How Has COVID-19 Impacted Research Production in Economics and Finance?

Abstract

Following the onset of COVID-19, research production in economics and finance (measured by the posting of working papers) increased by 29|$\%$|⁠. Production increases were widespread across geographies, job titles, departments, and ages with larger increases in top departments and for people under the age of 35. Men and women both experienced production increases with the exception of women between the age of 35 and 49, who experienced no production gains despite large increases for men in the same age group. COVID-19 increased reliance on past coauthorship networks, with larger production gains for authors that are more central to the network.

Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

COVID-19 upended the way people around the world live and work in unprecedented ways bringing with them potentially negative consequences. Some costs are likely borne by virtually all, such as a possible reduction in worker productivity associated with the transition from traditional office spaces to remote working arrangements. However, there is also significant concern that COVID-19 may have had a disproportionate impact on women, including a widely reported recent study by McKinsey and Lean In (2020) indicating that many working mothers are considering dropping out of the workforce or decreasing their work activity. Academic research faces many of these same challenges, and survey-based evidence (e.g., Myers et al. 2020; Barber et al. 2021; Deryugina, Shurchkov, and Stearns 2021) suggests that financial economists and academics more generally suffered a general reduction in researcher production, with a larger effect among women.

Academic research in economics and finance has unique features that make it difficult to predict whether COVID will have a positive or negative effect on research production. Feedback and constructive discussions are critical ingredients for academic inquiry, and these interactions traditionally revolve around in-person seminars, conferences, and informal office conversations, all of which suddenly ceased in March 2020. These negative impacts on research production possibly could be more severe for junior faculty. Not only are networking and presentation opportunities particularly valuable early in one’s career, but junior faculty are also more likely to have young children, which complicated working arrangements when schools and daycares closed. COVID-19 may also have a disproportionate impact on women due to unequal distributions of childcare and other household responsibilities. On the other hand, some aspects of the pandemic could lead to an increase in research production. COVID-19 presents new questions for potential research, many aspects of academic research make it a prime candidate for efficient teleworking, and COVID has eliminated many activities that typically compete with work time, commuting and travel for in-person conferences and seminars being the most obvious examples. A lack of social and recreational opportunities may also have spurred people to invest more time in their work.

Universities responded to concerns about reduced research production by almost universally extending tenure clocks in a gender-neutral manner.¹ COVID is likely to play an ongoing role in upcoming review, tenure, and promotion decisions, and if COVID had heterogeneous effects across different groups, one-size-fits-all tenure clock extensions may be insufficient. Gender-neutral policies could even have unintended negative consequences for some women, similar to the potentially inequitable effects of gender-neutral tenure clock extensions for new parents (Antecol, Bedard, and Stearns 2018). Equitable policy decisions require solid evidence on the actual impact of COVID on faculty research, which is the goal of this paper.

To assess COVID-19’s impact on research production with an observable measure of production available for virtually all faculty, we analyze working papers posted on the Social Science Research Network (SSRN) by faculty at top-50 U.S. economics and finance departments. The analysis includes detailed professional and demographic data hand-collected from CVs and author websites, and unique coauthor network analysis based on overlapping paper authorship. Defining and measuring research production is inherently challenging, and studying SSRN paper postings inevitably leaves out some dimensions of research production. However, creation of new papers is at the heart of research production, and the SSRN platform has the advantage of providing standardized information on new paper postings with wide usage among economics and finance researchers.

Our empirical measure (SSRN paper postings) is fundamentally a measure of worker output. This output stems from a combination of productivity and time spent working. COVID likely had important impacts on both channels, and some of our results are most consistent with changes in time available for research. Because we are not able to observe time spent working in order to distinguish between these channels, we describe our results as measuring production instead of productivity throughout the paper. This departs somewhat from most of the literature, which frequently refers to research output as productivity.

Counter to worries about a negative production shock, we find that paper posting rates rose significantly following the onset of COVID, increasing from a pre-COVID average of 1.1 papers per year to an annualized rate of 1.4 papers per year in March 2020 to February 2021.² This increase is partially due to COVID-related research, with authors posting an average of 0.21 COVID papers during the year following COVID. However, non-COVID research output also increased, especially early in the COVID time period, indicating that new COVID papers did not crowd out research on other topics.

While indicative of an increase in research production, SSRN posting rates only capture one part of the research pipeline.³ A possible concern is that researchers may sacrifice quality for the sake of speed, leading to more lower quality papers being posted. For robustness, we consider new postings of NBER working papers, which arguably face a higher bar with respect to quality, finding similar results. Moreover, an examination of papers posted to SSRN in a narrow window around the onset of COVID shows an increase in citations for post-COVID papers, inconsistent with a decrease in quality for post-COVID postings. We repeat our main analysis when considering citation-weighted paper postings with similar results.

Turning to the cross-section, production increases were broad-based, with similar initial increases in finance and economics, and across assistant, associate, and full professors. Production increases varied with department rank. Researchers in top-10 departments experienced an average production increase of 0.53 papers per year (37|$\%$| of their average production), compared to 0.33 papers per year (30|$\%$|⁠) for ranks 11–25 and 0.18 papers per year (19|$\%$|⁠) for ranks 26–50. To address the concern that our sample may not be representative, we analyze extended samples consisting of non-U.S. and lower-ranked departments with consistent results. Taken together, these findings are consistent with a reallocation of time away from other professional obligations, such as in-person seminars and conferences, which may have been particularly time-consuming for researchers in top departments.

Production changes also varied across age groups and gender. Professors under the age of 35 experienced production increases that were 0.19 to 0.23 papers per year larger than the production increases of older professors. Consistent with concerns that COVID disproportionately hurt women, we find that women experienced smaller production gains than men. This gender disparity is concentrated in women between the age of 35 and 49. Whereas women and men below 35 and above 50 experienced nearly identical production gains, women between the age of 35 and 49 experienced no detectable production gain, while men in this age group experienced a production increase of 0.38 papers per year (32|$\%$|⁠). This age range generally corresponds to faculty who are most likely to have young children, suggesting that post-COVID production differences between women and men may be due to different burdens at home, perhaps because of the closure of schools and daycares. However, we are not able to observe childcare duties across individuals and thus are unable to directly test this channel.

Coauthorship networks have the potential to both mitigate and propagate adverse effects of COVID. Consistent with the resilience afforded by one’s network, researchers relied more heavily on preexisting coauthor relationships after the onset of COVID with repeat coauthorship growing faster than new coauthorship. To assess the role of coauthor relationships in facilitating post-COVID production or alternatively spreading shocks to affected coauthors, we map pre-COVID coauthor networks and analyze how they relate to post-COVID production. In support of the resilience offered by one’s research network, researchers with broader, more central, and more diverse networks fared better post-COVID. This finding is potentially related to our other cross-sectional results, as network structure varies across groups (e.g., stronger coauthor relationships in top-10 departments vs. ranks 26–50). Instead, tests which include network centrality and author characteristics indicate that network centrality and author characteristics independently contribute to changes in post-COVID production. We also find support for negative consequences of research networks, with a negative spillover effect for coauthors of women age 35–49 that is similar in magnitude to the direct production effect for this group, even after controlling for researchers’ own demographics.

The impact of COVID was most pronounced soon after COVID hit the world on a widespread scale, with an average production increase of 0.58 papers per year in March to June of 2020 compared to increases of 0.24 papers per year in July to October of 2020 and 0.18 papers per year in November 2020 to February 2021, followed by a return to pre-COVID production levels in March to June 2021. Similar patterns are seen for the differential negative effect for women between the age of 35 and 49 and for the differential positive effects for top-10 departments and researchers with strong coauthor networks. The persistence of these results for 16 months indicates that COVID had a lasting impact on the production of new projects that is not explained by changes in the timing of paper postings or by differential focus on certain parts of research pipelines. Nonetheless, COVID possibly could have longer term effects on production or production quality that we are unable to detect. Concerns about these longer-term effects are mitigated by the length of our sample and the profession’s return to more normal operations soon after the end of our sample. Nonetheless, pausing seminars, conferences, and other in-person interactions and exchanges of ideas could have a longer-term impact on generation of new research, particularly with respect to follow-on ideas from existing projects, and our research design is not suited to picking up these potential long-term effects.

The gender differences we find for post-COVID production changes contribute to a broad literature on gender differences in the academic workplace. A general theme in this literature is that women are underrepresented (Chari and Goldsmith-Pinkham 2017; Lundberg and Stearns 2019; Sherman and Tookes 2022).⁴ In a notable exception using a randomized experiment, Williams and Ceci (2015) find a preference for female faculty hiring decisions in STEM fields. The importance of these issues likely explains why so much of the literature on COVID research production has focused on potential gender differences. Findings in this new literature are mixed, which may be partially explained by differences in methodology, production measures, and focus on different components of the production function (e.g., productivity vs. time allotment). Myers et al. (2020) present survey evidence of a post-COVID decrease in research time allotment of U.S. and European-based researchers in STEM fields. Deryugina, Shurchkov, and Stearns (2021) find that research time decreased post-COVID with more pronounced decreases for women in a broad survey of academic researchers.⁵ However, other studies find no difference for female researchers with related proxies for research production.⁶Fuchs-Schündeln (2020) finds a small but statistically significant drop in the share of papers submitted by female authors to the Review of Economic Studies. We differ from most of these papers in our focus on economics and finance, objective data on new research production for nearly all faculty at top-50 departments, and detailed data hand-collected from CVs and faculty websites. We also examine mechanisms and potential alternative explanations in more detail than most other papers.

The three papers most closely related to this paper are Barber et al. (2021), Amano-Patiño et al. (2020), and Cui et al. (2022). Barber et al. (2021) survey financial economists and find that reported productivity has decreased significantly since the onset of COVID, with particularly large decreases for researchers with young children. The overall negative effect contrasts with our findings and could reflect differences between perceived and actual productivity changes or could be due to sample composition, particularly if people more affected by COVID were more likely to complete a survey on its impact. Interestingly, we find that only 25|$\%$| of finance faculty experience a decrease in posting rates, whereas Barber et al. (2021) find that 78|$\%$| of faculty report a decrease in productivity.⁷ Larger production decreases for researchers with young children are consistent with our findings for women in the 35 to 49 age group and likely stem from the same mechanism.

