Survey design and distribution

During March–April 2021, we used an anonymous Qualtrics survey to ask academic trainees (i.e., graduate students and postdoctoral scholars [hereafter postdocs]) currently working at American and Canadian Universities within environmental biology fields to self-identify their demographic information, current peer-reviewed publication records, and the effects of the pandemic on their feelings towards writing habits and productivity (full survey available in S1 File). We chose to restrict the sample frame to environmental biology trainees because STEM sub-fields can have large differences in advisor and department involvement and expectations for publishing [6].

We were approved by Yale University’s Institutional Review Board as exempt human subjects research under 45CFR46.104 (2)(ii) and did not track any personal identifying information or geographic location (e.g., IP address) to allow for honest and open answers. Participants gave written informed consent at the start of the survey (full survey in S1 File). Survey completion was voluntary, and we advertised it via social media (Twitter, Reddit, Facebook, Instagram), targeted emails to colleagues, and posted the survey twice on the ECOLOG-L listserv hosted by the Ecological Society of America. To increase the geographic diversity of the sample frame, we emailed 98 different department chairs or graduate coordinators from at least one major public R1 university in each U.S. state and Puerto Rico, U.S. Ivy League institutions, and five R1 Canadian Universities and asked them to distribute our survey among trainees. It is probable that trainees from non-R1 institutions filled out our survey after finding it on social media; however, we cannot determine the percentage of trainees from R1 or non-R1 universities. We did not require respondents to answer all questions; therefore, some respondents skipped questions or left fields blank. We eliminated these blank or skipped responses and categorized them as “NA”, but as a result, the sample size differs between questions. We denote sample size either in text or in figure captions to account for this.

Data analysis

We fit Bayesian multiple linear regressions in R version 4.0.2 (R Core Team 2020) using ‘rstanarm’ [26] to estimate how training years and identity (see Box 1) affected total publications using a Gaussian distribution. We estimated whether identity impacted the probability of a respondent indicating the pandemic affected their writing using a binomial distribution and logit link function. We then used ‘bayesplot’ [27] for visualization. For each model, we used weakly informative normal prior distributions (as suggested by experts, e.g., [28]) with a mean of zero and standard deviation of 2.5 and then allowed ‘rstanarm’ to automatically scale and center predictors and adjust scales of the priors during each run, which are also the default settings in ‘rstanarm’ [26]. We ran four chains for 10,000 iterations and discarded the first half as warm-up to obtain 20,000 simulations for analysis. We confirmed convergence using the Gelman-Rubin statistic (R_hat < 1.01) and by examining trace plots. None of the models had influential outliers as assessed by leave-one-out cross-validation (“loo”) in the ‘rstan’ package [29]. We report model coefficients as the median (β_hat in text, point estimates in Figs 1 and 3) and credible intervals (e.g., a 95% credible interval indicates there is a 95% probability that the true parameter lies within that range). Bayesian posterior distributions are generally more intuitive because they are probabilistic [30], so we used the thousands of iterations per model (i.e., posterior distribution) to look at the probability of the coefficient being positive or negative, which we report in text as the % probability a coefficient is < or > 0.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1.

The effect of years in training and identity on total publications for A) graduate students (n = 229) and B) postdoctoral scholars (n = 79) leading up to the COVID-19 pandemic. Multiple regression models suggest that identity of graduate students and postdoctoral scholars predicts publication totals. Each point is the median parameter estimate, thick lines are 50% credible intervals (CRIs), and thin lines are 95% CRIs. Graduate and postdoc yrs indicate the number of years in training (as a continuous variable). First gen = first in family to graduate from college; female = female respondents; BIPOC = Black, Indigenous, and/or a person of color; chronic condition = chronic health condition or disability; and ESL = respondents with English as a second language as categorical variables (yes/no).

