gap in engineering relative to the non-science and non-engineering fields indicates that female engineers are relatively unlikely to work part-time (the difference-in-differences is 2.9 percentage points). Compared to economics and finance, women in both science and engineering are relatively unlikely to work part-time, although the standard errors are high. These results and those in Tables 3 to 5 show that excess female exits from science and engineering are confined to exits from engineering to unrelated full-time jobs; these excess female exits are apparent only in comparison with all non-science and non-engineering fields and not in comparison with economics and finance. The last result is consistent with the hypothesis that, rather than posing unique problems for women, engineering careers may reflect problems that grow as the male share of the field increases. Figure 4 gives a sense of the potential importance of the male share in the field of study of the highest degree. For each of the 143 detailed fields of study, I plot the excess female exit rate against the share of males. The relation is strongly positive, and the R2 of the regression line, weighted by the number of workers trained in the field, is fully 0.27. The slope of 0.18 indicates that an increase of 10 percentage points in the male share increases the gender gap in the exit rate by 1.8 percentage points.7 The positive relation holds within both science and non-science and non-engineering, but not within engineering. Few non-science and non-engineering fields are as male-dominated as the engineering fields, and most of the exceptions are the small technology fields. figure Figure 4. Excess Female Exits from Fields of Study by Share of Men in Field Notes: The exit rate from a field is the share of working respondents reporting their work was unrelated to the field of study of highest degree. The regression line is weighted by the number of workers trained in the field. Method The starting point for explaining excess female exits is the regression equivalent of the difference-in-differences performed previously, with controls added to determine whether individual characteristics can explain the patterns I have found. I then exploit respondents’ own explanations for no longer using what they learned while earning their highest degree, and I consider the role of the share of males in the degree field of study. I estimate a basic specification using linear probability regressions, weighted with survey weights, for the pooled years 2003 and 2010: Yit=β0+β1Sit+β2Eit+β3Fit+β4Sit×Fit+β5Eit×Fit+β6Xit+β7Fit×γt+γt+ϵit, where i indexes individuals and t indexes the year, S is a dummy for science as the field of study of highest degree, E is a dummy for engineering as the field of study of highest degree, F is a dummy for female, and Y is a dummy for an exit from science and engineering or from employment. Excess female exits from science and engineering would be reflected in positive values for β4 and β5, respectively. In a more general specification, I replace the main effects for science and engineering with a more detailed set of field-of-study dummies (35 for the sample that includes all fields of study; 16 for the sample in which only economics and finance are compared to science and engineering) while leaving the interaction terms the same. I allow the coefficient on the female dummy to vary by year, and I calculate robust standard errors. The first outcome Y of interest is a dummy for either working in a job unrelated to the field of study of the highest degree or not being employed, using the sample of all respondents (or respondents with specific highest-degree fields). The second outcome of interest is a dummy for the current job being unrelated to the field of study of the highest degree, using the sample of workers. The third outcome of interest is a dummy for not being employed, estimated using the sample of all respondents. Using the sample of workers, I also examine the probability of the current job being unrelated to the field of study for a specific reason. For example, I estimate a linear probability model for the probability of a worker who had left his or her field and cited family as the main reason, and a second model for the probability of a worker who had left his or her field and cited family as either the main reason or a lesser reason. For the main reason, only one of the seven possible reasons can be given; for ease of interpretation, I do not use a multinomial or nested logit to examine the choice. The covariates X comprise dummies for a master’s degree (including MBA), doctoral degree, or professional degree; five dummies for years since earning the highest degree; six dummies for age; dummies for black, Hispanic, and Asian; and dummies for being foreign born. I also control for three dummies each for the importance the respondent attached to the nine job attributes. In this way, I can control for the initial selection into different fields (although ideally the job preferences would be measured before the respondent graduated)—for example, engineering may attract women who care more about pay and promotion