key: cord-0061647-0ac3vmfz authors: Gallacher, Guillermo; Hossain, Iqbal title: Remote Work and Employment Dynamics under COVID-19: Evidence from Canada date: 2020-07-01 journal: nan DOI: 10.3138/cpp.2020-026 sha: b283c905ea8db8b20f49f6e6a5901ccb3601f6ee doc_id: 61647 cord_uid: 0ac3vmfz In this study, we find that 41 percent of jobs in Canada can be performed remotely, with significant variation across provinces, cities, and industries. We complement this finding with labour microdata and document facts on the relationship between the feasibility of remote work and income inequality, gender, age, and other worker characteristics. We then show that, under some of our specifications, workers in occupations for which the possibility of remote work is less likely experienced larger employment losses between March and April. This relationship however does not seem to hold for a different measure of the possibility of remote work or for employment losses across industries with different possibilities of remote work nor across provinces or cities with different possibilities of remote work. Social distancing is costly because some jobs cannot be performed at home. A key estimate to assess the economic impact of the coronavirus 2019 pandemic is the percentage of jobs that can be done remotely. With this motivation, in this article we focus on Canada and estimate the feasibility of working from home, as well as the heterogeneity of this variable along several dimensions. We then measure employment changes during March and April 2020 and compare these changes with the computed feasibility of working remotely. First, we provide a literature review and describe our contribution. We then compute the percentage of jobs that can be performed at home. We fi nd that 41 percent of jobs in Canada can be performed remotely, with signifi cant variation across provinces, cities, and industries. Next, we use labour microdata and document that poorer workers, male workers, workers without a college degree, private sector workers, single workers, small fi rm workers, seasonal or contractual workers, part-time workers, younger workers, and non-immigrant workers tend to be employed in jobs for which remote work is less possible. We then show that workers in occupations for which the possibility of remote work is less likely experienced larger employment losses between March and April. This relationship however does not seem to hold for a different measure of the possibility of remote work or for employment losses across industries with different possibilities of remote work nor across provinces or cities with different possibilities of remote work. We follow the methodology in Dingel and Neiman (2020 ) to measure the percentage of jobs that can be done at home. This methodology has since been applied to other countries. For instance, Saltiel (2020 ) focuses on a sample of classifi cation, photographers, painters, air traffi c controllers, administrative assistants, land survey technologists, technicians and forestry professionals can work from home, whereas in the alternative classifi cation they cannot. According to our estimates, 41 percent of jobs in Canada can be done from home. When weighted by wages, this percentage increases to 51 percent. The robustness check estimates for our alternative specifi cation are in line with these percentages but are somewhat lower-37 percent and 48 percent, respectively (all robustness checks are available in the replication package, available at https:// github.com/guillgall/remotework_dynamics ). The higher estimate when weighting by wages indicates that higher-wage jobs tend to be associated with jobs that can be more easily performed remotely. We return to this point later with the microdata. Dingel and Neiman (2020) estimate the percentage of jobs that can be done at home in the United States at 37 percent. Even though Dingel and Neiman do not report Canadian estimates, our results are consistent with their international evidence. 5 As can be seen in Table 1 , considerable heterogeneity exists among Canadian provinces. Ontario has a high share of jobs that can be done at home (44 percent), and Newfoundland and Labrador has the least amount of jobs can be done at home (32 percent). We also observe heterogeneity at the city level. In Table 2 , we report the estimates for the 10 largest cities. Ottawa, Toronto, and Montreal lead this remote work ranking (with roughly half of jobs being able to be done from home). We also include in the last row the aggregated smaller cities estimate. The full sample of cities can be seen in Table A .2 . Figure 1 shows the relationship between the estimated share of jobs that can be performed remotely and city size. As can be observed, the relationship is positive and non-linear. Last, Table A .3 reports the territorial distribution. The Northwest Territories, Yukon, and Nunavut do not show developing countries. We complement this methodology with worker characteristics. The worker characteristics and distributional aspect have been studied by, among others, Mongey, Pilossoph, and Weinberg (2020 ) and Yasenov (2020 ) in the United States, Lekfuangfu et al. (forthcoming ) in Thailand, and Foschiatti and Gasparini (2020 ) in Argentina. We contribute to this literature by adding Canadian estimates. Our results are in line with all these fi ndings. 1 In terms of employment dynamics, our article is closest to Mongey et al. (2020 ) , who report a link between remote work feasibility and changes in employment in March (in the United States). Other articles study job vacancies. Lange and Warman (2020 ) report that in Canada very rapid declines in vacancy postings occurred as the COVID-19 crisis brought the economy to a halt in mid-March. Postings declined by up to 50 percent until mid-April. Kahn, Lange, and Wiczer (2020 ) study the changes in job vacancies and unemployment insurance claims for different occupations and industries during the pandemic in the United States. We apply the methodology of Dingel and Neiman (2020 ) to Canadian data. The main idea in Dingel and Neiman is to classify the feasibility of working at home for all occupations and merge this classifi cation with occupational employment counts. The feasibility measure is based on responses to two Occupational Information Network (O*NET) surveys. 