https://doi.org/10.1371/journal.pone.0291124.g001

Our primary response of interest was the total number of trainee publications at time of survey. We have no way of distinguishing which publications were from pre-pandemic vs. during the pandemic, so we assume total publications represents a pre-pandemic metric for two reasons. First, our survey was open less than a year into the pandemic (~11 months since shut-downs began), so the pandemic represented a small fraction of total training time. Secondly, time from full manuscript to publication is often a year or more, so the total number of publications should mainly reflect pre-covid output. We used years as a graduate student and as a postdoc as continuous variables in the models to adjust for baseline differences in publication output due solely to career stage. For example, a first-year graduate student will likely have fewer publications than a third-year postdoc. Identity factors (defined fully in Box 1) were coded as whether the respondent was part of the group (yes = 1) or not (no = 0). Identity factors included whether someone was the first in their family to obtain a college degree (first generation, hereafter first-gen); gender identity; whether a trainee identified as BIPOC; whether the trainee had a chronic health condition or disability (hereafter, chronic condition); and whether the trainee’s first language was English or not. Unfortunately, we had too small of a sample size of individuals identifying as non-binary/other (e.g., non-binary, third gender) to get accurate model results (n = 8/311 respondents), so we eliminated all but those who identify as male or female from the analysis on publication totals. Recognizing that these individual survey respondents have shared insights from their experiences, we express deep gratitude for their voluntary contributions. Although these data were too minimal to statistically test how being gender non-binary affected publication outcomes, we encourage readers to view a summary of these responses (S6 Table).

Sample frame

We had 342 survey respondents—292 of whom finished the entire survey (85% completion rate)—from 149 subfields of environmental biology (S1 Fig). The most common subfield descriptions included “ecology” (n = 158), “biology” (n = 44), “evolutionary” (n = 34), and “plant” (n = 29). Our sample frame included graduate students from early to advanced stages (0–15 years, avg = 4.7 yrs, SD = 2.7, n = 231/311), and postdocs who had a similar range of experiences from early to advanced stages (0–9 yrs, avg = 1.9, SD = 1.9, n = 80/311).

Respondents represented a diversity of identities (S1 Table). The majority of graduate students (73%, n = 140/193) and postdoc respondents (61%, n = 46/76) self-identified as female (Box 1), while 24% and 37% (n = 47/193 and n = 28/76; graduate students and postdocs, respectively) self-identified as male, and a small number self-identified as non-binary, third gender, or other (3%, n = 6/193; and 3%, n = 2/76). Most respondents (92%) were previously or are currently at a university in the U.S. or Canada; therefore, our results mainly reflect trainee experiences in this region of North America.

Overall, 20% (n = 53/267) of respondents self-identified as BIPOC, comprising 24% of graduate students (n = 46/193) and 9% of postdocs (n = 7/74). First-gen college graduates made up 24% of both graduate student (n = 46/194) and postdoc (n = 19/78) respondents. Additionally, 18% of respondents reported having a disability or chronic health condition (Box 1), with 20% of graduate students (n = 39/193) and 13% of postdocs (n = 10/77). Finally, 18% (n = 36/199) and 23% (n = 18/78) of graduate students and postdocs, respectively, reported that English was their second language (ESL, Box 1).

Publishing output is predicted by training and identity

A substantial number (40%) of trainees had published prior to starting graduate school (n = 123/310). Postdocs had more first- (avg = 5.1 versus 1.7) and co-authored publications (avg = 5.6 versus 2.5) on average than graduate students and the variation in both metrics was higher among postdocs (S2 Fig). The strongest effect on publication total was the number of years spent as a trainee (Fig 1, S2 and S3 Tables). Each additional year in graduate training resulted in an additional half of a publication for graduate students (β_hat = 0.52, 100% of posterior samples [hereafter probability] > 0; Fig 1A) and slightly less for postdocs (β_hat = 0.31, 81% probability > 0; Fig 1B). Years spent as a postdoc had a strong effect. Each additional postdoc year resulted in two more publications (β_hat = 2.10, 100% probability > 0; Fig 1B).

