Identifying Effective Mentors in Scientific Communication: A Latent Profile Analysis of Mentor Beliefs

Abstract

The need to specifically mentor graduate and medical students performing biomedical and biobehavioral research in communication skills is increasingly being highlighted to increase intention to pursue academic research careers, including physician–scientist careers. This study used data collected from 354 research faculty in 33 states across the United States to examine beliefs and perceived barriers about mentoring in scientific communication (writing, presenting, and informal discussion about science), with the goal of advancing evidence-based recommendations for mentoring interventions. Latent profile analysis identified four mentor profiles, based on beliefs regarding mentoring responsibility, expected outcomes, and barriers in scientific communication mentoring. Problem solvers, who acknowledged trainee problems but reported high efficacy in overcoming them, offered the highest levels of supportive and instructive mentoring. Since mentoring messages and actions influence trainee career development significantly, our results have important implications for faculty development to advance effective mentoring, especially in scientific communication.

Keywords

latent profile analysis scientific communication mentoring research career social cognitive career theory

Mentoring in biomedical and biobehavioral research and in academic medicine has been acknowledged as a critical factor in the development of academic researchers and clinician-scientists. By facilitating trainees’ professional and personal development, career choice, and research productivity, mentors inspire and guide trainees toward career advancement and independence, including success in publications and grant funding (Longo et al., 2011; Sambunjak et al., 2006, 2010; Wolfson & Arora, 2018). Over the past decade, strong studies of mentorship have emerged with rigorous study designs, structured data collection, and systematic data analysis, supported by theory and informed scholarship, to enhance mentoring skills and guide intervention development (Pfund et al., 2014; Pfund et al., 2015). These include research to understand the factors that influence trainee skill development in scientific communication (SciComm) and the role that mentors play in teaching these skills (Cameron et al., 2015). SciComm includes writing, formally presenting, and informally conversing about science and medicine, and its mastery can become a critical bottleneck for career advancement. Since manuscript publication by trainees in medical school and research fellowships has been associated with increased intent for career-long research involvement (Wolfson & Arora, 2018), the need for effective mentorship in communication skills has become a growing area of concern.

Theoretical Framework for Mentoring in SciComm

Social cognitive career theory (SCCT; Lent et al., 1994) focuses on social cognitive mechanisms and processes in academic and career behavior. Based on the work of Bandura (1986, 1997) in explaining relationships between people’s beliefs and behaviors, SCCT was developed to address the role of motivational factors in career choices, considering both individual (e.g., self-efficacy, outcome expectations) and contextual (e.g., perceived barriers, supports) constructs as important variables in the development of career interests, choices, and performance in career paths. The application of SCCT has been supported in numerous studies in science, technology, engineering, and mathematics fields over a 30-year period (Lent et al., 2018) and more recently has been applied to the career development of physician-scientists (Bakken et al., 2006; Fernandez et al., 2019).

A small set of studies has used the SCCT framework to investigate SciComm skill development and career choice among doctoral and postdoctoral biomedical trainees. Trainees’ SciComm productivity, self-efficacy, and interest in SciComm activities, as well as outcome expectations relative to the impact of SciComm skills on career success, have been predictive of intention to pursue a research career (Cameron et al., 2015). In the latent variable structural models reported in these studies, both the previous SciComm products that trainees had produced (e.g., preparing a draft of a manuscript, presenting a poster or oral talk at a scientific meeting) and the mentoring in SciComm that trainees perceived they had received from their mentors were significant predictors of trainees’ appraisals of their abilities or self-efficacy in SciComm tasks, which served as a crucial mediator in predicting their career choice (Cameron et al., 2020). These results support three of the four hypothesized sources of self-efficacy in trainees, as conceived by Bandura (1986, 1997), within the task domain of SciComm: personal mastery experiences, represented by SciComm productivity; and verbal persuasion and vicarious learning, as reflected in the mentoring practices construct. Identifying the sources of self-efficacy beliefs has become a recent research focus in SCCT models, addressing an important gap in the literature on career decision making (Byars-Winston & Rogers, 2019; Lent et al., 2017).

With the findings of SciComm and career intention among trainees in mind, it also makes sense to study factors that motivate mentors to perform these role-specific tasks that influence trainees, again using the SCCT model, but from the mentors’ point of view. Since SCCT suggests that the interest–choice process relationship is weakened when proximal influences that constitute barriers are perceived to be high (Lent et al., 1994), it is important that research examine possible obstacles that may affect mentoring behaviors and possibly impair mentors’ self-efficacy in guiding their trainees in the skilled behaviors of writing, presenting, and impromptu scientific conversation.

A Person-Centered Approach to Studying Mentoring

Much of the growing body of empirical research on mentoring has focused on identifying mentor characteristics, behaviors, and skills that influence trainee research and career development (Sambunjak et al., 2006, 2010). In academic medicine, interest has focused on the benefits of mentoring and its importance in faculty development as well as the potential problems or barriers to mentoring that may interfere with the development of functional mentoring relationships (Rodenhauser et al., 2000; Sambunjak et al., 2010; Straus et al., 2013). The majority of the research on mentoring has used a variable-centered approach (e.g., regressions, path analysis) that describes how variables are related to other variables on average and across individuals (Aldenderfer & Blashfield, 1984). This approach assumes a homogeneous population of mentors who vary in attributes associated with the outcome of interest, for example, studies that examine behaviors of mentors associated with trainee outcomes without considering possible mentor subgroups that differ in their mentoring beliefs.

An alternative, less common strategy uses a person-centered approach that seeks to identify and group individuals with common attributes and then describes how individuals in those groups function (Aldenderfer & Blashfield, 1984). The primary goal in person-centered statistical approaches is to classify individuals into groups based on their patterns of responses to observed variables of interest, thereby maximizing the homogeneity within groups and the heterogeneity between groups (Hagenaars & McCutcheon, 2002; Lanza et al., 2003). Person-centered approaches that use latent class and latent profile analysis (LPA) are the focus of new research on career development (Vondracek & Porfeli, 2002) and may be useful in designing more tailored mentor-training interventions. Using a person-centered approach to identify distinct profiles among mentors in biomedical science could enhance our understanding of mentor motivations and behaviors, especially in the understudied domain of SciComm mentoring.

The Present Study

Given the growing number of previous quantitative studies linking mentor attitudes to mentor behaviors (Geraci & Thigpen, 2017), the purpose of the current study was to explore the belief profiles of faculty mentors regarding SciComm tasks and other research skills of trainees and to examine the association between these profiles and faculty perceptions of their mentoring practices. The current study makes additional contributions to SCCT by examining mentor perceptions of the positive and negative outcomes they expect from mentoring trainees at the doctoral and postdoctoral level actively pursuing research careers. Since contextual supports and constraints are hypothesized to be important influences on choice behaviors in the SCCT model, and sources of self-efficacy, the current study examines mentor perceptions of potential barriers to mentoring in the specific domain of SciComm. Perceived mentoring barriers could arise from SciComm deficits or relevant issues among trainees, the training/institutional environment, or even the mentors themselves. Mentors may feel they do not have the skill, interest, or time to provide targeted guidance to trainees in writing, giving oral presentations, or in teaching trainees how to engage well in impromptu conversations about science and research. They may also exhibit impatience when faced with SciComm mentoring tasks. Less SCCT research has focused on the role of personality and other individual differences, but the SCCT framework posits that such person inputs indirectly affect vocational interest formation through their influence on self-efficacy and outcome expectations (Lent et al., 1994). Recent work has supported this hypothesis and found additional direct effects on interest (Schaub & Tokar, 2005). In any case, regardless of their perceived barriers to mentoring in SciComm, mentors may not even believe it to be their responsibility to provide such guidance. If, however, mentors do feel responsible for SciComm mentoring and perceive few barriers to such tasks (i.e., they do have the skill, interest, time, and patience, and they can deal with trainee/institutional issues), this may imply that they are personally capable and willing to provide mentoring guidance in this domain. Since self-efficacy, interest, and outcome expectations affect goal-directed behavior (Bandura, 1986, 1997; Lent et al., 1994), which in this case would constitute mentoring practices to trainees in SciComm activities, the current study affords an important test of these relations among mentors in the understudied domain of SciComm within the SCCT framework.