than do other women. I do not control for fertility (or marriage) in these regressions because the fertility choice is made jointly with the decision about whether to remain in the field of study (or to remain employed). Nevertheless, the correlation between fertility and remaining in the field of study may be informative, so I examine it in additional regressions. Because I do not know the timing of respondents’ leaving the field of study, ideally I would use information on lifetime fertility to date. Unfortunately, I know only the number and ages of children of the respondents living in the household at the time of the survey, so I proxy lifetime fertility with a dummy Cit for whether any child of the respondent (of any age) was living in the household. I explore gender and field differentials in the relation between fertility and how closely related the respondent’s job was to his or her field of study by adding Cit, Cit × Fit, Cit × Sit, Cit × Eit, Cit × Sit × Fit, and Cit × Eit × Fit to Equation (1). A positive coefficient for the triple interactions in a regression for working in an unrelated job would suggest that women, compared to men, have more difficulty combining work and children in science and engineering than in other fields. I initially compare science and engineering to all non-science and non-engineering, a set of fields of study as disparate as business, teaching, and technology (i.e., technical training below the level of engineering). Although science and engineering may seem naturally distinct from other fields because of their mathematical nature and the use of equipment and laboratories, possibly what in fact distinguishes them is their high share of male workers. Female exits may increase relative to male exits as the share of male workers increases, and any apparent specificities in science and engineering may simply reflect this. One way to test this is to choose a comparison group that is heavily male. I therefore also present the results for a sample restricted to those whose highest degree is in science, engineering, economics, or financial management. Another way to test this is to allow for excess female exits from male-dominated fields by adding to Equation (1) the controls mj and mj × Fijt, where mj is the share of men in field of study j. For these regressions, I use the most detailed field-of-study categories (143 categories). If β4 and β5 were initially statistically significantly positive and they change little with the addition of these covariates, the share of males in a field of study is not relevant for gender differences between science and engineering and other fields. In contrast, if β4 and β5 become statistically insignificant, science and engineering exit rates merely reflect their male-dominated workforce.8 For these regressions, I cluster the standard errors by detailed field of study.9 Results Why Do Women Leave Science and Engineering? I begin the regression analysis by using the broad definition of exit from a field of study, which includes exits to non-employment in addition to exits to an unrelated field, and estimating Equation (1) with this as the dependent variable. Table 6, column (1), presents a specification almost equivalent to the simple difference-in-differences using all respondents in the Table 3, except that a year dummy and its interaction with Female are also controlled for. The coefficients on the two interaction terms represent the difference-in-differences effects, which are close to those in Table 3. In Table 6, column (2), I replace the dummies for science and engineering with the 35 field of study dummies, which increases the excess female exits from engineering slightly to 6.5 percentage points. Controlling for worker characteristics in column (3) increases this gap slightly, but controlling for what workers value in a job (job tastes) in column (4) reduces it slightly to 6.5 percentage points. The column (4) coefficient on the science interaction, meanwhile, indicates that excess female exits from science are 2 percentage points lower than exits from non-science and non-engineering fields; however, this is not statistically significant. Table Table 6. Effect of Field of Study and Gender on Having a Job Unrelated to Highest Degree or Being Non-employed Table 6. Effect of Field of Study and Gender on Having a Job Unrelated to Highest Degree or Being Non-employed View larger version In Table 6, column (5), I restrict the non-science and non-engineering fields to economics and finance only, and I use the specification of column (4). Whereas the excess female exit rate from engineering is a statistically insignificant 1.2 percentage points higher than from economics and finance, the excess female exit rate from science is a large 7.1 percentage points lower than from economics and finance. No excess female exits from science and engineering are apparent relative to economics and finance. Although I focus here on the differences in point estimates between column (5) and the rest of Table 6, the standard errors are sufficiently large in column (5) that the differences are not statistically significant. The patterns in Table 7, using