Occupation and working outcomes during the Coronavirus Pandemic

A first outcome of interest to develop our research question is the event “work interruption” experienced by the respondents during the first wave of the Pandemic. This outcome is elicited through the question: “Due to the Corona crisis have you become unemployed, were laid off or had to close your business?”. Note that in this question respondents are instructed to answer “yes” also when they have only temporarily suspended their working activity. In order to estimate the parameters of interest, we define a categorical variable, which takes value one if the respondent reports work interruptions and value zero otherwise. The descriptive table provided in the supplementary material show that 18% of individuals in our sample declared a work interruption. Figure 1 shows the fraction of work interruptions by gender and country: significant heterogeneity emerges among countries and unconditional frequencies do not show any clear gender patterns.

The fraction of women who temporarily or permanently stopped their activity is particularly high in Israel and Greece but lower than for men in Luxembourg, Latvia and Lithuania. As we argued, in order to explain these patterns, one needs detailed information on the characteristics of the labour market and individual characteristics, including demographics and type of activities performed at work.

A second outcome is the intensive margin: i.e. the length of a work interruption, based on the question: "How long were you unemployed, laid off or had to close your business?" –—measuring the number of weeks of interruption. This variable lends itself to different possible specifications, as we shall later explain. As a first approximation, we define a categorical variable taking three possible values: value zero if respondents continued working (82% of individuals), value one if they experienced a "short" interruption (around 11% of them experienced an interruption between 1 and 8 weeks) or value two if they stopped working for more than 8 weeks (7%).³ Figure 2 describes the distributions of the length of respondents’ work interruptions conditional on having undergone such events, by country and gender. The graph documents important differences between men and women—particularly strong in several countries such as the Czech Republic, Latvia, Spain or Sweden —as well as large heterogeneities among countries.

The descriptive evidence provided in Fig. 1 outlines the major role of the Pandemic in changing individuals’ working patterns. Differences in labour market experiences could be related to multiple factors: partly to the stringency of lockdown measures in a given country, partly to pre-pandemic socioeconomic conditions and largely to the intrinsic characteristics of the job. In this paper, we build a unique dataset by combining the SHARE COVID-19 Survey with information from the regular SHARE survey up to wave 8.⁴ This process allows us to create a detailed dataset of work characteristics before and during the Pandemic. The novelty relates to the use of the ISCO-08 3-digit codes associated with the job performed that are collected for working respondents in waves 6 through to 8.⁵ This linkage provides us with a very large set of occupations from which we infer jobs characteristics, as well as details of tasks that workers are involved in.

However, potential drawbacks may arise from this wealth of information. Due to the density of the ISCO-08 codes (3-digit), the set of possible occupational titles is so wide that some categories may not be well-represented. Moreover, the use of a full set of indicators in pooled estimations may also translate into a loss in degrees of freedom due to the large number of explanatory variables that need to be included (“curse of dimensionality”). It is worth recalling that we are looking at a sample of Europeans aged 50 and over, meaning that job characteristics potentially relevant for younger workers may not apply in our study.

To overcome these issues—while taking advantage of the richness of such a detailed job classification—we exploit multiple aspects related to occupations characteristics. First of all, we classify jobs according to a dimension that was deemed relevant during the COVID-19 Pandemic: the essential nature of goods or services produced and provided. This variable identifies workers who perform crucial tasks, spanning from highly skilled professionals such as doctors to low-skilled workers, like food processers. More precisely, we take advantage of the list of ISCO-08 3-digit codes as identified by Fasani and Mazza (2020) and available in Table 1.⁶ Secondly, by following Basso et al. (2020) methodology, we built two indexes meant to measure two key features of occupations: (i) the extent to which an occupation could be technically performed from home and (ii) a measure of the level of social interaction when performing the job. Both indices are generated at an ISCO-08 3-digit level, using pre COVID-19 data elicited from the Bureau of Labour Statistics (BLS) O*NET Survey data 2018. Thus, it is worth emphasising that they are built on pre-pandemic characteristics, that is, the way jobs were carried out in “normal times”.⁷

The Bureau of Labour Statistics (BLS) O*NET Survey data 2018 provides a detailed description of work traits for each job in the USA (“work activities” and “work context”).⁸ Based on a broad range of questions, Basso et al. (2020) define four groups of occupations, capturing different “degrees of safeness” in the workplace and classify each ISCO-08 3-digit code accordingly. The most restrictive definition of safeness includes jobs potentially performed remotely (category 1). The second and third categories relax the previous definition and incorporate, besides the occupations in the first class, also jobs with a ‘low physical proximity and limited exposure to customers and to the public’ (category 2) and jobs with a “higher degree of interactions with external customers, but the level of physical proximity remains low” (category 3).⁹ The fourth group is a residual class—labelled as “unsafe jobs”—including all the remaining jobs with a relatively high risk of contagion. Although we use the same set of O*NET questions and thresholds, we do not stick to their definition of safeness. We completely separate the concepts of remote work feasibility and level of social interactions by defining two distinct indexes ranging between 0 and 1:

