PaperlayerView exact revision on arXiv

Source-linked AI summary

Loss of New Ideas: Potentially Long-lasting Effects of the Pandemic on Scientists

Jian Gao, Yian Yin, Kyle R. Myers, Karim R. Lakhani, Dashun Wang

arXiv:2107.13073v1cs.DLphysics.soc-ph

TL;DR

Evidence remains limited on how pandemic-related declines in new research projects vary across scientists. The paper examines new collaborations in publication data and finds that scientists were less likely to pursue new projects even as work time recovered.

Abstract

from arXiv · show

Extensive research has documented the immediate impacts of the COVID-19 pandemic on scientists, yet it remains unclear if and how such impacts have shifted over time. Here we compare results from two surveys of principal investigators, conducted between April 2020 and January 2021, along with analyses of large-scale publication data. We find that there has been a clear sign of recovery in some regards, as scientists' time spent on their work has almost returned to pre-pandemic levels. However, the latest data also reveals a new dimension in which the pandemic is affecting the scientific workforce: the rate of initiating new research projects. Except for the small fraction of scientists who directly engaged in COVID-related research, most scientists started significantly fewer new research projects in 2020. This decline is most pronounced amongst the same demographic groups of scientists who reported the largest initial disruptions: female scientists and those with young children. Yet in sharp contrast to the earlier phase of the pandemic, when there were large disparities across scientific fields, this loss of new projects appears remarkably homogeneous across fields. Analyses of large-scale publication data reveal a global decline in the rate of new collaborations, especially in non-COVID-related preprints, which is consistent with the reported decline in new projects. Overall, these findings highlight that, while the end of the pandemic may appear in sight in some countries, its large and unequal impact on the scientific workforce may be enduring, which may have broad implications for inequality and the long-term vitality of science.

The pandemic’s impact remains significant but has shifted substantially in nature

These findings trace how the pandemic’s effects on scientists shifted from work-time and output disruptions toward fewer new projects and non-COVID collaborations. They also consider how stable short-term output measures may obscure lasting, unequal impacts.

The pandemic’s impact remains significant but has shifted substantially in nature

  • The average work-time shortfall narrowed from 7.1 hours per week in April 2020 to 2.2 hours per week in January 2021, while 2020 publication and submission counts were only moderately lower than in 2019.The work-time impact relative to pre-pandemic levels changed from -13.8% in April 2020 to -4.3% by January 2021.Source: The pandemic’s impact remains… · pp. 2–3

Fewer new projects initiated during the pandemic

  • The share reporting no new projects increased from about 8.9% in 2019 to about 27.0% in 2020, roughly threefold; new projects declined 36.2%, equivalent to about one project per scientist in 2020.Source: Fewer new projects initiated during… · p. 3

Field- and group-level differences

  • Being female and having young children were the strongest demographic features associated with project declines, matching groups reporting the largest initial disruptions.The paper suggests these declines may further exacerbate unequal pandemic effects, especially for scientists who did not pursue COVID-related research in 2020.Source: Field- and group-level differences · p. 4
  • Project declines were substantial across disciplines; only biochemists had significantly lower-than-average declines after adjustment for individual-level features.The authors report that no scientific field was immune to the loss of new projects.Source: Field- and group-level differences · pp. 4–5

An early sign for the loss of new projects: Decreases in new collaborations

  • New collaborations on COVID papers rose roughly 40% above 2019 levels, while new collaborations on non-COVID papers fell about 5% below 2019 levels.The publication-dataset measurements were broadly consistent with self-reported changes in new collaborations.Source: An early sign for the loss of new… · p. 5
  • Non-COVID preprints showed a stronger decline in new collaborations in the second half of 2020, and collaboration and new-project rates remained strongly correlated after controls.The authors note that other social and institutional factors may also contribute, so this pattern is supporting evidence rather than proof of cause.Source: An early sign for the loss of new… · p. 6

Concluding remarks and discussions

  • Scientists’ work time had largely recovered by early 2021 and submission and publication decreases appeared minor, yet they remained substantially less likely to pursue new projects.The authors suggest the pandemic’s impact may last longer than commonly imagined.Source: Concluding remarks and discussions · p. 6
  • Stable or rising submissions and publications may reflect work on existing topics, write-ups of prior research, or revived legacy projects, making output counts an incomplete measure of productivity.The authors caution that short-term metrics may mask longer-lasting effects, with implications for scientific inequality and long-term vitality.Source: Concluding remarks and discussions · pp. 6–8
  • Potential contributors include reduced access to facilities and field sites, fewer in-person interactions and training, and intermittent lockdowns, but the evidence is correlational.These disruptions could impede new hypotheses and projects, while the study leaves the causes of the decline unresolved.Source: Concluding remarks and discussions · pp. 7, 9
  • The authors argue that science policy should support in-person interactions and new collaborations, including through bridge funds, while addressing scientists’ mental-health challenges.They note that virtual events may not facilitate social functions involved in forming new ideas.Source: Concluding remarks and discussions · pp. 7–8
  • The decline in new projects was especially pronounced for women and caregivers of young children, while remaining homogeneous across almost all fields.The authors suggest this cross-field pattern may reflect disruptions common across disciplines.Source: Concluding remarks and discussions · pp. 7–8

Competing interests

The authors declare no competing interests; the study received IRB approval at Northwestern and Harvard, with informed consent from participants.

