Fluctuating Team Science: Perceiving Science as Collaborative Improves Science Motivation

The “team science revolution” suggests that people working together from different backgrounds and with different training make for the best innovation and discovery (Travis, 2011). Indeed, with the “revolution” underway there is a burgeoning science about team science – examining factors which enhance or impede team effectiveness including individual factors, team factors, and institutional factors (National Research Council, 2015). Important as each of these factors are in isolation and in tandem, we add to this list cultural factors, with basis in widely-held stereotypes in the United States (Brown, Steinberg, Lu, & Diekman, 2017). Rightly or wrongly, team science still conjures the image of each investigator working solo on his or her part of the problem, then meeting now and again to discuss the findings (Rey, 2008). The lone scientist stereotype is difficult to shake (Cheryan, Plaut, Handron, & Hudson, 2013; Diekman, Clark, Johnston, Brown, & Steinberg, 2011) and is doing a disservice to recruiting people into science and retaining them across the science pipeline (Diekman, 2015). People are inherently interpersonally motivated (Baumeister & Leary, 1995; Deci & Ryan, 1985) and working with, or even next to other people improves motivation (e.g., Isaac, Sansone, & Smith, 1999). Indeed, few people are truly “low” in interpersonal motives (Smith & Ruiz, 2007), and women and other underrepresented minorities (URMs) are especially likely to want to work with other people (Diekman, Brown, Johnston, & Clark, 2010; Diekman et al., 2011; Morgan, Isaac, & Sansone, 2001; NSF, 2015; Smith, Cech, Metz, Huntoon, & Moyer, 2014; Villarejo, Barlow, Kogan, Veazey, & Sweeney, 2008). For students beginning their science career, to what extent do they see science as providing opportunities to fulfill collaborative goals, and how does this perception of science as social-collaborative (or not) change over time, as students gain experience working in science labs? Do changes in perceptions of the social-collaborative nature of science predict intrinsic interest over time? How, if at all, does the collaborative nature of science especially impact students living at the intersection of disadvantaged identities?

Goal Congruity Model and Science Motivation

Classic motivation theories (Lewin, 1951) illustrate that a greater congruency between characteristics of the situation and the person maximizes task experiences and task outcomes. When the person and environment factors are based in socially constructed roles (e.g., gender), Social Role Theory (Eagly & Diekman, 2005) contends that people come to expect that demographic characteristics of the person can be matched or mismatched with the characteristics of the situation in a reciprocal process. When a person possesses the characteristics believed to match the characteristics needed for the role, a sense of congruity is established and in turn, both the person and environment factors are reinforced. By contrast, when a person does not possess the characteristics believed to match the characteristics needed for the role, incongruity results which can lead to feeling more negativity toward one's role. A contemporary variation of Social Role Theory is elaborated in Diekman and colleagues' Goal Congruity Model of Role Entry, Engagement, and Exit (Diekman, Steinberg, Brown, Belanger, & Clark, 2017), which suggests that person characteristics can extend beyond demographic variables, into one's personally-held values and goals, to connect explicitly with beliefs about one's career.

Evidence from the Goal Congruity Model shows people want careers that provide them the opportunity to meet their valued work and personal goals (Diekman, et al., 2010; Weisgram, Bigler, & Liben, 2010). Moreover, for adolescents and young adults who are starting to make important educational and career decisions, perceptions of science are often based on stereotypes (Cheryan, Siy, Vichayapai, Drury, & Kim, 2011; Cheryan et al., 2013) and are highly malleable (Brown, Thoman, Smith, & Diekman, 2015; Diekman et al., 2011). Importantly, the goal congruity perspective suggests that given an expectation or experience of goal incongruity, the resulting outcome is either to change one's goal or to make a different role decision, which should manifest into differential science motivation.

Past research points to communal perceptions of science as key to understanding goal-congruity (Boucher, Fuesting, Diekman, & Murphy, 2017; Brown, Smith, Thoman, Allen, & Muragishi, 2015; see Diekman, et al., 2017 for review). Communal perceptions of science are “other-focused” and include three dimensions: altruism, relationship- building, and social-collaborative engagement. This is typically compared to agentic perceptions of science, which are “self-focused” and include three dimensions: seeking power, achievement, and independence (Bakan, 1966). Supporting evidence for the Goal Congruity Model has unpacked various aspects of communal and agentic perceptions and the consequences of congruity and incongruity. For example, Brown and colleagues compared the relative strength of communal goals measured as a trait (e.g., what do I want from my job?) and communal goals measured as contextual-perceptions (e.g., what does this job allow me to do?). Results suggested that when both were considered independently and jointly, communal perceptions of the field emerged as the central factor influencing motivation for and positivity toward science (Brown, Thoman, et al., 2015). In addition, students who strongly endorse one or more types of communal goals lose interest in science when they perceive science as lacking opportunities to fulfill those goals (Allen, Muragishi, Smith, Thoman, & Brown, 2015; Diekman & Steinberg, 2013; Smith et al., 2014; Thoman, Brown, Mason, Harmsen, & Smith, 2015). Certainly, people differ in which specific communal aspect they might want fulfilled from a job or task. Thoman et al. (2015), for example, finds altruistic goals in particular are key for motivating URM undergraduate students in science, above and beyond other intrinsic and extrinsic goals (see also Allen et al., 2015; Smith et al., 2014).