Amano-Patiño et al. (2020) look at the impact of COVID on productivity for female economists using NBER and Center for Economic and Policy Research (CEPR) working papers from January to April of 2020 compared to previous years. While our papers are related, we address somewhat different questions and come to different conclusions. In particular, Amano-Patiño et al. (2020) mainly focus on participation in research related to the pandemic and find that post-COVID, women (especially mid-career women) are less likely to write COVID papers without any decrease in authorship of non-COVID papers, whereas we find that women between the age of 35 and 49 experience a smaller production gain post-COVID, including for non-COVID papers. We also differ from Amano-Patiño et al. (2020) in that we study a broader measure of research production that includes NBER papers, as well as papers posted to SSRN, we consider nearly all researchers in top finance and economics departments as opposed to just researchers who post to NBER and CPER, and we hand-collect graduation dates from CVs to infer age instead of relying on job titles, such as associate professor, to identify mid-career researchers.

Cui, Ding, and Zhu (2022) study SSRN paper postings across a wide range of social science fields, including economics and finance, and find that gender imbalances have generally grown since the onset of COVID. Our richer professional and demographic data allow us to identify the differential impact of COVID on women in the 35–49 age range compared to women of other ages. This leads to a somewhat nuanced conclusion that women in the age range most likely to have young children fell behind men in the same age group whereas younger and older women fared similarly to their male counterparts. We also examine production changes across a wide range of other characteristics, with the clear findings that researchers with stronger coauthor networks, younger researchers and faculty at top departments experienced larger production gains compared to other faculty.

Relative to the overall literature, our paper makes four primary contributions. First, we show that research production increased in finance and economics after the onset of COVID using objective data on new paper postings. This increased production was real and long-lasting as opposed to a shift in the timing or quality of projects. Second, whereas most of the literature focuses on differences between women and men, our hand-collected data allows us to infer age and examine the interaction between gender and age. Doing so yields the more nuanced result that women between the age of 35 and 49 experienced a relative decrease in production, while women and men in other age groups fared similarly after COVID. Third, our analysis indicates that differences across department ranks are even more important than gender differences. Finally, our analysis of coauthor networks is new to the COVID literature and indicates that more central and diversified authors differentially expanded research production after COVID, an insight that likely has implications beyond COVID.

1. Data and Sample Selection

We hand collect data on tenure track faculty at top-50 U.S. economics and finance departments. Economics department rankings are from the October 2020 IDEAS/RePEc ranking of U.S. economics departments. For finance departments, we use the Arizona State University (ASU) Finance Research Rankings based on top-four journal publications.⁸ The included departments and associated rankings are listed in Table IA.2. In both cases, the rankings are based on all available data as of October 2020, which entails all publication years for the IDEAS/RePEc economics rankings and publications in 1990 to 2019 for the ASU finance rankings. We opt for this long measurement period, because the infrequent nature of publications might result in a noisy measure over shorter periods. We also consider alternative rankings based on the most recent 10 years of data (see Table IA.3) with consistent findings, as discussed in Internet Appendix A. We focus on U.S. departments because the varied organizational structures of some non-U.S. departments make it more difficult to identify the relevant set of faculty. To confirm that our results are not sensitive to our sample definition, we also consider extended samples with non-U.S. departments and a random sample of lower-ranked departments (see Tables IA.4 and IA.5) with similar results.

We start by collecting names and titles of tenure track faculty from department websites as of the fall semester of 2020. We then collect professional and demographic data from CVs, university websites, and personal websites. At a minimum, we require information about when a person obtained their PhD or started their first academic job, which is available 99|$\%$| of the time. Undergraduate graduation year is also available 95|$\%$| of the time. We also collect gender, which we identify based on names, photos, and gender pronouns used on university and personal websites. We estimate each individual’s age as of 2020 using undergraduate graduation year, assuming undergraduate completion at 21 years of age.⁹

Our primary measure of research production is papers posted to SSRN, an online platform dedicated to the dissemination of early-stage research (i.e., working papers). SSRN is the most commonly used repository for economics and finance researchers, with 87|$\%$| of finance papers and 67|$\%$| of economics papers available on SSRN (as shown in Internet Appendix Figure IA.1).¹⁰ Moreover, Internet Appendix Figure IA.2 shows that SSRN contains the 60|$\%$|-70|$\%$| of NBER Summer Institute papers, over 80|$\%$| of papers presented at the Western Finance Association (WFA) meetings from 2018 to 2021, and 93|$\%$| of papers published in the Review of Financial Studies from January to September 2021.

To construct our measures of faculty research production, we first identify each professor’s SSRN author page, which we are able to find for 98|$\%$| of professors.¹¹ The SSRN author pages list details of all papers the author has posted on SSRN including the paper’s title, posting date, last revision date, number of pages, reference information, and a full list of author names. We supplement this data with abstract and keyword information from each paper’s SSRN paper page. We restrict the sample to research papers by excluding cases, appendices, presentation slides, and any postings that are under 10 pages or have more than five authors.¹² To mitigate concerns about including tenure-track faculty that are no longer research active, we remove authors with no postings on SSRN between July 2016 and February 2021, thereby dropping 55 individuals. The final sample consists of 2,118 faculty and 8,461 papers, including 1,315 economics professors and 803 finance professors.

Our main sample consists of papers posted between July 2016 and February 2021. The February 2021 end date corresponds to one year after the widespread onset of COVID. We also consider an extended sample through June 2021. The extended sample covers a period during which vaccination was becoming more widespread and life was starting to return to more normal patterns in most of the United States. This sample allows us to test whether any of our results reverse, which could be the case if they are due to shifts in the timing of postings as opposed to differences in production rates. Both samples end before the widespread onset of Delta and Omicron variants, which prolonged the COVID crisis beyond what most people expected in the spring of 2021.

Table 1 describes the faculty and papers in our main sample. Faculty posted an average of 1.15 papers per year during the sample period. Finance faculty represent 38|$\%$| of the sample, and 18|$\%$| of the sample is female, consistent with the underrepresentation of women in economics and finance academia. As an additional verification of the representativeness of our sample with respect to gender, we compare the female faculty rate at the economics departments in our sample with data from the American Economic Association Committee on the Status of Women in the Economics Profession (2021) survey. Figure IA.3 shows that women faculty percentages are similar in both samples, as well as within different department ranking categories. Finally, Table 1 also shows that the average faculty age is 47 years.

Table 1

Sample description 

A. Faculty characteristics
	Mean	SD	P25	P50	P75
Papers per year	1.15	0.95	0.43	0.86	1.50
Finance|$^a$|	0.38	0.49	-	-	-
Female|$^a$|	0.18	0.38	-	-	-
Age	47.12	12.80	36	45	57
Assistant|$^a$|	0.25	0.44	-	-	-
Associate|$^a$|	0.17	0.38	-	-	-
Full|$^a$|	0.57	0.50	-	-	-
Months in sample	53.90	7.37	56	56	56
B. Paper characteristics
	Mean	SD	P25	P50	P75
Pages	53.52	23.10	39.00	52.00	66.00
Number of authors	2.70	1.00	2.00	3.00	3.00
Citations	4.87	11.73	0.00	1.00	5.00
Downloads	318.90	1,088.88	24.00	82.00	274.00

This table reports summary statistics for the 2,118 authors and 8,461 papers in the sample. The sample consists of tenure-track faculty from top-50 U.S. economics and finance departments who posted at least one paper in SSRN from July 2016 to February 2021. Citations and downloads are numbers reported on SSRN as of July 2022. |$^a$|Dummy variable.

The unit of observation for most of our analysis is at the author-month level. Researchers enter the sample at the latter of July of 2016 or July of the year in which they earned their PhD. Most researchers are in the sample for the full 56 months from July 2016 to February 2021, for a total of 114,168 author-month observations. The table also reports summary statistics for the number of downloads and citations recorded in SSRN, measured as of July 2022.

2. Production Changes following COVID-19

As a first step in evaluating potential changes in production, we visually examine the time-series of working paper posting rates from July 2016 to July 2021. Figure 1 plots the average number of new papers posted to SSRN each month across all 2,118 individuals in our sample. The biggest takeaway from this plot is that production significantly increased after the widespread onset of COVID in March 2020. Average production, represented by the line with solid circles, increased from a pre-COVID rate of 0.09 papers per month to 0.12 papers per month between March 2020 and February 2021 with a peak of 0.15 papers in June 2020, reverting to pre-COVID levels in the last months of our sample. We further examine the increase in author production post-COVID in the Internet Appendix. In Figure IA.4, we plot the percentage of authors posting one or more papers by month and find similar increases after the onset of COVID. In Figure IA.5, we plot the distribution of the number of papers posted per author during the 12-month periods before and after the onset of COVID and find that 42.3|$\%$| of authors experienced an increase in postings post-COVID, with only 26.7|$\%$| experiencing a decrease. These results show that the increase in post-COVID production in Figure 1 is broad-based across most authors and is not driven by outliers. In Figure IA.6, to more directly compare our results to Barber et al.’s (2021) finding that 78|$\%$| of finance professors report having decreased productivity after the onset of COVID, we separately analyze economics and finance departments with the contrasting result that only 25.2|$\%$| of finance faculty had fewer paper postings in the year after COVID.

Figure 1

Monthly research production 

This figure plots the average number of papers posted by month. The solid circles represent the average number of papers when all papers in the sample are considered. The dashed line represents the average number of papers when the papers with COVID-related terms (“COVID,” “corona,” “SARS-COV-2,” and “pandemic”) in their titles are excluded. The solid line represents the average number of papers when COVID-related papers are excluded based on a broader definition that also considers COVID-related terms in the paper’s abstract and keywords. The vertical dashed line represents March 2020, and the vertical dotted line represents February 2021. Ninety-five percent confidence intervals are plotted for the average number of papers posted, including all papers.

