Article Figures & Data
Figures
- Figure 1.
Career interest has solidified by end of PhD. Graphs show changes in career interests over time, split by current position. Mean responses of PhD neuroscientists in four different career paths who were asked to rate their level of interest in four different career paths at three time points: start of PhD, end of PhD, and current, on a 4-point scale (where 1 represents “no interest” and 4 represents “strong interest”). RM MANOVA showed that mean retrospectively reported interest in all four career types changed over time for participants in all career types except academic research-focused respondents’ interest in academic research (A) and academic teaching-focused respondents’ interest in nonacademic research (B; Extended Data Figs. 1-1–1-4; standard error around the mean is indicated by colored shading; **p < 0.01, ***p < 0.001; for M, SD, and effect size detail, see Extended Data Fig. 1-4).
- Figure 2.
Gender differences among responses to variables capturing experiences, personal characteristics, and objective measures. Significance levels from F statistic (ANOVA, reported in Extended Data Figs. 2-1–2-3) comparing the means for women and men for each variable were all significant at p < 0.05, at least. A, Categorical variables. Responses on the x-axis were the percent of respondents in each group who indicated either that the response was important to them or agreed with the response. B, Interval variables. Responses on the x-axis are indicated to the right of the graph: interest levels in a given career (1–4, 4 being very interested), T2 interest minus T1 interest for change in interest, Z-score for “structural aspects of academia” factor scores, total publications/years of research for publication rate, level of confidence (centered −3.05 to 0.95, 0.95 being the most confident) for confidence in being an independent researcher, and years since completed PhD. Effect sizes are labeled when they reach at least “small” size. S, small effect size; M, medium effect size; *p < 0.05; **p < 0.01; ***p < 0.001. Error bars indicate standard deviation.
- Figure 3.
Differences between WR and UR responses to variables capturing experiences, personal characteristics, and objective measures. Significance levels from F statistic (ANOVA, reported in Extended Data Figs. 2-1–2-3) comparing the means for WR and UR responses for each variable were all significant at p < 0.05, at least. A, Categorical variables. Responses on the x-axis were the percent of respondents in each group who indicated either that the response was important to them or agreed with the response. B, Interval variables. Responses on the x-axis were T2 interest minus T1 interest for change in interest, total publications/years of research for publication rate, times supported by NIH, and years (centered) it took to complete PhD. Effect sizes are labeled when they reach at least “small” size. S, small effect size, M, medium effect size; *p < 0.05; **p < 0.01; ***p < 0.001. Error bars indicate standard deviation.
- Figure 4.
Logistic regression predicting academic versus nonacademic positions. Standardized regression coefficients and error bars for logistic regression predicting whether respondents were in academic or nonacademic positions. The dependent variable was the binary indicator of whether respondents’ positions were academic. Independent variables included career interest and change in career interest, experiences during PhD training and postdoctoral training, personal characteristics, objective measures, and interactions with gender and representation status. Factors are indicated by (Fac). The entire equation was significant at p < 0.001 and accurately predicted 89% of respondents in the analysis.
- Figure 5.
Logistic regression predicting research-focused versus teaching-focused academic positions. Standardized regression coefficients and error bars for logistic regression predicting whether respondents were in research-focused academic positions or teaching-focused academic positions. The dependent variable was the binary indicator of whether respondents’ positions were research-focused. Independent variables included career interest and change in career interest, personal characteristics, objective measures, NIH support, and interactions with gender. The entire equation was significant at p < 0.001 and accurately predicted 90% of respondents in the analysis. Factors are indicated by (Fac). See Extended Data Figure 5-1 for the list of explanatory/independent variables.
- Figure 6.
Logistic regression predicting nonacademic research versus science-related, nonresearch positions. Standardized regression coefficients and error bars for logistic regression predicting whether respondents were in nonacademic research positions or scientific nonresearch positions. The dependent variable was the binary indicator of whether respondents were in nonacademic research positions. Independent variables included change in career interest and whether respondents’ career goals had changed. The entire equation was significant at p < 0.001 and accurately predicted 84% of respondents in the analysis. See Extended Data Figure 6-1 for the list of explanatory/independent variables.