Our models found contrasting effects of identity on publication totals between graduate students and postdocs (Fig 1, S2 and S3 Tables). First-gen identity was associated with a slight increase in publication totals for graduate students (β_hat = 0.43, 77% probability > 0), but had little effect on postdocs (β_hat = -0.96, 71% probability < 0). Female identity was a neutral factor for graduate students (β_hat = 0.36, 77% probability < 0), but reduced postdoc publication total by 2.5 (β_hat = -2.47, 95% probability < 0). BIPOC graduate students had slightly fewer publications than non-BIPOC peers (β_hat = -0.29, 70% probability < 0), but BIPOC postdocs published more papers than non-BIPOC peers (β_hat = 1.47, 72% probability > 0). Graduate students with chronic conditions had approximately one fewer paper (β_hat = -0.78, 90% probability < 0), but postdocs with chronic conditions had more papers (β_hat = 1.74, 80% probability > 0). Having ESL had little effect on publishing productivity of graduate students (βhat = -0.22, 66% probability < 0) and postdocs compared to non-ESL peers (β_hat = -0.89, 68% probability < 0), although ESL trainees were the only identity to provide qualitative descriptions of their difficult experiences in an open response (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. In the optional open question “Is there anything else you would like to add about your writing experiences?” trainees with ESL were the only ones who responded about how their identity affected writing and publishing.

While the model indicated that having ESL was not detrimental to publication output, these stated experiences of trainees with ESL highlighted the challenges.

https://doi.org/10.1371/journal.pone.0291124.t001

Differential impact of the COVID-19 pandemic on perceived writing time, productivity, and motivation

The majority of respondents (70%, n = 197/279) reported that the pandemic impacted their writing habits (Fig 2A). When respondents answered “yes” to the parent question “Has the COVID-19 pandemic impacted your writing habits?” they were directed to a series of nested questions about perceived writing time, productivity, and motivation during the pandemic. Most trainees (52%, n = 101/195) reported more or much more time for writing during the pandemic (Fig 2B), but a combined 75% of respondents (n = 147/196) reported that they felt less or much less productive during the pandemic (Fig 2C). Similarly, 76% (n = 148/196) of respondents reported feeling less or much less motivated to write during the pandemic (Fig 2D).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. The impact of the COVID-19 pandemic on trainee writing habits and perceptions of productivity and motivation.

A) When answering the question: “Has the COVID-19 pandemic impacted your writing habits?” the majority of respondents said yes. We then asked follow-up questions on perceived effects of the pandemic on time for writing, productivity, and motivation. B) While many respondents reported having more or much more time for writing, C) most respondents reported that they felt less or much less productive during the pandemic. D) Similarly, the majority of respondents reported feeling less or much less motivated.

https://doi.org/10.1371/journal.pone.0291124.g002

Participants’ reported experience with their writing habits during the pandemic was unequal across identities. Specifically, most female respondents (74%, n = 131/176) reported that the pandemic impacted their writing habits, compared to 61% of those who identify as male (n = 46/75) and 63% (n = 5/8) of non-binary trainees. Strikingly, 85% of BIPOC trainees reported that their writing habits were impacted by the pandemic compared to 67% (n = 142/212) of white peers.

Our models that accounted for identity and years in graduate school supported these results and suggested with a high degree of statistical support (S4 Table) that first-gen, female, or BIPOC graduate students were more likely to have the pandemic impact their writing habits (Fig 3A). First-gen graduate students had a 71% probability and female graduate students had a 77% probability of answering that their writing habits were affected. BIPOC graduate students had the highest (88%) probability of reporting the pandemic affected their writing. Number of years spent in graduate school was unrelated to the impact of the pandemic on writing (1 yr = 51% versus 5 yrs in graduate school = 53% probability). Interestingly, graduate students with ESL were strongly unlikely to report that the pandemic affected writing habits (100% posterior < 0), with only a 25% probability (Fig 3A, S4 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3.