The increase in production since the onset of COVID is in part due to papers researching COVID. A conservative estimate of COVID papers based on COVID terms (“COVID,” “corona,” “SARS-COV-2,” and “pandemic”) in their titles indicates that 11.2|$\%$| of papers posted from March 2020 to February 2021 are focused on COVID. Figure 1 plots the time series of posting rates when excluding these papers as the dashed line. The share of COVID-related papers from March 2020 onwards increases to 17.8|$\%$| when considering a broader definition that also considers COVID terms in paper abstracts and keywords. This definition of COVID papers is likely an overestimate including some papers that are only marginally related to COVID as long as they mention COVID somewhere in their abstract. Figure 1 plots the posting rate when excluding the broader definition of COVID-related papers as the solid line without circles. The initiation of COVID-related research does not appear to have crowded out non-COVID research, which was up 24|$\%$| in March to June of 2020 and has been close to historical levels since then.

Next, we empirically test for a post-COVID change using multiple regression frameworks and production measures. We begin by estimating ordinary least squares (OLS) regressions of the form:

$$ \begin{equation} {\it production}_{\it it} = \alpha + \beta {\it COVID}_{t}+\phi_i+\theta_{m(t)}+\epsilon_{\it it}.\label{eq:COVIDregression} \end{equation}$$

(1)

Production is measured at the author-month level, and our primary variable of interest is an indicator for the post-COVID period (starting in March 2020). Monthly production observations are annualized by multiplying by 12 so that coefficients can be interpreted as changes in annual production. We include author (⁠|$\phi_i$|⁠) and month-of-year (⁠|$\theta_{m(t)}$|⁠) fixed effects to control for variation in average production across authors and seasonality in posting rates, respectively. Standard errors are double-clustered by author and month. The monthly clustering accounts for potential correlation across coauthors induced by posting joint papers. Table 2 reports the results.

Table 2

Changes in research production following COVID 

A. Alternative production measures
Regression model: OLS	(1)	(2)	(3)	(4)	(5)
	All papers	Non-COVID papers	NBER papers	Author-adj. papers	Cite-adj. papers
COVID	0.329|$^{***}$|	0.084*	0.273|$^{***}$|	0.099|$^{***}$|	3.416|$^{***}$|
	(0.054)	(0.042)	(0.045)	(0.021)	(0.644)
Author fixed effects	Yes	Yes	Yes	Yes	Yes
Month of year fixed effects	Yes	Yes	Yes	Yes	No
Linear time trend	No	No	No	No	Yes
Data frequency	Monthly	Monthly	Monthly	Monthly	Monthly
Sample time window	Full	Full	Full	Full	8 months
Observations	114,168	114,168	114,168	114,168	16,864
|$R^2$|	.060	.055	.089	.054	.162
Mean production (annualized)	1.146	1.089	0.526	0.472	3.643
B. Alternative regression models
Dependent variable: Papers	(1)	(2)	(3)	(4)	(5)
			Zero-
	OLS with		inflated	50th	75th
	time trend	Poisson	Poisson|$^a$|	percentile	percentile
COVID	0.438|$^{***}$|	0.268|$^{***}$|	0.332|$^{***}$|	0.209|$^{***}$|	0.477|$^{***}$|
	(0.060)	(0.039)	(0.032)	(0.027)	(0.047)
Author fixed effects	Yes	Yes	No	Yes	Yes
Month of year fixed effects	Yes	Yes	No	No	No
Linear time trend	Yes	No	No	No	No
Data frequency	Monthly	Monthly	Monthly	Annual	Annual
Observations	114,168	114,168	114,168	10,220	10,220
|$^a$|Marginal effect

This table reports coefficients (marginal effects in the case of column 3 of panel B) from regressions of the form described in Equation (1). Observations are at the author-month level, except where noted. The sample consists of economics and finance faculty at top-50 U.S. departments and spans July 2016 to February 2021 (November 2019 to June 2020 in the case of column 5 of panel A). Panel A reports results with alternative production measures as the dependent variable as described in the column headers, where Author-adj papers divides each paper by the total number of authors and sums over all papers posted in an author-month, and Cite-adj papers accounts for the number of citations of papers until July 2022. All outcomes are annualized by multiplying by 12. |$COVID$| is an indicator variable that takes the value of one starting in March 2020. All regressions in the panel include author and month-of-year (seasonality) fixed effects. Column 5 of panel A also includes a linear time trend. Panel B considers alternative regression models with number of papers posted as the dependent variable. Column 1 includes a linear time trend. Column 2 estimates a Poisson model. Column 3 estimates a zero-inflated Poisson model. Columns 4 and 5 estimate quantile regressions where observations are aggregated to the author-year level. Standard errors double-clustered by author and month are reported in parentheses, with the exception of panel A, column 5, and panel B, columns 3–5 which are clustered by author. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

Our baseline specification with number of papers posted as the outcome variable is shown in panel A, column 1. We focus on this outcome because economics and finance faculty are typically credited for the magnitude of their research with little if any adjustment for number of coauthors. Paper production increased after the onset of COVID by a highly significant 0.33 papers per year, a 29|$\%$| increase relative to the sample mean.

The remaining columns consider alternative production measures. Column 2 repeats the previous regression analysis with non-COVID papers as the dependent variable, which excludes papers with a COVID-related term in the title, abstract, or keywords. Even with this conservative definition, non-COVID paper production increased by 0.08 papers per year (significant at the 10|$\%$| level). This indicates that instead of COVID papers crowding out non-COVID research, non-COVID research also shared in the production increase.

In column 3, we analyze NBER working papers as of the date they were posted by the NBER.¹³ This analysis misses many papers from our main sample, particularly for authors who are not NBER members. As a result, mean NBER postings are 0.53 papers per year, compared to mean SSRN postings of 1.15 papers per year. On the other hand, NBER papers are potentially more polished and higher quality. The regression results indicate that NBER paper postings increased by a highly significant 0.27 papers per year, which is 52|$\%$| of the sample mean. Thus, NBER production increases even more than SSRN.

In column 4, we adjust papers for their number of authors to account for potential changes in coauthor numbers over time. COVID’s impact remains highly significant with a similar percentage effect (21|$\%$|⁠). Finally, column 5 takes a first step at evaluating a potential change in quality for post-COVID posted papers, with a more thorough treatment in Section 2.1. Here, the outcome variable is the number papers posted adjusted by for the number of citations they received as of July 2022 (i.e., a paper with twice as many citations counts twice as much). To reflect the fact that a more recently posted paper has less time to accumulate cites, we include a linear time trend and consider a narrow window of 4 months on either side of March 2020.¹⁴ In contrast to the hypothesis that researchers are trading off quantity with quality, cite-adjusted paper production increased by 94|$\%$| post-COVID.¹⁵

In panel B of Table 2, we consider alternative model specifications, with the number of papers posted as the outcome variable. In column 1, we consider the same OLS model as panel A but add linear time trends to account for the possibility of a more general change in paper posting rates over time. This results in an even larger COVID impact of 0.44 papers per year. In column 2, we consider Poisson regressions as recommended by Cohn, Liu, and Wardlaw (2022). The coefficient in this regression can be interpreted as a semielasticity, indicating that paper production increased by a highly significant 27|$\%$| following the onset of COVID. In column 3, we consider a zero-inflated Poisson model that allows for the possibility of extra zero observations beyond what the Poisson model would predict. We drop the author and month of year fixed effects from this model and cluster standard errors by author because of limitations of the Stata routine, and report average marginal effects. The estimated effect of COVID is a highly significant 0.33 papers per year, almost identical to the baseline specification in column 1 of panel A. In columns 4 and 5, we consider quantile regressions using recentered influence function regressions for the median and 75th percentiles of annual paper postings to confirm the results are not driven by the right tail of the distribution. These regressions use an annual data frequency due to the sparsity of paper postings at a monthly frequency.¹⁶ Consistent with the other specifications, research production increases after the onset of COVID, with highly significant increases of 0.21 papers per year for the median and 0.48 papers per year for the 75th percentile.

2.1 Has paper quality changed?

One potential concern is that the quality of research being conducted and posted may have changed after the onset of COVID, particularly if researchers started posting papers sooner than they had posted in the past or if COVID-related research differs from traditional research. While the results for NBER papers and citation-adjusted papers shown in Table 2 are inconsistent an influx of lower-quality, earlier-stage papers, we now undertake a more thorough examination of the possible change in paper quality after March 2020. Throughout this section, we evaluate two metrics related to a paper’s potential impact: citations and downloads. We begin with a graphical representation of general trends over time. Figure 2 plots the average number of citations (panel A) and downloads (panel B) as of July 2022 by date of paper postings separately for COVID and non-COVID papers.

Figure 2

Citations and downloads 

This figure plots quality-based characterstics by posting month for papers in the sample. The panels plot the average number of citations (panel A) and cumulative downloads (panel B) by month of paper posting based on SSRN data as of July 2022. The solid circles represent non-COVID papers, and the dashed line represents papers with COVID-related terms (“COVID,” “corona,” “SARS-COV-2,” and “pandemic”) in their titles, abstracts, or keywords. COVID downloads are omitted in March 2020, and COVID citations are omitted in March and April of 2020 due to high values that are greater than the scale of the plots. The vertical dashed line represents March 2020, and the vertical dotted line represents February 2021. Ninety-five percent confidence intervals are plotted for non-COVID paper averages.

Older papers have more time to accumulate citations and downloads, resulting in downward slopes as a function of paper posting month. However, COVID-related papers clearly differ from non-COVID papers with more citations and downloads, especially shortly after the onset of COVID. Moreover, non-COVID paper characteristics are fairly stable before and after COVID. The graphical trends depicted in the figure provide an initial indication that papers posted after COVID are not of lower quality than those posted before COVID, at least as measured by citations or downloads.