Competing interests

Human research participants

Data availability

  • Individual-level survey data are restricted; researchers may contact the authors about access, while raw Web of Science and Dimensions.ai data must be requested from those sources.Source: Data availability · pp. 9–10
  • Code to reproduce all plots and statistical analyses will be made freely available.Source: Data availability · p. 10

Additional information

Supplementary Information · S1. Survey Sampling and Recruitment

The surveys drew on randomly sampled Web of Science corresponding-author email addresses from U.S. and European scientists, then used personalized email invitations with a consistent recruitment process across survey waves.

1.1 Web of Science corresponding authors

  • Web of Science was chosen for its broad publication and citation coverage and systematic corresponding-author email records.Source: §1.1 · p. 16
  • The sampling frame began with 2010-2019 papers, excluded journals in the bottom 25% of their WoS category, and retained emails tied to at least two papers and a recent U.S. or European affiliation.Source: §1.1 · p. 16
  • Filtering yielded about 1.5 million unique email addresses: about 521,000 in the U.S. and 938,000 in Europe.Source: §1.1 · p. 16
  • For April 2020, researchers sampled 280,000 U.S. and 200,000 European addresses, oversampling the U.S.; for January 2021, they sampled 140,000 and 100,000, respectively.Source: §1.1 · p. 16

1.2 Participant recruitment

  • Both surveys used the same invitation-based recruitment process, with personalized email text tailored to each survey’s timing.Source: §1.2 · p. 16
  • The January 2021 email described the study’s aim and requested a five-minute survey; Northwestern’s IRB approved protocol STU00212699.The email also referenced prior Nature Human Behavior research with Kyle R. Myers and Karim R. Lakhani at Harvard’s Laboratory for Innovation Science.Source: §1.2 · p. 16

S2. Survey Instrument and Sampling Approach · 2.1 Survey questions

The surveys collected demographic, professional, and work-time information; the January 2021 survey retained these items and added questions about COVID-related research and changes in research outputs. They also asked respondents to compare research activity and publication measures across 2019 and 2020.

2.1 Survey questions

  • The January 2021 survey retained the April 2020 survey’s demographic, professional, and time-allocation questions, adding COVID-related research and 2020-versus-2019 output questions; demographic responses were optional.Source: §2.1 · p. 17
  • Respondents reported their position, field, primary institutional affiliation, institution closure status, and tenure status.Source: §2.1 · p. 17
  • Demographic items asked about gender, age, number and ages of dependents, and cohabitation status.Source: §2.1 · p. 17
  • Respondents estimated weekly job-related hours around January 2020 and current weekly hours in January 2021.Source: §2.1 · pp. 17–18
  • The January 2021 survey asked whether respondents worked on COVID-19-related research in 2020 and how many new projects they started in 2019 and 2020.Source: §2.1 · p. 18
  • Respondents reported how many people they collaborated with for the first time in 2019 and 2020.Source: §2.1 · p. 18
  • For research publications, defined to include research-focused works such as journal articles, conference proceedings, patents, and books, respondents reported submissions in 2019 and 2020.Source: §2.1 · p. 18
  • Respondents also reported how many peer-reviewed research publications they published in 2019 and 2020.Source: §2.1 · p. 18

2.2 Research field definitions

Survey research fields follow national survey classifications, including the U.S. Survey of Doctorate Recipients, aggregated to ensure adequate sample sizes and expanded to cover additional university disciplines.

2.2 Research field definitions

  • Research fields were based on classifications used in national surveys such as the U.S. Survey of Doctorate Recipients and aggregated to ensure sufficient sample sizes.Source: §2.2 · p. 18
  • The field scheme added Business Management, Education, Communication, and Clinical Sciences.Source: §2.2 · p. 19
  • These additions reflected major university schools or disciplines that did not map readily to default SDR fields, such as medical specialties excluded from SDR Health Sciences.Source: §2.2 · p. 19

2.3 Survey data sampling approach and basic statistics … 4.1 The Dimensions publication data

The section describes how the two survey samples were selected, how regressions were specified, and how publication records and COVID-related papers were identified in Dimensions. It reports a base sample of 2,447 January 2021 survey respondents and 216,187 COVID-related papers indexed for 2020.