Tables
Demographic variable Value n (%) Current position Research-focused academic 396 (51%) Teaching-focused academic 101 (13%) Nonacademic research 134 (17%) Science-related, nonresearch 150 (19%) Gender Woman 422 (54%) Man 359 (46%) Representation status WR 658 (84%) UR 123 (16%) Social identity WR man 298 (38%) WR woman 360 (46%) UR man 61 (8%) UR woman 62 (8%) Do you have a disability? No 758 (97%) Yes 23 (3%) First person/generation to graduate from 4-year college? No 602 (77%) Yes 179 (23%) PhD field (recoded) Neuroscience 349 (45%) Biological sciences 281 (36%) Other sciences/engineering 76 (10%) Other social sciences 75 (10%) Age range 20–24 19 (2%) 25–29 270 (35%) 30–34 377 (48%) 35–39 77 (10%) 40–44 21 (3%) 45–50 11 (1%) 51 or older 5 (1%) Prefer not to answer 1 (0%) Median years since PhD 6 Undergraduate institution in National Science Boarda top 50 research university? No 661 (85%) Yes 120 (15%) Master's degree in biomedical research discipline before PhD program? No 658 (84%) Yes 123 (16%) Doctoral institution in top 50 research university? No 383 (49%) Yes 398 (51%) Ever been a postdoctoral fellow? No 203 (26%) Yes 578 (74%) Career goal changed from research-based to outside of research? No, still is research-based 509 (65%) No, never was research-based 47 (6%) Yes, changed away from research 225 (29%) Primary way respondent found current position (recoded) Directly contacted by employer/recruiter 32 (4%) Former advisor/supervisor 79 (10%) Job posting 292 (37%) Previous position at same organization 103 (13%) Professional networking (other than advisor) 217 (28%) Other 58 (7%) Basic demographic information about the sample of 781 PhD neuroscientists who responded to the survey and labeled themselves as currently “professionals in the field.” Totals may not equal to 100% due to rounding.
↵a Top 50 research university is measured by research and development expenditures (National Science Board, 2016).
Current position Social identity Total UR women*** WR women** UR men WR men*** Research-focused academic 22 (35%)* 157 (44%)* 30 (49%) 187 (63%)*** 396 (51%) Teaching-focused academic 9 (14%)* 56 (16%) 8 (13%) 28 (9%) 101 (13%) Nonacademic research 6 (10%) 54 (15%) 17 (28%) 57 (19%) 134 (17%) Science-related, nonresearch 25 (40%)*** 93 (26%)** 6 (10%) 26 (9%)*** 150 (19%) Total 62 (100%) 360 (100%) 61 (100%) 298 (100%) 781 (100%) Proportion of “professionals in the field” respondents in each of the four “current positions” by gender and representation status. UR, underrepresented; WR, well represented. Significance level indicators (*p < 0.05, **p < 0.01, ***p < 0.001) in column headings show subset chi-square goodness-of-fit tests comparing the distribution of current positions for that social identity and the overall distribution of current positions. Significance level indicators in individual cells show Z-score tests for adjusted standardized residuals computed for that social identity group alone.
Independent variables Unstd. coeff. Adj. sig. Odds ratio Effect size Inverse odds ratio T1 career interest in research, academia 0.80 2.22 (S) T1 career interest in nonacademic research 0.30 1.35 - T1 career interest in science-related nonresearch −0.60 ** 0.55 (S) 1.82 (Fac) PhD belongingness department/social −0.22 0.80 - T1 minus T2 interest change in research, academia −0.31 * 0.74 - T1 minus T2 interest change in teaching, academia 0.39 1.47 - T1 minus T2 interest change in nonacademic research 0.36 1.44 - T1 minus T2 interest change in science nonresearch −0.60 ** 0.55 (S) 1.82 (Fac) Postdoc advisor relationship −0.15 0.86 - Postdoc support, faculty at primary institution 0.28 1.32 - Postdoc career advice, advisor −0.53 ** 0.59 (S) 1.69 Postdoc career advice, institution −0.42 * 0.66 (S) 1.52 Years it took to complete PhD? −0.13 0.88 - Years since completed PhD 0.09 1.09 - First-author publication rate 0.54 1.72 (S) Important aspects of careers: high autonomy 0.87 ** 2.38 (S) Important aspects of careers: work–life balance −0.46 0.63 (S) Important aspects of careers: job security 0.92 ** 2.51 (M) Important aspects of careers: monetary compensation −1.91 *** 0.15 (L) 6.67 Important aspects of careers: varied, diverse work −1.06 ** 0.35 (M) 2.86 (Fac) Like structural aspects of academia 1.00 *** 2.72 (M) (Fac) Like academic teaching/mentoring 0.41 * 1.50 (S) Career goal changed? No, still is research-based 1.15 3.17 (M) Career goal changed? Yes, changed from research −0.10 0.90 - Gender 0.40 1.50 (S) (Fac) Postdoc belongingness department/social 0.11 1.11 - PD support, faculty at primary institution 0.75 *** 2.12 (S) Gender BY (Fac) PD belongingness department/social −0.64 * 0.53 (S) 1.89 Gender BY PD support, faculty at primary institution −0.41 0.66 (S) Results of logistic regression predicting whether respondents were in academic or nonacademic positions. Full equation statistics: n = 781; accuracy = 89%; likelihood ratio test χ2 = 566.29; and p = 0.0000. (Fac), variable is a result of factor analysis; Unstd. coeff., unstandardized coefficient; Adj. sig., FDR-adjusted significance. Effect sizes are labeled when they reach at least “small” size. S, small effect size; M, medium effect size; L, large effect size.