To better evaluate a potential difference in paper quality, as measured by citations or downloads as of July 2022, we turn to a regression framework. To more sharply identify a discrete change in either metric, we consider a narrow band of posting dates around the onset of COVID, comparing papers posted in the 4-month stretch ending in February 2020 to the 4-month segment starting in March 2020. Here, the unit of observation is at the paper level, with observations winsorized at the 5|$\%$| level to account for the potential effect of outliers. Panel A of Table 3 evaluates a potential change in citations. In the first pair of specifications, we measure the post-COVID change in citations for all papers. The specifications indicate a post-COVID increase in citations, with a similar point estimate when accounting for time trends with a linear (column 1) or quadratic (column 2) term. In addition, the estimated changes are economically significant, with a post-COVID increase of between 1.43 and 1.58 citations per paper relative to an unconditional mean of 2.65 citations per paper.

Table 3

Changes in quality metrics following COVID 

A. Citations per paper
	All papers			Non-COVID papers
	(1)	(2)		(3)	(4)
COVID	1.580|$^{***}$|	1.426|$^{***}$|		0.273	0.215
	(0.470)	(0.459)		(0.388)	(0.395)
Linear time trend	Yes	Yes		Yes	Yes
Quadratic time trend	No	Yes		No	Yes
Sample time window (months)	8	8		8	8
Observations (papers)	1,375	1,375		1,222	1,222
|$R^2$|	.010	.013		.006	.008
Mean citations per paper	2.653	2.653		1.949	1.949
B. Downloads per paper
	All papers			Non-COVID papers
	(1)	(2)		(3)	(4)
COVID	43.184	42.543		18.535	19.892
	(33.406)	(33.737)		(32.501)	(32.803)
Linear time trend	Yes	Yes		Yes	Yes
Quadratic time trend	No	Yes		No	Yes
Sample time window (months)	8	8		8	8
Observations (papers)	1,375	1,375		1,222	1,222
|$R^2$|	.001	.001		.002	.002
Mean downloads per paper	212.9	212.9		191.8	191.8

This table reports coefficients from OLS regressions where the outcome variable is citations (panel A) or downloads (panel B) from SSRN as of July 2022. Observations are at the paper level, and winsorized at the 5|$\%$| level. The sample consists of papers by economics and finance faculty at top-50 U.S. departments which were posted from November 2019 to June 2020. Regressions include either linear or quadratic time trend controls. Columns 1 and 2 include all papers, while columns 3 and 4 include only non-COVID research papers. |$COVID$| is an indicator variable that takes the value of one starting in March 2020. Heteroscedasticity-robust standard errors are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

Importantly, Figure 2 depicts a noticeable increase in citations among COVID-related research. As such, this difference might mask a decrease in quality among non-COVID research, perhaps because of the accelerated release of early research. In response to this possibility, columns 3 and 4 report analogous specifications when excluding COVID-related papers. Coefficients for the post-COVID indicator become statistically insignificant while retaining small positive point estimates. Taken together, these results suggest that while the increase in citations is being driven by COVID-related research, non-COVID paper do not exhibit a decrease in quality as measured by citations. Panel B of Table 3 turns to the change in downloads by paper. In contrast to the previous panel, while point estimates are positive, post-COVID papers do not demonstrate a statistically significant change in downloads relative to their pre-COVID counterparts.

Finally, Internet Appendix B outlines additional tests of a change in paper quality using alternative transformations of the outcome variable (e.g., inverse hyperbolic sine), alternative models (e.g., median regressions), a wider sample window, and panel regressions using citations and downloads as of multiple snapshot dates. Results are somewhat mixed but consistently show no evidence of a discrete decrease in paper quality, with some specifications indicating a statistically significant increase in citations, while others show no statistical significance.

Taken together, the results in this subsection are consistent with either an increase or no change in paper quality following the onset of COVID, as measured by either paper downloads or citations.

3. Cross-Sectional Differences in Production Changes

COVID created new research opportunities and reduced competing time commitments for some researchers but also created profound challenges for many individuals. Exposure and resilience to these challenges and ability to exploit the opportunities created by COVID likely vary across researchers. In this section, we examine heterogeneity in the effect of COVID across subgroups, with a particular focus on department rank and two subgroups of researchers who potentially bear an outsized burden of the pandemic: ages typically associated with child raising, and female researchers.

Figure 3 plots production by month for different populations. To minimize noise, we plot 3-month moving averages of average papers posted per month. The plots clearly show that production increases were broadly shared across the profession. Panel A shows similar initial increases for economics and finance academics. Production among finance faculty remains elevated, while individuals in economics departments exhibit a reduction in production that approaches pre-COVID rates around August 2020. Panel B shows increases across all department ranks with particularly large increases for top-10 departments. Panel C shows similar production increases for assistant, associate, and full professors. Finally, Panel D shows similar production increases for both men and women.

Figure 3

Monthly production by subpopulations 

This figure shows the 3-month moving average of the average number of papers (with 95|$\%$| confidence intervals) posted by month for different subpopulations. Panel A splits the sample by whether the faculty serves in a finance or an economics department. Panel B splits the sample by three groups of the department’s ranking (i.e., top-10, 11–25, and 26–50). Panel C splits the sample by job title (i.e., assistant, associate, and full professors). Panel D splits the sample by gender. The vertical dashed line in all panels represents March 2020, and the vertical dotted line represents February 2021.

We investigate these patterns in more detail with difference-in-differences regressions of production at the author-month level on an indicator for the post-COVID period and interactions between the post-COVID indicator and author characteristics. Table 4 reports results for job-related characteristics, estimating regressions of the following form:

$$ \begin{equation} production_{it}=\alpha + \beta COVID_{t} + COVID_{t}X_{i}\Gamma+\phi_i+\theta_{g(i),m(t)}+\epsilon_{it},\label{eq: dd regression} \end{equation}$$

(2)

where our primary variables of interest are the interactions between |$COVID$| and indicator variables for various subpopulations. Our estimating equations continue to include author fixed effects (⁠|$\phi_i$|⁠). We also include month-of-year fixed effects, which we now allow to vary by subpopulation (⁠|$\theta_{g(i),m(t)}$|⁠) to ensure the interaction with |$COVID$| is not picking up differential seasonality in posting rates across subpopulations.¹⁷ The unanticipated nature of the pandemic and the focus on ex ante characteristics assuages some endogeneity concerns. However, we cannot rule out a differential effect across subgroups (e.g., gender) that is caused by an omitted variable rather than the onset of COVID. For such a case to hold true, one would need the variable to produce a differential change in the production across subgroups, which is not present prior to the onset of COVID and instead materializes in March 2020.¹⁸

Table 4

Research production changes by department ranking and job title 

	(1)	(2)	(3)	(4)
COVID	0.254|$^{***}$|	0.183|$^{***}$|	0.381|$^{***}$|	0.172|$^{***}$|
	(0.060)	(0.051)	(0.071)	(0.059)
Finance |$\times$| COVID	0.200|$^{**}$|			0.206|$^{**}$|
	(0.081)			(0.078)
Top-11 to 25 department |$\times$| COVID		0.145|$^{**}$|		0.123^*
		(0.060)		(0.063)
Top-10 department |$\times$| COVID		0.345|$^{***}$|		0.334|$^{***}$|
		(0.096)		(0.094)
Associate professor |$\times$| COVID			-0.165^*	-0.152^*
			(0.092)	(0.089)
Full professor |$\times$| COVID			-0.038	-0.050
			(0.074)	(0.073)
Author fixed effects	Yes	Yes	Yes	Yes
Month of year |$\times$| covariates fixed effects	Yes	Yes	Yes	Yes
Observations	114,168	114,168	114,168	114,168
|$R^2$|	.061	.061	.060	.062
Mean production (papers per year)	1.146	1.146	1.146	1.146

This table reports coefficients for regressions estimating Equation (2) with professional characteristics. Observations are at the author-month level. The sample consists of economics and finance faculty at top-50 U.S. departments and spans July 2016 to February 2021. The dependent variable is number of papers posted, which we annualize by multiplying by 12. |$COVID$| is an indicator variable that takes the value of one starting in March 2020. All regressions include author and month-of-year (seasonality) interacted with covariate fixed effects. Standard errors double-clustered by author and month are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

We begin by comparing economics departments to finance departments in column 1 of Table 4, with point estimates indicating a larger production increase in finance departments. Column 2 compares top-10 departments and departments ranked 11–25 to departments ranked 26–50 (the omitted category). Rank 11–25 departments experienced larger production gains than rank 26–50 departments, with even larger production gains for faculty in top-10 departments. Column 3 shows that production increases were similar across job titles (with somewhat lower gains for associate professors), and column 4 shows similar results when all covariates are considered jointly. In Internet Appendix Tables IA.6–IA.10, we consider alternative regression model specifications with similar results.¹⁹ Results are also consistent with alternative production measures.²⁰

The implication of Table 4 is that while production gains were broad across the economics and finance professions, faculty at top departments increased their production much more than other faculty. Whereas departments 26–50 increased their production by 0.18 papers per year, departments 11-25 and top-10 departments increased their production by 0.33 and 0.53 papers per year, respectively. Compared to mean production rates for both groups, this is a 37|$\%$| increase for top-10 departments and a 30|$\%$| increase for departments 11–25 compared to a 19|$\%$| increase for departments 26–50. The differential increase could be due to many factors including better resources, more developed networks from which to solicit feedback on research in the absence of in-person seminars and conferences, and greater success at identifying and executing new research ideas. Alternatively, to the extent that faculty from higher ranked departments face more professional obligations (e.g., in-person seminars and conference presentations/discussions) this result might represent a tilting of temporal budgets away from such duties and toward research.

Next, we turn to potential heterogeneity in production rates across age and gender. Table 5 examines the differential effect across these groups using the same regression specifications as before. In column 1, we report results comparing women and men. Coefficient estimates in panel A show that COVID production gains were smaller for women than for men by 0.11 papers per year. The negative interaction term (significant at the 10|$\%$| level) means that the total effect of COVID on women’s production (reported in panel B) is 0.24 papers per year, compared to the 0.35 papers per year effect for men. Compared to the mean production for each group, these effects represent a 23|$\%$| increase for women and a 30|$\%$| increase for men.

