Technology and evolution of occupational employment in Egypt (1998–2018): a task-based framework Open Access

The accelerating pace of technological advancements has raised many concerns worldwide regarding the future of work. With the beginning of the fourth industrial revolution in the 21st century, technology has become increasingly capable of undertaking many work activities that were previously confined to humans (Frey and Osborne, 2017). A recent estimation of automation potential in the Middle East documented that Egypt comes first concerning automation potential, with around 48% of its work activities being automatable, placing nearly 12 million workers at high risk of being displaced by technology (Moore et al., 2018). Given the high automation potential in Egypt, understanding the mechanism through which technology displaces workers and the magnitude of its role in shaping the structure of occupational employment are essential in expecting future demand for skills and in tailoring economic policies toward employment creation.

The theoretical framework on the impact of technology on workers witnessed major developments, shifting from a deskilling impact in the late 18th to the early 20th century to a skill-biased impact during the second half of the 20th century. Autor et al. (2003) introduced a relatively recent hypothesis which is the Routine-biased Technological Changes (RBTC). In their view, routine intensity is the degree to which tasks follow explicit rules that make them easily expressed in a computer code and repetitively executed identically by machines. Based on the hypothesis, technology is expected to replace workers performing routine tasks, whereas it is expected to complement those engaged in nonroutine cognitive and interactive tasks. While the hypothesis does not posit a direct link between technology and workers performing nonroutine manual tasks, previous RBTC literature predicts an employment expansion of manual workers if displaced routine workers largely move to manual rather than nonroutine cognitive and interactive tasks. Additionally, as routine tasks are more likely to be found in the middle of the skill distribution, several studies have documented that the RBTC hypothesis is significantly responsible for polarizing employment structure in many countries (Adermon and Gustavsson, 2015; Bisello, 2013; Das and Hilgenstock, 2022; Goos and Manning, 2007; Kampelmann and Rycx, 2013; Sebastian, 2018).

This study aims to validate the prevalence of the RBTC hypothesis in Egypt over the period (1998–2018) by testing it and its main employment implications. The study first undertakes an exploratory analysis to check employment evolution over time and its relationship with occupational task content. Second, it moves to a more formal assessment of the contribution of the task content of occupations in accounting for overall changes in occupational employment over the period. Third, relying on longitudinal data, the study examines individuals' mobility across occupational task categories over the period to validate the presumption that routine workers have a relatively higher tendency to switch their occupational task category than their abstract and manual counterparts and that they predominantly move to manual occupations.

Most empirical evidence on this hypothesis has been directed to developed countries, with limited attention being paid to a developing country like Egypt. Although few studies have addressed the effect of technology on occupational employment (Badran, 2019; El-Hamidi, 2020; Helmy, 2015), many gaps still exist (e.g. crudely allocating occupations into task categories, relying on highly aggregated occupational data, using relatively old occupational classification, and employing insufficient methodologies to control for competing explanations to the hypothesis). The study’s contribution lies in the following: First, it uses direct measures of occupational task content provided by Mihaylov and Tijdens (2019). Second, given that the task content of occupations could differ considerably within aggregate occupational categories, occupational data are disaggregated up to the four-digit level, following the most recent occupational classification (the ISCO-08). Third, the study applies different methodologies that allow for controlling the effect of other competing explanations for the hypothesis.

The rest of the paper proceeds as follows: Section 2 provides a theoretical and empirical overview of the relationship between technology and employment; Section 3 indicates data sources related to labor market information and task content of occupations that the study relied on; Section 4 explains the empirical strategy followed in the paper; Section 5 tackles the empirical findings, and Section 6 presents the main concluding remarks of the paper, highlighting some policy recommendation and suggestions for future research.

Although technological change throughout history has not been associated with mass unemployment (Autor, 2015), its impact on workers with different skills has been well documented and has developed considerably since the early industrialization era. The mainstream thought during the first and second industrial revolutions (late 18th – early 20th century) supported the idea that technological change was deskilling. Specifically, it reduced the skill content of most production activities and consequently favored unskilled workers over skilled ones. This is because the gradual mechanization of production during this period induced extreme labor division, which simplified many complex tasks previously performed by higher-skilled workers and made them doable by lower-skilled workers (Brugger and Gehrke, 2018).