the samples of workers only, are similar, indicating that exits to other jobs are driving the results in Table 6, not exits to non-employment. Compared to all non-science and non-engineering fields of study, excess female exits from engineering are indeed apparent (6.4 percentage points higher; Table 7, column (4)) that are not explained by a distinctive pattern in education, experience, demographics, or job tastes. The much narrower (and statistically insignificant) gap between engineering and economics and finance, however, hints that the excess engineering exits could be related to the male domination of the field rather than to field-specific characteristics, such as working with equipment. No excess exits from science are apparent, although this could reflect the opposing effects of unmeasured factors, such as more difficult working conditions in science for women than men and greater science interest and aptitude of women than men. Table Table 7. Effect of Field of Study and Gender on Having a Job Unrelated to Highest Degree Table 7. Effect of Field of Study and Gender on Having a Job Unrelated to Highest Degree View larger version Even though Table 5 indicates that differences in the female excess exits to non-employment are not statistically significant across fields, I probe this further in Table 8, estimating Equation (1) using non-employment as the dependent variable. The specifications in Table 8 are the same as those in Tables 6 and 7, and, as in Table 5, none of the coefficients is statistically significant. The point estimates in Table 8, columns (1) to (4) are very small; the point estimates in column (5), the comparison with economics and finance, indicate that both science and engineering have smaller excess female exits by 2.4 to 2.7 percentage points. Table Table 8. Effect of Field of Study and Gender on Probability of Non-employment Table 8. Effect of Field of Study and Gender on Probability of Non-employment View larger version Possibly, the lack of differential employment patterns by gender and field masks informative differences in the reasons for non-employment; however, unreported results indicate this is not the case. I have repeated the regressions (with respondents from all fields of study) seven times, each time using as the dependent variable the probability of not working for one of seven possible reasons that respondents could give (see Hunt 2010). I have found that the excess female exits to retirement are smaller in engineering than in non-science and non-engineering, and that the excess female exits for “Other reason” are smaller in science than in non-science and non-engineering (significant at the 10% level). Otherwise, I found no statistically significant coefficients on the interaction terms. I have also estimated the regressions for workers, with full-time work as the outcome, to determine whether women in science and engineering are pushed to work part-time, thereby exploiting their science and engineering human capital only incompletely. The results (not shown) do not support this hypothesis. I conclude that excess female exits from science and engineering are present only in exits to another full-time job. I pursue the analysis of the causes of women leaving science and engineering for a job in another field using the reasons given by respondents whose job was not related to their field of study and the sample with all fields of study. In Table 9, panel A, I consider the probability of leaving the field of study for each of the possible main reasons, after controlling for the full set of covariates (as in Table 7, column (4)). Table 9, column (1) shows that no statistically significant family-related excess female exits from science or engineering are present. Column (2) shows that pay and promotion opportunities play an important role in excess female exits from engineering; the coefficient on the engineering interaction term indicates an effect of 3.2 percentage points, accounting for 51% of the total conditional excess exits of 6.4 percentage points (in Table 7, column (4)). No excess exits of women from science are present because of pay and promotion opportunities. Table Table 9. Effect of Field of Study and Gender on Job Unrelated to Highest Degree for Various Reasons Table 9. Effect of Field of Study and Gender on Job Unrelated to Highest Degree for Various Reasons View larger version Table 9, columns (3) to (7), suggest that working conditions, the unavailability of a job in the field, changes in career interests, and job location play no role in excess female exits from either science or engineering; changes in career interests and other reasons have coefficients statistically significant at the 10% level, although small, for engineering. I conclude from Table 9, panel A, that the most important reason that women leave engineering at higher rates than men, relative to other fields of study, is pay and promotion opportunities. In Table 9, panel B, I search for more minor causes by estimating the probability of each of the seven reasons being mentioned at all, whether it was the most important reason or not. For engineering, the difference-in-differences