With this social cognitive perspective in mind, the first goal of the study was to determine whether distinct mentor profiles could be identified. In our effort to distinguish subgroups of mentors based on their beliefs on mentoring trainees in SciComm skills, we expected to identify at least two profiles: (1) mentors reporting high levels of responsibility for SciComm instruction, many benefits or positive outcomes from mentoring, low perceptions of barriers to SciComm mentoring, and few costs or negative outcomes associated with mentoring in general; and (2) mentors reporting low responsibility for SciComm instruction, few mentoring benefits, more SciComm barriers, and mentoring costs.
Hypothesis 1: Two or more mentor belief profiles that are quantitatively and qualitatively distinct will be identified.

A second goal of the study was to examine among research mentors the relationships between different belief profiles and mentoring practices that apply broadly across disciplines. Reports on the practices of successful research faculty have led to general measures of mentoring behaviors (Berk et al., 2005; Longo et al., 2011), but these have not been applied to the domain of SciComm. Our recent qualitative studies of faculty describing their own mentoring experiences have led to development of new measures focused on SciComm. These allow more comprehensive analysis of the mentoring received by trainees destined for research careers (Anderson et al., 2020; Cameron et al., 2013; Lee et al., 2018). By applying a person-centered approach to analysis of mentor belief profiles, we can now go beyond the typical variable-centered research on mentor differences to compare mentoring practices across groups of mentors with different profiles, identified by their personal beliefs on a variety of issues relevant to their mentoring role, including their beliefs about SciComm.
Hypothesis 2: Mentor practices will differ across mentors with different belief profiles. Higher levels of mentoring support and instruction in general research and SciComm will be associated with profiles that reflect more positive, motivated attitudes toward mentoring and lower perceptions of mentoring barriers.

Method

Participants and Procedure

Participants were part of two sequential, online, survey studies of mentors and trainees. Mentors from both samples were PhD, MD, or MD/PhD faculty in the fields of basic biomedical, clinical, population, or biobehavioral science. Study 1 was a cross-sectional study with participant mentors from research and clinical institutions and medical schools throughout the Texas Medical Center in Houston, TX, recruited for an online survey via institutional email and person referrals. Of those showing initial interest, 194 met eligibility criteria, that is, currently mentoring at least one doctoral (PhD or MD) or postdoctoral trainee or having mentored one or more within the last 3 years. Study 2 was a 2-year, national, longitudinal study of 185 participant mentors from 71 universities in 33 states in the United States who gave informed consent. In Study 2, participants were recruited via mailing lists of relevant research networks, contacts at institutions associated with our research, and flyers distributed at professional meetings. Potential participants were asked to forward information about the study to acquaintances who might be interested in participating. A total of 208 mentors responded to our emails, expressing initial interest. Mentors were considered eligible for the study if they were currently mentoring one or more doctoral or postdoctoral trainees. Participants agreed to participate by online informed consent in both studies, and the institutional review board of MD Anderson approved both study protocols.

The current study combined the mentor data from Study 1 (Houston sample) and Wave 1 (of four waves) from Study 2 (national sample; Waves 2–4 asked mentor attitudes about a specific trainee). Both Study 1 and Study 2-Wave 1 questionnaires instructed mentors to report their beliefs concerning their current and past trainees in general, which allowed the data to be combined. Of the eligible faculty mentors from Study 1 (N = 194) and Study 2 (N = 185) who agreed to participate, 15 individuals completed only demographic information (Study 1), and 10 never started (Study 2); hence, data were available for analysis from 354/379 mentors (93%).

The two samples did not differ on mentor age, but significant differences were found on mentor gender (p = .001), race (p < .001), primary language (p < .001), academic rank (p = .001), working discipline (p = .011), and current number of trainees (p = .008). Study 1 (Houston sample) included more male faculty (59% vs. 42%); Asian American (24% vs. 14%), Hispanic (9% vs. 5%), and ethnic-Other/Mixed faculty (11% vs. 4%); and fewer White faculty (53% vs. 74%), compared to national Study 2. Both samples were 3% African American. Study 1 also had more faculty whose primary language was not English (41% vs. 24%), more instructors (9% vs. 0.6%) and full professors (36% vs. 31%) and fewer assistant professors (25% vs. 35%) and associate professors (30% vs. 33%). Study 1 had more clinical faculty (27% vs. 14%) but slightly fewer in basic science (55% vs. 59%) and population science research, for example epidemiology (16% vs. 25%) than Study 2. Study 1 faculty were currently mentoring fewer trainees than the mentors in Study 2: 1–5 trainees (69% vs. 83%) and 6–10 trainees (9% vs. 17%), except for four Study 1 mentors who had more than 10 trainees under their care (2% vs. 0%). The Houston sample (Study 1) also included 38 mentors (20%) who had mentored within the last 3 years but had no current trainees (this option was not allowed in Study 2). Combining the two samples increased the size as well as the heterogeneity and representativeness of the final sample for the current analysis. The statistical technique we used, LPA, assumes that the population of study is not homogeneous and then proceeds to classify individuals into homogeneous subgroups based on their responses to the study variables of interest.

The final combined sample was 51% female, with a mean age of 46.7 years (standard deviation [SD] = 9.6). The majority of participants described themselves as White (68%) or Asian American (20%), and participants were fairly evenly distributed across academic ranks (31% assistant, 30% associate, and 34% full professors). The majority were experienced mentors of ethnically diverse doctoral and postdoctoral trainees; 62% had past experience mentoring African American trainees, 66% Hispanic trainees, and 18% trainees of other ethnicities. Mentors were currently mentoring a mean of 1.8 doctoral students (SD = 1.7) and 1.3 postdoctoral fellows (SD = 1.8).

