The impact of COVID-19 on digital communication patterns

The COVID-19 pandemic forced a rapid shift to remote work for many knowledge workers. This paper examines how this abrupt transition altered formal digital communication patterns, specifically meeting and email activity. The analysis uses de-identified metadata from an IT services provider, encompassing data from 3,143,270 users across 21,478 firms in 16 major metropolitan areas. The data includes information on email and meeting frequency, size, and duration, allowing for a detailed analysis of communication changes before and after government-mandated lockdowns in these areas. The lockdowns provided a clear breakpoint to assess the shift to remote work. The study aims to provide systematic evidence on how day-to-day work activities changed due to the pandemic and the subsequent shift to remote work, addressing the lack of prior research on this large-scale, forced transition.

Existing research on remote work transitions often focuses on voluntary, less widespread shifts under less dramatic circumstances, making it difficult to generalize to the COVID-19 situation. Studies examining forced transitions typically involve smaller workforces over shorter durations. Furthermore, prior research often lacks the scale and comprehensive digital communication data analyzed in this paper. The existing literature offers inconclusive predictions regarding changes in meeting frequency, size, and length, and the interplay between these factors. There is also limited understanding of how email activity might complement or substitute for meeting activity during a remote work transition. This study, therefore, doesn't present specific hypotheses but rather explores the changes observed in the data.

The study employs an OLS regression-based event study to analyze changes in digital communication patterns before and after government-mandated lockdowns in 16 international cities. The data includes communication frequency for email (internal and external) and meetings (count, duration, attendees), along with workday length and emails sent outside business hours. The data provider cleaned the data by excluding meetings with one attendee, those lasting over 8 hours, and those with more than 250 attendees. Only internal emails (between employees sharing the same domain) were included. The 16 cities were selected based on criteria including active user count, government-mandated work-from-home orders, and order implementation timing. The analysis uses two main regression specifications: one comparing pre- and post-lockdown periods using a single dummy variable, and another analyzing weekly changes relative to the lockdown week using week indicator variables. MSA-level fixed effects control for average differences across MSAs. Robustness checks were performed using different weighting and aggregation schemes, considering industry and firm size, and evaluating whether weekly or daily aggregation produces consistent results. Additionally, supplementary analyses examined potential sectoral differences and the changes in communication patterns up to nine months after lockdown.

The study finds a 12.9% increase in meeting count and a 13.5% increase in attendees per meeting after lockdowns. However, average meeting duration decreased by 20.1%, leading to an overall 11.5% reduction in daily meeting time. Internal emails increased by 5.2%, and the average number of recipients per email increased by 2.9%. External emails did not show significant changes. The average workday length increased by 8.2% (+48.5 min), and emails sent outside of business hours increased by 8.3%. Weekly analysis reveals consistent increases in meeting size and count, and decreases in meeting length after lockdown. The reduction in total meeting time persists weekly. Even nine months after lockdowns, internal email volume remains significantly higher than pre-lockdown levels, while the increase in email recipients diminishes. Robustness checks confirm the results are generally consistent across different weighting, aggregation methods, industries, and sizes of organizations, with the exception of email recipients. The only industry which shows a significant difference is the services industry (excluding financial services), in which email communication recovery post-lockdown is slower than in other industries.

The findings suggest organizations made communication trade-offs in response to the pandemic. Increased meeting and email frequency, and larger meetings and emails, were observed, potentially replacing in-person interactions and freeing up time for individual work. The increase in internal emails but not external emails might reflect a shift toward internal coordination and accountability needs during remote work. The increase in meeting frequency alongside decreased meeting length might indicate coordination challenges associated with remote work. The longer workdays and increased after-hours emails suggest a spillover of work demands beyond normal working hours, potentially due to challenges balancing work and personal life during remote work. The proactive nature of organizational adaptation to the pandemic, even before official lockdowns, is noteworthy.

This study provides valuable insights into the changes in digital communication patterns during the COVID-19 pandemic, showing significant adaptations in meeting and email activity in response to the widespread shift to remote work. The findings highlight the trade-offs organizations made regarding communication volume, frequency, and duration. The increased workday lengths also reveal a potential negative consequence of remote work. Future research could explore the long-term effects of these changes, the impact of different remote work models, and the role of technology choices in shaping communication patterns.

The study's data only represent a subset of organizational communication, excluding non-email communication channels and informal meetings. It also cannot disentangle the effects of remote work, macroeconomic shocks, and non-work-related behavior changes associated with lockdowns. Variations in lockdown stringency across locations are not fully accounted for. While efforts were made to ensure data accuracy, measurement errors in meeting metadata remain a possibility. These limitations should be considered when interpreting the results.