The current research isolates the “social-collaborative” aspect of communion. Social-collaborative engagement is the degree to which the collaborative and interactive nature of the discipline involves meaningful work with others (Diekman & Steinberg, 2013; Pohlmann, 2001). Experimentally, adding a social-collaborative dimension to descriptions of science improves students' motivation and positivity toward science. For example, describing a scientist's daily schedule collaboratively produced greater science career positivity for women, whereas men showed equal career positivity regardless of how the scientist daily schedule was framed (Diekman et al., 2011). Rather than experimentally manipulating the presence or absence of social-collaborative perceptions of science, the current research examines the natural fluctuation of social-collaborative perceptions in science over time, during a critical gateway to a science career; namely when undergraduates start working in faculty labs. In addition, we examined whether those fluctuations predict intrinsic motivation, which we define as the anticipated, actual, or sought experience of interest (Sansone & Smith, 2000; Sansone & Thoman, 2005). Understanding the experience of interest is important because it is paramount to persistence and career pursuit, above and beyond performance outcomes (e.g., Renninger & Hidi, 2015; Thoman, et al., 2013). Indeed, a meta-analysis of over 60 years of interest-vocational research shows that intrinsic interest is a central positive predictor of academic and career selection, performance, and persistence (Nye, Su, Rounds, & Drasgow, 2012).

Intersecting Gender and Ethnic Minority Identities

The “gender problem in STEM (science/technology/engineering/math) pursuits can be at least partially considered a communion problem” (Diekman et al. 2011, p. 11, emphasis added). To the extent that underrepresented ethnic minority men and women (e.g., Native Americans and Latino/as) also highly endorse communion (Jackson, Galvez, Landa, Buonora, & Thoman, 2016; Smith et al., 2014; Thoman et al., 2015), we argue that considering the intersection of gender and ethnic identities is an important part of understanding broad participation in the scientific workforce. For example, O'Brien, Blodorn, Adams, Garcia, and Hammer (2014) examined how gender and race contributed to African American and White men and women's science participation and stereotypes. Their research suggested that African American women were more likely to pursue majoring in science compared to White women – whereas this difference was smaller in men. Furthermore, African American men and women similarly held weak implicit gender-stereotypes about science, yet only for African American women did this tendency explain greater rates of participation for women compared to men. Likewise, research shows that ethnic minorities who are also first-generation college students are especially at risk for poor performance in biology classes, and they benefit the most from interventions that focus on the utility value of science (Harackiewicz, Canning, Tibbetts, Prinisky, & Hyde, 2016). Finally, an intersectionality lens illustrated that the common finding showing men are less likely to value communion than women was true, but only when comparing White men; Native American men valued communion as much as Native American women and White women in STEM (Smith et al., 2014). We thus add to this growing work on appreciating multiple identities and apply a similar approach to test differences and similarities in whether and how social-collaborative perceptions of science change over time for men and women who are URM.

Project Overview

Given the importance of social-collaborative perceptions about science to student motivation (e.g., Brown, Smith, et al., 2015; Diekman et al., 2011), we aimed to understand whether and how social-collaborative perceptions about science changed over time and whether greater perceptions correspond to enhanced science motivation. Our work challenges assumptions that social-collaborative perceptions of science are stable over time, and we argue that fluctuations may map onto students' likelihood to persist in the science pipeline. Identifying when, for whom, and with what motivational consequences perceptions of science develops has the potential to help create enhanced learning and working environments (see also Canning & Harackiewicz, 2015). Our project goal aligns well with a multilevel modeling approach – allowing for (a) within-person fluctuations over time (i.e., how an individual RA's perceptions vary), (b) between-person differences (i.e., how gender and/or ethnicity group perceptions vary), and (c) prediction of immediate and future science motivation outcomes.

We tested our research questions within a specific subpopulation of scientists – biomedical RAs. Given the diversity within fields of science studies (from physics, to earth science, to agricultural science, etc.) we identified biomedical sciences as our focus of study. Biological and biomedical sciences degrees lack ethnic and gender diversity across the scientific pipeline. For instance, whereas 59.4% of B.S./B.A. degrees in biological sciences were awarded to Whites, only 10.8% awarded to Latino/as, 7.2% were awarded to African Americans, and 0.3% awarded to Native Americans (NSF, 2015).