2.3 Survey data sampling approach and basic statistics

  • The April 2020 survey analysis retained 4,535 respondents after restricting eligibility to faculty or research-engaged retired PIs and excluding missing key variables and for-profit employees.The survey received 480,000 email invitations; 8,447 people continued past consent. Missing variables were not imputed.Source: §2.3 · p. 19
  • The January 2021 survey’s base sample included 2,447 faculty or PI respondents after excluding missing working-time data; analyses further excluded missing outcome or key variables as needed.The survey was sent to 240,000 email addresses, and 4,672 people continued past consent.Source: §2.3 · p. 19

3.1 Lasso selection and post-Lasso regression

  • For field or demographic comparisons, Lasso selected controls from field, professional, and demographic indicators and expanded continuous variables using inverse hyperbolic sine, square, and cubic transformations and interactions.The selection process aimed to exclude covariates that did not improve model fit.Source: §3.1 · p. 19
  • The two-step adaptive Lasso used ten-fold cross-validation, then selected variables entered post-Lasso OLS with heteroskedastic-robust standard errors.The implementation used Stata 16 defaults for initial guesses, tuning parameters, and stopping criteria.Source: §3.1 · pp. 19–20

3.2 Ordinary Least Squares (OLS) and probit regression

  • OLS with robust standard errors examined associations between changes in new projects and collaborators, while probit modeled the probability of reporting COVID-related research using professional, demographic, and field variables.The probit outcome was a binary indicator of whether scientists reported working on COVID-related research in 2020.Source: §3.2 · p. 20

4.1 The Dimensions publication data

  • The Dimensions [10] corpus, retrieved in March 2021, covered papers and preprints through 2020 with publication, author, venue, field, DOI, and publication-type metadata; authors were pre-disambiguated.Dimensions was used for its systematic coverage and relatively short publication-data lag.Source: §4.1 · p. 20
  • Dimensions indexed 216,187 COVID-related papers published in 2020, identified using a search query following prior work and a query suggested by Dimensions [12].The search covered all papers indexed by Dimensions.Source: §4.1 · p. 20

4.2 Matching survey respondents to Dimensions authors … 7.1 Group-level and field-level changes in new projects

New research projects declined in 2020, with larger reported declines among scientists not pursuing COVID-related research and some demographic groups. Work time showed signs of recovery, and survey publication counts aligned with Dimensions data.

4.2 Matching survey respondents to Dimensions authors

  • Matching survey respondents to Dimensions authors linked 2,141 of 2,447 January 2021 respondents, an approximately 87% match rate.The matching process used DOI-linked papers from WoS and matched first and last names; publication counts were then calculated for matched respondents.Source: §4.2 · pp. 20–21

5.1 Measuring new collaborations in publication data

  • The new-collaboration rate is the fraction of author pairs on a paper with no prior collaboration, measured for teams of 50 or fewer authors.Pairs were classified using their earliest collaboration year, based on disambiguated Dimensions author pairs from publications since 1950; publication month and type were also used in further analyses.Source: §5.1 · p. 21

6.1 Changes in total work hours

  • Weekly work time rose from about 44 hours in April 2020 to 47 in January 2021, while the average individual decline narrowed from about 7 to 2 hours.The percentage change in work hours rose from about -14% to about -4%.Source: §6.1 · p. 21

6.2 Changes in publication, submission, and projects

  • New projects declined 26% in 2020 versus 2019, more than submissions (-5%) or publications (-11%), with all three output metrics decreasing on average.Average new projects fell by 0.7 (about 2.5 to 1.8), submissions by 0.9 (about 6.2 to 5.3), and publications by 0.4 (4.5 to 4.1); logged-value and absolute-change analyses also supported the declines.Source: §6.2 · pp. 21–22

6.3 Comparison of the publication data in survey and Dimensions

  • Survey and Dimensions publication counts had Spearman rank correlations of 0.71 for 2019 and 0.75 for 2020.A linear fit explained about 57% of variance; Dimensions counts tended to be higher for more productive respondents, and positive correlations remained robust using logged values.Source: §6.3 · p. 22

7.1 Group-level and field-level changes in new projects

  • Scientists not pursuing COVID-related research reported a 57% decline in new projects versus the overall 27%; female scientists and those with young dependents were also more affected.Scientists in laboratory-dependent fields reported declines of 38–44%, while social science fields reported smaller declines.Source: §7.1 · p. 22
  • A Lasso regression selected features predictive of changes in new projects while controlling for other factors.The regression results are reported in Fig. 2 of the main text and S8.2.Source: §7.1 · pp. 22–23

7.2 Robustness checks using alternative measures … 9.1 COVID-related research by survey data

Across surveys and publication records, scientists reported declines in new collaborators and projects, while publication trends differed between aggregate output and individual authors. COVID-related research participation varied more across fields than across professional or demographic groups.