effect is positive and statistically significant for every reason except family-related reasons. This points to women having many reasons for leaving engineering at higher rates than men; however, the conclusion based on panel A, that the most important reason is pay and promotion, is reinforced by the largest coefficient in Panel B being for pay and promotion (4.6 percentage points, column (2)). The next largest coefficients are for changes in interests and job location (3.2 and 3.0 percentage points, respectively). For science, the only statistically significant coefficients indicate that the gender gaps in exits because no job was available in the field and because of a change in interests (significant at the 10% level) are smaller than for non-science and non-engineering by 1.7 and 1.9 percentage points, respectively. Considering the coefficient on the female dummy in the two panels of Table 9 is also interesting. In non-science and non-engineering occupations, women are more likely than men to give family-related reasons as the main reason (by 1.9 percentage points, Table 9, panel A) or any reason (by 3.9 percentage points, panel B) for having a job unrelated to field of study. This remains true qualitatively for women whose field of study is science or engineering in that the interaction terms are of the same sign as the female dummy. In contrast, women in non-science and non-engineering occupations are less likely than men to give pay and promotion opportunities as the main reason (by 2.2 percentage points, panel A) or any reason (by 2.6 percentage points, panel B) for having a job unrelated to the field of study. Based on the interaction term, women whose field of study is science are qualitatively similar. In contrast, the interaction term for engineering is sufficiently positive in both panels of Table 9 that we can conclude that women engineers are more likely than men generally to leave their field of study for pay and promotion opportunities. Table 9, panel B, shows that women were more likely than men to give working conditions or job location as a reason for no longer working in the field of study. I approach the importance of family from another angle, using information on the presence of children in the household, despite the likelihood that having children and leaving a field of study are often decided jointly. In Table 10, column (1), I reproduce the results of Table 7, column (4), for the probability of having a job unrelated to field of study. In column (2), I add a control for having a child in the household; as we can see, having a child is associated with a probability 1.8 percentage points lower of having a job unrelated to field of study. In column (3), I add the double interactions of having a child with female, science, and engineering; the coefficients on the interaction terms (not shown) are statistically insignificant, and the addition of the child covariates does not change the difference-in-differences coefficients compared to column (1). In column (4), I control for the triple interactions of female and child with science and engineering. The coefficients are small and statistically insignificant, and the difference-in-differences coefficients are essentially unchanged. Children do not appear to be relevant for excess exits of women from engineering. Table Table 10. Effect of Children on Having a Job Unrelated to the Highest Degree Table 10. Effect of Children on Having a Job Unrelated to the Highest Degree View larger version It is possible that women reported dissatisfaction with pay and promotion opportunities in engineering because they are more likely to be sidelined on returning from a career break than in other fields because of the rapid advancement of technology. Preston (2004) showed that among former scientists and engineers who re-entered science and engineering, those re-entering the fastest-evolving fields had the lowest wages compared to those who never left science and engineering. If this explanation were correct, pay and promotion dissatisfaction should be equally salient in science; however, this is not the case. The absence of information on actual experience in my samples precludes an analysis of the effect of career breaks. In Hunt (2010), I used an equivalent 1993 sample and found that controls for a career break (including its interaction with dummies for science and engineering) do not affect the results. Do Science and Engineering Differ because of Their High Share of Men? I now investigate whether excess female exits from engineering might merely be a manifestation of the effect on women of being in any male-dominated field; I do this by controlling directly for the share of the field of study that is male and its interaction with a female dummy. The baseline in Table 11, column (1), is the same as the specification in Table 7, column (4), except that the field-of-study controls are just two dummies for science and engineering. We can see an excess female exit rate of 5.9 percentage points for engineering. In Table 11, column (2), I add the controls for the male share; this causes the excess exit rates for engineering to flip