Table 5

Research production changes by gender and age 

A. Regression coefficients
	(1)	(2)	(3)
COVID	0.349|$^{***}$|	0.502|$^{***}$|	0.479|$^{***}$|
	(0.059)	(0.103)	(0.126)
Female |$\times$| COVID	-0.113^*
	(0.061)
Age 35–49 |$\times$| COVID		-0.192|$^{**}$|	-0.104
		(0.092)	(0.113)
Age 50+ |$\times$| COVID		-0.228|$^{**}$|	-0.207^*
		(0.091)	(0.115)
Female |$\times$| age under 35 |$\times$| COVID			0.098
			(0.178)
Female |$\times$| age 35–49 |$\times$| COVID			-0.314|$^{***}$|
			(0.090)
Female |$\times$| age 50+ |$\times$| COVID			0.014
			(0.104)
Author fixed effects	Yes	Yes	Yes
Month of year |$\times$| covariates fixed effects	Yes	Yes	Yes
Observations	114,168	114,168	114,168
|$R^2$|	.060	.061	.061
Mean production (papers per year)	1.146	1.146	1.146
B. Marginal effect of COVID
Female = 1	0.236|$^{***}$|
	(0.058)
Age 35–49 = 1		0.310|$^{***}$|
		(0.060)
Age 50+ = 1		0.274|$^{***}$|
		(0.053)
Female = 1, age under 35 = 1			0.577|$^{***}$|
			(0.135)
Female = 1, age 35–49 = 1			0.061
			(0.066)
Female = 1, age 50+ = 1			0.286|$^{**}$|
			(0.120)

Panel A reports coefficients for regressions estimating Equation (2) with professional characteristics. Panel B reports the marginal effect of |$COVID$|⁠, an indicator variable that takes the value of one starting in March 2020, on different subpopulations. Observations are at the author-month level. The sample consists of economics and finance faculty at top-50 U.S. departments and spans July 2016 to February 2021. The dependent variable is number of papers posted, which we annualize by multiplying by 12. All regressions include author and month-of-year (seasonality) interacted with covariate fixed effects. Standard errors double-clustered by author and month are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

Column 2 of Table 5 considers differences across age groups with the result that faculty between the age of 35 and 49 and faculty who are at least 50 years old experienced smaller production increases compared to younger faculty.²¹ The marginal effects reported in panel B indicate that faculty age 35–49 had an increase in production of 0.31 papers per year, and faculty age 50 and over had an increase in production of 0.27 papers per year compared to an increase of 0.50 papers per year for faculty under the age of 35. Relative to mean production by age group, these represent increases of 44|$\%$| for faculty under the age of 35, 27|$\%$| for faculty ages 35–49, and 24|$\%$| for faculty ages 50 and older.

Next, we consider interaction effects between age and gender to assess whether the differential effect of COVID on women’s production is consistent across age groups. Column 3 of Table 5 shows that, whereas women who are under 35 or at least 50 had similar production increases compared to their male counterparts, women between the age of 35 and 49 experienced a production increase that is 0.31 papers per year smaller than men in the same age group, a difference that is statistically significant at the 1|$\%$| level. On average, women in this age group experienced no production gain with an insignificant point estimate of only positive 0.06 compared to a production increase of 0.38 papers per year for men in this age group. Relative to mean group production, these effects represent a 6|$\%$| increase for women age 35–49 compared to a 32|$\%$| increase for men age 35–49. To visualize this difference, Figure 4 plots post-COVID production relative to pre-COVID production for each age-gender group with a clear difference between men and women in the 35–49 age group. The figure also shows somewhat higher production gains for women under 35 compared to their male counterparts, but this difference is not statistically significant. To visualize how post-COVID research production changes varied by age and gender more generally, Internet Appendix Figure IA.8 plots binscatters of the impact of COVID (measured as average papers per year after COVID minus average papers per year before COVID at the author level) by age separately for women and men. While the figure is somewhat noisy, women throughout the 35–49 aged group had lower production gains compared to their male counterparts, consistent with Table 5 and Figure 4. Results are generally robust to alternative regression models (see Tables IA.16–IA.20) and alternative production measures (see Tables IA.21–IA.25).²²

Figure 4

Production changes by subpopulations of gender and age 

This figure plots post-COVID production relative to pre-COVID production for different subpopulations of gender and age. The bars distinguish between post-COVID papers unrelated to COVID (dark section of the bars) and papers related to COVID (light section of the bars). Three age groups are considered (34 years old or younger, between 35 and 49 years old, and 50 years or older).

The overall result is clear and robust. COVID differentially affected the production of women between the age of 35 and 49. This corresponds to women who are most likely to have young children, adding to the growing evidence that professional women continue to face unequal burdens in their family life that were exacerbated by the sudden onset of COVID. Because we lack data on family status, our average effects for women between the age of 35 and 49 likely understate the burden of COVID on researchers who were most directly affected COVID school and daycare closures. Observed production may also understate the differential impact of COVID to the extent that parents of young children sacrificed on other margins, such as sleep, to keep up some production. Despite this differential burden and the fact that production almost certainly decreased for some individuals, the lack of a difference for women in other age groups suggests that aside from family-related factors, COVID-induced increases in research production were shared by both genders.

Relying on data from professional websites and CVs has significant benefits, particularly by avoiding the potential for sample-selection bias, which is a frequent concern for survey data. Nonetheless, a notable drawback of our approach is the inability to observe the presence of young children at the individual level and directly test a specific childcare channel. To assess the extent to which the 35–49 age group is likely to correlate with having young children, we estimate the likelihood of having a young child (under 12 years old) using American Community Survey (ACS) data from 2010 to 2020 for a comparable sample of individuals who report an occupation of “Postsecondary Teacher,” hold a doctorate degree, and earn an income of at least $70,000.²³Figure IA.9 plots the probability of having a young child as a function of age separately for males and females. The 35–49 age group corresponds to the peak probability of having a young child for both genders. Perhaps confirming priors that males are more likely to delay having children until later in life, the distribution of having a young child is moderately shifted to the left and exhibits a thinner right tail for females relative to males. Given this finding, Internet Appendix Table IA.26 reexamines the results in Table 5 when replacing the interaction of gender and age bins with the imputed likelihood of having a young child (as a function of age and gender) reported in Figure IA.9. Results from OLS and Poisson regressions are consistent with previous findings. Specifically, the post-COVID interaction between young child probability and female is negative (significant at the 10|$\%$| level). By contrast, the direct post-COVID effect for women relative to men is positive and slightly positive, indicating that COVID similarly affected women and men other than the differences when probability of having a young child is high.²⁴

3.1 Sample extensions

The sample considered up to this point consists of tenure track professors at top-50 U.S. economics and finance departments. While this is a large sample that is of interest in its own right, it may not be representative of economics and finance research more generally. To address potential external validity concerns, we consider two extended samples: non-U.S. departments that are ranked in the top-50 internationally and a random sample of U.S. departments ranked 51–100.²⁵

Table 6 reports results for the extended samples. Column 1 regresses production on interactions between the post-COVID indicator variable and indicator variables for department ranks. As before, the omitted category is departments ranked 26–50. The coefficient for |$top \ 51-100 \times COVID$| is small and statistically insignificant, indicating that lower-ranked departments experienced production gains that are similar to departments ranked 26–50 (the omitted category). While production gains were largest in top-10 and top-25 departments, there is no evidence of a decrease or reversal from top 26–50 departments to lower-ranked departments that are outside of the main sample. In column 2, we also find consistent results internationally with no evidence of any difference between U.S. and non-U.S. departments.

Table 6

Sample extensions 

	(1)	(2)	(3)	(4)	(5)
Sample departments	Top-100 U.S.	Top-50 international	U.S. rank 51-100	Top-50 non-U.S.	Top-50 U.S. (recent ranking)
COVID	0.183|$^{***}$|	0.329|$^{***}$|	0.570|$^{***}$|	0.291|$^{**}$|	0.247|$^{***}$|
	(0.051)	(0.054)	(0.123)	(0.121)	(0.050)
Top-51 to 100 department |$\times$| COVID	0.032
	(0.056)
Top-11 to 25 department |$\times$| COVID	0.145|$^{**}$|				0.077
	(0.060)				(0.068)
Top-10 department |$\times$| COVID	0.345|$^{***}$|				0.265|$^{***}$|
	(0.096)				(0.069)
Non-U.S. |$\times$| COVID		-0.104
		(0.072)
Age 35–49 |$\times$| COVID			-0.422|$^{***}$|	0.057
			(0.127)	(0.149)
Age 50+ |$\times$| COVID			-0.382|$^{***}$|	-0.185
			(0.125)	(0.163)
Female |$\times$| age under 35 |$\times$| COVID			-0.422|$^{**}$|	0.365
			(0.196)	(0.273)
Female |$\times$| age 35–49 |$\times$| COVID			-0.021	-0.354|$^{***}$|
			(0.132)	(0.130)
Female |$\times$| age 50+ |$\times$| COVID			-0.082	-0.204|$^{*}$|
			(0.241)	(0.117)
Author fixed effects	Yes	Yes	Yes	Yes	Yes
Month of year |$\times$| covariates fixed effects	Yes	Yes	Yes	Yes	Yes
Observations	145,808	159,904	31,640	45,736	113,788
|$R^2$|	.058	.064	.043	.074	.064
Mean production (papers per year)	1.084	1.088	0.862	0.946	1.168

This table reports coefficients for regressions estimating Equation (2) with extended samples. Column 1 extends the sample to include a random sample of 25 economics departments and 25 finance departments ranked 26–100. Column 2 extends the sample to include non-U.S. departments that are in the top 50 internationally. Columns 3 and 4 consider the sample extensions on their own, separate from the baseline sample of top-50 U.S. departments. Column 5 considers alternative department rankings based on the most recent 10 years of data. Observations are at the author-month level. The dependent variable is number of papers posted, which we annualize by multiplying by 12. |$COVID$| is an indicator variable that takes the value of one starting in March 2020. All regressions include author and month-of-year (seasonality) interacted with covariate fixed effects. Standard errors double-clustered by author and month are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

In columns 3 and 4 of Table 6, we test for gender differences in the extended samples. For U.S. departments outside of the top-50 (column 3), the gender difference is concentrated in younger women in the under 35 age group with similar production changes for older women and men. In the non-U.S. sample (column 4), women between the ages of 35 and 49 have lower post-COVID production compared to their male counterparts (significant at the 1|$\%$| level), and there is also a negative effect for women age 50 and up (significant at the 10|$\%$| level).²⁶

In addition to the extended samples, column 5 of Table 6 also considers alternative department rankings based exclusively on publications over the most recent 10 years. Consistent with the main results in Tables 2 and 4, COVID significantly increases research production in this regression, with a particularly large production boost for top-10 departments. The differential production increase in departments ranked 11–25 compared to lower ranked departments is no longer statistically significant.²⁷

Overall, the extended samples indicate that the production results in this paper are general features of finance and economics research as opposed to being specific to top-50 departments in the U.S. This also indicates that differences between our production increases and the negative productivity survey results documented by Barber et al. (2021) are unlikely to be explained by our sample selection.

