1.1 Data sources

The main data sources of this paper are the World Bank Group’s Enterprise Surveys (ES) and Covid-19 follow-up surveys (COV-ES). The ES is a nationally representative dataset of registered firms in the private sector with five or more employees. The ES excludes firms in the agricultural, mining and several service sectors such as health and social work, real estate or research and development.4 As the agricultural sector is excluded, the data is naturally not representative of all key sectors within the countries. Even though agriculture employs more than 60% of the workforce in many developing countries, the dataset allows the most comprehensive analysis of firms in key sectors across countries of different income levels. Within each country, the sample of firms is stratified by industry, size and location. The data includes sampling weights to correct for selection probability. The COV-ES is a rapid phone survey in response to the pandemic (Apedo-Amah et al. 2020). The sample of the COV-ES consists of enterprises in a baseline ES between 2016 and 2020, which were re-interviewed in up to three waves (at the time of writing) since May 2020. This study focuses on the first wave of the COV-ES data, which has the largest coverage and is best comparable between countries. Besides Armenia, Azerbaijan, Bosnia-Herzegovina, Montenegro and Serbia, for which the first wave of data was collected in early 2021, the data stems from 2020. The questionnaire consists of items about the impact of the pandemic on a firm’s operating status, its sales or employment levels, any operational adjustments and government assistance. The COV-ES data furthermore includes adjusted sampling weights. To ensure equal weights across countries, within country survey weights are re-scaled to sum to 1 (Muzi et al. 2021). This ensures that firms in one country are not over-proportionally represented in the cross-country sample.

At the heart of this paper is the classification of firms into key and non-key sectors by each country. The sources of these lists vary from official government gazettes with essential sectors defined by sector-level codes to newspaper articles. For each country, these lists were translated into the ES data’s ISIC Rev. 3.1. sectoral coding, either through crosswalks or by matching sectoral descriptions with the ISIC rev. 3.1. sector definitions. Manual adjustments were implemented in cases when the lists were not directly translatable to the sectoral classification.5 For instance, certain keywords, such as “hygiene products” or “pharmaceuticals”, were cross-referenced with a variable in the ES data containing a description of the main product or main service of the firm. Responding business owners or managers wrote down a sentence or several keywords to describe the main operation of the firm. Based on the list of key sectors, a variable was then created which indicates whether the firm operates in a key sector of the respective country, or if it produces a good/ offers a service which is defined to be key. An exception here is Lebanon, where the survey data includes a variable indicating whether the enterprise is key or not.

Nevertheless, the quality of some of the countries’ lists does not allow a meaningful classification into key and non-key sectors. Of a total of 46 countries with COV-ES data6, for 12 countries no list could be found, for 6 countries the quality of the list was too poor for classification or the share of firms in key sectors would have been larger than 95%, and one country had no available baseline ES data (see Table A.1 in the Appendix). Countries for which there is no meaningful list of key sectors are relatively evenly spread geographically. The final sample consist of 27 countries.

1.2 Descriptives

To get an overview over how the pandemic affected firms in key and in non-key sectors, this section provides descriptive evidence before turning to the empirical analysis in the subsequent section. Table 1 contains summary statistics of the sample of 27 countries, which consist of countries in Africa, Latin America, Central and East Asia, Western and Eastern Europe and the Middle East. The share of firms in the respective country’s key sectors differs substantially between countries from 20 per cent up to 90 per cent. Regional patterns are only evident up to a certain extent. For instance, while Latin American countries all exhibit a low share of firms in key sectors, the share differs strongly between Eastern European countries. Similarly, across-countries there is substantial variation with regard to which sectors are defined as key. Table 2 displays the share of key firms in each aggregated sector. In most sectors, the share of key firms lies between 35 and 55 per cent. Thus, within-sector variation comes from different definitions of key sectors between countries. Not surprisingly, the food and beverages as well as the coke and chemicals (which includes pharmaceuticals) sectors exhibit the largest share of key firms, in contrast to the accommodation and restaurant sector which has the smallest share.