Starting from the second half of the 20th century, after the diffusion of Information and Communication Technologies (ICT), there was strong empirical support for the skill-biased effect of technology in the USA (Katz and Murphy, 1992) and many other developed and developing countries (Berman and Machin, 2000; Conte and Vivarelli, 2011). During this period, particularly in developed countries, there was a persistent increase in wage inequality between low- and high-skilled workers, albeit a continuous increase in the supply of the latter group. Empirical findings of these studies returned this increase to the bias of technology toward higher-skilled workers, or the Skill-biased Technological Change (SBTC) hypothesis, which was strong enough to more than compensate for the effect of the increase in their relative supply on their wages.

Despite the wide support of the SBTC hypothesis, several studies pointed out that it becomes unable to explain patterns of employment polarization that have started to emerge since the 2000s, shifting the scholarly focus to a more refined version which is the RBTC hypothesis (Adermon and Gustavsson, 2015; Autor et al., 2003; Bessen, 2011; Kampelmann and Rycx, 2013; Salvatori, 2018; Sebastian, 2018). This hypothesis is traced back to the study of Autor et al. (2003) who explained the mechanism through which technology, mainly computer technology, substitutes or complements workers in performing different tasks from a completely new perspective that relies on the routine intensity of tasks rather than the skill level required to perform them (Autor et al., 2003).

Tasks can be classified into two broad categories based on their susceptibility to computerization (i.e. routine and nonroutine tasks). Routine tasks are procedural, well-defined and rule-based which makes them easily performed by computers at a relatively lower cost. On the contrary, non-routine tasks involve greater complexity and depend heavily on individuals’ tacit knowledge, making humans still highly needed in their performance. Therefore, computers are more likely to substitute workers performing routine tasks and less likely to substitute those involved in nonroutine tasks. For nonroutine tasks, particularly nonroutine cognitive and interactive tasks, computers are expected to complement workers by increasing their productivity (Autor et al., 2003; Fernández-Macías, 2012; Frey and Osborne, 2017).

While the RBTC literature places the susceptibility of tasks to computerization on their routine intensity regardless of the skill level required for performing them, it posits that the link between computerization and skills is indirect and depends mainly on how routine and non-routine tasks are spread across the skill distribution. Namely, since routine tasks are characteristic of the majority of middle-skilled jobs, while non-routine tasks are characteristic of the majority of low- and high-skilled jobs, computerization is expected, though not necessarily, to result in a hollowing out to the middle of the employment structure in terms of skills or employment polarization (Adermon and Gustavsson, 2015; Autor et al., 2003; Bessen, 2011; Goos and Manning, 2007; Kampelmann and Rycx, 2013; Salvatori, 2018; Sebastian, 2018).

Two main perspectives could be adopted while addressing the effect of task content on employment. The first is to assume that the task content of occupations is constant over time and examine how it impacts employment change between occupations (the extensive margin) (Goos and Manning, 2007; Kampelmann and Rycx, 2013; Sebastian, 2018). The second is to allow task content to change over time to showcase the extent to which the nature of tasks performed within a given occupation is changing over time (the intensive margin). Though the first perspective is the most dominant among the RBTC literature, the second has recently gained increasing attention. Several studies found that routine employment has also been declining within occupations, as tasks performed within occupations have increasingly become nonroutine (Acemoglu and Autor, 2011; Akçomak et al., 2016; Atalay et al., 2020). The study’s scope is confined to addressing the extensive margin, given that there is still no detailed workers’ survey data on the nature of tasks performed in different occupations over time in Egypt.

The link between the RBTC hypothesis and employment polarization is heavily addressed by empirical studies, especially in developed countries. Although the pioneering study of Autor et al. (2003) on the US labor market over the period (1960–1998) is the first to introduce the concept of routine task intensity, it was found to be associated with employment upgrading rather than polarization. However, subsequent studies that extended the period of analysis suggested strong empirical support for the effect of the RBTC hypothesis on both employment and wage polarization. Polarization is found to be driven mainly by the striking growth of workers in low-skilled jobs as a result of the displacement of those in middle-skilled jobs (Autor et al., 2006; Autor and Dorn, 2013).