4. Coauthorship Networks

COVID had a large impact on the way people work together with an abrupt halt to in-person collaboration and relationship formation. This transition may have been more disruptive for some relationships than for others, and there is a large concern that COVID may have hurt researchers’ ability to establish new relationships. Preexisting coauthor relationships may have also helped researchers exploit new research opportunities and could have helped mitigate negative burdens associated with COVID to the extent one had relationships with other researchers who were less affected. In this section, we explore how COVID affected coauthorship formation and how preexisting coauthorship networks influenced COVID’s impact on production.

During the pre-COVID sample period of July 2016 to February 2020, researchers in our sample had an average of 5.8 coauthors, 1.6 of whom are in-sample tenure track faculty at top-50 U.S. economics and finance departments, with significant variance across researchers. Figure IA.10 plots coauthor relationships between professors at top-50 finance departments, and summary statistics describing pre-COVID coauthorship networks are reported in Internet Appendix Table IA.28.

We begin by considering how COVID changed coauthorship patterns, with results reported in Table 7. Column 1 considers how sole-authorship changed after the onset of COVID, with no evidence of a post-COVID change. Columns 2 and 3 consider changes in repeat and new coauthorships. On a nominal basis, both types of coauthorships exhibit a comparable post-COVID increase. However, researchers in the sample exhibit a lower rate of repeat coauthorships, with an average of 0.94 repeat coauthorships per year compared to 1.22 new coauthorships per year. Given this, repeat coauthorship exhibits a larger relative increase of 48|$\%$| post-COVID compared to 34|$\%$| for new coauthorship.

Table 7

Changes in coauthorship following COVID 

	(1)	(2)	(3)
Dependent variable:	Sole-authored papers	Repeat coauthorships	New coauthorships
COVID	-0.002	0.451|$^{***}$|	0.415|$^{***}$|
	(0.008)	(0.063)	(0.067)
Author fixed effects	Yes	Yes	Yes
Month of year fixed effects	Yes	Yes	Yes
Observations	114,168	114,168	114,168
|$R^2$|	.040	.065	.043
Mean dependent variable	0.0866	0.938	1.219

This table reports coefficients for regressions estimating Equation (1) with coauthorship measures as the dependent variable. Observations are at the author-month level. The sample consists of economics and finance faculty at top-50 U.S. departments and spans July 2016 to February 2021. The dependent variables, indicated at the top of each column, are annualized by multiplying by 12. |$COVID$| is an indicator variable that takes the value of one starting in March 2020. All regressions include author and month-of-year (seasonality) fixed effects. Standard errors double-clustered by author and month are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

Next, we examine how the characteristics of an individual’s coauthor network affected post-COVID production, as some coauthor relationships could have been more productive than others after the onset of COVID. In particular, lockdowns may have had a negative relative effect on same-department collaboration, which previously benefitted from in-person meeting not shared by different-department counterparts. Column 1 of Table 8 tests this possibility, where the variable of interest is the interaction of a post-COVID indicator and the share of a researcher’s pre-COVID coauthors that are in the same department. Consistent with COVID limiting opportunities for in-person collaboration with departmental colleagues, eliminating the previous advantage they enjoyed relative to distant coauthors, the interaction coefficient is negative and large. In fact, the magnitude (⁠|$-$|0.27 papers per year) is almost as large as the overall |$COVID$| coefficient (0.31 papers per year), indicating that exclusively coauthoring with departmental colleagues offsets most of COVID’s production gains.

Table 8

Coauthor characteristics 

	(1)	(2)	(3)	(4)	(5)
COVID	0.308|$^{***}$|	0.208|$^{**}$|	0.124|$^{*}$|	0.393|$^{***}$|	0.394|$^{**}$|
	(0.057)	(0.081)	(0.073)	(0.103)	(0.157)
Same department coauthor share |$\times$| COVID	-0.267|$^{**}$|
	(0.128)
Finance coauthor share |$\times$| COVID		0.005
		(0.083)
Top-11 to 25 department coauthor share |$\times$| COVID			-0.049
			(0.101)
Top-10 department coauthor share |$\times$| COVID			-0.087
			(0.131)
Associate professor coauthor share |$\times$| COVID				-0.097
				(0.113)
Full professor coauthor share |$\times$| COVID				-0.189|$^{**}$|
				(0.084)
Age 35–49 coauthor share |$\times$| COVID					0.040
					(0.127)
Age 50+ coauthor share |$\times$| COVID					-0.023
					(0.143)
Female under 35 coauthor share |$\times$| COVID					0.401
					(0.272)
Female 35–49 coauthor share |$\times$| COVID					-0.274|$^{*}$|
					(0.142)
Female 50+ coauthor share |$\times$| COVID					-0.254
					(0.294)
Author fixed effects	Yes	Yes	Yes	Yes	Yes
Month of year fixed effects	Yes	Yes	Yes	Yes	Yes
Finance |$\times$| COVID fixed effects	No	Yes	No	No	No
Department rank |$\times$| COVID fixed effects	No	No	Yes	No	No
Associate and full prof. |$\times$| COVID fixed effects	No	No	No	Yes	No
Gender |$\times$| age |$\times$| COVID fixed effects	No	No	No	No	Yes
Observations	106,928	82,648	82,648	82,872	82,872
|$R^2$|	.059	.057	.057	.057	.057
Mean productivity (papers per year)	1.196	1.315	1.315	1.313	1.313

This table reports coefficients for regressions estimating Equation (1) with pre-COVID coauthor characteristics. Observations are at the author-month level. The sample consists of economics and finance faculty at top-50 U.S. departments and spans July 2016 to February 2021. The dependent variable is number of papers posted, which we annualize by multiplying by 12. |$COVID$| is an indicator variable that takes the value of one starting in March 2020. Same department coauthor share is the share of pre-COVID coauthors that are in the researcher’s own department. The coauthor share variables in columns 2 through 5 are the share of in-sample coauthors that are in the indicated group. All regressions include author and month-of-year (seasonality) fixed effects. Columns 2 through 5 include fixed effects for the researcher’s own field, department rank, title, gender, and age interacted with |$COVID$| as indicated. Standard errors double-clustered by author and month are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

Additionally, researchers whose coauthors were negatively affected by COVID may have suffered a negative spillover effect. Columns 2 to 5 of Table 8 considers this possibility. The regressions follow the same specification as before to test whether coauthor professional and demographic characteristics influence post-COVID production after controlling for the researcher’s own characteristics with |$COVID \times author \ characteristic$| fixed effects. For finance coauthors (column 2) and coauthor department rank (column 3), there is no discernible spillover. Column 4 reports results based on the share of coauthors that are associate or full professors, showing that coauthors of full professors tended to experience smaller production gains. This is somewhat surprising because the direct production change to full professors is similar to assistant professors in Table 4. Finally, turning to the effect of age and gender we see that the largest spillover effect in Table 8 comes from previously coauthoring with women between the age of 35 and 49. Here, the negative spillover coefficient of |$-$|0.27 (significant at the 10|$\%$| level) is almost as large as the |$-$|0.31 direct effect for women age 35–49 reported in Table 5, suggesting that previously coauthoring exclusively with age 35–49 women largely offsets post-COVID production gains.²⁸

Network structure could also play a role in how researchers responded to COVID, with potential benefits to researchers who are more central and who have more coauthor connections. One common way of measuring how central a node (in this case a researcher) is in a network is to calculate its betweenness centrality, which is the number of shortest paths between other nodes that pass through the node, scaled by the total number of shortest paths between all other node pairs. Table IA.31 summarizes betweenness centrality overall, as well as by department rank and gender.²⁹

Table 9 examines how preexisting coauthor networks influenced COVID’s impact on production using the same regression structure and fixed effects used in previous regressions. Column 1 estimates a regression with a post-COVID dummy variable and the interaction between this variable and the author’s pre-COVID network betweenness centrality. For ease of interpretation, we standardize the network centrality measure to be mean zero with a variance of one. Researchers that are more central to the network experienced a larger production increase. The coefficient of 0.30 implies that a one-standard-deviation increase in centrality is associated with 0.30 more papers per year post-COVID. To verify that these results are not specific to a particular measure of network connections, Tables IA.32 and IA.33 (discussed in Internet Appendix C) estimate equivalent regressions for alternative measures of centrality and coauthor research diversity. More central researchers and those with a more research-diverse set of coauthors exhibit a larger increase in post-COVID paper production.