↵* p < 0.05.
↵** p < 0.01.
↵*** p < 0.001.
- Table 4.
Logistic regression predicting research-focused versus teaching-focused academic positions
Independent variables Unstd. coeff. Adj. sig. Odds ratio Effect size Inverse odds ratio (Intercept) −0.11 0.90 - T1 career interest in research academic 1.23 *** 3.41 (M) T1 career interest in teaching academic −1.06 *** 0.35 (M) 2.86 T1 minus T2 interest change in research academia 0.68 * 1.97 (S) T1 minus T2 interest change in teaching academia −1.25 *** 0.29 (M) 3.45 Important aspects of careers: collaboration 0.68 1.97 (S) Important aspects of careers: job security −0.75 0.47 (S) (Fac) Like academic teaching/mentoring −1.13 *** 0.32 (M) 3.13 Career goal changed? No, still is research-based 1.78 *** 5.93 (L) Times supported by NIH, pre-Curr Pos −0.27 0.76 - Gender −0.88 0.41 (S) Aspects of careers: intellectually stimulating −0.47 0.62 (S) Gender: aspects of careers—intellectually stimulating 1.93 * 6.90 (L) Results of logistic regression predicting whether respondents were in research-focused academic positions or teaching-focused academic positions. Full equation statistics: n = 497; accuracy = 90%; Akaike information criterion (AIC) = 274.75; likelihood ratio test χ2 = 253.05; and p = 0.0000. (Fac), variable is a result of factor analysis; Unstd. coeff., unstandardized coefficient; Adj. sig., FDR-adjusted significance. Effect sizes are labeled when they reach at least “small” size. S, small effect size; M, medium effect size; L, large effect size.
↵* p < 0.05.
↵*** p < 0.001.
- Table 5.
Logistic regression predicting nonacademic research versus science-related, nonresearch positions
Independent variables Unstd. coeff. Adj. sig. Odds ratio Effect size Inverse odds ratio (Intercept) −1.66 *** 0.19 (L) 5.26 T1 minus T2 interest change in nonacademic research 0.89 *** 2.44 (S) Career goal changed? No, still is research-based 3.35 *** 28.37 (L) Results of logistic regression predicting whether respondents were in nonacademic research positions or science-related, nonresearch positions. Full equation statistics: n = 284; accuracy = 84%; AIC = 240.32; likelihood ratio test χ2 = 158.48; and p = 0.0000. Unstd. coeff., unstandardized coefficient; Adj. sig., FDR-adjusted significance. Effect sizes are labeled when they reach at least “small” size. S, small effect size; L, large effect size.
↵*** p < 0.001.
- Table 6.
Contingency table analysis of current position by how participants found current position
What was the primary way you found your current position? Current position Total (for row) Sig. of follow-up χ2 Research-focused academic Teaching-focused academic Nonacademic research Science-related, nonresearch n %Col n %Col n %Col n %Col n %Col Contacted by employer/recruiter 13 3% 1 1% 10 8% 8 5% 32 4% n.s. Former advisor/supervisor 611 15% 22 2% 92 7% 72 5% 79 10% *** Job posting 1301 33% 642 63% 451 34% 531 35% 292 37% *** Previous position at same organization 831 21% 52 5% 62 5% 92 6% 103 13% *** Professional networking (not advisor) 761 19% 283 28% 562 42% 572 38% 217 28% *** Other 33 8% 1 1% 8 6% 16 11% 58 7% Not performed Total (for column) 396 100% 101 100% 134 100% 150 100% 781 100% *** Results of contingency table analysis relating respondents’ current positions and how they found their current positions. Rows show the distribution of people in current positions who found their positions through the primary method listed. Full table statistics: Monte Carlo simulated chi-square test of independence (2,000 runs) = 123.01 and simulated p = 0.0005. The total (for row) column provides the overall distribution of ways participants found their current position. Superscripts show row groups that had similar (not significantly different) rates. %Col, column percentages; Sig. of follow-up χ2 indicates FDR-adjusted significance of χ2 for that row. Superscripts represent groups (by row) that are significantly different at an FDR-corrected p value of 0.05 or lower.