There are also plenty of studies that addressed the nexus between the RBTC hypothesis and employment polarization in other developed countries such as the UK (Bisello, 2013; Goos and Manning, 2007; Salvatori, 2018), Germany (Kampelmann and Rycx, 2013), Spain (Sebastian, 2018), Sweden (Adermon and Gustavsson, 2015), a sample of European countries (Fernández-Macías and Hurley, 2017; Goos et al., 2014), and a sample of Organization of Economic Cooperation and Development (OECD) countries (Haslberger, 2021). Although a negative relationship between the routine intensity of jobs and employment growth is widely supported by these studies, some studies stressed that the RBTC hypothesis is neither necessarily associated with employment polarization nor is it the main driver behind it wherever it exists.

For instance, while employment in high-skilled jobs has witnessed substantial growth in almost all European countries, the growth patterns of employment in middle- and low-skilled jobs differ considerably depending on labor market policies and institutional settings (Fernández-Macías and Hurley, 2017; Haslberger, 2021). Even among studies that confirmed a link between the RBTC hypothesis and employment polarization, they highlighted the importance of many other factors in explaining employment polarization in Europe, such as the increasing supply of high-skilled workers (Salvatori, 2018; Sebastian, 2018), the influx of low-skilled migrant workers (Bisello, 2013), female labor force participation (Adermon and Gustavsson, 2015; Haslberger, 2021), and offshoring routine jobs and shifts in product demand across different economic activities (Goos et al., 2014).

Empirical evidence conducted in developing countries tends to be less supportive of the link between the RBTC hypothesis and employment polarization. The relatively lower support for this link is attributed to numerous factors such as: (1) low elasticity of substitution between technology and labor; (2) low technology absorptive capacity; (3) low exposure to routinization due to the concentration of employment in low-skilled manual occupations that are intensively non-routine and (4) offshoring many routine jobs from developed to developing countries wherein labor input is cheaper (Das and Hilgenstock, 2022; Maloney and Molina, 2019; Martins-Neto et al., 2021; Pena and Siegel, 2023).

Using job-level data on the Egyptian economy, Helmy (2015) examined sources behind skill demand polarization in Egypt over the period (2000–2009), proxied by the change in the share of each skill group in the wage bill. Polarization was supported and attributed mostly to the effect of the product demand shift which increased the demand for low- and high-skilled workers relative to their middle-skilled counterparts. She also found a significant and larger increase in the demand for low- and high-skilled workers within economic activities than for middle-skilled workers, suggesting a significant, but minor, role played by the RBTC hypothesis. However, the study highlighted that the expansion of the private sector over the period increased the demand for all skill groups, especially the middle-skilled ones, mitigating overall polarization.

Badran (2019) also investigated employment polarization in Egypt by relying on data from the ELMPS for 1998, 2006, 2012, and 2018. Applying a quadratic regression of employment on wage and wage squared as a proxy of skills, the study found that results vary depending on the specified model. While a U-shaped relationship is evident between employment and wages using pooled OLS estimation, a J-shaped relationship is confirmed when the panel data estimation is used. Although the study considered technology as a potential explanation for the observed employment patterns, no formal tests were conducted to validate its role.

In another descriptive study on Egypt using the ELMPS for the same years, El Hamidi (2020) addressed the RBTC hypothesis by crudely allocating the one-digit ISCO-08 occupations to task categories. Results reveal that employment in non-routine cognitive and manual occupations experienced growth, while it experienced a decline in routine occupations, supporting employment polarization. Additionally, manual occupations gained the most in terms of wage growth, followed by routine occupations, then by cognitive occupations, signaling evidence of wage downgrading. These findings were justified by the massive increase in the supply of high-skilled workers who are mainly concentrated in non-routine cognitive occupations.