Our sample included students currently working in research assistant roles working with biomedical faculty at 10 institutions, which varied in research focus, size, geographic location, and science departments. Among the biomedical RAs, we tested for effects of gender and/or ethnic minority status. Drawing on Goal Congruity Model (Diekman et al., 2010; 2011; 2017), we predicted that (a) URM women would report the highest levels of social-collaborative science perceptions and motivation due to the congruity between communal goals and science, and (b) non-URM men would report the lowest levels of social-collaborative science perceptions and motivation due to the incongruity between agentic goals and science. We expected that non-URM women and URM men would fall in between, given the combined influence of gender norms and culture ethnicity norms informing perceptions and experiences.

Method

Participants

Undergraduate biomedical RAs (N = 522, 59.8% female, Mage = 22.44 years, SD_age = 4.36 years) were recruited for a larger study¹ through their (randomly selected, NIH funding-eligible) faculty mentors working at one of 10 Mountain West and West coast universities/tribal colleges. Overall, 147 of students (86 women and 61 men) from the sample self-identified with an URM group in science, with 12% Latino/a, 8% Native American, and 3% African American.

Forty-eight faculty mentors (M = 46.55 years, SD = 8.89 years, Range = 29-73 years) were represented across 14 departments, including biological sciences, chemistry/biochemistry, engineering, life sciences, psychology, microbiology and immunology, nursing, and health and human performance (see Online Supplement for faculty demographics).

Procedure

Time (0-4) was measured in academic terms. Research assistants were contacted by email approximately 5-6 weeks into the term to complete a baseline survey with demographic information (Time 0), and then approximately 10-12 weeks later to complete an end-of-term follow-up survey about their research experience (Time 1). At the end of each subsequent academic term, RAs received an email invitation to complete an additional follow-up survey, regardless of whether they were still working in the research laboratory (Times 2-4). Table 1 presents the number of responses for each measure at each time point (see Online Supplement for additional analyses to test for and rule out alternative explanations for findings based on missing data/participant attrition). Social-collaborative science perceptions were assessed at Times 0-4, and immediate intrinsic interest in science and future interest in attending a biomedical graduate program were assessed at Times 1-4. For each completed survey, RAs were compensated with a $35.00 online gift card.

Table 1.

Means and standard errors for collaboration goals, intrinsic interest, and interest in biomedical graduate programs.

	Collaboration Goalsa			Intrinsic Interestb			Interest in Biomedical Graduate Programc
	N	M	SE	N	M	SE	N	M	SE
Middle of Semester 1	493	8.09	0.08	-	-	-	-	-	-
End of Semester 1	252	7.59	0.11	254	34.60	0.33	278	17.61	0.44
End of Semester 2	171	7.61	0.13	170	34.23	0.38	200	17.00	0.47
End of Semester 3	127	7.49	0.15	128	34.06	0.40	165	15.92	0.51
End of Semester 4	124	7.53	0.13	123	33.72	0.47	158	15.73	0.57

Note. Only Collaboration goals had measurements at the middle of the first semester.

^aFit statistics: -2ResLL = 4422.4, AIC = 4430.4, BIC = 4447.4

^bFit statistics: -2ResLL = 3963.9, AIC = 3983.9, BIC = 4026.4

^cFit statistics: -2ResLL = 5111.3, AIC = 5131.3, BIC = 5173.9

Measures

Complete information for all measures is presented in the Online Supplement.

Social-collaborative science perceptions

Students' perceptions of biomedical research were measured using a modified version of the Work Values Scale (Johnson, 2002). Social-collaborative science perceptions were assessed with two items: “The research work I do in this lab gives me a chance to make friends” and “The research work I do in this lab permits contact with a lot of people” on 1 (not at all) to 5 (very much) Likert-scales. Responses for social-collaborative perceptions were summed creating a range of 2-10 (M = 7.75, SD = 1.88). Students indicated their agreement with social-collaborative perception items at Times 0-4 (rs ranged from 0.62 - 0.68). Overall, 521 students provided 1,321 total observations for social-collaborative science perceptions.

Intrinsic interest in science

Students' intrinsic interest in science was measured using six items adapted from Smith, Sansone, and White (2007) (e.g., “I would describe my research lab as very interesting” and “I become very absorbed with work my research lab while I am doing it”) on 1 (strongly disagree) to 7 (strongly agree) Likert-scales. Responses for intrinsic interest in science were summed creating a range of 10-42 (M = 34.51, SD = 5.70). Students rated their perception of intrinsic interest in science at Times 1-4 (αs ranged from 0.87 - 0.93). Overall, 304 students provided 832 total observations for intrinsic interest in science.