Measures

Barriers to mentoring in SciComm

Perceived barriers to mentoring trainees in SciComm were measured with a 22-item scale assessing trainee, institutional, and mentor-self barriers (Anderson et al., 2020). Trainee barriers (9 items) represent contextual influences that SCCT proposes relate to choice actions directly or indirectly through self-efficacy, that is, trainee issues that make it difficult for mentors to help trainees with SciComm tasks (e.g., trainee weakness in basic English writing skills, insufficient effort, dread of negative feedback or criticism). Institutional barriers (6 items) also represent contextual influences, specifically barriers within the institution and work/training environment that make it difficult to mentor in SciComm (e.g., lack of institutional resources for trainees and mentors, lack of mentor incentives). Mentor barriers were represented with a two-factor construct that reflected (1) lack of self-efficacy (4 items) and interest in teaching SciComm (1 item) and (2) lack of time, as a contextual influence (1 item), and impatience (1 item), as a person input or individual difference variable exacerbated by lack of time. Items were rated on a 5-point Likert-type scale (1 = very insignificant barrier to 5 = very significant barrier). A confirmatory factor analysis of the hypothesized four-factor structure for barriers (trainee, institution, mentor: self-efficacy/interest, and mentor: time/impatience) was conducted using the combined sample of 354 mentors. Model fit was assessed using indices recommended by Hu and Bentler (1999): the comparative fit index (CFI), the Tucker–Lewis index (TLI), the root mean squared error of approximation (RMSEA), and the standardized root mean squared residual (SRMR). SRMR values ≤.08 in combination with CFI and TLI values ≥.95 or RMSEA values ≤.06 indicate excellent model fit, but more liberal values are also used (e.g., CFI ≥ .90, Hoyle & Panter, 1995, RMSEA ≤ .08, Browne & Cudeck, 1993). The four-factor model for barriers showed an acceptable fit in the combined sample of $χ_{S B}^{2} (180) = 376.29$ , CFI = .92, TLI = .91, RMSEA = .056, and SRMR = .058. Cronbach’s αs for the trainee, institution, and mentor (items together) factors were .87, .84, and .77, respectively, in the combined sample.

Expected outcomes from mentoring

Perceived positive expected outcomes or benefits from mentoring and perceived negative expected outcomes or costs of being a mentor to trainees were measured with items from a 13-item scale developed by Anderson and colleagues (2020). This scale measures positive outcomes (6 items; e.g., “My own productivity is increased by mentoring”) and negative outcomes (7 items; e.g., “Mentoring takes valuable time away from my own tasks”) on a 5-point Likert-type scale ranging from 1 (strongly disagree) to 5 (strongly agree). Anderson et al. (2020) reported Cronbach’s α reliabilities of .69 for positive and .81 for negative expected outcomes from mentoring. Only 2 positive and 3 negative items were measured in both Sample 1 and Sample 2, so 5 expected outcomes items were included in the combined analysis.

Prevalence of SciComm problems among trainees

The perceived prevalence of problems in SciComm skills among trainees was measured with 1 item, “How prevalent a problem in your research field are trainees’ weaknesses in scientific communication skills,” rated on a 5-point scale (1 = not at all a problem to 5 = very significant problem). In the combined sample, validity for this single-item measure is supported by positive correlations with mentor perceptions of trainee barriers in SciComm (skills, effort, and attitude), as shown in Table 1.

Table 1.
Means, Standard Deviations, and Pearson’s Correlations for Study Variables.

M SD 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Expected mentoring outcomes

1 Positive 3.67 .67 1

2 Negative 3.17 .68 .03 1

Trainee barriers

3 Skills 3.47 .92 −.14 .16 1

4 Effort 2.99 .86 −.13* .23 .61 1

5 Attitude 2.98 .88 .05 .31 .45 .70 1

Mentor barriers

6 Skills/interest 2.06 .71 .03 .21 .19 .35 .29 1

7 Time/impatience 3.09 .87 .02 .37 .26 .34 .34 .44 1

Institutional/training environment barriers

8 Institutional/training barriers 3.16 .82 .07 .19 .17 .23 .21 .22 .21 1

SciComm problem prevalence and responsibility for instruction

9 Prevalence 3.67 .86 −.03 .18 .44 .32 .27 .06 .24 .17 1

10 Responsibility 4.05 .98 .15** −.11 −.02 −.04 .04 −.06 −.14* −.14* .03 1

Mentoring practices

11 Career 3.75 .62 .24** −.07 −.04 .01 .03 −.11* −.20 −.04 −.00 .07 1

12 Psychosocial 4.14 .55 .22 −.03 −.04 .01 .04 −.13* −.14 −.02 .08 .07 .61 1

13 Writing 3.79 .63 .10 −.13* .08 .05 .07 −.18 −.16 −.05 .06 .09 .57 .55 1

14 Presenting 4.18 .68 .09 .02 .17** .14* .11* −.24** −.12* −.02 .07 .02 .49 .46 .54 1

15 Speaking 3.84 .82 .06 −.01 .18 .11* .10 −.20 −.17 −.02 .11* −.02 .53 .44 .63 .64 1

Note. SciComm = scientific communication.

*p < .05. **p < .01.

Responsibility for teaching trainees SciComm skills

Mentors’ perceived responsibility for teaching SciComm was measured with 1 item, “Teaching basic SciComm skills to trainees is the responsibility of the mentor,” rated on a 5-point scale (1 = strongly disagree to 5 = strongly agree). In the combined sample, validity for this single-item measure is supported by positive correlations with mentor perceptions of positive expected outcomes from mentoring and negative correlations with negative expected mentoring outcomes, lack of time/impatience for mentoring in SciComm, and institutional/training environment barriers to SciComm mentoring, as shown in Table 1.

Mentoring practices in research and SciComm

Mentor perceptions of their mentoring practices were measured with a 21-item scale assessing mentoring in general research and the three modes of SciComm (Anderson et al., 2020; Lee et al., 2018). The five-factor scale measured (1) research technical-career mentoring related to research skill, productivity, and networking (5 items), for example, “Introduce my trainees to important people in the field”; (2) research psychosocial mentoring related to personal coping (4 items), for example, “Foster strategies for developing resilience, e.g. ability to recover from disappointments or problems”; (3) mentoring in writing (5 items), for example, “Give trainees confidence regarding their skills in scientific writing”; (4) mentoring in oral presentations (4 items), for example, “Set aside time to help trainees rehearse their oral presentations”; and (5) mentoring in impromptu speaking/conversation (3 items), for example, “Give useful feedback on the general speaking skills of trainees (e.g., grammar, vocabulary, pronunciation).” Perceived frequency of behaviors was rated on a 5-point scale (1 = never to 5 = always). Confirmatory factor analyses showed a satisfactory fit for the five-factor structure in the current combined sample, $χ_{S B}^{2} (177) = 354.87$ , CFI = .92, TLI = .90, RMSEA = .055, and SRMR = .052. Cronbach’s αs for the research and SciComm domains of the scale were .78 and .89 in the current combined sample.

Statistical Analysis

LPA (Hagenaars & McCutcheon, 2002; Muthén & Muthén, 1998–2014) was conducted with Mplus Version 7.3 (Muthén & Muthén, Los Angeles, CA) to create latent profiles of mentors. LPA is a person-centered, latent class approach used with continuous data, where nonobservable profiles of respondents are generated based on similar response patterns across multiple indicators.

Data from the Study 1 and Study 2 samples were combined and 10 indicators were used for the LPA, including only items used in both the Study 1 and 2 surveys. The indicators used were the averages of items for SciComm trainee barriers: skills (2 items), effort (4 items), and attitude (3 items); SciComm mentor barriers: skills/interest (5 items) and time/impatience (2 items); training environment barriers (6 items); the SciComm problem prevalence measure (1 item); the SciComm teaching responsibility measure (1 item); and mentoring expected outcomes, including positive outcomes or benefits (2 items) and negative outcomes or costs (3 items).