Previous RBTC literature has emphasized the importance of accounting for occupational-specific characteristics while addressing the role of occupational task content. This is because the growth or decline of an occupation might not be attributed to the effect of technology, as reflected by occupational task content, but rather to changes in specific aspects of employment that are predominant in that occupation. Those characteristics are typically grouped based on whether they are related to the demand or supply sides of the labor market (Acemoglu and Autor, 2011; Adermon and Gustavsson, 2015; Brambilla et al., 2023; Cirillo et al., 2021; Fernández-Macías and Hurley, 2017; Goos and Manning, 2007; Guarascio et al., 2018; Kampelmann and Rycx, 2013; Salvatori, 2018).

The employment sector, formality, and economic activity are key demand-side factors that are commonly considered in the literature. Occupations dominated by public employment have less potential to grow, especially in light of the general contraction of public sector employment (Kampelmann and Rycx, 2013). Also, widespread informality, particularly in developing countries, could direct occupational growth toward dominantly informal occupations rather than their formal counterparts (Brambilla et al., 2023; Martins-Neto et al., 2021). Additionally, countries undergoing structural transformation typically experience a demand shift toward specific economic activities based on their economic development stage (Adermon and Gustavsson, 2015; Goos and Manning, 2007; Kampelmann and Rycx, 2013; Salvatori, 2018).

Supply-side factors, most importantly, education, age, and gender are also highly relevant in accounting for occupational growth. The surge in the supply of educated graduates significantly contributed to employment growth in high-skilled occupations in the USA and Europe (Acemoglu and Autor, 2011; Goos and Manning, 2007; Guarascio et al., 2018; Katz and Murphy, 1992). The increasing participation of women also significantly explained gendered sectoral and occupational segregation in developing countries, leading to the growth of employment in relatively low-skilled jobs (Borrowman and Klasen, 2020). Similarly, increasing the supply of workers of a certain age range could be associated with higher employment growth in occupations dominated by this age group (Adermon and Gustavsson, 2015; Guarascio et al., 2018; Kampelmann and Rycx, 2013; Salvatori, 2018).

This study uses data from the ELMPS for all labor market information. To date, there are four waves of this survey, covering the years 1998, 2006, 2012, and 2018. This study relies on waves of years 1998 and 2018 so that the relationship between technology and employment is addressed over a sufficiently long period, as recommended by many studies (Goos et al., 2014; Kampelmann and Rycx, 2013; Sebastian, 2018). The analysis is limited to workers aged 16–64 who are usually employed based on the market definition. Since occupational classifications are not the same in the two waves, occupational data up to the four-digit level for the year 1998 were transformed into the recent occupational classification adopted in 2018, which corresponds to the ISCO-08. All labor market variables created from the dataset in each year are created using expansionary weights to ensure the representativeness of the results.

The fact that occupational data are very detailed in these two waves makes it straightforward to crosswalk between occupational classifications. However, as the ISCO-08 disaggregates some occupations that are merged in the 1985 occupational codebook applied for 1998, observations with merged occupations are dropped from the analysis. The study also dropped occupations that do not show up in all waves and those that have less than 10 workers in all waves. This ensures that occupations included are consistently represented across the two years to reduce the sampling error (Adermon and Gustavsson, 2015; Bisello, 2013; Sebastian, 2018). The study ended up with an initial sample of 18,104 individuals (5,465 individuals in 1998 and 12,639 individuals in 2018) distributed across 141 four-digit occupations (OAMDI, 2019).

In Egypt, till now there is no standard evaluation of occupational task content available. Thus, this paper relies on measures of occupational task content computed by Mihaylov and Tijdens (2019). Their study utilized the detailed description of tasks performed in different occupations found in the ISCO-08 classification. In this classification, each occupation is identified by a title, a code, a definition of its scope, and a detailed description of tasks performed within it. Mihaylov and Tijdens (2019) focused on the four-digit level in which an overall 427 four-digit occupations are described by a total of 3,264 occupational tasks (Mihaylov and Tijdens, 2019).