Binomial multiple regression models suggest identity of graduate students and postdoctoral scholars influenced yes/no responses of A) graduate students (n = 229) and B) postdoctoral scholars (n = 79) to the question “Has COVID-19 impacted your writing habits?”. All estimates are in logit scale for ease of comparison. More positive values indicate a higher probability of answering “yes.” Points are the parameter estimate medians, thick lines are 50% credible intervals (CRIs), and thin lines are 95% CRIs. Graduate and postdoc yrs indicate the number of years in training (as a continuous variable). First gen = first in family to graduate from college; female = female; BIPOC = Black, Indigenous, and/or a person of color; chronic condition = chronic health condition or disability; and ESL = English is the second language of respondents coded categorically (yes/no).

https://doi.org/10.1371/journal.pone.0291124.g003

Among postdocs, those who have spent longer in postdoctoral training and who identify as female were the most likely to report that the pandemic affected their writing, whereas postdocs with chronic conditions and ESL were statistically unlikely to report an effect on writing habits (Fig 3B, S5 Table). Each additional year of postdoc training increased the probability of the pandemic affecting writing habits by 6% (96% posteriors > 0; e.g., 1 yr postdoc prob = 57% versus 4th yr prob = 76%). Female postdocs had a 71% probability of being affected by the pandemic (96% posteriors > 0). Postdocs identifying as first-gen and BIPOC reported little to no effect of the pandemic on writing (Fig 3B, S5 Table). Postdocs with ESL had only a 22% probability (95% posteriors < 0) of saying the pandemic affected their writing, and postdocs with chronic conditions a 21% probability (97% posteriors < 0).

Diversity, equity, and inclusion in scientific writing

Our results strongly demonstrate that female postdoctoral trainees and graduate students with chronic conditions had lower publication output pre-pandemic compared to their peers (Fig 1). The emphasis on high publication output in academia makes it likely that lower outputs at some point early in one’s career will have a lingering effect regardless of future progress. Interestingly, graduate students and postdocs often had contrasting results with respect to how identity affected publication output. Graduate students had similar publication records across genders, but female postdocs had 1.4 fewer publications than their male counterparts. Postdocs are more likely to have children than graduate students given that they tend to be older [31–33], and this could disproportionately affect the child-carrying and primary caregiving partner, although further research is needed to explore this. However, even for women without children, gender schemas [34] and deliberate and systematic obstacles (i.e., the hostile obstacle course [12]) make it harder for women to succeed in academia. This causes women to leave during the transition from graduate student to postdoc [19] and from postdoc to faculty [35].

First-gen graduate students had more publications than their non-first-gen peers but there was no difference at the postdoctoral level (Fig 1). Others have found first-gen biology doctoral students have similar experiences and outcomes as non-first-gen peers, matching our results [36], but our weak trend at the postdoctoral level shows that we need more data to understand how first-gen postdocs might respond to the pressures of postdoctoral positions and uncertainty in the job market. BIPOC trainees and trainees with chronic conditions tended to have fewer publications as graduate students but more as postdocs (Fig 1). People from HEGs finishing STEM PhDs are half as likely to have submitted a paper in the previous year compared to people from non-HEGs [6] and in-depth surveys suggest high writing anxiety among BIPOC biomedical graduate students [8]. Graduate students experience major institutional sources of stress during their degrees including role strain (e.g., being a teacher but also a student), mentor relationships, isolation, and funding concerns [37]; how these graduate school stressors may build upon the strain of being part of a HEG needs to be explored and addressed.

Trainees with ESL had slightly fewer publications than trainees with English as a first language. Although these trends were weak, many respondents explicitly mentioned the struggles of English as the default language of science in the open response section (Table 1). Research shows that trainees with ESL face enormous obstacles to scientific publishing [38] but writing intervention programs have helped this group gain confidence and skills in scientific writing [8]. Universities should invest in extra resources for trainees with ESL, including discipline-specific writing courses [39,40].

Our results, alongside previous work [6], demonstrate that certain groups of trainees had lower publication outputs than their peers leading up to the pandemic. Academia is currently structured to differentially impact identity groups under baseline conditions [15,22]. The pandemic has highlighted and, in some cases, accentuated the unjust differences that exist for some community members. Although significant and global disturbances like the pandemic are uncommon, the productivity disruptions mirror more singular events that individuals experience such as grief, illness, or unpredictable childcare, so equitably adjusting academic metrics of success will have long-reaching positive effects on future trainees seeking tenure-track positions.