The number of profiles was determined using multiple fit indices: the Akaike information criteria (AIC; Akaike, 1987), the Bayesian information criterion (BIC; Schwarz, 1978) and the sample size–adjusted BIC, and the Lo–Mendell–Rubin (LMR) likelihood ratio test (Lo et al., 2001) and the bootstrap likelihood ratio test (BLRT; McLachlan & Peel, 2000). The latent profile models were built iteratively, allowing one additional class at a time. Improved fit was determined by examining a decreasing AIC, BIC, and adjusted BIC as well as a statistically significant (p < .05) LMR and BLRT, indicating a nonsignificant improvement in fit. Entropy, a measure of classification certainty, was also considered, where scores closer to 1.00 are more favorable (Wang et al., 2017). The number of members per class was considered to ensure that all classes included at least 1% of the sample to support generalizability and replicability, and the theoretical meaning of the classes was considered (Nylund et al., 2007).

To determine whether the final mentor profiles were related to mentor behavioral practices, the three-step approach in Mplus with the Bolck, Croon & Hagenaars (BCH) option (Asparouhov & Muthén, 2014) was used to compare the identified classes on five continuous perceived mentor practice outcome variables (i.e., mentoring on technical-career issues, psychosocial/personal coping support, and mentoring in scientific writing, presenting, and impromptu, informal speaking about science). The BCH function provides equality tests of class-specific means of the distal outcomes across classes while ensuring the stability of the initial class. χ² analyses were used to determine whether class membership differed by categorical, demographic variables; Analysis of variance was used for continuous demographic variables.

Results

Latent Profiles of Mentor Beliefs

Means, SDs, and Pearson’s correlations for the study variables are shown in Table 1. In the series of nested models that were fit and compared, to evaluate the number of mentor profiles or classes that best represented the study sample, four profiles (i.e., a four-class model) provided the best, meaningful fit to the data (see Table 2). Although the LMR test became nonsignificant at the four-class model (p = .641), indicating no significant improvement over the three-class, the four-class model provided a slightly smaller AIC, BIC, and adjusted-BIC than the three-class model, a significant BLRT value, and a meaningful sample size distribution across mentor profiles (13%, 29%, 11%, and 48%). It also provided valuable information about mentor differences that the three-class model did not capture. Thus, the four-class model was chosen over the more parsimonious three-class model and also over the more difficult to interpret five-class model.

Table 2.
Fit Statistics for Latent Profile Analysis of Mentor Beliefs in Scientific Communication.

Classes Log Likelihood AIC BIC (n Adjusted BIC) LMR (BLRT)p Values Entropy ClassProportions (%)

1 −5,884.29 11,828.58 11,944.66 (11,849.48)

2 −4,004.90 8,071.79 8,191.74 (8,093.39) <.001 (<.001) .77 36; 64

3 −3,928.41 7,940.83 8,103.34 (7,970.10) .005 (<.001) .80 13; 58; 30

4 −3,895.09 7,896.19 8,101.26 (7,933.12) .641 (<.001) .77 13; 29; 11; 48

5 −3,861.871 7,851.742 8,099.377 (7,896.342) .219 (<.001) .76 8; 27; 12; 18; 36

Note. AIC = Akaike information criteria; BIC = Bayesian information criterion; n adjusted BIC = sample size–adjusted BIC; LRT = likelihood ratio test; LMR = Lo–Mendell–Rubin LRT; BLRT = bootstrap LRT. The optimal number of classes is presented in bold.

We labeled the first mentor belief profile the low-problem group, which comprised 13% of the sample. The mentors in this belief profile reported the lowest prevalence of SciComm problems among trainees in their research field and the lowest levels of specific SciComm issues in trainees that would serve as barriers to mentoring; thus, they did not have problems with trainees in this area or were unaware of these types of problems. They also reported the most favorable outcome expectations (highest benefits and lowest costs) associated with mentoring trainees and few personal barriers to SciComm mentoring. The second mentor profile was labeled the high-problem group, and it comprised 29% of the sample. This group of mentors had the opposite pattern of perceptions of high problems, barriers, and costs. They reported high prevalence of trainee SciComm problems in their fields; the highest levels of trainee-related barriers, personal barriers as mentors, and institutional barriers; and the highest level of mentoring negative outcomes/costs. These negative beliefs were coupled with the lowest level of SciComm teaching responsibility among the four groups and a lower level of perceived positive outcomes/benefits from mentoring, compared to the low-problem group.

The third profile comprised 11% of the sample and was labeled the Problem-solver group. These mentors resembled the low-problem group in perceiving few personal barriers but were like the high-problem group in endorsing a high prevalence of trainee SciComm problems in their research field and high levels of trainee-related barriers. Among the four groups, they reported the highest level of SciComm teaching responsibility. Thus, they acknowledged the prevalence and scope of SciComm skill problems among trainees, in contrast to the low-problem group, but like the low-problem group, they had high self-efficacy regarding their SciComm teaching skill, interest, and time to solve or address the trainee problems.

The fourth mentor profile was the most common, comprising almost half the sample (48%). It was labeled the middle-of-the-road group because they showed an intermediate pattern of responses, positioned between the other groups on all 10 belief indicators. They acknowledged trainee problems (i.e., prevalence of SciComm problems and trainee-skill barriers) and their own personal barriers (i.e., mentor time), but they also perceived high benefits and lower costs to mentoring, with response levels in-between the ratings of the other groups. Figure 1 shows the means for each of the 10 indicators by mentor profile.

Figure 1.
Latent profiles of mentor beliefs. Note. Values on the y-axis represent mean scores on 1- to 5-point Likert-type scales on mentor beliefs. The x-axis shows the 10 indicators of mentor beliefs. TB = trainee barriers; MB = mentor-self barriers. Mentor-friendly index: high values desirable for mentor; all others are mentor-adverse indices: high values are not desirable for mentor.

Demographic Differences of Mentor Belief Profiles

Mentor profile groups did not differ by gender, primary language, discipline, or the number of trainees they were currently mentoring (Table 3). However, there were significant differences by race/ethnicity (p = .009), where more Hispanic and Asian mentors were in the high-problem or Middle-of-the-road groups, more Black/African American mentors were in the low-problem or Problem-solver groups, and more White and Other/Mixed ethnicity mentors were in the Middle-of-the-road group. There were also group differences on academic rank (p = .002) and age (p = .003). Most instructors were in the high-problem group, and most assistant professors were in the Middle-of-the-road group. Across rank, associate and full professors tended to be in either the high-problem or Middle-of-the-road group; however, the Problem-solver group contained more full professors. Sample differences were also present (p < .001), where more Sample 2 participants (national survey) were in the Middle-of-the-road group, while more Sample 1 participants (Houston survey) were in the high-problem or Middle-of-the-road groups.

Table 3.
Demographics of Each Profile Group.

