The COVID-19 pandemic significantly accelerated the shift to work from home (WFH), with lasting impacts globally. In the US, WFH days leapt from 7% in 2019 to nearly 60% in spring 2020, then fell back to stabilise at around 28% by 2023 (Barrero et al. 2021, 2023). Similar patterns are evident internationally. In Germany, the percentage of employees working from home rose from about 5% pre-pandemic to roughly 25% since early 2022, as noted in the ifo business survey (Alipour 2023). Statistics derived from online job ads and office occupancy also indicate that WFH rates have stabilised since 2022 (Hansen et al. 2023, Alipour et al. 2021, Adrjan et al. 2022, Bloom et al. 2023).
However, there are significant cross-country variations in WFH practices. In the third wave of our Global Survey of Working Arrangements, collected in April and May 2023, we sampled 34 countries. In Figure 1, we plot WFH rates for graduates, who are potentially the most comparable across countries. We find that WFH rates are highest at an average of 1.8 days per week in English-speaking countries, with European and Latin American countries lower (1.1 days) and Asian countries the lowest (0.9 days). These patterns match Workplace Mobility data published by Google until October 2022, which track the frequency of workplace visits by country and month.
Figure 1 Average paid full days worked from home per week, college graduates
Notes: Responses to the question “For each day last week, did you work 6 or more hours, and if so where?”. Sample of respondents with at least a college degree in the Global Survey of Working Arrangements (G-SWA) from 34 countries surveyed in April-May 2023.
Why does work from home vary so much across countries?
There are several reasons why WFH varies so much across countries. WFH intensity varies significantly across industries, occupations, and regions due to the nature of different jobs (Dingel and Neiman 2020, Alipour et al. 2023). Factors such as industry composition, workplace culture, managerial styles, and perceptions of remote work productivity influence the adoption of WFH arrangements (Hansen et al. 2023). Urban areas, which often have higher population densities and a greater prevalence of knowledge-based industries suitable for remote work, show higher WFH rates (Barrero et al. 2023, Althoff et al. 2022). Additionally, experiences during the pandemic, including stricter, longer lockdowns, led to a more entrenched WFH culture (Adrjan et al. 2023, Aksoy et al. 2022).
In Zarate et al. (2024), we use data from the third wave of the Global Survey of Working Arrangements to examine the factors influencing WFH intensity across countries as of April-May 2023. We regress the average number of full paid WFH days by country on several predictors including GDP per capita, which reflects labour productivity and educational attainment; cumulative lockdown stringency during the pandemic; population-weighted density; and the proportion of jobs suitable for remote work based on Dingel and Neiman (2020).
Additionally, we explore the impact of individualism on WFH adoption, a cultural dimension that values personal freedom and autonomy (Heine 2008, Triandis 1994, 1995). For this purpose, we rely on Hofstede’s (2011) index of individualism, which measures the extent to which individuals in a society prioritise their own ambitions and independence above the collective goals and unity of the group. Given that WFH success often relies on minimal direct supervision, we hypothesise that more individualistic cultures may be more conducive to adopting remote work.
Predictors of remote-work adoption
We estimate specifications that include all five regressors (lockdown stringency, GDP per capita, weighted population density, individualism, and industry mix) at the same time. To ease interpretation, we standardise each of the regressors to have mean zero and unit standard deviation.
Individualism has the strongest association with WFH intensity: moving from the 10th percentile country (China) to the 90th (Netherlands) by individualism score implies 0.6 more WFH days per week, 63% as large as the average number of full paid days of WFH across countries. Indeed, simply plotting WFH levels against individualism shows a surprisingly good fit, as highlighted in Figure 2.
Figure 2 Individualism and WFH
Notes: The figure plots the average full paid days WFH and Hofstede’s individualism index. Average full paid days WFH calculated among college graduates only.
Lockdown stringency and population-weighted density also have statistically significant coefficients. Thus, countries with stricter and longer lockdowns during COVID-19 and with a higher population-weighted density tend to have higher WFH levels. But the magnitude is smaller than that predicted by the individualism scores. Moving from the 10th percentile (Sweden) to the 90th (Austria) of our lockdown stringency index implies an increase of 0.22 WFH days per week. The implied increase is smaller, at 0.15 days WFH per week, when moving from the 10th percentile country (Poland) to the 90th (Türkiye) by population density. In contrast, we do not find statistically significant coefficients for GDP per capita and industry mix, and the coefficients are also smaller in magnitude.
We also calculate our country-level average level of WFH separately for each sex and re-run the specification with the full set of regressors. Individualism predicts higher levels of WFH for both men and women, but lockdown stringency and population-weighted density yield statistically significant estimates only for women.
How much of the international variation in WFH levels can our explanatory variables account for, collectively and individually?
Figure 3 plots the R-squared from our regressions. Collectively, our regressors account for 56% of the variation in WFH levels across our 34 countries. Individualism has the largest explanatory power among our variables, with a univariate R-squared of 37%. Other variables account for much less of the cross-country variation. Even combining (naïvely) the univariate R-squared values of these other variables yields a lower figure than the R-squared value for the regression of WFH rates on individualism.