from positive to negative and statistically significant; engineering loses 4.4 percentage points fewer women than we would expect given how male it is. Science loses 5.1 percentage points fewer women. Men are much less likely to leave a male-dominated field of study (an increase of 10 percentage points in the male share reduces the male exit rate by 1.7 percentage points), and women’s exit rates are positively affected by the male share (significant at the 10% level, Table 11, bottom row), which means that excess female exits from male-dominated fields exist generally. Once the share of males in a field of study is appropriately controlled for, the relative female exit rates from science and engineering look favorable compared to other fields. Table Table 11. Effect of Male Share in Field of Study on Having a Job Unrelated to Highest Degree Table 11. Effect of Male Share in Field of Study on Having a Job Unrelated to Highest Degree View larger version I next investigate whether the share of male workers in a field may be proxying for other underlying characteristics of jobs in the field (a full treatment is beyond the scope of the article because of the limitations of the cross-section data, and I do not show the results here). The addition of some characteristics of people who stay (stayers) in fields of study that are positively correlated with the share of males renders the coefficient on the main effect of the male share small and statistically insignificant, but it leaves the coefficient on the interaction term with gender unchanged. This leaves the sum of the two coefficients—the total effect for women—twice as large and strongly statistically significant. The characteristics that have this effect are average hours in the field, hierarchy in the career path (as captured by the share of stayers who are supervisors and who are supervisors of supervisors), and the share in the field who answered “Very important” to each of the job attributes. Characteristics that have little effect on the coefficients of the male share are share in the field working more than 45 or 50 hours per week and the average wages in the field. For none of these field characteristics does the interaction with gender have a statistically significant coefficient. In Table 11, columns (3) to (6), I relate the male share covariates to the main reason given for leaving the field. Columns (3) and (4) show that the male share covariates have only small effects on the difference-in-differences coefficients for leaving the field for family-related reasons (as the main reason). In contrast, in columns (5) and (6), we can see that because of the large effects of the male share covariates on leaving a field for pay and promotion opportunities, their inclusion fully explains the 2.6 percentage point excess exit rate from engineering for this reason. Women’s concerns about pay and promotion are, therefore, not an engineering-specific issue but, instead, an issue general to male-dominated fields. No excess female exit rate is present for the unreported reasons for doing unrelated work, although the coefficients of the male share are often statistically significant (Hunt 2010 provided these results for the older data). The results suggest that we should not look for explanations connected with the nature of scientific and engineering work (e.g., the use of labs and equipment and the study of natural phenomena) but, instead, for explanations of the female retention difficulties that become more severe as the share of men in the workforce increases and that affect women’s pay and promotion. Conclusion I have demonstrated that the exit rate for women compared to men is indeed higher from engineering than from other fields, resulting from excess female exits to jobs in another field, but that no similar pattern exists for science. Neither worker characteristics nor worker preferences about job attributes, including salary and opportunities for advancement, contribute to explaining the excess female exits from engineering, and I find no differential impact of having children for women trained as engineers. Furthermore, I have shown that the problems are not the family-related ones emphasized in most of the previous literature. Rather, I find that the most important driver of excess female exits from engineering is dissatisfaction over pay and promotion opportunities, a factor explaining half of the differential gender gap in exit rates. Family-related constraints are not a factor; although many more women than men cited family issues as the reason for leaving engineering, the gender gap is just as large in non-science and non-engineering fields. I find that working conditions, the unavailability of a job in the field, changes in professional interests, and job location play statistically significant but secondary roles. The results appear to point to problems for women that are specific to the engineering profession. Nevertheless, I show that the excess exits of women trained as engineers, as well as their excess exits because of pay and promotion opportunities, are no larger than would be expected given the share of men in the field; compared