Low-Problem Group, N (%) High-Problem Group, N (%) Problem-Solver Group, N (%) Middle-of-the-Road Group, N (%) p Value

Gender .144

Female 24 (14) 58 (34) 15 (9) 76 (44)

Male 21 (12) 44 (24) 24 (13) 91 (51)

Primary language .693

English 34 (14) 68 (29) 26 (11) 111 (46)

Other languages 11 (10) 34 (30) 13 (11) 56 (49)

Academic rank .002

Instructor 2 (13) 8 (50) 3 (19) 3 (19)

Assistant professor 9 (8) 25 (23) 8 (7) 66 (61)

Associate professor 16 (15) 34 (32) 6 (6) 50 (47)

Professor 16 (13) 34 (29) 22 (19) 47 (39)

Race/ethnicity .009

Hispanic 3 (14) 9 (41) 3 (14) 7 (32)

Asian American 7 (10) 30 (43) 9 (13) 24 (34)

Black/African American 3 (27) 2 (18) 4 (36) 2 (18)

White 29 (13) 54 (24) 20 (9) 124 (55)

Other 3 (13) 7 (29) 3 (13) 11 (46)

Sample <.001

Sample 1 14 (8) 62 (35) 27 (15) 76 (42)

Sample 2 31 (18) 40 (23) 12 (7) 92 (53)

Age in years

Mean (SD) 47.02 (9.93) 46.93 (9.80) 51.87 (10.79) 45.41 (8.78) .003

Current trainees mentoring

Mean (SD) 3.29 (3.64) 2.78 (2.21) 3.69 (3.29) 2.88 (1.93) .180

Note*. p Values for categorical variables were derived by χ² tests and those for continuous variables were derived by analysis of variance.

Associations of Mentor Belief Profiles With Mentoring Practices

Using the equality tests of means across classes, we observed significant differences by mentor belief profiles in mentoring practices, both in general research mentoring of trainees and in practices specific to SciComm (see Table 4; only significant between-profile comparisons are presented). The Problem-solver group reported providing all five types of mentoring (i.e., research technical-career, research psychosocial, writing, presenting, and speaking) significantly more frequently than did high-problem or Middle-of-the-road mentors. The Problem-solver mentors also reported significantly more mentoring for trainees in oral presentations and spontaneous speaking than did low-problem mentors. However, low-problem mentors reported more mentoring in general research, both technical-career and psychosocial, than did high-problem mentors, and more than Middle-of-the-road mentors in psychosocial, personal strategies. Finally, high-problem mentors reported significantly more frequent mentoring in impromptu, spontaneous speaking with trainees than mentors in the Middle-of-the-road group.

Table 4.
Between-Profile Comparisons on Mentor Practices.

Low- Problem Group High- Problem Group Problem-Solvers Group Middle-of- the-Road Group Overall Test Between-Profile Comparisons

Class 1 (N = 45) Mean (SE) Class 2 (N = 102) Mean (SE) Class 3 (N = 39) Mean (SE) Class 4 (N = 168) Mean (SE)

Research/career mentoring 3.92 (.11) 3.63 (.08) 4.07 (.12) 3.69 (.06) 12.38 3 > 2, 4; 1 > 2

Psychosocial mentoring 4.30 (.09) 4.06 (.07) 4.40 (.10) 4.08 (.05) 11.65 3 > 2, 4; 1 > 2, 4

Scientific communication mentoring

Writing 3.91 (.11) 3.75 (.08) 4.22 (.12) 3.66 (.06) 17.68 3 > 2, 4

Presenting 4.21 (.11) 4.23 (.08) 4.57 (.12) 4.02 (.07) 15.07 3 > 1, 2, 4

Spontaneous speaking 3.93 (.14) 3.89 (.09) 4.44 (.13) 3.60 (.08) 25.54*** 3 > 1, 2, 4; 2 > 4

Note. N = 336. Analyses were performed with the BCH procedure in MPlus Version 7.3. The overall significance test values are χ² values with degrees of freedom (df) = 3.

p < .05. p < .01. p < .001. Profile comparisons differ significantly at p < .05.

Discussion

The majority of mentors in academic medicine and scientific research work diligently to launch their protégés into their careers, but we have shown that mentors vary in their beliefs about problems among their trainees, in the mentoring barriers they themselves face, and in their mentoring practices, especially in the domain of SciComm. Using data from a national sample of mentors from 33 states and data from one of the largest medical centers in the United States, we found that faculty mentors displayed distinguishable patterns of beliefs that we have described in four mentor profiles: low-problem, high-problem, Problem-solvers, and middle-of-the-roaders.

Our findings indicate that, as expected, mentors’ attitudes and belief profiles were related to their mentoring behaviors and practices. This observation is consistent with psychological theory that explains how attitudes and beliefs form the basis of motivation for behaviors and practices (Bandura, 1977, 1997; Higgins & Sorrentino, 1990). Our findings also support the predicted relationships of SCCT, where perceived barriers, outcome expectations, interests, and self-efficacy beliefs influence choice behaviors (Lent et al., 1994). Mentors are likely to perform more mentoring behaviors when they hold more positive beliefs about the value and benefit of mentoring and see fewer personal barriers to their efforts (i.e., have higher self-efficacy or possess the necessary skills, interest, and time to specifically mentor in SciComm). This pattern was observed with Problem-solver mentors compared to those who are less positive about mentoring and see more barriers. Both high-problem and Middle-of-the-road mentors perceived having more personal mentoring barriers (i.e., have fewer skills, lower interest, and less time), institutional barriers, and high costs or negative outcomes from mentoring. As we would expect, these two profiles reported mentoring their trainees less often than the Problem-solver group in all five categories of mentoring practices. The high-problem group reported working with trainees more on spontaneous speaking (i.e., give feedback on grammar, vocabulary, and pronunciation; guide how to speak up and participate in scientific discussions) than Middle-road mentors. This finding is consistent with the belief of high-problem mentors that SciComm was a more prevalent problem among trainees in their research field, a view less common among Middle-of-the-road mentors. Post hoc analyses indicated that this difference was not due to the presence of more non-native English-speaking (NNE) trainees among mentors in the high-problem group. In fact, the Problem-solver group had the highest percentage of NNE trainees (Problem-solvers: 53% with >half of their trainees being NNE compared to 42% in High, 42% in Middle, and 17% in Low group; data not shown).

One interesting and useful finding was the emergence of the Problem-solver and low-problem profiles, two smaller groups of mentors that both reported few personal barriers in mentoring trainees about SciComm. They appear to have higher self-efficacy in overcoming mentor-related issues and hold high-benefit and low-cost beliefs about mentoring in general. However, the two groups differed notably in their acknowledgment of trainee SciComm problems: Problem-solvers reported high levels of trainee problems, second only to the high-problem group, while mentors with the low-problem profile either did not have these types of problems with their trainees or were unaware of them. This difference was related to mentoring practices in that low-problem mentors reported significantly less frequent mentoring in the five mentoring categories than Problem-solver mentors. Thus, it appears that either their perceptions of trainee SciComm problems as less serious or their perception of the lower frequency of these types of problems in their trainees influenced their mentoring behavior. Because the Problem-solver profile was associated with the highest levels of mentoring practices, we might conclude that the most beneficial attitude profile for mentors of trainees in the skill domain of SciComm appears to be a high recognition of trainee problems coupled with low perceived mentor-related barriers. In contrast, the attitude profiles for the large Middle-of-the-road group and the sizable high-problem group may underachieve the potential skill mastery of their trainees in SciComm and in other training domains for which mentoring behaviors are critical for development.