Mihaylov and Tijdens (2019) allocated the related tasks of each occupation into one or more of the five task categories proposed by Autor et al. (2003) (i.e. routine cognitive, routine manual, non-routine analytic, non-routine interactive, and non-routine manual tasks). Their task allocation is based on an in-depth case-by-case evaluation to decide whether a specific task is routine or non-routine on the one hand and whether it is cognitive or manual on the other hand. Then, the shares of each of the five task categories are calculated so that they sum up to one for each occupation. In addition, an overall routine task intensity index is calculated for each occupation by subtracting the total sum of non-routine task shares from the total sum of routine task shares, with values ranging between −1 and 1.

To simplify our empirical analysis, the task measures are aggregated into three categories: (1) abstract which is the sum of shares of both non-routine cognitive and non-routine interactive tasks; (2) routine which is the sum of shares of routine cognitive and routine manual tasks; and (3) manual which is the share of non-routine manual tasks. The analysis is further simplified by turning task shares into task dummies such that each four-digit occupation is classified into only one task category which is the one in which it has the highest share. The total number of four-digit occupations is reduced to 136, as one occupation does not have a task content measure, and three others have mixed task content. The final sample used in the study includes 16,569 individuals (4,938 individuals in 1998 and 11,631 individuals in 2018).

The Mihaylov and Tijdens (2019) measures of occupational task content have several advantages: First, they are calculated for the same occupational classification (ISCO-08) and disaggregation level (four-digit-level) as the current study which makes them directly employed in the analysis. Second, since occupational task measures are standard and not computed for a specific country or year, this helps avoid imposing any assumption regarding the similarity of task content of occupations across countries or over time. Third, rather than ad hocly selecting tasks to define each occupational task category, the entire occupation-specific tasks for each four-digit occupation in the ISCO-08 classification are allocated to each occupational task category, making their measure less biased.

Both parametric and nonparametric graphical inspection methods are applied to examine the employment evolution pattern across the skill distribution (1998–2018) by estimating quadratic and Kernel-smoothed fits, respectively. While the former tests for a U-shape relationship by estimating an OLS regression of changes in occupational employment on a skill measure and its quadratic term, the latter fits a smoothed curve for the relationship without imposing any restrictions on its type (Salvatori, 2018). In both methods, the 136 four-digit occupations are ranked ascendingly based on their initial skill level, and percentage point changes in their employment shares are calculated and plotted against their skill level. To avoid any bias in results that could originate from occupations with the smallest initial employment shares driving the overall conclusion, changes in employment shares are weighted by their initial levels.

The occupational evolution pattern relies heavily on the selected skill measure. While the ISCO-08 classification groups occupations based on their skills into low-, medium-, and high-skilled occupations ^[1], this ranking is confined to the one-digit level. Thus, using the ISCO-08 ranking will not utilize the disaggregated advantage of the data, as it will categorize occupations into only these three categories. To circumvent this issue, the ISCO-08 aggregate ranking is compared with aggregate rankings obtained from other skill proxies such as wage, education, and the International Socio-Economic Index of Occupational Status (ISEI-08) to select the proxy that has the strongest correlation with the ISCO-08. These proxies are commonly used by previous literature to account for skills and could be disaggregated to the four-digit occupational level (Adermon and Gustavsson, 2015; Bisello, 2013; Fernández-Macías, 2012; Fernández-Macías and Hurley, 2017; Kampelmann and Rycx, 2013; Oesch and Piccitto, 2019; Salvatori, 2018; Sebastian, 2018).

Namely, the four-digit occupations are grouped into employment-weighted tertiles based on these proxies. So, in the end, each four-digit occupation could be classified into a low-skilled, middle-skilled, or high-skilled occupation based on the ISCO-08, our reference ranking, and based on the other three proxies. The skill measure with the highest significant positive Spearman rank correlation coefficient with the ISCO-08 is used in the parametric and nonparametric analyses, which require a detailed ranking of each four-digit occupation. After exploring employment evolution, a preliminary check of its link to the occupational task content is investigated before the formal analyses. This is done by tracking the employment evolution of each occupational task category to ensure that routine employment is declining compared to abstract and manual ones and identifying their location across the skill distribution.