Table 9

Research production changes by coauthor network, demographic, and professional characteristics 

	(1)	(2)	(3)	(4)	(5)
COVID	0.271|$^{***}$|	0.155|$^{***}$|	0.387|$^{***}$|	0.329|$^{***}$|	0.271|$^{**}$|
	(0.056)	(0.057)	(0.134)	(0.106)	(0.121)
Between centrality |$\times$| COVID	0.297|$^{***}$|	0.276|$^{***}$|	0.298|$^{***}$|		0.277|$^{***}$|
	(0.063)	(0.059)	(0.063)		(0.058)
Top-11 to 25 department |$\times$| COVID		0.141|$^{**}$|		0.137|$^{**}$|	0.134|$^{**}$|
		(0.058)		(0.059)	(0.058)
Top-10 department |$\times$| COVID		0.241|$^{***}$|		0.339|$^{***}$|	0.233|$^{***}$|
		(0.081)		(0.094)	(0.079)
Age 35–49 |$\times$| COVID			-0.095	-0.091	-0.084
			(0.123)	(0.111)	(0.120)
Age 50+ |$\times$| COVID			-0.136	-0.205|$^{*}$|	-0.138
			(0.116)	(0.115)	(0.116)
Female |$\times$| age under 35 |$\times$| COVID			0.118	0.113	0.117
			(0.184)	(0.176)	(0.184)
Female |$\times$| age 35–49 |$\times$| COVID			-0.279|$^{***}$|	-0.297|$^{***}$|	-0.271|$^{***}$|
			(0.092)	(0.089)	(0.092)
Female |$\times$| age 50+ |$\times$| COVID			0.043	0.009	0.037
			(0.108)	(0.101)	(0.106)
Author fixed effects	Yes	Yes	Yes	Yes	Yes
Month of year fixed effects	Yes	Yes	Yes	Yes	Yes
Month of year |$\times$| covariates fixed effects	No	Yes	Yes	Yes	Yes
Observations	108,552	108,552	108,552	114,168	108,552
|$R^2$|	.060	.060	.060	.061	.061
Mean productivity (papers per year)	1.183	1.183	1.183	1.146	1.183

This table reports coefficients for regressions estimating Equation (1) with pre-COVID coauthor network, demographic, and professional characteristics. Observations are at the author-month level. The sample consists of economics and finance faculty at top-50 U.S. departments and spans July 2016 to February 2021. The dependent variable is number of papers posted, which we annualize by multiplying by 12. Between centrality is the betweenness centrality for each author (node) where edges represent coauthor relationships in the pre-COVID period, standardized to have a mean of zero and standard deviation of one. All regressions include author and month-of-year (seasonality) fixed effects (interacted with covariates where indicated). Standard errors double-clustered by author and month are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

On average, network centrality is higher for researchers in top departments, and men tend to have somewhat higher network centrality than women (see Table IA.31). It is possible that some cross-sectional results could be manifestations of preexisting differences in coauthorship networks. The remainder of Table 9 simultaneously estimates the effects of network centrality and characteristics previously considered. Column 2 focuses on centrality and department rank. Both effects are robust to controlling for one another with coefficients that are similar to previous regressions. A one-standard-deviation increase in network centrality is associated with a post-COVID production increase of 0.28 papers per year, similar to the incremental production increase of 0.24 papers per year for researchers in top-10 departments. Columns 3 to 5 also consider gender and age characteristics. Network centrality and department rank continue to be significant. In particular, column 5 includes network centrality, department rank, and interactions of age and gender, with strong positive effects for network centrality and highly ranked departments and a strong negative effect for women age 35–49.³⁰ In short, COVID tended to be a differentially burdensome challenge for women age 35–49 rather than a reflection of differential coauthor networks, and being in a highly ranked department and having a strong coauthor network helped researchers mitigate COVID burdens and exploit new opportunities. These effects are independent of one another and highlight the large heterogeneity in how COVID affected individual researchers.

5. Impact of COVID over Time

COVID took the world by surprise with widespread shutdowns. As time progressed, some things opened back up and people adjusted to the new constraints imposed by COVID. It is also possible that some of the initial changes we observe in paper postings could be changes in timing or differential focus on certain parts of research pipelines.

To assess the changing impact of COVID over time, including the possibility of a reversal, we differentiate between four time periods: (1) early (March to June of 2020), (2) middle (July to October of 2020), (3) late (November 2020 to February 2021), and (4) extended (March to June 2021). Column 1 of Table 10 estimates the overall impact of COVID with positive and significant coefficients in the early, middle, and late COVID time periods. COVID had a bigger impact in the early time period with a production increase of 0.58 papers per year in March to June 2020 compared to increases of 0.24 papers per year in July to October 2020 and 0.18 papers per year in November 2020 to February 2021. In the extended time period, research production returned to its pre-COVID mean with a coefficient of 0.01 that is statistically insignificant. The decreasing effects indicate a gradual return to more normal patterns. Importantly, average production was still significantly enhanced throughout the entire first year of COVID, never falling below pre-COVID levels, even 16 months after the onset of COVID. The persistent positive effect and the lack of any subsequent reversal indicate that COVID had a long-lasting impact as opposed to merely changing the timing of projects in researchers’ pipelines.

Table 10

Impact of COVID over time 

	(1)	(2)	(3)	(4)	(5)
COVID (early)	0.584|$^{***}$|	0.602|$^{***}$|	0.309|$^{***}$|	0.959|$^{***}$|	0.529|$^{***}$|
	(0.111)	(0.084)	(0.113)	(0.253)	(0.113)
COVID (middle)	0.242|$^{***}$|	0.153|$^{***}$|	0.134|$^{**}$|	0.311|$^{*}$|	0.184|$^{***}$|
	(0.039)	(0.035)	(0.060)	(0.163)	(0.042)
COVID (late)	0.182|$^{***}$|	0.033	0.112	0.241|$^{**}$|	0.120|$^{*}$|
	(0.058)	(0.026)	(0.092)	(0.099)	(0.066)
COVID (extended)	0.010	-0.024	-0.012	0.096	-0.017
	(0.136)	(0.135)	(0.143)	(0.134)	(0.139)
Finance |$\times$| COVID (early)		-0.046
		(0.141)
Finance |$\times$| COVID (middle)		0.239|$^{***}$|
		(0.088)
Finance |$\times$| COVID (late)		0.395|$^{***}$|
		(0.126)
Finance |$\times$| COVID (extended)		0.091
		(0.092)
Top-10 department |$\times$| COVID (early)			0.666|$^{***}$|
			(0.093)
Top-10 department |$\times$| COVID (middle)			0.340|$^{**}$|
			(0.138)
Top-10 department |$\times$| COVID (late)			0.057
			(0.131)
Top-10 department |$\times$| COVID (extended)			0.040
			(0.078)
Female |$\times$| age 35–49 |$\times$| COVID (early)				-0.593|$^{***}$|
				(0.161)
Female |$\times$| age 35–49 |$\times$| COVID (middle)				-0.197|$^{**}$|
				(0.085)
Female |$\times$| age 35–49 |$\times$| COVID (late)				-0.170
				(0.140)
Female |$\times$| age 35–49 |$\times$| COVID (extended)				-0.186|$^{*}$|
				(0.110)
Between centrality |$\times$| COVID (early)					0.557|$^{***}$|
					(0.020)
Between centrality |$\times$| COVID (middle)					0.231|$^{**}$|
					(0.089)
Between centrality |$\times$| COVID (late)					0.102|$^{***}$|
					(0.017)
Between centrality |$\times$| COVID (extended)					0.086
					(0.133)
Author fixed effects	Yes	Yes	Yes	Yes	Yes
Month of year fixed effects	Yes	Yes	Yes	Yes	Yes
Month of year |$\times$| covariates fixed effects	No	Yes	Yes	Yes	No
Top-11 to 25 department |$\times$| COVID subperiod fixed effects	No	No	Yes	No	No
Gender |$\times$| age |$\times$| COVID subperiod fixed effects	No	No	No	Yes	No
Observations	122,640	122,640	122,640	122,640	116,496
|$R^2$|	.061	.062	.062	.062	.061
Mean productivity (papers per year)	1.142	1.142	1.142	1.142	1.179

This table repeats regressions reported in Tables 2, 4, 5, and 9 with separate indicator variables for early, middle, late, and extended COVID subperiods. |$COVID \ (early)$| is an indicator variable that takes the value of one from March to June of 2020. |$COVID \ (middle)$| is an indicator variable that takes the value of one from July to October of 2020. |$COVID \ (late)$| is an indicator variable that takes the value of one from November 2020 to February 2021. |$COVID \ (extended)$| is an indicator variable that takes the value of one from March to June of 2021. The regressions are the same as the regressions in the previous tables except that they are interacted with COVID subperiods instead of the overall COVID indicator. Reported coefficients are limited to the variables that are of most interest because of space considerations. Standard errors double-clustered by author and month are reported in parentheses. *|$p<$|⁠.1; **|$p<$|⁠.05; ***|$p<$|⁠.01.

In columns 2 to 5 of Table 10, we assess differential production increases across field, department rank, age, and gender in the four COVID periods. Differential effects for top departments, women age 35–49, and coauthor network centrality are larger during the early period and dissipate during the middle and late periods, mirroring results from column 1.³¹ The implication is that top-10 departments and researchers with strong coauthor networks experienced a production boost that has not reversed, and women between the age of 35 and 49 continued to fall behind throughout most of the COVID time period. In contrast, differences between finance and economics grew over time, indicating that economics production returned to normal levels sooner than finance production. We find consistent results when only considering non-COVID papers (Table IA.38), which shows that COVID research has not crowded out non-COVID research.

6. Conclusion

In the face of an unprecedented public health crisis and complete reinvention of working patterns around the world, finance and economics professors responded by increasing their production of new working papers by 29|$\%$| with larger increases for top-10 departments and young faculty but smaller increases for women between the age of 35 and 49. Broader coauthor networks helped researchers take advantage of COVID production gains, and repeat coauthorship grew more than new coauthorship after the onset of COVID. However, production gains were large enough that new coauthorship also increased despite the challenges COVID created for forming new relationships.