Our demographic findings suggest a relationship between mentor experience and their belief profile patterns. Problem-solvers, the seemingly optimal mentor group, included more full professors than other profiles. The Middle-of-the-road group, a less optimal group for mentoring trainees, contained the highest proportion of assistant professors, who are not only less experienced, but are arguably under intense pressure to advance their own careers, as manifest by their perception of greater personal barriers. Assistant professors also made up 25% of the less optimal, high-problem profile. Our findings support previous research showing that past experience as a mentor is associated with more mentoring (as reported by mentors), especially in regard to mentoring trainees in career advancement (Allen & Eby, 2004; Fagenson-Eland et al., 1997).

Limitations and Future Research

This study has a number of unique strengths, including the use of data from a sizable, national sample of faculty mentors in academic medicine and a focus on mentors of more advanced trainees in biomedical and biobehavioral research (i.e., doctoral students and postdoctoral fellows) than those included in previous studies. Our study expands the empirical evidence in mentoring research and the application of SCCT models to mentoring behaviors. It also builds upon our previous work, contributing unique insights from the perspective of mentors about mentoring beliefs and practices related to developing trainee SciComm skills.

The study also has limitations. Our use of cross-sectional data limits the interpretation of our results to associations rather than causal pathways between beliefs and behaviors. Combining the two data sets from Study 1 and the first wave of Study 2 greatly expanded the size and representativeness of our sample but prevented us from using the longitudinal design of Study 2 in analyses to determine whether the four profiles evolved or changed over time, as would be possible with latent transition analysis (Collins & Wugalter, 1992).

Self-reports from mentors may be biased or may not reflect their actual behaviors as observed by others (e.g., mentor practices, prevalence of SciComm problems in mentors’ research fields). It is also important to understand that the score differences found between groups are all relative, that is, there is no concrete score baseline. The mean rating level differences are not large, and the means for the “lower” groups are not particularly low, given our use of a 1–5 response scale. Hence, one could argue that all of the mentor groups are really “doing their job” fairly well. Moreover, the level of mentor experience was also hard to quantify since faculty rank and number of trainees do not necessarily define “seasoned” mentors.

Nonetheless, we found significant differences in important mentoring attitudes and in self-reported mentoring practice behaviors in the understudied domain of SciComm. These differences may suggest where interventions will have their greatest impact on mentor effectiveness, an important topic for exploration in future research.

Implications for Career Counseling

The implications of our findings for developing interventions to aid mentors as they guide trainees in their careers may be useful for faculty mentors, training programs in biomedical and biobehavioral research and academic medicine, career counselors, and others responsible for guiding trainees in their research career paths. Our data and profile analysis suggest that particular types of mentors may be more effective in facilitating this process, especially in the critical domain of SciComm. Specifically, our findings suggest the value of learning from the perspectives and strategies of Problem-solver type mentors, who acknowledge trainee SciComm problems, but display confidence in coping with barriers they and their trainees face in improving SciComm skills, as these may be useful aspects to address in mentoring interventions for all types of mentors. As trainees must be involved in successful research and mentored effectively to sustain their commitment to science (Straus et al., 2006), mentors are a crucial factor to bridge trainees from contemplating research careers that depend on publishing and presenting their work to actually initiating, pursuing, and persisting in those career paths successfully. This would be true for mentors of trainees pursuing academic careers and for trainees pursuing careers in other science-related sectors.

To supplement current, evidence-based mentor training strategies (Pfund et al., 2014; Pfund et al., 2015), interventions are needed to assist and inspire mentors in teaching and improving trainee SciComm skills and productivity. For example, instructional strategies are needed to inform mentors of the urgency of trainee communication challenges, to increase the awareness of mentors who may not recognize trainee problems, to encourage them to assess and address issues authoritatively, and to teach High-, Middle-, and low-problem mentors new mentoring strategies. Mentors may not be familiar with how to teach trainees the newer communication formats required for career success today (e.g., technology, entertainment, design (TED) talks, elevator speeches). The high level of responsibility and perceived mentoring benefits that we found across all groups is meaningful (see the “mentor-friendly” indices, Figure 1). Even the high-problem and Middle-of-the-road groups displayed these positive attitudes toward mentoring, although they were held back by perceived barriers. This may indicate that they are ripe for training and likely to benefit from more experience. Although teaching new strategies and improving mentors’ beliefs may not improve behavioral outcomes among all mentors, mentor training based on profiles of their beliefs and perceptions offers promise to improve the support available to trainees and, thereby, advance their career outcomes. Some new interventions are currently being developed and evaluated (e.g., NIGMS R25 GM12564).

Wide acceptance of the importance of written and oral communication skills for trainee success in academic and nonacademic research careers and the rapid advance of biomedical and clinically relevant research in the 21st century have produced an urgent need for effective strategies to help trainees develop SciComm skills and help mentors guide them in this challenging process (Cameron et al., 2013; Chang et al., 2015). In the context of dwindling financial resources in academia and increasing time demands on faculty, we need to help mentors shape their mentoring ability and commitment by teaching them to advise more effectively and efficiently.

	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
Expected mentoring outcomes
1	Positive	3.67	.67	1
2	Negative	3.17	.68	.03	1
Trainee barriers
3	Skills	3.47	.92	−.14**	.16**	1
4	Effort	2.99	.86	−.13*	.23**	.61**	1
5	Attitude	2.98	.88	.05	.31**	.45**	.70**	1
Mentor barriers
6	Skills/interest	2.06	.71	.03	.21**	.19**	.35**	.29**	1
7	Time/impatience	3.09	.87	.02	.37**	.26**	.34**	.34**	.44**	1
Institutional/training environment barriers
8	Institutional/training barriers	3.16	.82	.07	.19**	.17**	.23**	.21**	.22**	.21**	1
SciComm problem prevalence and responsibility for instruction
9	Prevalence	3.67	.86	−.03	.18**	.44**	.32**	.27**	.06	.24**	.17**	1
10	Responsibility	4.05	.98	.15**	−.11	−.02	−.04	.04	−.06	−.14*	−.14*	.03	1
Mentoring practices
11	Career	3.75	.62	.24**	−.07	−.04	.01	.03	−.11*	−.20**	−.04	−.00	.07	1
12	Psychosocial	4.14	.55	.22**	−.03	−.04	.01	.04	−.13*	−.14**	−.02	.08	.07	.61**	1
13	Writing	3.79	.63	.10	−.13*	.08	.05	.07	−.18**	−.16**	−.05	.06	.09	.57**	.55**	1
14	Presenting	4.18	.68	.09	.02	.17**	.14*	.11*	−.24**	−.12*	−.02	.07	.02	.49**	.46**	.54**	1
15	Speaking	3.84	.82	.06	−.01	.18**	.11*	.10	−.20**	−.17**	−.02	.11*	−.02	.53**	.44**	.63**	.64**	1

Classes	Log Likelihood	AIC	BIC (n Adjusted BIC)	LMR (BLRT)p Values	Entropy	ClassProportions (%)
1	−5,884.29	11,828.58	11,944.66 (11,849.48)
2	−4,004.90	8,071.79	8,191.74 (8,093.39)	<.001 (<.001)	.77	36; 64
3	−3,928.41	7,940.83	8,103.34 (7,970.10)	.005 (<.001)	.80	13; 58; 30
4	−3,895.09	7,896.19	8,101.26 (7,933.12)	.641 (<.001)	.77	13; 29; 11; 48
5	−3,861.871	7,851.742	8,099.377 (7,896.342)	.219 (<.001)	.76	8; 27; 12; 18; 36