to men, women have relatively higher total exit rates and pay- and promotion-motivated exit rates from male-dominated fields of study. This result is robust to controls for the field’s working hours, wages, and share of workers in management, all of which are positively correlated with the male share. This is consistent with my finding that excess female exits from engineering are much smaller (and statistically insignificant) compared to the male-dominated economics and financial management fields than compared to non-science and non-engineering fields generally. The implication is that a lack of mentoring and networks, or discrimination by managers and coworkers, are the more promising of the existing explanations for excess female exits, and that explanations hinging on the precise nature of engineering work, such as use of equipment, labs, and field work, and consideration of natural phenomena, should be discarded. Appendix I use the 2003 and 2010 waves of the NSCG, data collected under the auspices of the NSF. The data may be downloaded from http://sestat.nsf.gov/datadownload/. The 2003 survey is a stratified random sample of respondents to the 2000 Census long form who reported having a bachelor’s degree or higher; the sampling frame of the similar 2010 survey is the 2008 and 2009 American Community Survey. I drop respondents who lived outside the United States or in U.S. territories and those who were ages 65 and older. I include in all the samples those who were self-employed in their principal job. I exclude from the analysis of the probability of employment those who said they were either working part-time or not working because they were students, and I exclude the former group also from the analysis of working in the field of study of highest degree. I scale the weights from the 2003 and 2010 surveys so that the sum of weights is equal for each year. I define the science and engineering fields of study to be those coded as such by the NSF in the data, with the important exception of social science, which I classify as a non-science and non-engineering field. NSF’s classification excludes health fields as well as fields considered to be technician preparation, including computer programming and engineering technology (the latter field is tiny). When distinguishing between the fields of science and engineering, I include computer science with engineering and mathematics with science. Field of study is provided in 30 or 143 categories. To use 142 field dummies as controls may seem excessively detailed, but the more aggregate categorization is rather coarse for fields outside science, engineering, and social science. For these fields, I therefore examine the finer categorization for my sample of workers and select the six finer fields studied by more than 2% of workers. I use these and the 30 aggregate categories to create 36 field-of-study categories, listed in Table A.1. I could have followed the existing literature by basing exit from a field on occupation rather than the relation of the job to highest degree earned; if I had done this, however, I could not then associate a reason with the field change and I could not determine an obvious equivalent to leaving science and engineering for workers trained in other fields. Of the workers who had a highest degree in science and engineering and who said their current job was unrelated to their field of study, 80% and 77%, respectively, were not working in a science or engineering occupation. Appendix A Table Table A.1. Share of Workers Leaving Field of Highest Degree, by Field and Gender (%) Table A.1. Share of Workers Leaving Field of Highest Degree, by Field and Gender (%) View larger version Table Table A.2. Fields of Study of Highest Degree, by Gender (%) Table A.2. Fields of Study of Highest Degree, by Gender (%) View larger version Table Table A.3. Means of Covariates by Gender Table A.3. Means of Covariates by Gender View larger version Table Table A.4. Workers Attaching Very High Importance to Particular Job Attributes (%) Table A.4. Workers Attaching Very High Importance to Particular Job Attributes (%) View larger version Acknowledgements I thank Leah Brooks, Daniel Parent, and participants in seminars at Bocconi University, the Canadian Labour Market and Skills Researcher Network (CLSRN) 2010 annual conference, Hunter College (CUNY), McGill University, the National Bureau of Economic Research (NBER), University of British Columbia (UBC), and the University of Milan for comments. I am grateful to David Munroe and Marjolaine Gauthier-Loiselle for research assistance and to the Social Science and Humanities Research Council of Canada for financial support. This article was written while I was a visiting professor at UBC. I am also affiliated with the Centre for Economic Policy Research (CEPR; London) and the German Institute for Economic Research (DIW; Berlin). A data appendix with additional results, and copies of the computer programs used to generate the results presented in the article, are available from the author at jennifer.hunt@rutgers.edu. 1 See also Xie and Shauman (2003); Rosser and Taylor (2009). 