2 Dingel and Neiman estimate that as much as 37 percent of jobs can plausibly be done from home in the United States, and they provide estimates for a large sample of countries (but not Canada). We use Statistics Canada's Employment Income Statistics (EIS), a tabulation from the 2016 Census that contains four-digit occupation classifi cation employment counts at the provincial, city, and territorial levels (51 geographic areas in total). Classifi cation codes correspond to the National Occupational Classifi cation (NOC). The EIS also contains average income, which we use to compute percentage of wages (in addition to percentage of jobs). We use the Brookfi eld Institute for Innovation + Entrepreneurship (BII+E) cross-walk of O*NET with NOC. We then merge the occupations in NOC with O*NET's binary remote work index computed by Dingel and Neiman (2020) . We call this the benchmark remote work index. 3 As a robustness check, we then manually assign values for the NOC occupation categories, using introspection (as in the robustness check in Dingel and Neiman 2020) and call this the alternative remote work index. 4 Appendix A, Table A .1 reports the four-digit occupation codes for which the two measures differ the most. On the one hand, according to our benchmark classifi cation, engineers and journalists cannot work from home, whereas in our alternative classifi cation they can. On the other hand, in our benchmark Next, we explore the Labour Force Survey (LFS), which contains Canadian labour microdata at the two-digit occupation level. We assign remote index values by taking a weighted average across the four-digit subcategories. In Tables A.4 and A.5 , we document that the provincial and city estimates using the LFS are consistent with those reported earlier. In addition, in Table A .6 we report the estimation by industry: as can be observed, sectors such as fi nance and insurance report the largest percentage of ability to perform remote work, whereas agriculture reports the lowest percentage. Remote Work Index focused on the United States. We run a simple regression as in Mongey et al. Let y ij be a characteristic of a worker i who reports working in occupation j in 2019. We construct binary variables based on the worker characteristics of LFS 2019; for example, we construct a variable y ij if the continuous variable wage is above the median. We then estimate the following regression for each of our observables, where S j is the two-digit occupation remote work score: We then plot the values for ˆy  in Figure 3 . As an example to understand the interpretation of ˆy  , let us take the case of below-median income: because ˆ1 y   , then workers with below-median income tend to be in occupations that are less likely to be performed remotely. This What type of workers are more vulnerable to this labour market shock? We use LFS 2019 data to document correlations between worker characteristics and the possibility of working from home. We fi rst explore the distributional dimension. In Figure 2 , we report the remote work index by percentile of the income distribution. This fi gure shows that higher-income workers are on average more likely to be able to work from home. This would suggest that social distancing is regressive, in the sense that poorer workers tend to work in jobs that are more diffi cult to perform remotely. Note that this is consistent with the fi ndings in the previous section on the higher share of jobs that can be done at home when weighting by wages. 7 We then explore other worker characteristics. Here we follow the methodology in Mongey et al. (2020 ) , who is consistent with the fi ndings reported earlier (share of wages is higher than share of jobs) as well as with the discussion of Figure 2 . Our results suggest that poorer workers, male workers, workers without a college degree, private sector workers, single workers, small fi rm workers, seasonal or contractual workers, part-time workers, younger workers, and non-immigrant workers tend to be employed in jobs for which remote work is less possible. These results are overall consistent with the fi ndings of Mongey et al. (2020) for the United States. We then obtain the March and April 2020 LFS data and compare them with the February 2020 data (before the social distancing measures in Canada). In Figure 4 , we can see the correlations between the monthly percentage change in employment and the remote work index at the two-digit occupation level. Between March and April 2020 (Figure 4b) , the relationship was positive: the two-digit occupations that had higher remote work scores experienced smaller (less negative) employment declines. However, no such relationship exists for February-March 2020 ( Figure 4a ). Here we are using the alternative remote work index. We then control for jobs deemed essential by public policy. For example, front-line medical workers have a low remote work index (0), but because they have been declared essential, they can continue to work. We construct an essential service dummy variable to take this important caveat into account. 8 We also construct a continuous essential service variable: we use the four-digit essential service binary classifi cation done by the Labour Market Information Council (LMIC) and aggregate it into two digits by taking employment weighted averages, thus making the index no longer binary. 