	Low-Problem Group, N (%)	High-Problem Group, N (%)	Problem-Solver Group, N (%)	Middle-of-the-Road Group, N (%)	p Value
Gender					.144
Female	24 (14)	58 (34)	15 (9)	76 (44)
Male	21 (12)	44 (24)	24 (13)	91 (51)
Primary language					.693
English	34 (14)	68 (29)	26 (11)	111 (46)
Other languages	11 (10)	34 (30)	13 (11)	56 (49)
Academic rank					.002
Instructor	2 (13)	8 (50)	3 (19)	3 (19)
Assistant professor	9 (8)	25 (23)	8 (7)	66 (61)
Associate professor	16 (15)	34 (32)	6 (6)	50 (47)
Professor	16 (13)	34 (29)	22 (19)	47 (39)
Race/ethnicity					.009
Hispanic	3 (14)	9 (41)	3 (14)	7 (32)
Asian American	7 (10)	30 (43)	9 (13)	24 (34)
Black/African American	3 (27)	2 (18)	4 (36)	2 (18)
White	29 (13)	54 (24)	20 (9)	124 (55)
Other	3 (13)	7 (29)	3 (13)	11 (46)
Sample					<.001
Sample 1	14 (8)	62 (35)	27 (15)	76 (42)
Sample 2	31 (18)	40 (23)	12 (7)	92 (53)
Age in years
Mean (SD)	47.02 (9.93)	46.93 (9.80)	51.87 (10.79)	45.41 (8.78)	.003
Current trainees mentoring
Mean (SD)	3.29 (3.64)	2.78 (2.21)	3.69 (3.29)	2.88 (1.93)	.180

	Low- Problem Group	High- Problem Group	Problem-Solvers Group	Middle-of- the-Road Group	Overall Test	Between-Profile Comparisons
Research/career mentoring	3.92 (.11)	3.63 (.08)	4.07 (.12)	3.69 (.06)	12.38**	3 > 2, 4; 1 > 2
Psychosocial mentoring	4.30 (.09)	4.06 (.07)	4.40 (.10)	4.08 (.05)	11.65**	3 > 2, 4; 1 > 2, 4
Scientific communication mentoring
Writing	3.91 (.11)	3.75 (.08)	4.22 (.12)	3.66 (.06)	17.68**	3 > 2, 4
Presenting	4.21 (.11)	4.23 (.08)	4.57 (.12)	4.02 (.07)	15.07**	3 > 1, 2, 4
Spontaneous speaking	3.93 (.14)	3.89 (.09)	4.44 (.13)	3.60 (.08)	25.54***	3 > 1, 2, 4; 2 > 4

Footnotes

Authors’ Note

Cheryl B. Anderson and Shine Chang contributed equally and should be considered co-first authors.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by a grant from the National Institutes of Health, National Institute of General Medical Sciences, R01 GM085600, “Improving Retention of Minority Trainees: Mentoring in Scientific Communication Skills.”

ORCID iD

Cheryl B. Anderson

References

Akaike

(1987). Factor analysis and AIC. Psychometrika, 52(3), 317–332. https://doi.org/10.1007/bf02294359

Aldenderfer

M. S.

Blashfield

R. K.

(1984). Cluster analysis. Sage.

Allen

T. D.

Eby

L. T.

(2004). Factors related to mentor reports of mentoring functions provided: Gender and relational characteristics. Sex Roles, 50(1), 129–139. https://doi.org/10.1023/b:sers.0000011078.48570.25

Anderson

C. B.

Chang

Lee

H. Y.

Baldwin

C. D.

(2020). Mentoring barriers, expected outcomes, and practices in scientific communication: Test of the social cognitive model [Manuscript submitted for publication].

Asparouhov

Muthén

B. O.

(2014). Auxiliary variables in mixture modeling: 3-step approaches using Mplus (Mplus Web Notes: No. 15, Version 8). https://www.statmodel.com/download/webnotes/webnote15

Bakken

L. L.

Byars-Winston

Wang

M. F.

(2006). Viewing clinical research career development through the lens of social cognitive career theory. Advances in Health SciencesEducation: Theory and Practice, 11(1), 91–110. https://doi.org/10.1007/s10459-005-3138-y

Bandura

(1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84, 191–215. https://doi.org/10.1037/0033-295x.84.2.191

Bandura

(1986). Social foundations of thought and action: A social cognitive theory. Prentice Hall.

Bandura

(1997). Self-efficacy: The exercise of control. W.H. Freeman.

10.

Berk

R. A.

Berg

Mortimer

Walton-Moss

Yeo

T. P.

(2005). Measuring the effectiveness of faculty mentoring relationships. Academic Medicine, 80(1), 66–71.

11.

Browne

M. W.

Cudeck

(1993). Alternative ways of assessing model fit. In Bollen

K. A.

Long

J. S.

(Eds.), Testing structural equation models (pp. 136–162). Sage.

12.

Byars-Winston

Rogers

J. G.

(2019). Testing intersectionality of race/ethnicity × gender in a social-cognitive career theory model with science identity. Journal of Counseling Psychology, 66(1), 30–44. https://doi.org/10.1037/cou0000309

13.

Cameron

Collie

C. L.

Baldwin

C. D.

Bartholomew

L. K.

Palmer

J. L.

Greer

Chang

(2013). The development of scientific communication skills: A qualitative study of the perceptions of trainees and their mentors. Academic Medicine, 88(10), 1499–1506. https://doi.org/10.1097/ACM.0b013e3182a34f36

14.

Cameron

Lee

H. Y.

Anderson

Byars-Winston

Baldwin

C. D.

Chang

(2015). The role of scientific communication skills in trainees’ intention to pursue biomedical research careers: A social cognitive analysis. CBE: Life Sciences Education, 14(4), ar46. https://doi.org/10.1187/cbe.14-09-0152

15.

Cameron

Lee

H. Y.

Anderson

C. B.

Trachtenberg

Chang

(2020). The role of scientific communication in predicting science identity and research career intention. PLoS ONE, 15(2), e0228197. https://doi.org/10.1371/journal.pone.0228197

16.

Chang

Baldwin

C. D.

Cameron

(2015). Mentoring mentors in scientific communication for trainees. Academic Medicine, 90(3), 265. https://doi.org/10.1097/acm.0000000000000629

17.

Collins

L. M.

Wugalter

S. E.

(1992). Latent class models for stage-sequential dynamic latent variables. Multivariate Behavioral Research, 27(1), 131–157. https://doi.org/10.1207/s15327906mbr2701_8

18.

Fagenson-Eland

E. A.

Marks

M. A.

Amendola

K. L.

(1997). Perceptions of mentoring relationships. Journal of Vocational Behavior, 51(1), 29–42. https://doi.org/10.1006/jvbe.1997.1592

19.

Fernandez

Chen

Quan

Martinez

Flowers

Aronson

(2019). Evaluation of a medical student research and career development program to increase diversity in academic medicine. Academic Medicine, 94(8), 1220–1228. https://doi.org/10.1097/acm.0000000000002760

20.

Geraci

S. A.

Thigpen

S. C.

(2017). A review of mentoring in academic medicine. The American Journal of the Medical Sciences, 353(2), 151–157. https://doi.org/10.1016/j.amjms.2016.12.002

21.