Table 2 displays the values of these two indexes for each ISCO-08 3-digit job. Interestingly, we do not find a defined pattern characterising these measures: while some occupations display high remote work feasibility and low social contacts (e.g. legal professionals), others are technically teleworkable but would normally require intensive social interactions (for example 233 — secondary education teachers). This is also easily observed in Fig. 3, which matches the two indexes for a selection of ISCO-08 sub-majors in the left panel and for several ISCO-08 3-digit codes in the right panel. This comparison gives an idea of the significant heterogeneity, which exists among occupations even within the same sub-major. For instance, important differences appear when looking at professional teachers (23) and their detailed decomposition by 3-digit codes. University and higher education teachers (231) display high remote work feasibility and mild social contacts, but primary school and early childhood teachers (234) on the other hand, require strong interactions with children. Furthermore, secondary school teachers (233) would normally require intensive social interactions yet have features which are well suited to home working.

All these examples reveal the importance of using detailed job information in order to better understand the role of job characteristics during the first wave of the Pandemic.

Table 3 reports the marginal effects of the probability of work interruptions for two specifications: the first column—model 1—is a parsimonious specification in which, besides the occupation specific variables (i.e. the essential nature of a job, the remote work feasibility index and the social interaction index), we include gender, age and country of residence. In model 2, we also control for education, information technology skills, type of employment and health status.

Our results point out job characteristics as major determinants of the probability to experience work interruptions during the first—unexpected—wave of the Pandemic. Indeed, individuals employed in “essential” activities were 3.5 percentage points less likely to have gone through work breaks than those working in “unessential” jobs. In relative terms, with respect to the average sample probability of 17.9%, individuals employed in essential jobs were 19.5% less likely to experience interruptions with respect to non-essential employees. The marginal effects of the two indexes reveal that increasing suitability of remote work is associated with a significantly lower probability of work interruptions, while the higher the level of social interaction in the workplace, the larger the likelihood of experiencing work breaks.¹¹ Figure 4 provides additional insights by displaying the marginal effects of an essential job at various levels of remote work feasibility index and social interaction index.

It is worth noticing that, for values of the remote working feasibility index smaller than 0.6 (representing little or modest home working suitability), being an essential occupation is associated with significantly lower probabilities of experiencing work interruptions. The positive slope suggests that, as the technical teleworkability of a job increases, the gap between essential and unessential occupations gets smaller with respect to the likelihood of work breaks during the Pandemic. An opposite relationship is found when looking at the social interaction index: jobs characterized by a low intensity of interaction between people display no significant differences between essential and unessential activities. On the contrary, as the level of social contact becomes more important, the difference between the two categories increases (i.e. essential jobs display lower probabilities of work interruption).

One could argue that the three selected job characteristics (i.e. essential/unessential plus the two indexes) might be arbitrary and conceal useful information because they are based on criteria reflecting the COVID-19 Pandemic. Indeed, as the Pandemic occurred, some jobs became more relevant than others and at the same time, some occupations were more prone to home working or less risky in terms of social interaction intensity. In order to show that our proposed measures preserve the value of the original information, we carried out a robustness check (see Table 4) by estimating Eq. (1) with forty dummy variables, one for each job sub-major. Note that in this setup sometimes we treat in a unique group rather heterogeneous occupations, due to how the ISCO-08 2-digit classification clusters jobs. For example, we cannot distinguish between sellers of food (“essential” goods) and those vending other commodities.¹²

We choose “teaching professionals” as the baseline group due to their fairly homogeneous nature in terms of work arrangements options during the Pandemic: most teaching activities continued remotely in almost every European country. With respect to the baseline group, the coefficients show that jobs belonging to other sub-majors had significantly higher probabilities of temporary or permanent work interruptions. Larger and statistically significant effects are associated with occupations related to tourism and hospitality, while jobs in “subsistence agricultural activities” were found to have a lower probability of interruptions. These results are in line with our main specifications.

Table 5 reports the marginal effects of two ordered probit models for the length of work interruptions. Individuals working in essential occupations were about 1.3–1.6 percentage points less likely to experience longer work interruptions (columns 2 and 3 respectively) and more likely to go through brief episodes (less than 1 week) or no activity stop (column 1 of each specification), with respect to the “unessential” ones. In relative terms, being employed in an essential activity determines a reduction in the probability of a brief or long interruption of about 12.15% and 22.22%, respectively. Instead, the magnitude of the effect when considering the probability of zero weeks of interruption is much smaller, i.e. no interruptions at all (+ 3.53%). Jobs with high suitability to remote work display significantly lower probabilities of longer work breaks, while those with a large intensity of social interactions have higher likelihoods of prolonged interruptions. The results are consistent with those found in the estimation of the probability of stopping work. As a robustness check, we also perform a Tobit regression model using the number of weeks of interruption as a continuous dependent variable. The results support our findings and are available as supplementary information.