To further understand the link between the task content of occupations and changes in occupational employment the study first conducts a decomposition analysis, which is commonly used by previous studies (Bisello, 2013; Goos and Manning, 2007; Salvatori, 2018; Sebastian, 2018). Its idea is to decompose overall changes in employment shares in each occupational task category into between-group and within-group changes. If the RBTC hypothesis prevails, the within-group changes should be higher and negative for routine occupations and higher and positive for abstract occupations. While there is no direct theoretical assumption made for manual occupations which are less likely to be affected by technology, their employment share is expected to increase if displaced routine workers switch to manual occupations.

The decomposition analysis is conducted for each demand-side and supply-side factor discussed in Section 2.2. It is also undertaken once by considering each factor individually and once by considering them all together. The baseline decomposition is carried out on 2 employment sector groups, 2 employment formality groups, 10 economic activity groups; 3 education groups, 3 age groups, and 2 gender groups. All those factors comprise 720 groups, resulting from the interaction of all factors’ groups (detailed groups per factor as well as changes in their employment composition are listed in Table A1 in the  Appendix).

Mathematically, changes in the overall employment share of each occupational task category [$ΔEjt]$ can be decomposed into between-group change [$ΔEjtB$] and within-group change [$ΔEjtW$] as described in Equation (1):

where,

1. $ΔEjt=Ejt1−Ejt0:$ is the change in the overall employment share of occupational task category j in total employment over the period (⁠$t0$ – $t1$⁠); where j includes abstract, routine and manual occupations and (⁠$t0$ – $t1$⁠) refers to the time frame of the analysis (1998–2018).

2. $ΔEjtB=∑kΔEktδjk$⁠: is changes in employment shares that happened between different groups, where $ΔEkt=Ekt1−Ekt0$ is the change in the employment share of group k between years $t0$ and $t1$⁠, and $δjkt=(δjkt1+δjkt0)/2$ is the average share of occupational task category j in the total employment of group k for the years $t0$ and $t1$⁠.

3. $ΔEjtW=∑kΔδjktEk$⁠: is changes in employment shares that happened within the same group, where $Δδjkt=δjkt1−δjkt0$ is the change in the share of occupational task category j in the employment of group k between the years $t0$ and $t1$⁠, and $Ek=(Ekt1$ + $Ekt0)/2$ is the average employment share of group k for the years $t0$ and $t1$⁠.

To check whether routine workers are more likely to switch their occupational task category than abstract and manual workers and whether they typically move to abstract or manual occupations, the occupational task mobility of workers who are observed in the two waves (925 workers) is examined by calculating the transition probability matrix for each occupational task category. The transition probability from state i to j is represented as: $pij=Pr⁡(Xt=j|Xt=i$⁠), and could be calculated through the formula of the Equation (2) (Bisello, 2013; Sebastian, 2018):

where, $Nij$ is the number of workers who changed their occupational task from category i to j between 1998 and 2018 (the cell count), and $∑j=1nNij$ is the total number of workers working in a certain occupational task category in 1998 (the row count).

In order to validate the statistical significance of the relationship between task content of occupations and occupational employment, a logistic regression is estimated. The model’s specification relies on specification adopted by Adermon and Gustavsson (2015), Brambilla et al. (2023), Cirillo et al. (2021), Guarascio et al. (2018), and Kampelmann and Rycx (2013). Based on it, a dummy of employment growth or decline is regressed on task content of occupations as the core explanatory variable in addition to changes in other controlling variables that capture the main occupation-specific controls that are used in the baseline decomposition analysis, as presented in Equation (3):

where, subscripts j, t₀, and t refer, respectively, to occupations, the initial year (1998) and the end year (2018).

In this model, the unit of analysis is the 136 four-digit occupations. The dependent variable is a dummy variable that captures occupational employment growth over the study’s period. The explanatory variable of interest $(Tk,j,t0)$ is the task content of occupation j in the initial year t_0, which is assumed to be constant over time. It is captured by three task dummies k (i.e. abstract, routine, and manual), with routine considered the reference category. If the RBTC hypothesis holds, the coefficients of the abstract and manual tasks should be significantly positive. The regression is weighted by the initial employment share of each occupation, as this ensures that results reflect the relative importance of occupations in the labor market (Bisello, 2013; Kampelmann and Rycx, 2013).