↵*** p < 0.001.
Figure 1-1
Omnibus repeated measures MANOVA results. Results from omnibus repeated measures MANOVA to ascertain whether there were differences in the 4 Career Interest ratings over 3 Time points (within-subjects ordinal dependent variables) by Gender, UR Status, and Current Position (categorical independent variables). UR=Under-Represented, ATS BS=ANOVA-Type Statistic Bootstrap, BH Adj=Benjamini and Hochberg adjusted, Sig=Significance. *** = p < 0.001. Download Figure 1-1, DOCX file.
Figure 1-2
Four 2-way follow-up repeated measures (Time by Type of Interest) MANOVAs separately by Current Position. Results from separate repeated measures MANOVAs to ascertain whether there were differences in the 4 Career Interest ratings over 3 Time points (within-subjects ordinal dependent variables) for each of the 4 Current Positions. ATS BS=ANOVA-Type Statistic Bootstrap, BH Adj=Benjamini and Hochberg adjusted, Sig=Significance. * = p < 0.05, *** = p < 0.001. Download Figure 1-2, DOCX file.
Figure 1-3
Sixteen 1-way follow-up repeated measures (Time) ANOVAs separately by Current Position and Type of Interest. Results from separate repeated measures ANOVAs to ascertain whether there were differences over 3 Time points (within-subjects ordinal dependent variable) for each of the 4 Current Positions and 4 Career Interest ratings. ATS BS=ANOVA-Type Statistic Bootstrap, Sig=Significance. *** = p < 0.001. Download Figure 1-3, DOCX file.
Figure 1-4
Mean Career Interest ratings by Current Position, Type of Interest, and Time. Means and standard deviations of Career Interest ratings with indication of significance from follow-up ANOVAs (Figure 1-3). The highest ratings for each time point are indicated by ⴕ. Sig=Significance, SD=Standard Deviation. *** = p < 0.001. Download Figure 1-4, DOCX file.
Figure 2-1
ANOVA Results for Continuous Dependent Variables by Gender and UR Status. Results from thirty-five two-way (Gender by UR Status) ANOVAs on continuous explanatory variables (dependent variable) to ascertain whether there were Gender or UR Status differences in the explanatory variables. UR=Under-Represented, BH Adj=Benjamini and Hochberg adjusted, Sig=Significance. Effect size: (-) = negligible effect size, (S) = small effect size. * = p < 0.05, ** = p < 0.01, *** = p < 0.001. Download Figure 2-1, DOCX file.
Figure 2-2
Logistic Regression Results for Binomial Categorical Dependent Variables by Gender and UR Status. Results from eighteen two-way (Gender by UR Status) logistic regressions on dichotomous explanatory variables (dependent variable) to ascertain whether there were Gender or UR Status differences in the explanatory variables. UR=Under-Represented, BH Adj=Benjamini and Hochberg adjusted, Sig=Significance. Effect size: (-) = negligible effect size, (S) = small effect size, (M) = medium effect size. * = p < 0.05, *** = p < 0.001. Download Figure 2-2, DOCX file.
Figure 2-3
Multinomial Logistic Regression Results for Multinomial Categorical Dependent Variables by Gender and UR Status. Results from three two-way (Gender by UR Status) multinomial logistic regressions on multinomial categorical explanatory variables (dependent variable) to ascertain whether there were Gender or UR Status differences in the explanatory variables. UR=Under-Represented, Sig=Significance. * = p < 0.05, *** = p < 0.001. Download Figure 2-3, DOCX file.
Figure 4-1
Lasso regression predicting Academia vs. Not Academia. Results of 5-fold cross-validated lasso logistic regression predicting whether respondents were currently in academic or non-academic positions from all explanatory variables and their interactions with Gender and/or UR Status. Conf=confidence, CV=cross-validation, sd=standard deviation. Download Figure 4-1, DOCX file.
Figure 5-1
Lasso regression predicting Research Academia vs. Teaching Academia. Results of 5-fold cross-validated lasso logistic regression predicting whether respondents were currently in research or teaching academic positions from all explanatory variables and their interactions with Gender and/or UR Status. Conf=confidence, CV=cross-validation, sd=standard deviation. Download Figure 5-1, DOCX file.
Figure 6-1
Lasso regression predicting Non-academic Research vs. Science/Non-research. Results of 5-fold cross-validated lasso logistic regression predicting whether respondents were currently in non-academic research or scitentific non-research positions from all explanatory variables and their interactions with Gender and/or UR Status. Conf=confidence, CV=cross-validation, sd=standard deviation. Download Figure 6-1, DOCX file.
In this issue