9 We formally test this relationship. For an occupation j , let employment be q j and denote ∆ q j,t, τ as the percentage of change in employment between period t and τ . We then run the following Ordinary Least Squares regression, where S j is the remote work score of occupation j and ES j is the "essential-service" continuous variable (or the "essential service" dummy variable as described in the previous paragraph): When we run this regression with the benchmark remote work index, this index is not signifi cant for any of the month-to-month changes. We report those results in our online Appendix B. In Table 3 , we report the results using our alternative remote work index and the essential service variable. As can be observed, the remote work index score variable is positive and statistically signifi cant only for the March-April change. The lack of statistical signifi cance for the January-February change (the before-social-distancing observation) is expected but is more surprising in the case of February-March. As we report in the online Appendix B, a signifi cant relationship also holds at the ten NOC broad occupation level for the March-April variation, under both the benchmark and alternative remote work index. The relationship is however no longer statistically signifi cant under other specifi cations for the two-digit occupation level: in online Appendix B, we provide all other specifi cations (benchmark remote work index, essential service dummy variable, both benchmark remote work index and essential service dummy variable), as well as the regression tables at the industry, province and city level. The magnitude of the COVID-19 shock on labour markets is historic. Policy-makers around the world are responding as fast as possible to mitigate this shock. Understanding and designing policy during the crisis and after it (ease of lockdown and relaxing of social distancing) is crucial. This article has three main results: 1. We fi nd that 41 percent of jobs in Canada can be performed remotely, with signifi cant variation across provinces, cities, and industries. 2. We complement this fi nding with labour microdata and document facts on the relationship between remote work and worker characteristics; our results suggest that poorer workers, male workers, workers without a college degree, private sector workers, single workers, small fi rm workers, seasonal or contractual workers, part-time workers, younger workers, and non-immigrant workers tend to be employed in jobs for which remote work is less possible. 3. We show that, under some of our empirical specifi cations, workers in occupations with less possibility for remote work experienced larger employment losses between March and April. This relationship however does not seem to hold for a different measure of Future work with updated data could extend and improve on the empirical fi ndings. For instance, measuring the share of productivity or output (potentially produced remotely) instead of jobs seems important. It will also be relevant to study the impact of policy, such as the Canada Emergency Response Benefi t fi nancial support. 1 Our overall remote-work estimate for Canada (41 percent) seems to also be in line with the estimations by Chen and Mehdi (2019) , who calculate that 41.5 percent of workers in Canada can choose their start and end time at work. Chen and Mehdi (2019) provide other important stylized facts on the labour market. The estimates by Statistics Canada (which were published after our paper was accepted) for remote work at the aggregate and provincial level are in line with our estimates, although their estimates at the in-dustrial level show more dispersion than ours. They also study heterogeneity along income, education, gender and age. See Deng, Morissette, and Messacar (2020) . 2 These are the Work Context and Generalized Work Activities surveys, which provide data on whether an occupation requires daily work outdoors, operation of vehicles, mechanized devices, equipment, physical activities, and other characteristics relevant for the possibility of remote work. 3 One issue that we faced was that the O*NET occupation classifi cation in BII+E and Dingel and Neiman (2020) have some discrepancies. After merging with the NOC data, 17 occupations were dropped. Please refer to the replication package for details. 4 Two of our research team manually assigned values of 0, 0.5, and 1.0 to each four-digit NOC code, which we then averaged. 5 See Figure 2 in Dingel and Neiman (2020) and recall that Canada's gross domestic product per capita is around US$46,000. 6 The rural estimate was computed indirectly, because we only observe urban and aggregate employment: employment in Canada is the sum of rural and urban employment, E CAN = E Rural + E Urban . Moreover, the remote work share for Canada, S CAN , is a weighted average of the urban and rural shares, Notes: q j,t,τ is the percentage of change in employment between month t and τ in occupation j, Sj (Alternative) is the remote work index, and ES j is the essential service variable. * p < 0.1; ** p < 0.05; *** p < 0.01. Source: Authors' calculations. 8 We intuitively match the two-digit occupations to the essential service sectors as listed by Public Safety Canada: energy and utilities, information and communication technologies, fi nance, health, food, water, transportation, safety, government, and manufacturing. We are aware of the subjectivity involved in the construction of this dummy variable. We thus complement this with the alternative essential service variable as robustness check (as described next). 9 These two essential service measures turn out to have some differences. For instance, occupations linked with fi nance and manufacturing were labelled as essential by us but had a low LMIC score. We treated them as essential because Public Safety Canada linked both fi nance and manufacturing as essential service sectors. El impacto asimétrico de la cuarentena Labor Demand in the Time of COVID-19: Evidence from Vacancy Postings and UI Claims Vacancy Posting in 2020: Estimates Based on Job Bank and External Providers On Covid-19: New Implications of Job Task Requirements and Spouse's Occupational Sorting Which Workers Bear the Burden of Social Distancing Policies Home Working in Developing Countries IZA Discussion Paper No. 13197 Assessing Job Quality in Canada: A Multidimensional Approach StatCan COVID-19: Data to Insights for a Better Canada. Cat. No. 2020001. Ottawa: Statistics Canada How Many Jobs Can Be Done at Home ?" NBER Working Paper No. 26948 . Cambridge The replication package for this project is available at https://github.com/guillgall/remotework_dynamics . We thank Umut Oğuzoğlu, Janelle Mann, Ian Hudson, an anonymous guest editor, and participants at the University of Manitoba online brownbag for comments. Mahesh Acharya and Shantanu Debbarman provided excellent research assistance. We also thank Mary Lui (University of Toronto Press) for feedback and suggestions. Any errors are our own.