These findings give reasons for both optimism and concern. Economics and finance professors seem to have navigated the transition to virtual work relatively well, with significant increases in production (at least as measured by new paper postings) that are broadly shared across most observable characteristics. However, the magnitudes of these gains were not equal. Top departments enhanced their production more than other departments, particularly with respect to COVID-related research, and women in the age range most associated with young children had lower production gains than their male counterparts. As the economics and finance professions continue to grapple with the fallout from COVID and gender challenges more generally, we hope that our evidence will help inform these deliberations.

Acknowledgement

We thank Itay Goldstein (the editor), two anonymous referees, Tetyana Balyuk, Scott Bauguess, Jonathan Cohn, Willie Fuchs, Clifton Green, John Griffin, Stefan Jaspersen, Christoph Herpfer, Nadya Malenko, William Mann, Kyle Myers, Andrey Ordin, Mahyar Sefidgaran, Laura Starks, René Stulz, Heather Tookes, and Wuyang Zhao and seminar participants at the University of Texas for helpful comments. We thank Jireh Cen, Monica Hebner, Zoey Hu, Cangyuan Li, Scarlett Li, William Robinson, and Ronak Shah for excellent research assistance. The authors were affected by production shocks examined in this paper and potentially stand to benefit from university policies related to issues studied in this paper. In particular, two of the authors have young children, and one served as a primary caregiver and homeschool instructor for much of Spring 2020. This paper previously circulated under the title “How Has COVID-19 Impacted Research Productivity in Economics and Finance?” Supplementary data can be found on The Review of Financial Studies web site.

Footnotes

¹ See Internet Appendix Table IA.1 for information on tenure clock extension policies at specific universities.

² Regression analysis that controls for author fixed effects and seasonality confirms that paper production increased by 0.33 papers per year following the onset of COVID-19, a 29|$\%$| increase relative to the mean production rate of 1.15 papers per year.

³ Interestingly, many economics and finance journals also reported an increase in submissions after the onset of COVID-19, suggesting that production also increased for these fields at a later stage of the research pipeline (e.g., McKenzie 2021; Nagel 2020; Chan et al. 2021).

⁴ Hurdles faced by women include access to smaller coauthor networks (Boschini and Sjögren 2007; Ductor, Goyal, and Prummer 2021), higher service loads (Guarino and Borden 2017), unequal credit attribution from coauthorship (Sarsons et al. 2021), gender bias in teaching evaluations (Mengel, Sauermann, and Zölitz 2019), and male bias against female leadership (Husain, Matsa, and Miller 2021).

⁵Staniscuaski et al. (2021) find similar results when surveying researchers across various fields in Brazil, and Minello, Martucci, and Manzo (2020) provide supplementary evidence from a collection of interviews with academic mothers. Several studies find evidence of decreased research production for women using proxies, such as Twitter posts (Kim and Patterson 2022) and share of biomedical research by female authors (Muric, Lerman, and Ferrara 2021; Andersen et al. 2020).

⁶ For example, there is no dropoff in female research based on preprints in nonmedical fields (Vincent-Lamarre, Sugimoto, and Larivière 2020), Chilean grant applications (Garrido-Vásquez, Vaccari-Jiménez, and Villagrán-Valenzuela 2020), or aquatic science journal submissions (Hobday, Browman, and Bograd 2020). Vincent-Lamarre, Sugimoto, and Larivière (2020) find a decrease in female first author share for medical research preprints, but not preprints in other fields, with gender disparities in authorship of COVID-related papers across fields (though not for NBER papers).

⁷ Our sample of top departments is also narrower than Barber et al.’s (2021) broad survey of the American Finance Association, and posting papers on SSRN could differ from other forms of production. We address these concerns with sample extensions, relatively long time series, and analysis of paper quality, but we cannot fully rule out the possibility that results could be different for other production measures.

⁸ We exclude non-U.S. departments, and recalculate rankings based solely on U.S. departments. The ASU rankings include several business school departments that combine economics and finance. We include these departments and classify everyone in them as finance. We do not include a finance department for Princeton even though it is in the ASU rankings because Princeton does not have a finance department and most financial economists at Princeton are included in its economics department.

⁹ For the 5|$\%$| of cases with missing undergraduate data, we calculate age based on PhD graduation year assuming PhD completion at age 27.

¹⁰Figure IA.1 assesses the popularity of the SSRN, NBER, CEPR, and ArXiv research repositories for papers released between January 2019 and September 2021 for a random sample of 200 researchers. For this comparison, we use Google Scholar as a benchmark because it includes multiple research repositories. While Google Scholar is useful to identify the universe of papers is does not contain standardized data, such as paper posting dates, that are necessary for our main analysis.

¹¹ Some professors have multiple SSRN profiles and thus multiple author pages. We include all of the researcher’s author pages and remove duplicate papers.

¹² Cases are identified based on reference information which are standardized and include terms such as “Case Number XX-XXX.” Appendices and slides are identified based on keywords in titles, combined with manual reviews. Postings less than 10 pages are frequently cases, conference panels, opinion pieces, or other nonresearch content.

¹³ In some cases, earlier versions of these papers were previously posted on SSRN. For this analysis, we exclusively use the date of the paper’s first NBER posting.

¹⁴ Because of the limited number of monthly observations used in this analysis, we cluster standard errors by author instead of double clustering by author and month, and we do not include month-of-year fixed effects.

¹⁵ This large increase in cite-adjusted production is driven by two factors. First, production increases were largest during the first few months after COVID. Second, citations per paper also increased after the onset of COVID. We estimate both of these effects in more detail in subsequent analysis.

¹⁶ Standard errors are clustered by author because there are not enough annual observations to also cluster by year.

¹⁷ While this is not a major concern in our primary sample (which includes 12 months of post-COVID observations), variation in seasonality has the potential to affect inferences in later tests, which either focus on subperiods or extend the primary sample.

¹⁸Figure 3 depicts an almost instantaneous impact on production and divergence across subgroups, a pattern the omitted variable would also need to generate.

¹⁹ Results are generally consistent (with a few exceptions) when considering Poisson regressions (Table IA.6), zero-inflated Poisson regressions (Table IA.7), median regressions (Table IA.8), 75th percentile regressions (Table IA.9), and OLS regressions that accommodate potentially differential pre-COVID trends in posting rates across groups with linear time trends (Table IA.10). Effects are statistically insignificant in the median regressions, and for rank 11–25 departments when including time trends. For the remainder of the paper, we refer to this series of robustness tests collectively as “alternative regression models.”

²⁰ The alternative production measures are non-COVID papers (Table IA.11), NBER papers (Table IA.12), papers adjusted for number of authors (Table IA.13), total number of pages posted (Table IA.14), and papers adjusted for number of citations (Table IA.15). To account for accumulation of citations over time, we ensure that the citation-adjusted regressions include a linear time trend and are limited to a narrow window of 4 months before and after the onset of COVID. The production increase at top-10 departments is consistent across measures, and there is a differential production increase for finance professors across all measures except NBER papers and citation-adjusted papers.

²¹Internet Appendix Figure IA.7 plots the distribution of faculty across age groups. For women, 97 individuals are under 35, 191 are 35–49, and 91 are 50 or older. For men, 300 individuals are under 35, 732 are 35–49, and 707 are 50 or older.

²² The main exception is that the interaction between the female and age 35–49 coefficient is not significant in the median regression (Table IA.18), similar to median results for finance and top-10 departments, and results are not significant in the citation-adjusted regressions, which have less power because of being restricted to a narrow window of 4 months before and after the onset of COVID.

²³ The probability distribution is calculated by performing locally weighted scatterplot smoothing (Lowess), where the outcome is an indicator variable taking on a value of one for individuals with a child under 12.

²⁴ The post-COVID coefficient for having a young child is positive for males. However, we caution against drawing strong inferences from this because the imputed probability of having a child is correlated with other potential contributing factors (e.g., job title and age).

²⁵ The non-U.S. sample includes 20 economics departments and 8 finance departments with a total of 845 researchers. For the extended U.S. sample, to keep data collection manageable we randomly sample 25 finance departments and 25 economics that are ranked between 51 and 100, resulting in 590 additional researchers.

²⁶Internet Appendix Table IA.27 reports Poisson regressions equivalent to Table 6 with similar results.

²⁷ In Internet Appendix A, we consider the sensitivity of our findings more broadly to basing rankings on more recent data, repeating all of our main tables. This change generally yields similar results, with exceptions discussed in Internet Appendix A.

²⁸Internet Appendix Table IA.29 repeats Table 8 with Poisson regressions with similar results except that the negative same-department and full-professor coauthor coefficients are no longer significant. To check that the spillover effect for coauthoring with age 35–49 women is not driven by unobserved characteristics, such as from men of the same age or women in other age groups who also have children, Table IA.30 reestimates the final specification from Table 8 in samples that are restricted to male researchers, both overall and in particular age groups. The spillover restricted to men in general is larger and more significant than in the baseline specification in Table 8, and spillovers to men in particular age groups are all negative, though with less statistical significance.

²⁹ Reflecting the overall sparseness of the network, more than 50|$\%$| of researchers have a network centrality of zero, indicating that they are peripheral to the overall network.

³⁰ To account for potential changes in paper quality, Table IA.34 repeats Table 9 with citation-adjusted papers in the narrow event window used in previous citation regressions. Results are similar for network centrality and top-10 departments but insignificant for women age 35–49, consistent with previous citation-adjusted regressions. Table IA.35 estimates network centrality and department rank regressions in a sample that is limited to women age 35–49. Coefficients for centrality and department rank are positive but statistically insignificant in the restricted sample. Tables IA.36 and IA.37 repeat all regressions from Tables 4 and 5, respectively, when controlling for betweenness centrality. The previous results are confirmed, and betweenness centrality is positive and statistically significant in all specifications.

³¹ Columns 3 and 4 of Table 10 include all of the same variables as previous specifications, interacted with the COVID subperiods. For brevity the reported coefficients are limited to the main coefficients of interest.