Future interest in attending a biomedical graduate program

Future motivation assessed students' commitment to the next step of their research training with science, stemming from their immediate experience of interest. Students' future motivation to apply to a biomedical graduate program was measured using items modified from Carroll, Sheppard, and Arkin (2009). Interest in attending a biomedical graduate program was assessed with four items (e.g., “How interested are you in learning more about graduate programs related to biomedical research?” and “How willing would you be in the future to apply to graduate school for biomedical research?”) on 1 (not at all) to 7 (extremely) Likert-scales. Responses for interest in attending a biomedical graduate program were summed creating a range of 4-28 (M = 16.79, SD = 7.86). Students rated their future motivation for biomedical graduate programs at Times 1-4 (αs ranged from 0.97 - 0.98). Overall, 323 students provided 1,006 total observations for interest in attending a biomedical graduate program.

Results

Analytic Approach

Due to the nesting of repeated occasions within students, multilevel models were estimated to account for non-independence across observations. All models were estimated using restricted maximum likelihood (REML) estimation in SAS v. 9.4; as a result, statistical significance of fixed effects were evaluated using Wald p-values, whereas statistical significance of random effects were evaluated using deviance difference test (likelihood-ratio test; −2ΔLL) and information criteria (AIC and BIC). For all outcomes, time was centered at the end of the first term. To describe random variation around a given fixed effect, 95% random effect confidence intervals were calculated as: fixed effect ± $1.96\sqrt{\text{random effect varience}}$. Further, effect sizes from multilevel models are reported as proportion of variance explained for a specific variance component using pseudo-R² calculated as: (variance_larger – variance_smaller) / variance_larger, in which variance_smaller indicates the model with the additional predictor variable and variance_larger indicates the comparison model (Raudenbush & Bryk, 2002, p. 79). The specific variance component explained will be identified in the subscript for R², such that $R_{\text{Res}}^2$ indicates explained residual variance, $R_{\text{RI}}^2$ explained random intercept variance, and $R_{\text{RS}}^2$ explained random slope variance. Effect sizes for mean differences are reported as Cohen's d.

First, unconditional univariate models identified the pattern of change over time for each outcome. Next, conditional univariate models for each outcome were estimated and evaluated whether URM status, gender, or the URM × Gender interaction moderated change over time. Finally, two sets of multilevel structural equation models (SEM) were estimated to evaluate whether change in collaboration goals predicted intrinsic interest and whether intrinsic interest predicted interest in attending a biomedical graduate program; both models also evaluated whether the prediction was moderated by URM status, gender, or both. Multilevel SEMs were required because intrinsic interest, social-collaborative science perceptions, and interest in attending a biomedical graduate program were found to have significant random variability in intercept and change over time. Thus, the multilevel SEMs explicitly separated residual and random slope variance allowing the evaluation of the mean and variance in intrinsic interest, social-collaborative science perceptions, and interest in attending a biomedical graduate program across terms (see Hoffman, 2015; Lüdtke, Marsh, Robitzsch, Trautwein, Asparouhov, & Muthén, 2008; Preacher, Zyphur, & Zhang, 2010).

As students are nested within labs and these labs are nested within institutions, we also estimated a four-level model to evaluate whether the demographics of the lab and institution explained significant variability in social-collaborative science perceptions. Results indicated no meaningful differences (see Online Supplement); thus, when using the model for social-collaborative science perceptions in the multilevel SEM analysis below, the lab- and school-level variability are excluded for parsimony purposes.

Social-Collaborative Science Perceptions

Unconditional model

The intra-class correlation indicated that approximately 50% of the variability in social-collaborative science perceptions was due to level-2 person mean differences; 95% of the sample was expected to have social-collaborative science perceptions measured at the end of the first term ranging between 5.28 and 10.00. Occasion-specific variances and between-occasion correlations as estimated by a saturated means, unstructured variance model are provided in Table 2.

Table 2.

Variances and correlations for collaboration goals, intrinsic interest, and interest in biomedical graduate programs.

	Middle of Semester 1	End of Semester 1	End of Semester 2	End of Semester 3	End of Semester 4
Collaboration Goals
Mid-Semester 1	2.96
End of Semester 1	0.48	3.50
End of Semester 2	0.47	0.59	3.70
End of Semester 3	0.46	0.47	0.55	3.64
End of Semester 4	0.50	0.58	0.63	0.67	3.22
Intrinsic Interest
End of Semester 1	-	29.95
End of Semester 2	-	0.74	33.37
End of Semester 3	-	0.60	0.73	29.04
End of Semester 4	-	0.57	0.63	0.75	37.74
Interest in Biomedical Graduate Program
End of Semester 1	-	57.15
End of Semester 2	-	0.76	57.75
End of Semester 3	-	0.70	0.80	61.75
End of Semester 4	-	0.60	0.66	0.78	68.24

Note. Only collaboration goals had measurements at the middle of the first semester. Variances presented on the diagonal, with correlations in off diagonals.