COVID-19 pandemic and identity

Graduate students and postdocs have navigated the impacts of the pandemic at a critical time in their careers when publishing matters most for their future career success. While recent studies have examined the effect of the pandemic on early-career faculty [24,25,41], our study is the first to our knowledge that quantitatively examines the impact of the pandemic on trainees (but see Suart et al. [42] for a qualitative overview). The differential writing outputs across trainees may be exacerbated by the ongoing pandemic. The majority of trainees reported that the pandemic has affected their writing habits, with most also feeling less productive and motivated to write (Fig 2). Thus, while these trainees felt that they had more time for writing during the pandemic, it was not actually conducive to writing productivity or motivation. This finding is not surprising; other studies have explored the connection between negative mental health, feelings of apathy, and lower productivity and motivation [42]. Mental health was already a concern for trainees [43], and the pandemic will likely worsen anxiety and depression [44,45]. Furthermore, 70% of respondents preferred working in non-home environments that were unavailable during lockdowns. Many trainees were likely adjusting to working from home while living in smaller shared spaces with roommates or family. Trainees who are parents had the additional challenge of caring for their children, and women in particular have been disproportionately affected by increased childcare or other family caretaking responsibilities stemming from the pandemic [24,25,46,47].

However, the pandemic has not affected all trainees equally. While there was no effect of years spent in graduate school, each additional year of postdoctoral work strongly increased the reported effect of pandemic on writing habits (Fig 3). Most postdoc positions in environmental biology fields are for 2–3 years at most, so losing 1.5 years of access to networking, lab work, and field sites was devastating, especially when paired with grim job prospects [48,49] and the importance of publishing during postdoctoral work [2]. Despite calls for more support and contract extensions [50,51], many universities did not accommodate postdocs. Postdocs are deeply concerned that the pandemic has worsened their overall career prospects [51,52].

We also found evidence that female trainees were disproportionately affected. The experiences of non-binary and other gender trainees were similar (S5 Table), but more data are needed to more fully characterize their experience throughout the pandemic. Other studies have also reported gender inequalities in research productivity during the pandemic, including women having less time for research [25] and being underrepresented in COVID-19-related research and authorship [46,53,54]. Actions such as creating niche funding opportunities to support women and gender minorities and developing flexible working schedules for those with childcare responsibilities have been proposed as strategies for supporting these early-career scientists [24]. Additionally, creating institutional resources for writing (e.g., courses, workshops, support groups) could help mitigate the effects of unexpected disturbances.

Disparities in the pandemic’s impact on writing were most pronounced for BIPOC respondents. A striking 85% of BIPOC trainees reported that the pandemic affected their writing habits, and overwhelmingly felt unproductive and unmotivated to write. The timing of the pandemic also coincided with the collective trauma felt by Black Americans in response to police brutality and the murders of Breonna Taylor and George Floyd, among others. Furthermore, violence directed at Asian Americans has been on the rise in the U.S. during the pandemic [55]. We recognize that this violence transcends workspaces and can manifest in field work settings as well [15,56]. Although our data do not differentiate between the public health and social justice crises, the compounding issues of the pandemic, racism, and the rise of racially-charged violence likely contributed to the decreased writing motivation felt by BIPOC trainees. It is critical for BIPOC trainees to take care of their mental health [57], and for academic communities to support and retain BIPOC trainees by dismantling institutional white supremacy and creating inclusive environments where BIPOC scholarly excellence is celebrated [15,21,22,58]. Future research should aim to disentangle the role of longstanding inequities in academia and the effects of recent events (e.g., the pandemic and the social justice movement of 2020) on work output.