2 My analysis goes beyond that of Morgan (2000), who used the 1993 National Survey of College Graduates (NSCG) to compare the exits from full-time employment of women from different fields. 3 In an earlier version of this article (Hunt 2010), I used the 1993 and 2003 surveys. The results using newer data are very similar. 4 Only a tiny share of the exiters sample worked in what NSF classifies as science and engineering management. The NSF may use the answers to the job relatedness to categorize the occupations. 5 5Immigrants are overrepresented in both these groups, but nevertheless they represent only one-quarter of each. 6 A comparison with the equivalent graph in Hunt (2010), which used 1993 and 2003 data, suggests cohort effects are important. The change to more recent data shifts the points of divergence to the right for the three fields of study, causing them to disappear in the cases of science and non-science and non-engineering. 7 The share of males is not directly affected by exits because it refers to a time before any exits occurred. 8 As can be seen in Figure 4, the relationship is fairly linear, and a quadratic in the male share does not improve the fit. 9 The share male mj is not directly affected by exits because it refers to a time before any exits occurred. Omitted variables may be correlated with both the pre-labor-market entry rates (and hence the share of males) and exit rates, and I explore which variables the share of males could be proxying for. References Antecol, Heather, Cobb-Clark, Deborah. 2013. Do psychosocial traits help explain gender segregation in young people’s occupations? Labour Economics 21: 59–73. Google Scholar, Crossref, ISI Fouad, Nadya A., Singh, Romila, Fitzpatrick, Mary E., Liu, Jane P. 2012. Stemming the tide: Why women leave engineering. Center for the Study of the Workplace Report, University of Wisconsin–Milwaukee. Accessed at http://studyofwork.com/files/2011/03/NSF_Women-Full-Report-0314.pdf (May 2015). Google Scholar Frehill, Lisa M. 2007. Are women more or less likely than men to be retained in engineering after college? Society of Women Engineers Magazine 53(4): 22–25. Google Scholar Frehill, Lisa M. 2008. Why do women leave the engineering work force? Society of Women Engineers Magazine 54(1): 24–26. Google Scholar Frehill, Lisa M. 2009. SWE retention study and work/life balance. Society of Women Engineers Magazine 55(4): 34–40. Google Scholar Frehill, Lisa M. 2012. Gender and career outcomes of U.S. engineers. International Journal of Gender, Science and Technology 4(2): 149–66. Google Scholar Hall, Linley Erin. 2007. Who’s Afraid of Marie Curie? The Challenges Facing Women in Science and Technology. Emeryville, CA: Seal Press. Google Scholar Hewlett, Sylvia Ann, Luce, Carolyn Buck, Servon, Lisa J., Sherbin, Laura, Shiller, Peggy, Sosnovich, Eytan, Sumberg, Karen. 2008. The Athena Factor: Reversing the brain drain in science, engineering, and technology. Harvard Business Review Research Report No. 10094. Cambridge, MA: Harvard Business Publishing. Google Scholar Hunt, Jennifer. 2010. Why do women leave science and engineering? NBER Working Paper No. 15853. Cambridge, MA: National Bureau of Economic Research. Google Scholar, Crossref Hunt, Jennifer. 2011. Which immigrants are most innovative and entrepreneurial? Distinctions by entry visa. Journal of Labor Economics 29(3): 417–57. Google Scholar, Crossref Hunt, Jennifer, Garant, Jean-Philippe, Herman, Hannah, Munroe, David J. 2013. Why are women underrepresented amongst patentees? Research Policy 42(4): 831–43. Google Scholar Morgan, Laurie A. 2000. Is engineering hostile to women? An analysis of data from the 1993 national survey of college graduates. American Sociological Review 65(2): 316–21. Google Scholar, Crossref Preston, Anne. E. 1994. Why have all the women gone? A study of exit of women from the science and engineering professions. American Economic Review 84: 1446–62. Google Scholar Preston, Anne. E. 2004. Leaving Science: Occupational Exit from Science Careers. New York: Russell Sage Foundation. Google Scholar Preston, Anne. E. 2006. Women leaving science. Working paper. Haverford College, Haverford, PA. Google Scholar Rosser, Sue V., Taylor, Mark Zachary. 2009. Why women leave science. Technology Review, January/February. Accessed at http://www.technologyreview.com/article/21859/ (May 2015). Google Scholar Sonnert, Gerhard, Holton, Gerald. 1995. Gender Differences in Science Careers. New Brunswick, NJ: Rutgers University Press. Google Scholar Stinebrickner, Todd, Stinebrickner, Ralph. 2011. Math or science? Using longitudinal expectations data to examine the process of choosing a college major. NBER Working Paper No. 16869. Cambridge, MA: National Bureau of Economic Research. Google Scholar, Crossref Stephan, Paula E., Levin, Sharon G. 2005. Leaving careers in it: Gender differences in retention. Journal of Technology Transfer 30: 383–96. Google Scholar, Crossref Xie, Yu, Shauman, Kimberlee A. 2003. Women in Science. Cambridge, MA: Harvard University Press. Google Scholar Zafar, Basit. 2013. College major choice and the gender gap. Journal of Human Resources 48(3): 545–95. Google Scholar, Crossref