Additional insights into the impact of job features and their magnitude are provided by Figs. 5 and 6: they show the average marginal effects of an essential occupation on the probability of having experienced 0, 1 to 8 and more than 8 weeks of work interruption at different values of the two indexes. Workers employed in “essential” occupations unsuited to remote work (index values smaller than 0.5) display a significantly lower probability of having prolonged work interruptions with respect to workers performing “unessential” jobs. Such a difference vanishes as home work feasibility increases. A similar impact—but opposite in sign—is observed for the social interaction index: jobs characterised by intensive social contacts but regarded as “crucial” in society, reveal a reduction in the likelihood of experiencing longer work interruptions with respect to non-essential ones. This effect disappears at lower values of the social interaction index.

In addition to the previous finding, our results add salient evidence on several other issues. We find that education has a clear mitigating role for the negative labour market effects of the Pandemic, even when controlling for occupation features. Respondents holding higher levels of education (vis-a-vis the reference category “high school degree”) display a 3.5 percentage points lower likelihood of work interruption and about 2 percentage points smaller probability of undergoing prolonged interruption spells. We speculate that educational attainment plays a relevant role per se, both because workers with higher education are often associated with “higher quality jobs”, and also because education is related to the specific tasks required in a job. The idea is that the human capital of highly educated workers may be more flexible in terms of tasks performed. By recalling the basic characteristics defined by Autor and Dorn, 2009, Autor and Dorn 2013 and Deming 2017, the exogenous shock generated by the Pandemic has probably affected more jobs involving tasks of high routine intensity, i.e. tasks that involve a well-defined repetitive set of procedures. In fact, during economic downturns, sizeable employment losses mainly appear among the more routine-intensive middle-skilled occupations, some of these jobs eventually disappear and are not retrieved when the economy recovers (Jaimovich and Siu, 2020).

Finally, our results also highlight differences between workers in different types of employment. With respect to the baseline category of the private employees, public employees were 8.3 percentage points less likely to have experienced work interruption whereas self-employed workers had significantly larger probabilities of such an event. Moreover, public sector employees are characterised by a 4.1 percentage points lower probability of having experienced work interruptions between 1 and 8 weeks, and 4.2 percentage points smaller likelihood of breaks longer than 8 weeks. We find an opposite and significant effect for self-employed workers.

The previous models allow us to address several questions that are currently the object of debate for researchers and policy makers. Did women pay a higher price than men in terms of work interruptions during the Pandemic? Are there heterogeneities in terms of job characteristics useful to build more targeted (and potentially more effective) support measures?

By recalling that particular care should be paid in drawing general conclusions—our sample looks at workers aged 50 and over—we attempt to provide answers to the above questions. When introducing a “female dummy” in the above models, we find that women in our age groups are more likely to experience work interruptions with respect to men (about 3.9 percentage points more), and longer work breaks (by 1.8 pp more for interruptions between 1 to 8 weeks, and by 2.1 pp more for episodes longer than 8 weeks). In relative terms, women have been 21.79% more likely to experience working breaks than men. Moreover, by looking at the duration of such interruptions, they also display higher probabilities of short and long breaks of about + 16.5% and + 29.16%, respectively. In order to get further insights, we run the regressions separately by gender. Table 6 reports the results of these estimations both for the probability and for the length of work interruptions.

It is easy to observe that a large part of the effect captured in all the main specifications by the essential nature of a job is mostly driven by women. Female workers employed in essential activities are 5.7 percentage points less likely to experience interruptions than those employed in non-essential ones. Differently, male workers seem more vulnerable as the level of social interaction at the usual workplace increases. As expected, the remote work feasibility of a job has been a crucial determinant during the first wave of the COVID-19 Pandemic, irrespective of gender. Figs 7, 8, 9, 10 depict the average marginal effects of being employed in essential occupations (with respect to non-essential ones) at different levels of remote work feasibility and social interaction, for men and women separately. The first type of interaction points out the home work feasibility as the prevailing dimension among women: at lower levels of the remote work feasibility index the essentiality of tasks performed by women is highly significant to avoid work interruptions as well as longer breaks, while this is not the case for men. As regards our second index, for high levels of social interaction at work, the essential nature of an occupation represents a deterrent against job interruption mainly for women, while the opposite is found for men.

Several aspects such as gender and age composition of specific jobs can partially explain the previous results. For instance, the prevalence of women tends to be higher among essential but more exposed to contagion (intensive social interaction) activities (i.e. nursing and midwifery professionals (222) or primary school and early childhood teachers (234)), while male workers prevail among essential but lower risk occupations (i.e. heavy truck and bus drivers (833) or mixed crop and animal producers (613)). No less important is the role played by intergenerational differences and thus, the representativeness of our sample of 50 + workers: the gender selection into specific jobs—more or less demanding in terms of tasks—might be highly pronounced among older cohorts compared to younger ones.

Overall, the previous findings show that the negative effects of the Pandemic on workers were harsher on women. However, the results also reveal that gender differences in labour market outcomes are driven by the intrinsic characteristics of the jobs/occupations they are involved in.