Table 1: Sample composition

Table 2: Share of Key firms by sector

The choice of governments on which sectors and firms to define as key is by nature not random. Governments needed to consider whether a sector is vital for the functioning of the society, but also if it is of paramount importance for the country’s economy. Table 3 displays average firm characteristics by key and non-key firms, and t-statics of the difference between the two.7 As could be expected, the characteristics of key firms differ from non-key firms. Key firms are larger, both in terms of baseline sales and employment, and older. Furthermore, key firms are more likely to export. These differences show the necessity to control for baseline firm characteristics in the empirical estimation. In Table 4, the summary statistics of all dependent variables are presented. Seventy per cent of firms reported a decline in sales, while slightly less than half had to close at some point after the start of the pandemic. On average, firms reported fewer employees than before the pandemic, with male employment responding stronger than female employment. The number of furloughed workers is larger than that of workers whose wage was cut and the number of workers that took leave. A quarter of firms started or increased online activities and deliveries, while a third started or increased remote working arrangements. In terms of channels through which the pandemic affected firms, two-thirds experienced a decline in demand and slightly more than half a decline in input supplies or working hours. Lastly, a third of firms have received government support during the pandemic.

Table 3: Mean differences between key and non-key firms

Estimations are weighted with the re-scaled sampling weights and include country fixed-effects, 1-digit sector fixed-effects and baseline year fixed-effects.

Table 4: Descriptive statistics of all dependent variables

Descriptive statistics of all dependent variables.

Figure 1 displays how the pandemic affected the operational status and sales of firms. Panel a) shows that 72 per cent of firms in key sectors remained open throughout the pandemic, which is true for only 28 per cent of non-key firms. While this draws a stark contrast between key and non-key firms, it also shows that there is not a one-to-one correlation between the operational status and being in a key sector. Panel b) of the same figure shows that the majority of firms, regardless of being deemed key, experienced declining sales during the pandemic. Nevertheless, key firms less often reported declining sales (62 per cent vs. 82 per cent) and more often reported unchanged (25 per cent vs. 12 per cent) or even increasing sales (11 per cent vs. 5 per cent), as compared to non-key firms. Whether this relationship holds in an empirical model and controlling for a number of characteristics along different dimensions will be investigated in the next section. This shows that there are also winners in the pandemic, albeit the share being small. Figure A.1 in the Appendix shows the same trends but with respect to sectors instead of key firms.8 Interestingly, most sectors have a relatively equal share of firms that closed at some point and those that remained open throughout. This again shows that it is not necessarily only sectors across countries which are decisive for how firms are affected by the pandemic, but that being deemed key marks a starker contrast. The accommodation and restaurant sector exhibits the largest decline in sales and the largest proportion of firms that closed at some point (Aga and Maemir 2021). Firms in the coke, petroleum and chemicals sector as well as the vehicles and other transport sector were most often able to increase their sales.

Figure 1: Impact of the Covid-19 pandemic on firms

Notes: The sample consists of 9169 firms in Panel A and 9873 firms in Panel B. COV-ES sampling weights are applied

While declines in sales have been substantial, fewer firms have decreased their number of employees (Figure 2). Furthermore, the difference between key and non-key firms here is smaller than for sales. Two-thirds of both types of firms reported that the number of permanent employees remained the same, 29% of non-keys and 23% of keys reported decreases (panel a)). The share of firms that stated an increasing number of employees is similar to those who reported increasing sales (11% vs. 6%). With respect to temporary employment, key firms more often reported that the number of employees remained the same and fewer firms overall increased the number of employees (a larger share of firms reported “Don’t know”).

Figure 2: Impact of the Covid-19 pandemic on employment

Notes: The sample consists of 9602 firms in Panel A and 9420 firms in Panel B. COV-ES sampling weights are applied.

A similar picture arises with respect to channels through which the pandemic has affected firms the most, namely hours worked, demand and supply of inputs. Panel a) of Figure 3 shows that key firms less often report decreases in all three categories. While hours worked affected less than half of key firms, the fact that still almost 40% of key firms reported a decrease in the category indicates that worker supply was not only affected through government mandated closures. Declining demand is the strongest negative channel, both for key and non-key firms. However, 14 per cent of key firms reported increases in demand, which is in line with the share of firms reporting increasing sales. Roughly the same share of key firms reported that the supply of inputs decreased or remained the same, while two-thirds of non-key firms experienced declining supplies. Turning to firm-level adjustments in the aftermath of the pandemic, a different picture emerges (panel b)). Non-key firms actually more often adjusted their mode of operation as compared to key firms. The differences are starkest with respect to online activities. One in three non-key firms increased their online performance, in contrast to only one in five key firms. In fact, there is evidence that the crisis accelerated the expansion of e-commerce, which is likely to sustain in the long-term (OECD 2020a). Remote work was adopted by a similar share of both types of firms.