Hagenaars

J. A.

McCutcheon

A. L.

(2002). Applied latent class analysis. Cambridge University Press.

22.

Higgins

E. T.

Sorrentino

R. M.

(1990). Handbook of motivation and cognition: Foundations of social behavior (Vol. 2, pp. xviii, 621). Guilford Press.

23.

Hoyle

R. H.

Panter

A. T.

(1995). Writing about structural equation models. In Hoyle

R. H.

(Ed.), Structural equation modeling: Concepts, issues, and applications (pp. 158–176). Sage (Reprinted from In File).

24.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6(1), 1–55. https://doi.org/10.1080/10705519909540118

25.

Lanza

S. T.

Flaherty

B. P.

Collins

L. M.

(2003). Latent class and latent transition analysis. In Schinka

J. A.

Velicer

W. E.

(Eds.), Handbook of psychology: Research methods in psychology (pp. 663–685). Wiley.

26.

Lee

H. Y.

Anderson

C. B.

Trachtenberg

Chang

Cameron

(2018). Validation of a scale for mentoring practices specific to scientific communication among trainees. Understanding Interventions, 9(2). https://www.understandinginterventionsjournal.org/issue/1000

27.

Lent

R. W.

Brown

S. D.

Hackett

(1994). Toward a unifying social cognitive theory of career and academic interest, choice, and performance. Journal of Vocational Behavior, 45(1), 79–122. https://doi.org/10.1006/jvbe.1994.1027

28.

Lent

R. W.

Ireland

G. W.

Penn

L. T.

Morris

T. R.

Sappington

(2017). Sources of self-efficacy and outcome expectations for career exploration and decision-making: A test of the social cognitive model of career self-management. Journal of Vocational Behavior, 99, 107–117. https://doi.org/10.1016/j.jvb.2017.01.002

29.

Lent

R. W.

Sheu

H.-B.

Miller

M. J.

Cusick

M. E.

Penn

L. T.

Truong

N. N.

(2018). Predictors of science, technology, engineering, and mathematics choice options: A meta-analytic path analysis of the social-cognitive choice model by gender and race/ethnicity. Journal of Counseling Psychology, 65(1), 17–35.

30.

Mendell

N. R.

Rubin

D. B.

(2001). Testing the number of components in a normal mixture. Biometrika, 88(3), 767–778. https://doi.org/10.1093/biomet/88.3.767

31.

Longo

D. R.

Katerndahl

D. A.

Turban

D. B.

Griswold

Hewett

J. E.

Dougherty

T. W.

Schubert

(2011). The research mentoring relationship in family medicine: Findings from the grant generating project. Family Medicine, 43(4), 240–247.

32.

McLachlan

Peel

(2000). Finite mixture models. Wiley.

33.

Muthén

L. K.

Muthén

B. O.

(1998–2014). Mplus user’s guide (5th ed.). Muthén & Muthén.

34.

Nylund

K. L.

Asparouhov

Muthén

B. O.

(2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14, 535–569. https://doi.org/10.1080/10705510701575396

35.

Pfund

House

S. C.

Asquith

Fleming

M. F.

Buhr

K. A.

Burnham

E. L.

Eichenberger Gilmore

J. M.

Charles Huskins

McGee

Schurr

Shapiro

E. D.

Spencer

K. C.

Sorkness

C. A.

(2014). Training mentors of clinical and translational research scholars: a randomized controlled trial. Academic Medicine, 89(5), 774–782. https://doi.org/10.1097/acm.0000000000000218

36.

Pfund

Spencer

K. C.

Asquith

House

S. C.

Miller

Sorkness

C. A.

(2015). Building national capacity for research mentor training: An evidence-based approach to training the trainers. CBE: Life Sciences Education, 14(2), ar24. https://doi.org/10.1187/cbe.14-10-0184

37.

Rodenhauser

Rudisill

J. R.

Dvorak

(2000). Skills for mentors and proteges applicable to psychiatry. Academic Psychiatry, 24(1), 14–27.

38.

Sambunjak

Straus

S. E.

Marusic

(2006). Mentoring in academic medicine: A systematic review. JAMA, 296(9), 1103–1115. https://doi.org/10.1001/jama.296.9.1103

39.

Sambunjak

Straus

S. E.

Marusic

(2010). A systematic review of qualitative research on the meaning and characteristics of mentoring in academic medicine. Journal of General Internal Medicine, 25(1), 72–78. https://doi.org/10.1007/s11606-009-1165-8

40.

Schaub

Tokar

D. M.

(2005). The role of personality and learning experiences in social cognitive career theory. Journal of Vocational Behavior, 66(2), 304–325. https://doi.org/10.1016/j.jvb.2004.09.005

41.

Schwarz

(1978). Estimating the dimension of a model. Annals of Statistics, 6(2), 461–464. https://doi.org/10.1214/aos/1176344136

42.

Straus

S. E.

Johnson

M. O.

Marquez

Feldman

M. D.

(2013). Characteristics of successful and failed mentoring relationships: A qualitative study across two academic health centers. Academic Medicine, 88(1), 82–89. https://doi.org/10.1097/ACM.0b013e31827647a0

43.

Straus

S. E.

Straus

Tzanetos

, & International Campaign to Revitalise Academic Medicine. (2006). Career choice in academic medicine: Systematic review. Journal of General Internal Medicine, 21(12), 1222–1229. https://doi.org/10.1111/j.1525-1497.2006.00599.x

44.

Vondracek

F. W.

Porfeli

(2002). Integrating person- and function-centered approaches in career development theory and research. Journal of Vocational Behavior, 61(3), 386–397. https://doi.org/10.1006/jvbe.2002.1881

45.

Wang

M.-C.

Deng

Yang

(2017). Performance of the entropy as an index of classification accuracy in latent profile analysis: A Monte Carlo simulation study. Acta Psychologica Sinica, 49(11), 1473–1482. https://doi.org/10.3724/sp.j.1041.2017.01473

46.

Wolfson

R. K.

Arora

V. M.

(2018). The physician-scientist pipeline: Return on investment. Medical Education, 52(5), 466–468. https://doi.org/10.1111/medu.13556

	Low- Problem Group	High- Problem Group	Problem-Solvers Group	Middle-of- the-Road Group	Overall Test	Between-Profile Comparisons
	Class 1 (N = 45) Mean (SE)	Class 2 (N = 102) Mean (SE)	Class 3 (N = 39) Mean (SE)	Class 4 (N = 168) Mean (SE)	Overall Test	Between-Profile Comparisons
Research/career mentoring	3.92 (.11)	3.63 (.08)	4.07 (.12)	3.69 (.06)	12.38**	3 > 2, 4; 1 > 2
Psychosocial mentoring	4.30 (.09)	4.06 (.07)	4.40 (.10)	4.08 (.05)	11.65**	3 > 2, 4; 1 > 2, 4
Scientific communication mentoring
Writing	3.91 (.11)	3.75 (.08)	4.22 (.12)	3.66 (.06)	17.68**	3 > 2, 4
Presenting	4.21 (.11)	4.23 (.08)	4.57 (.12)	4.02 (.07)	15.07**	3 > 1, 2, 4
Spontaneous speaking	3.93 (.14)	3.89 (.09)	4.44 (.13)	3.60 (.08)	25.54***	3 > 1, 2, 4; 2 > 4