Figure 3 Percentage of cross-country variation accounted for by each variable
Notes: The figure plots the R2 of the cross-country regression of the average full paid days WFH on each variable. The first bar in each figure shows the R2 of the cross-country regression including all five variables. Average full paid days WFH calculated among college graduates only.
How much of the individual-level variation in WFH rates within the US can we account for?
We complement our cross-country analysis by investigating whether we see similar predictors of WFH across individual workers in the US. We measure the number of full paid days of WFH among respondents in the Survey of Working Arrangements and Attitudes in 2023. We regress this WFH measure on state-level income, state-level lockdown stringency during the pandemic, local population density, the WFH propensity of the individual’s industry, and voting patterns as a measure of politics and culture. As with our cross-country analysis, we standardise our explanatory variables to ease interpretation.
We show these results in Figure 4. Local population density has the largest coefficient among the set of regressors, followed by WFH propensity in the worker’s industry, lockdown stringency, and Joe Biden’s county-level vote share in the 2020 presidential election.
As with our cross-country results, we find population density is a stronger predictor of WFH levels among women than among men. For men, the political-cultural environment is relatively more important, especially in the full sample, but this pattern contrasts with roughly equal explanatory power that individualism has for men and women’s WFH share in the cross-country analysis. Not surprisingly, our explanatory variables explain much less of the individual-level variation in US data than do a similar set of variables when looking at mean WFH outcomes across our 34 countries.
A comparison of which variables have the largest univariate R-squared is broadly consistent with our analysis of the regression coefficients. Population density explains 1.8% of the individual-level variation. The local Biden vote share and the WFH propensity of the worker’s industry complete the top three. Thus, partisan affiliation, which reflects cultural factors as well, is a predictor of individual-level WFH when looking across individual Americans.
Figure 4 Percentage of individual-level variation accounted for by each variable
Notes: The figure plots the R2 of the individual-level regression of full paid days WFH on each variable. The first bar in each figure shows the R2 of the individual-level regression including all five variables. Average full paid days WFH calculated among college graduates only.
Concluding remarks
We document some interesting patterns in WFH intensity across countries and people. Clearly, we need deeper investigations to explore the causal determinants of these differences. As businesses and policymakers navigate the post-pandemic world, we also need to understand why workers and firms opt for particular working arrangements and to assess the resulting implications for productivity, wages, work-life balance, workplace cultures, and urban planning.
This column only scratches the surface of the evidence and analysis in our larger body of work. All Global Survey of Working Arrangements and Survey of Working Arrangements and Attitudes data are freely available for use by researchers at https://wfhresearch.com/.
References
Adrjan, P, G Ciminelli, A Judes, M Koelle, C Schwellnus, and T M Sinclair (2023), “Unlocked potential: Working-from-home job postings in 20 OECD countries”, AEA Papers and Proceedings 113: 604–8.
Adrjan, P, G Ciminelli, A Judes, M Koelle, C Schwellnus, and T M Sinclair (2022), “Working from home after COVID-19: Evidence from job postings in 20 countries”, SSRN, 18 March.
Aksoy, C G, J M Barrero, N Bloom, S J Davis, M Dolls, and P Zarate (2022), “Working from home around the world“, Brookings Papers on Economic Activity, Fall 2022.
Alipour, J-V (2023), “Kein Homeoffice ist auch keine Lösung”, ifo Schnelldienst 10/2023: 35–38.
Alipour, J-V, O Falck, and S Schüller (2023), “Germany’s capacity to work from home”, European Economic Review 151.
Alipour, J-V, C Langer, and L O’Kane (2021), “Is working from home here to stay? A look at 35 million job ads”, CESifo Forum 22(6): 44–46.
Althoff, L, F Eckert, S Ganapati, and C Walsh (2022), “The geography of remote work”, Regional Science and Urban Economics 93: 103770.
Barrero, J M, N Bloom, and S J Davis (2021), “Why working from home will stick”, NBER Working Paper 28731.
Barrero, J M, N Bloom, and S J Davis (2023), “The evolution of work from home”, Journal of Economic Perspectives 37(4): 23–50.
Bloom, N, J M Barrero, S J Davis, B Meyer, and E Mihaylov (2023), “Survey: Remote work isn’t going away – and executives know it”, Harvard Business Review, 28 August.
Dingel, J I, and B Neiman (2020), “How many jobs can be done at home?”, Journal of Public Economics 189.
Hansen, S, P J Lamber, N Bloom, S J Davis, R Sadun, and B Taska (2023), “Remote work across jobs, companies, and space”, NBER Working Paper 31007.
Heine, S J (2008), Cultural psychology, W.W. Norton and Company.
Hofstede, G (2011), “Dimensionalizing Cultures: The Hofstede Model in Context”, Online Readings in Psychology and Culture 2(1).
Triandis, H C (1994), Culture and social behavior, McGraw-Hill.
Triandis, H C (1995), Individualism and collectivism, Westview Press.
Zarate, P, M Dolls, S J Davis, N Bloom, J M Barrero, and C G Aksoy (2024), “Why does working from home vary across countries and people?”, CEPR Discussion Paper 19003.