Figure 3: Channels and adjustments

Notes: The sample consists of 9602 firms in Panel A and 9420 firms in Panel B. COV-ES sampling weights are applied.

Lastly, panel a) of Figure 4 indicates that key firms more often received government support than non-key firms. Interestingly, the type of support that key and non-key firms received differs (see panel b)). While key firms more often received cash transfers or wage subsidies, i.e. immediate liquidity boosts, non-key firms more often benefitted from more indirect measures, namely payment deferrals and better access to credit. Therefore, governments apparently differed between firms in key and non-key sectors with respect to the type of assistance issued.

Figure 4: Government support

Notes: The sample consists of 9509 firms in Panel A and 4579 firms in Panel B. COV-ES sampling weights are applied

2.1 Estimation strategy

While the COVID-19 pandemic can safely be viewed as an external and unexpected shock, firms are not randomly sorted into key and non-key sectors. Therefore, it is necessary to control for firm-level as well as sector-level characteristics, in order to properly identify the effects of being deemed key. The baseline linear probability model regression specification is as follows:

${\mathrm{Y}}_{\mathrm{i}\mathrm{s}\mathrm{c}}=\propto +{\mathrm{\beta }}_1{\mathrm{K}\mathrm{e}\mathrm{y}}_{\mathrm{i}\mathrm{s}\mathrm{c}}+{\mathrm{\beta }}_2{\mathrm{C}}_{\mathrm{i}\mathrm{s}\mathrm{c}}+{\mathrm{\theta }}_{\mathrm{s}}+{\mathrm{\delta }}_{\mathrm{c}}+{\mathrm{\tau }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{i}\mathrm{s}\mathrm{c}},$ (1)

where subscript i denotes the firm, s denotes the sector and c the country the respective firm operates in. ${\mathrm{Y}}_{\mathrm{i}\mathrm{s}\mathrm{c}}$ is the outcome of interest, such as the reported pandemic-induced change in sales or change in employment. The coefficient ${\mathrm{\beta }}_1$ measures the effect of the main explanatory variable, the firm being regarded as key.9 Country fixed-effects are denoted by ${\mathrm{\delta }}_{\mathrm{c}}$, which account for differences in country-level conditions, how hard the pandemic hit the country of operation and responses to the pandemic (OECD 2021). The term ${\mathrm{\tau }}_{\mathrm{t}}$ indicates a fixed-effect for the year in which the baseline ES survey was conducted. The latter variable accounts for differences in timing between the ES and COV-ES surveys. Further, ${\mathrm{\theta }}_{\mathrm{s}}$ represents sector fixed-effects (either at the 2- or 4-digit level), which control for all sectoral characteristics. Therefore, the regression results will not be driven by the pandemic differentially affecting single sectors, e.g. benefitting the chemicals sector or damaging the accommodation and restaurant sector. Vector ${\mathrm{C}}_{\mathrm{i}\mathrm{s}\mathrm{c}}$ represents a set of control variables, which are based on literature using the ES data to study firm-level outcomes (Amin and Viganola 2021; Aga and Maemir 2021).