Next, a series of unconditional piecewise (i.e., discontinuous) time models were estimated to describe the pattern of change in social-collaborative science perceptions over five repeated occasions given that an obvious breakpoint was observed at the end of the first term. Thus, this piecewise change model was estimated using two separate linear slopes before and after the end of the first term (i.e., slope01 = change before the end of the first term; slope14 = change after the end of the first term) and fit the data significantly better than both the fixed linear time and fixed quadratic time growth model (AIC for piecewise: 4430.4 vs. AIC for linear and quadratic time: 4454.2 and 4447.8, respectively; note that the random linear time and random quadratic time effects were not estimable). The model indicated a significant decrease in social-collaborative science perceptions before the end of the first term by 0.50 points, 95% CI = [−0.69, −0.31], t(842) = −5.07, p < .001, $R_{\text{Res}}^2=5.84\%$. Descriptively, the model also showed a trend for a decrease after the end of the first term by 0.03 points per term although this effect was not significant, 95% CI = [−0.12,0.06], t(788) = −0.69, p = 0.49, $R_{\text{Res}}^2=0.00\%$.

Adding random slope variance before the end of the first term (and covariance with random intercept variance) significantly improved model fit, −2ΔLL(2) = 10.6, p = 0.01. Thus, change prior to the end of the first term varied randomly across students, as the 95% of students had time slopes ranging from −3.57 to 1.59. Therefore, the final unconditional univariate model for social-collaborative science perceptions included two piecewise linear time slopes, with the breakpoint at the end of the first term, and a random linear time slope variance before the end of the first term. Estimated means, standard errors, and fit statistics for this final model are provided in Table 1.

Conditional models with URM and Gender

A significant URM × gender × linear time after the end of the first term interaction was identified, B = 0.38, 95% CI = [0.03,0.73], t(690) = 2.13, p = 0.03, $R_{\text{RI}}^2=0.79\%$ (see Figure 1). Specifically, URM women averaged significantly greater social-collaborative science perceptions at all occasions compared to non-URM women, but this difference was smallest at the middle of the first term, M_diff = 0.47, 95% CI = [0.07,0.88], t(531) = 2.29, p = 0.02, d = 0.20, and increased across observations such that the largest difference was observed at the end of the fourth term, M_diff = 0.94, 95% CI = [0.22,1.66], t(713) = 2.56, p = 0.01, d = 0.19. By contrast, the difference in social-collaborative science perceptions between URM men and non-URM men was initially significant and decreased across terms. That is, although URM men averaged significantly greater social-collaborative science perceptions compared to non-URM men at the middle and end of the first term, M_diff = 0.74, 95% CI = [0.26,1.22], t(527) = 3.00, p = 0.003, d = 0.26 and at the end of the second term, M_diff = 0.51, 95% CI = [0.02,1.01], t(530) = 2.03, p = 0.04, d = 0.18, the two groups had similar social-collaborative science perceptions at the end of the third and fourth terms. Neither the URM × gender interaction nor the URM × gender × linear time before the end of the first term interactions were significant, Bs < .34, ps > .37.

Figure 1.

Predicted trajectories showing three-way interaction between URM status, gender, and the linear time effect after the end of the first term (change from middle of semester 1 to end of semester 1 is constant across groups). Note that 95% confidence intervals are not presented because the overlap of intervals across groups at each observation would not allow identification of intervals specific to a given group.

Intrinsic Interest

Unconditional model

The intra-class correlation indicated that approximately 67% of the variability in intrinsic interest was due to level-2 person mean differences; 95% of the sample was expected to have intrinsic interest scores at the end of the first term ranging between 24.78 and 42.00. There was no estimable variability in intrinsic interest between labs or between institutions. Occasion-specific variances and between-occasion correlations as estimated by a saturated means, unstructured variance model are provided in Table 2.

A series of polynomial change models were estimated for intrinsic interest. Adding a fixed linear time effect indicated a significant decrease in intrinsic interest scores by an average of 0.30 points per term, 95% CI = [−0.55,−0.06], t(443) = −2.41, p = 0.02, $R_{\text{Res}}^2=1.64\%$. A random linear time slope variance (and covariance with random intercept variance) significantly improved model fit, −2ΔLL(2) = 22.4, p < 0.001, indicating that linear change across terms varied significantly between students, with the 95% of students having linear time slopes ranging from −2.92 to 2.30. Thus, the final model for intrinsic interest was a random linear time model. Estimated means, standard errors, and fit statistics for this final model are provided in Table 1.