7.2 Robustness checks using alternative measures

  • Alternative logged and raw-change measures largely reproduce project-loss results: demographic differences remain sharp, field differences small, with biochemists the exception.With absolute changes, Lasso additionally selected having 6–11-year-old dependents together with having 0–5-year-old dependents.Source: §7.2 · p. 23

7.3 Changes in new collaborators

  • 3.9 new collaborators per scientist in 2019 fell to 2.9 in 2020, a mean relative decline of 17%; the share reporting no new collaborators rose from about 15% to 35%.Source: §7.3 · p. 23
  • Among non-COVID scientists, average new collaborators halved from about 4.4 in 2019 to 2.2 in 2020, while COVID scientists averaged a 15% increase.The reported relative change was a 32% reduction for non-COVID scientists and a 15% increase above the 2019 level for COVID scientists.Source: §7.3 · p. 23

7.4 Associations between new projects and new collaborators

  • New collaborator counts significantly and positively predicted new project counts after controls, and absolute- and percentage-change measures showed the same positive association.Source: §7.4 · p. 24

8.1 Trends of total and average publications

  • Dimensions data show total articles and preprints increased in 2020, including after COVID-related publications were excluded.Source: §8.1 · p. 24
  • Among authors publishing yearly in 2015–2019, COVID authors increased article output in 2020 while non-COVID authors' average fell slightly; results held across 1-, 5-, and 10-year windows.Across all authors, non-COVID authors' average article output in 2020 was very close to 2019.Source: §8.1 · p. 24

8.2 Changes in new collaborations measured by publication records

  • Publication records show 2020 new-collaboration fractions fell for non-COVID papers and rose for COVID papers; non-COVID preprint ratios declined in the year's second half.Earlier annual ratios were largely stable and slightly above 1; the pattern was robust across team sizes and tended to be stronger in smaller groups.Source: §8.2 · pp. 24–25

9.1 COVID-related research by survey data

  • About 34% of respondents worked on COVID-related research in 2020, with participation varying more by field than by professional or demographic factors.Social and clinical scientists had higher participation, physicists lower participation, and probit results supported the field-associated pattern.Source: §9.1 · p. 25

SI Figures

The supplementary figures document survey- and publication-based measures of scientists’ work time, research output, new projects, collaborations, and COVID-related research activity. The captions describe the comparisons and analyses shown but do not provide numerical results.

Work time

  • Figure S1 charts pre- and post-pandemic work hours and their changes across the April 2020 and January 2021 surveys.It includes weekly totals, changes from pre-pandemic levels, and average percentage changes.Source: Figure S1 · p. 26

Research output

  • Figures S2 and S3 compare 2019 and 2020 publications, submissions, and projects using relative and absolute changes.S2 also shows logged-value analyses; S3 caps displayed changes at −5 and 5.Source: Figure S2 · p. 26, Figure S3 · p. 27

New projects

  • Figures S5 and S6 examine changes in new projects across demographic, professional, and research-field groups, including adjusted regressions and alternative measures.The analyses include a non-COVID indicator; Figure S6 also presents logged and raw-value differences, with significance levels described in its caption.Source: Figure S5 · p. 28, Figure S6 · p. 28, SI Figures · p. 29

Survey-based collaborations

  • Figures S7 and S8 show survey-based changes in new collaborators, including comparisons between scientists who did and did not work on COVID-related research.The displays include distributions, average counts, absolute changes, and percentage changes.Source: Figure S7 · p. 29, Figure S8 · p. 30

Publication-based analyses

  • Figures S9–S11 present publication-based measures of output and new collaborations, including comparisons by COVID-related publication status and preprint group size.The captions describe article and preprint counts, temporal changes in new-collaboration fractions, year-to-year comparisons, and monthly preprint analyses.Source: Figure S9 · p. 31, Figure S10 · p. 32, Figure S11 · p. 32

COVID-related research

  • Figures S12 and S13 describe COVID-related research engagement by professional, demographic, and field group, and model whether scientists worked on it.The figures show group-level fractions and probit regression coefficients.Source: Figure S12 · p. 33, Figure S13 · p. 33

SI Tables

The supplementary tables report regression analyses linking new research projects with collaborators, controlling for professional, demographic, and field variables. They cover project counts and changes in project counts, with a non-COVID research indicator and robust standard errors.

SI Tables

  • Table S1 reports regressions of new-project counts on collaborator counts, controlling for professional and demographic variables and research fields.It includes a non-COVID research indicator and reports robust standard errors.Source: Table S1 · p. 34
  • Table S2 reports regressions relating absolute and percentage changes in new projects to corresponding changes in collaborators, with professional, demographic, and field controls.It includes a non-COVID research indicator and reports robust standard errors.Source: Table S2 · p. 35
Loading 2107.13073v1…