The control variables aim to capture underlying firm and economy characteristics, without which the estimation would suffer from omitted-variable-bias.10 Firstly, larger, older and more productive firms may have a better capacity to cope with the negative consequences of the pandemic (Belitski et al. 2021; James Hurley et al. 2021; Amin and Viganola 2021; Kozeniauskas, Moreira, and Santos 2020; Liu et al. 2021). Firm level performance is controlled through the number of permanent employees, sales in the baseline year (both in logs) and the age of the firm. Furthermore, dummies indicating whether the firm belongs to a multi-establishment enterprise or whether the firm is to some part owned by the government or state are included. Subsidiary firms are likely to benefit from their parent company through transfers and shock-absorbing capacities (Liu et al. 2021; Amin and Viganola 2021). Similarly, it has been shown that firms with partial state-ownership have larger sales and employment rates during the pandemic, which is explained by soft-budget constraints (Liu et al. 2021; Megginson, Ullah, and Wei 2014). More ambiguous is the relationship between foreign-ownership or export orientation and the consequences of the COVID-19 pandemic. On the one hand, foreign-owned and exporting firms tend to be larger and more resilient (Aga and Maemir 2021). On the other hand, these firms are more reliant on global supply chains, which have undergone tremendous shocks during the pandemic (Guan et al. 2020; Janzen and Radulescu 2021; Liu et al. 2021). Furthermore, recent evidence suggests that the gender of business-owners or managers plays a role with regard to how a firm deals with the pandemic, both in terms of adjustments and responses (Liu, Wei, and Xu 2021). Therefore, a dummy for the manager or business owner being female is included. Lastly, with respect to the business environment, variables indicating whether the establishment experienced any power outages and whether it has a line of credit or loan are added. Additional controls are added for robustness tests. The first is the management or business-owner experience, as the quality of management influences how it responds to the effects of the pandemic (Grover and Karplus 2021). Furthermore, baseline differences in e-commerce use are controlled for with a dummy of the firm having a website. In reference to the business and regulatory environment, a dummy for whether the management spends time on any regulations is added. Lastly, the COV-ES interview month captures the fact that interviews were held at different points in time after the start of the pandemic.

2.2 Estimation results

Table 5 provides baseline regression results, based on a linear probability model. The dependent variable is an indicator whether the firm experienced a decline in sales since the start of the pandemic. Throughout, country fixed-effects and the baseline year are controlled for. The first three columns include 1-digit sector fixed-effects and the last two columns 4-digit sector fixed-effects. Column 1 displays the coefficient of the Key firm dummy, without further controls. It is negative and statistically significant, indicating that firms deemed as essential to the economy and society were less likely to experience a decline in sales. Column 2 adds baseline firm-level characteristics, which account for endogeneity concerns with respect to firms being valued as key. Notably, firms with larger sales a priori are less likely to have decreasing sales through the pandemic. Thereafter, additional control variables are included, which characterize the firm’s organizational structure, exporting-status and infrastructure and credit access. Multi-establishment firms appear to have been able to better deal with the pandemic than individual enterprises. Using 4-digit sector fixed-effects reduces the size of the main coefficient, however it remains statistically significant at the 1 per cent level. The fact that the coefficient remains highly significant is reassuring, as all industry-specific characteristics on a highly disaggregated level are controlled for. Therefore, variation in the Key variable does not appear to be driven by single industries, which are always deemed essential across countries, even when disaggregated at the 4-digit level. The country-level classification of key sectors appears to matter in terms of how strong the pandemic impacted single firms. However, one could argue that the Key variable simply reflects the fact that non-key firms were mandated to close during the pandemic, and thereby by construction showing a negative sales effect of these firms. Column 5 controls for the firm having had to close at some point since the start of the pandemic. While this further reduces the size of the coefficient, it does not change its statistical significance. The inclusion of this variable furthermore diminishes concerns that the definition of key sectors is an ad-hoc measure of the operational status of establishments during the pandemic. Overall, the table presents robust evidence that key firms were able to better deal with the consequences of the pandemic. It should be emphasized, however, that majority of key firms nonetheless experienced declines in sales. On a further note, firms with female managers as well as multi-establishment firms less often reported decreasing sales.

Table 5: Estimation results - decline in sales

Standard errors, in parentheses, are clustered on the country level. Estimations are weighted with the re-scaled sampling weights

* p<0.10, ** p<0.05, *** p<0.01

In Table 6, the operational-status is used as the outcome variable in columns 1 and 2. The results indicate that key firms were less likely to close at some point, even with 4-digit sector fixed-effects. Columns 3-6 switch to changes in the number of employees (measured as the inverse hyperbolic sine to account for negative values) as outcome: In columns 3 and 4 total permanent employees, in column 5 female permanent employees and in column 6 male permanent employees. The Key variable is only significantly different from zero with 1-digit sector fixed-effects for total employment. This would indicate that non-key firms had a stronger decrease in overall employment, which however is not robust to 4-digit sector fixed effects. The Key coefficient is of larger magnitude in the estimation with female employment as the outcome as compared to male employment, but both are not statistically different from zero. Note however that due to missing values of either baseline employment in the ES data or missing values of employment in the COV-ES data, the sample size is starkly smaller for the employment outcomes than for the question on whether sales have declined.