Conditional models with URM and Gender

Results indicated a significant main effect of URM status where URM students averaged 1.70 points higher intrinsic interest compared to non-URM students across all terms, 95% CI = [0.38,3.02], t(291) = 2.53, p = 0.01, $R_{\text{RI}}^2=3.17\%$. URM did not moderate the effect of linear time, B < .01, p > .99. The main effect of gender was not statistically significant nor did gender moderate the effect of linear time, Bs > − .48, ps > .43. Finally, neither the URM status × gender interaction nor URM status × gender × linear time interaction were not significant for intrinsic interest, Bs < .59, ps > .67.

Interest in Attending a Biomedical Graduate Program

Unconditional model

The intra-class correlation indicated that approximately 70% of the variability in interest in attending a biomedical graduate program was due to level-2 person mean differences; 95% of students were expected to have interest in attending a biomedical graduate program rated at the end of the first term ranging between 3.88 and 28.00. Relative to this unconditional two-level model, neither the addition of between-lab variability nor between-institution variability significantly improved model fit, −2ΔLL(1) = 3.7, p = .054 and −2ΔLL(2) = 4.9, p = .086, respectively. Occasion-specific variances and between-occasion correlations as estimated by a saturated means, unstructured variance model are provided in Table 2.

A series of polynomial models of change were estimated for interest in attending a biomedical graduate program. Adding a fixed linear time effect indicated a statistically significant decrease in biomedical graduate school interest by an average of 0.68 points per term, 95% CI = [−0.96,−0.39], t(539) = −4.69, p < 0.001, $R_{\text{Res}}^2=8.40\%$. A random linear time slope variance (and its covariance with random intercept variance) significantly improved model fit, −2ΔLL(2) = 33.7, p < 0.001 (with smaller AIC and BIC), which indicated linear change across occasions varied significantly between students, with 95% of students expected to have linear time slopes ranging from −4.13 to 2.75. No additional fixed or random polynomial effects were statistically significant. Estimated means, standard errors, and fit statistics for this final model are provided in Table 1.

Conditional models with URM and Gender

Results indicated a statistically significant URM status × linear time interaction, in which URM students had less negative linear time slope by 0.93 points per observation, 95% CI = [0.18,1.67], t(208) = 2.45, p = 0.02, $R_{\text{RS}}^2=5.59\%$ (see Figure 2). Although the difference in interest in attending a biomedical graduate program between students of URM and non-URM status was not statistically significant at the end of the first term, M_diff = 1.67, 95% CI = [−0.20,3.57], t(311) = 1.76, p = 0.08, d = 0.20, URM students averaged a non-statistically significant 0.04 point decrease per term, whereas non-URM students averaged a statistically significant 0.97 point decrease per term. Thus, statistically significant differences were observed between students of URM and non-URM status at all occasions beginning at the end of the second term, with the largest difference observed at the end of the fourth term, M_diff = 4.47, 95% CI = [2.17,6.77], d = 0.50. Interest in attending a biomedical graduate program did not differ between men and women, and gender did not moderate the effect of URM status (Bs < .39, ps > .28).

Figure 2.

Predicted trajectories showing two-way interaction between URM status and the linear time effect. Error bars represent 95% confidence intervals.

Social-Collaborative Science Perceptions predicting Intrinsic interest

As described above, for social-collaborative science perceptions, two fixed piecewise slopes were estimated (break point at the end of the first term; Slope01_tiColl and Slope 14_tiColl) with a random slope variance before the end of the first term, and the three-way interaction between URM status, gender, and the linear time slope after the end of the first term. For intrinsic interest, a random linear time model was estimated (as indicated by Slope14_tiInt given that the first measurement of intrinsic interest occurred at the end of the first term) with a main effect of URM status (see Online Supplement for multilevel SEM equation).

The fixed effect for social-collaborative science perceptions (level-2 random intercept; γ_02Int), predicted intrinsic interest, such that for every point greater social-collaborative science perceptions at the end of the first term, intrinsic interest at the end of the first term was significantly greater by 0.96 points, 95% CI = [0.19,1.72], t(267) = 2.47, p = 0.01, which explained approximately 3.28% of the random intercept variance for intrinsic interest. Further, the fixed effect for the level-1 residuals for social-collaborative science perceptions, γ_20Int, was statistically significant indicating that for every one-point increase in social-collaborative science perceptions more than predicted by the student's own trajectory at a given measurement, intrinsic interest at that same measurement was also expected to increase by 0.73 points, 95% CI = [0.44,1.02], t(93) = 4.99, p < 0.001, which explained approximately 11% of intrinsic interest residual variance. Statistically significant random variation was also indicated for this effect, −2ΔLL(3) = 15.6, p = 0.001 (with smaller AIC), with 95% of students expected to range from −0.90 to 2.36. Finally, we tested whether URM status or gender moderated the relationship between social-collaborative science perceptions and intrinsic interest. Results indicated that neither URM status nor gender moderated any prediction of intrinsic interest by social-collaborative science perceptions in the multivariate models for intrinsic interest, Bs > −.64, ps > .16.