First-gen graduate students were also more likely to report a disruption to writing during the pandemic, which may be due to the mental toll of having more family and friends with low socioeconomic status who struggled more throughout the pandemic [59]. Surprisingly, trainees with ESL and those with chronic conditions were statistically less likely to report the pandemic affected their writing habits (Fig 3). In the case of trainees with ESL— many of whom are likely international trainees— the pandemic prevented many from being able to see their families. However, these trainees already were physically distanced from their families even prior to the pandemic, which could explain why they were less likely to report the pandemic affected their writing habits given that they could have already been experiencing isolation. More research is needed to understand why these trainees reported less interruption to writing habits.

Limitations of the study

Given differences in publishing and research between STEM fields, our results most prominently reflect the experiences of environmental biology trainees. While our findings can inform our understanding of the pandemic’s impact, future research is needed to explicitly explore the effects of the pandemic on trainees across a wider range of identities and STEM fields. We chose an anonymous survey mechanism to maximize and standardize our sample size, but in doing so, we lack the nuance we could have included if our study had also allowed for follow-up interviews. The survey questions were subject to interpretation and reporting differences among respondents. Because of this, our analyses had to be neutral; for example, we asked trainees whether the pandemic impacted their writing, however we cannot fully conclude it was a negative impact, despite follow-up questions showing a perceived lack of motivation and productivity. Additionally, although ESL trainees did not report effects of the pandemic on writing habits, it could very well be that cultural stigmas around discussing difficulties prevented them from answering our questions honestly, especially considering the free responses indicated writing in general was much more challenging for ESL trainees (Table 1). The survey also lacked the ability to differentiate between ESL trainees who reside in their home country and international students. Furthermore, caretaker status would have been an informative metric due to school and care center closure during the pandemic, as this likely had a very strong influence on productivity. Future research should disentangle these effects further.

The intersection of identities, such as race and gender, is important [17,60] and undoubtedly would also add new insight and a fine-tuned understanding of how different identities were affected by the pandemic. To adequately explore intersectionality, we would have had to specifically target responses from people at the intersection of the identities of interest. As presented here, the five aspects of identity in our survey (race, gender, first-gen, ESL, and chronic conditions/disability) could be assessed additively. Although this does not quantify interactive effects, the sum of identities provides the minimum effect one could expect on publishing output. We encourage future studies to quantitatively assess intersectionality of different identities and include more nuance in their assessments.

Equity in academia

We show that the pandemic disproportionately affected historically excluded environmental biology trainees. Female, BIPOC, and first-gen trainees were much more likely to indicate that the pandemic impacted their writing habits. These writing interruptions may linger on the CVs of those who were trainees during the pandemic, manifesting as disparities in publication counts compared to more productive colleagues and competitors. To recover from the pandemic, we should reconsider how to evaluate job candidates in future hiring. The current academic evaluation metrics are not equitable because they do not (1) account for the different input into individual publications, (2) give equal weight to the non-publishing aspects of academia (e.g., teaching, mentoring, JEDI), or (3) account for the means by which academics achieve their outputs [15]. Research projects that are theory-, lab-, or field-based require different inputs from inception to publication. Furthermore, some scientists may encounter hostile situations during field work that can delay progress [56]. Search committees should consider these project inputs when evaluating candidates beyond total publication counts and journal impact factors.

Thus, while peer-reviewed publications maintain the quality and knowledge-base of science, the non-publishing aspects of academia should be given more weight in the evaluation process. For example, on its own, a candidate’s publication output does little to impact the issue of undergraduates from HEGs leaving STEM degrees [61] or graduate students leaving academia. However, a candidate with training and experience in inclusive teaching practices and active participation in JEDI initiatives is likely going to improve their prospective institution by being both a good instructor and mentor. Finally, while a candidate can have an exceptional track record of publications and grants, they may have achieved those outputs while harming community members around them [15]. The differential impacts of the pandemic on HEGs, along with the call to create and retain a diverse and equitable faculty is an opportunity for universities to reevaluate and adapt their current metrics. For example, non-white, non-male, and first-gen faculty are more likely to participate in JEDI activities [62]; if trainees engaging in and prioritizing institutional service and JEDI initiatives have similar identities, these individuals may have less time to dedicate to writing. To diversify academia, including non-publishing aspects of a candidate as measures of success would help create a more equitable evaluation process during hiring.