Table 6: Estimation results - firm status and employment change

Standard errors, in parentheses, are clustered on the country level. Estimations are weighted with the re-scaled sampling weights.

* p<0.10, ** p<0.05, *** p<0.01

Table 7 in turn looks at different types of labour adjustments. The first two columns show a significant difference between the number of furloughed workers in key and non-key firms. To avoid large losses in permanent employees, non-key firms appear to have more often resorted to furloughing as compared to key firms, making use of government supported short-working time arrangements. Furthermore, firms deemed non-key appear to more often have reduced number of temporary workers (column 3), the effect however not holding with the inclusion of 4-digit sector fixed-effects11. Lastly, there is a significant difference between the two types of firms with respect to the number of workers that took leave or quit, with 4-digit sector fixed-effects. One can only hypothesize about why this is the case. The opposite effect could have been expected, as workers in key firms oftentimes faced large burdens and psychological stress (Béland, Brodeur, and Wright 2020). A possible explanation could be that key firms in some countries had better access to PPE and vaccination, as compared to non-key firms. The results should be treated with caution though, due to the large number of missing values.

Table 7: Estimation results - employment adjustments

Standard errors, in parentheses, are clustered on the country level. Estimations are weighted with the re-scaled sampling weights.

* p<0.10, ** p<0.05, *** p<0.01

The pandemic has induced firms to adjust their modus operandi. With stores and firms having to close, many companies both in developing and developed countries have moved their operations online, digitalized their working process, resorted to deliveries and initiated remote working possibilities (OECD 2020a). Table 8 shows that the difference in uptake of these modes of operation is relatively small between key and non-key firms. Even though non-key firms had to close more often, they do not appear to more often have resorted to deliveries (columns 3 and 4) or remote work (columns 5 and 6). However, there does seem to be a difference with respect to moving business activities online, though the Key coefficient is statistically significant only at the 10 per cent level (column 2). Increasing usage of online activities may well be driven out of necessity for non-key firms. Similarly, these firms more often converted their production or service, as is displayed in Table 9. This is however only true for firms outside of retail and wholesale trade. Only when these firms are excluded in columns 3 and 4, the Key coefficient becomes statistically significant. Therefore, what has been shown anecdotally for individual firms (Manolova et al. 2020; Laato et al. 2020), is here confirmed for non-key firms outside the retail and wholesale sectors. Non-key firms thus have in parts been innovative in their reaction to the pandemic. Some of these firms have started or increased the use of online platforms and have changed the main product or service that they offer.

Table 8: Estimation results - operational adjustments

Standard errors, in parentheses, are clustered on the country level. Estimations are weighted with the re-scaled sampling weights.

* p<0.10, ** p<0.05, *** p<0.01

Table 9: Estimation results – change in products or services

Standard errors, in parentheses, are clustered on the country level. Estimations are weighted with the re-scaled sampling weights.

* p<0.10, ** p<0.05, *** p<0.01

Regarding channels affecting firms during the pandemic, a clear distinction between key and non-key firms arises across all categories. Table 10 displays regression results with the outcomes being indicator variables for having experienced declines in input supply (columns 1 and 2), declines in demand (columns 3 and 4) and declines in hours worked (columns 5 and 6). The Key dummy is negative and statistically significant in all columns, indicating that all channels had a stronger impact on non-key firms than on key firms. The largest difference between the two types of firms lies in the supply of inputs, closely followed by the demand for the firm’s products or services. That non-key firms experienced stronger declines in demand is not surprising. On the one hand, these firms more often had to close their stores, especially affecting firms in retail. Moreover, the pandemic caused a shift in consumers’ preferences and lower spending especially of high-income individuals on non-necessary goods (Chetty et al. 2020). That key firms had lesser trouble acquiring inputs is most likely driven by the fact that firm’s further upstream the value chain of key firms are also deemed as key. Non-key firms therefore faced a double penalty. Consumers spent less on the goods that they produce or services they provide, while at the same time they faced difficulties in the supply chain (Guan et al. 2020).

Table 10: Estimation results - impact channels

Standard errors, in parentheses, are clustered on the country level. Estimations are weighted with the re-scaled sampling weights.

* p<0.10, ** p<0.05, *** p<0.01

2.3 Robustness tests and extensions

In this section, robustness tests and several extensions to the main results are presented.