Intrinsic Interest predicting Interest in Attending a Biomedical Graduate Program

As described above, random linear time models were estimated for both intrinsic interest and interest in attending a biomedical graduate program. Although both outcomes included a main effect of URM status, only interest in attending a biomedical graduate program required a URM status × linear time interaction effect (see Online Supplement for multilevel SEM equation).

The fixed effect for intrinsic interest (level-2 random intercept; γ_02Grad), predicted interest in attending a biomedical graduate program, such that for every point greater intrinsic interest at the end of the first term, interest in attending a biomedical graduate program at the end of the first term was significantly greater by 0.53 points, 95% CI = [0.32,0.73], t(293) = 5.07, p < 0.001, which explained approximately 6% of random intercept variance for interest in attending a biomedical graduate program. No other effects were statistically significant; neither URM status nor gender moderated any direct effects of intrinsic interest predicting interest in attending a biomedical graduate program (Bs > −.42, ps > .06).

General Discussion

Our research mapped the natural fluctuation of students' social-collaborative perceptions about science and why these changes matter for understanding group differences biomedical RAs' motivation. Importantly, we gained insight into a critical period of students' career (and identity) development—that is, after they have been exposed to science courses, during their exploration of biomedical research in practice, but before making decisions about whether to pursue graduate training in biomedical research or a career in biomedicine. We simultaneously investigated research questions involving intrapersonal fluctuations in social-collaborative science perceptions relative to self (level 1) and between-person differences in social-collaborative science perceptions relative to others (level 2). We believe such an approach offers important theoretical and methodical advances for Goal Congruity research (Diekman et al., 2010; 2011; 2017; Weisgram et al., 2010) and diversity-science efforts more generally.

Theoretical and Methodological Advances

The Goal Congruity Model of Role Entry, Engagement, and Exit (Diekman et al., 2017) posits that in creating a more inclusive and welcoming scientific community, we must first understand the function of communal goals for people living and working within science. This is paramount because STEM careers are perceived to offer few opportunities to fulfill communal goals– even compared to other male-dominated careers outside of STEM (e.g., business, law, medicine) (Diekman, et al., 2010). We documented when, for whom, and with what motivational consequences the social-collaborative perception of science develops. Students' social-collaborative science perceptions did fluctuate over the course of four academic terms. Results show a stark realization about the experience of scientific research: students perceived the lab environments to be less of a “team science” environment than they expected. During the initial weeks and months working on biomedical research projects, there was a decrease in social-collaborative science perceptions. This represents an example of a level 1 research question: intrapersonal fluctuations in social-collaborative science perceptions that are higher or lower than one's usual level. Overall, the trend between the first two time points was negative, but some students experienced higher than usual social-collaborative science perceptions within that same time frame. Understanding (and capitalizing on that initial incline or minimally, compensating for the decline) appears to be an important component of motivating students from all gender and ethnic backgrounds in the science pipeline.

Notably, fluctuations in social-collaborative science perceptions were less uniform between the end of the first and fourth academic terms when taking into account a RAs' gender and ethnicity. URM women's social-collaborative science perceptions were most stable over the next three academic terms (compared to other intersecting cultural groups). This represents an example of a level 2 research question: interpersonal comparisons of social-collaborative science perceptions that are higher or lower compared to others. The cultural group hypothesized to have the strongest communal goal endorsement (URM women) reported the highest naturally-occurring social-collaborative science perceptions. Are URM women then creating and seeking out lab tasks or experiences which will allow for the opportunity to create and strengthen social ties? Although we can only speculate about why these particular women demonstrated the most stable social-collaborative science perceptions, it does point to creating a research infrastructure which bolsters social-collaborative science perceptions among non-URM women and men, and especially URM men who showed a sharp decline during this same three-semester time period.

Of course, fluctuations in social-collaborative perceptions are only meaningful if they map onto motivational and/or career outcomes. Results further showed that increases in social-collaborative perceptions of science, relative to students' usual levels, corresponded to greater immediate intrinsic interest across time points. This is encouraging news for intervention researchers, as it replicates past research which aims to increase the social-collaborative utility value at static time points (Diekman et al., 2011) but importantly extends this finding to RAs over multiple semesters of lab work. Given that immediate intrinsic interest is a proximal predictor of future career interest (Nye et al., 2012; Renninger & Hidi, 2015; Thoman et al., 2013), we also considered whether increases in intrinsic interest for one's research would translate into greater intentions to pursue graduate training in biomedical research. Results showed that students' increases in immediate intrinsic interest, relative to others, predicted greater future motivation to pursue graduate biomedical training. Is such a motivational benefit only true for URM students, or does the pathway between social-collaborative science perceptions ➔ immediate intrinsic interest ➔ intention to attend graduate training in biomedicine extend to all students? No differences emerged based on intersecting disadvantaged identities – which clearly points to increasing social-collaborative science perceptions as a facilitating factor in increasing all students' biomedical research motivation.

A methodological contribution of this paper is showcasing the use of multilevel growth models to disentangle the implications of complex changes in motivational processes within individuals over time, across social groups. We not only documented group differences in changes of social-collaborative goals over time, but also why these changes matter for important career outcomes. The match between our research questions about students' individual development of social-collaborative science perceptions and resulting biomedical science motivation were well-suited to such a statistical test—and opens up the possibility of novel research hypotheses for researchers considering longitudinal designs (see also Fishbach & Hofmann, 2015).

Limitations and Future Directions

Our unique sample of ethnically diverse men and women RAs stemming from several different biomedical labs and academic institutions carries both strengths and weaknesses. The inherent nature of the sampling procedure required that students already be members of a biomedical research laboratory to be eligible to participate in this research. As such, our Time 0 (i.e., 5-6 weeks into the academic term) does not coincide with zero experience in the laboratory. Nonetheless, by establishing this baseline and subsequently examining change in social-collaborative science perceptions within and between individuals over four academic terms, we can begin to understand when, for whom, and with what motivational consequences perceptions of science develops from that baseline.

Further, addressing the intersection of disadvantaged identities in and of itself can raise a number of concerns. Although we allowed for self-identification of gender and ethnicity to construct the identity categories, we acknowledge that this initial test of social-collaborative science perceptions may have obscured important heterogeneity within those cultural groups. We certainly do not want to claim that all URM women have the same disadvantaged experience in science, or alternatively that all majority men have the same advantaged experience with science. There are likely several variables not assessed in the current research which would capture some of this rich cultural heterogeneity including gender expression, tribal affiliation, immigration status, social class, first-generation college status, among others – and acknowledging that diverse social identities accumulate into an experience of privilege and/or oppression allows for a more situated understanding of a cultural group's experience within science.

Although unrelated to study predictions, a surprising pattern that emerged in these data is that, as a main effect, URM students reported greater intrinsic and career interest than non-URM students. Data collected from freshmen and sophomore undergraduate typically demonstrate similar mean levels of interest (e.g., Jackson et al., 2016), but the current sample of RAs is much more advanced. Perhaps because science attrition is higher among URMs than non-URM students during the foundational levels of undergraduate science education, it is possible that URM students who persist to being an RA are those who are highly interested in research and biomedical careers.

Future research would do well to investigate the precise nature and source of the social-collaborative perceptions of science. Interviews and open-ended prompts would allow for additional insights regarding the social-collaborative nature of research tasks themselves, lab meetings, communication protocols, dissemination of the work, and differences in peer vs. mentor interactions. For instance, Thoman, Sansone, and Pasupathi (2007) identified talking with others about an activity as a self-regulatory mechanism to maintain intrinsic interest. Surveys about the extent to which students share their research experience with social support networks on-campus (e.g., academic advisors, affinity groups) and off-campus (e.g., community events, family) may be influential in predicting later social-collaborative science perceptions. Such qualitative data could create a deeper understanding of the antecedents leading to social-collaborative science fluctuations.

Implications

Consistent with national task force efforts and funding agency priorities (President's Council of Advisors on Science and Technology, 2012; Valantine & Collins, 2015), our data suggests that changing the social-collaborative science culture is one pathway to enhance biomedical research motivation. Given that social-collaborative perceptions of science are linked with immediate and future motivation, future interventions might focus on establishing team-based approaches that integrate communal (e.g., group-based practices such as paired student projects) and agentic (e.g., recognition programs that reward scientific discovery) elements. For example, new mentoring models might infuse research experiences with shared goals and team-based skill development, opportunities for social interaction during the research process, and a more diverse scientific pipeline (see also Diekman, Weisgram, & Belanger, 2015).

Taken together, is possible that science is impeding its own progress by unintentionally burying the social-collaborative engagement in this work. As the scientific workforce continues to become more diverse, learning and working environments can be enhanced by fostering a team science atmosphere. For instance, mentors can foster such social-collaborative perceptions of science among students to enhance immediate interest and persistence within science. In addition, professors and universities can support community-outreach opportunities, through innovative pedagogy and community partnerships. A rich opportunity exists to retain a diverse pool of future researchers and professionals in science, who may be at risk for opting out of the science pipeline, by restoring the social-collaborative spirit advocated within team science.