Table A.2 in the Appendix adds additional controls to the model. Firstly, the manager’s years of work experience in the sector control for the notion of more experienced managers can better deal with economic downturn. Secondly, a dummy for the firm having a website is included, as firms with experience in e-commerce may be better equipped to transform their business practices. Thirdly, a dummy whether the firm had to spend time on regulations controls for institutional differences and bureaucratic requirements. Lastly, the COV-ES interview month accounts for the interview being held at different points in time after the start of the pandemic. The outcomes in the table are those for which significant effects were previously found and which are most interesting for the analysis. The dummy for the firm being valued as key for the economy remains statistically significant for all outcomes even with the additional control variables. Firms which had a website before the start of the pandemic were more likely to start or increase online business activities. The inclusion of the variable however slightly increases the size of the Key variable’s coefficient. Firms in a more regulated environment are also more likely to increase online business activities, as well as to furlough workers. Most likely, countries with regulated environments have better infrastructure, which therefore correlates with an increase in online activities. Government programs such as furloughing workers by nature requires bureaucratic steps.

While key firms have evidently faced fewer negative consequences of the pandemic, government support may differentially enable key or non-key firms to better cope with the economic downturn. Table A.3 in the Appendix tests these differential effects, by interacting the Key dummy with a dummy of having received government support. The interaction term is not statistically significantly different from zero for any of the outcomes though. This indicates that government support did not benefit key firms more than non-key firms, or vice versa. Furthermore, the coefficient of the government support dummy is positively related to the decline in sales and to the decline in demand. This could indicate that governments targeted firms which faced a larger shock in terms of sales and demand. Recent evidence suggests that, in the case of European countries, government subsidies were relatively efficiently distributed and not over proportionally targeted firms which were unproductive already before the pandemic (Kozeniauskas, Moreira, and Santos 2020; Bighelli and Lalinsky 2021).

Several studies have highlighted that firm size matters with respect to how firms cope with the pandemic. Smaller firms were disproportionally affected by lower demand and disrupted supply chains. Table A.4 tests whether there are any differential effects by firm size between key and non-key firms. The firm size at the time of sampling is used, such that there is no endogeneity effect between the pandemic and firm size. The results exhibit some interesting patterns. For instance, larger key firms are less likely to have experienced a decline in sales, while there is no difference between non-key firms of different sizes. There is also no differential effect with respect to the number of furloughed workers. The previous results showed that non-key firms are more likely to increase online activities, which appears to be driven by larger firms. This may especially be true for less developed countries, where barriers are larger especially for smaller firms. Surprisingly, larger non-key firms more often faced input supply shortages, while the opposite is true for key firms. In contrast to declining sales, declines in demand were more often reported by larger than smaller non-key firms.

Next, we investigate whether the difference between key and non-key firms varies between regions. The countries in the sample are split into four regions:12 Africa and the Middle East, Central and East Asia, Europe and Latin America. As above, the Key dummy is interacted with the regional indicator in Table A.5 in the Appendix. The baseline category, omitted in the table, is the region Africa and the Middle East. The interaction coefficients are negative and statistically significant in most cases, besides for the online business activity outcome. These findings suggest that the categorization of key and non-key firms in Africa and the Middle East is less distinctive relative to other regions of the world. This could mean that firms deemed key in the region had fewer advantages in terms of government support over non-key firms or that government regulations towards non-key firms were less stringent. For instance, much fewer firms in Sub-Saharan Africa received government support during the pandemic as compared to other regions in the world (Aga and Maemir 2021). Therefore, it can be expected that governments in general mandated fewer policies related to the business environment and regulations for firms in the region.

Lastly, Table A.6 tests whether the number of workers which took leave, possibly due to home-care responsibilities or fear of or actual infection, is correlated with any of the other observed factors and a firm being key. For example, key firms which did not offer the possibility of remote work may have had more employees taking leave. Frontline workers saw themselves confronted with large levels of psychological and physical stress especially at the start of the pandemic, leading to exhaustion and burnout symptoms (Béland, Brodeur, and Wright 2020). The results of the table however do not provide evidence for any such effects. Each column interacts the Key dummy variable with an observed adjustment or government support. The only statistically significant interaction is government support, which suggests that firms receiving support had fewer workers taking leave. This could potentially mean that these firms were able to provide better conditions for workers.