To nullify the potential effect of occupational composition, a vector of occupation-specific control variables $(Xl,j,t)$ are added to the regression, considering their initial values in 1998. This vector includes the sector of employment as measured by the share of public sector workers; the formality of employment as measured by the share of formal workers; the economic activity of employment as measured by the share of workers in growing economic activities between 1998 and 2018 (see Table A1); education as measured by the share of workers with below intermediate education; age as measured by the share of young workers (15–29); and gender as measured by the share of male workers.

Panels A, B, and C of Figure 1 show the evolution pattern of occupational employment over the period (1998–2018) against aggregate occupational rankings obtained from wages, education, the ISEI-08, and the ISCO-08 skill measures. Apparently, it is difficult to reach a common conclusion. Employment polarization is evident when wages are used, but education and ISEI-08 suggest upgrading-middle and downgrading employment patterns, respectively. Although wages are an important indication of occupational skill level, sometimes they do not align together (Oesch and Piccitto, 2019). For instance, 23.6% of occupations are found in relatively higher wage tertiles (overpaid) and 21.3% are found in relatively lower wage tertiles (underpaid) than what is expected based on their typical position according to the ISCO-08. Education also could be a distorted skill measure, given the evidence of education-occupation mismatch in Egypt (El-Hamidi, 2009, 2020; Sadeq, 2014). Comparing the education grouping with the ISCO-08 reveals that 11.8% of occupations are overeducated and 19.1% are undereducated. This is manifested particularly in clerical occupations of which 60% exist in the highest educational tertile, albeit normally considered as middle-skilled.

The Spearman rank correlation coefficient in Table 1 shows that the ISEI-08 has the highest significant positive correlation with the ISCO-08 (0.83) at 1%, 5% and 10% significance levels. Thus, it is used in parametric and nonparametric analyses. This measure was introduced firstly by Ganzeboom et al. (1992) and updated by Ganzeboom (2010) to align with the ISCO-08, up to the four-digit level. The ISEI-08 measure serves as a good skill proxy, as its calculation rests on education and earnings associated with each occupation. It assesses the educational impact on earnings by emphasizing the intermediary role of occupations. Thus, high index values mean that an occupation not only requires a high education level but also translates educational attainment into high earnings. Moreover, as this index is standard and not specific to a particular county or country group, it helps avoid any distortions associated with using country-specific measures like wages and education (Ganzeboom, 2010).

The parametric and nonparametric representations of the relationship between employment change and skills at the four-digit occupational level are presented in panels A and B of Figure 2. As found in Figure 1, polarization is not supported, and employment structure is closer to a downgrading pattern. The unsupportive evidence of employment polarization does not necessarily refute the RBTC hypothesis. Routine employment may indeed be declining, but it may also be either not large enough to drive a notable employment decline at the middle of the skill distribution or not concentrated in the middle. The RBTC literature has emphasized two main conditions for routinization to be associated with employment polarization: (1) the high exposure to routinization which means having a sizable share of workers working in routine occupations; and (2) the concentration of routine employment in the middle of the skill distribution which could also be affected by the first condition (Das and Hilgenstock, 2022; Maloney and Molina, 2019; Martins-Neto et al., 2021; Pena and Siegel, 2023).

Tracking the evolution of routine employment shows that it witnessed a notable decline over the period (by 7.70% points), unlike both abstract and manual employment which experienced an increase (3.20 and 4.49% points, respectively). However, none of the abovementioned conditions are satisfied. For the first condition, Table 2 illustrates that while routine occupations represent more than one-third of total occupations (35.29%), they have the lowest employment share from total employment (17.56%). This suggests low initial exposure to routinization, which could be justified by the fact Egypt, like many other developing countries, has been undergoing structural transformation, with a sizable share of its workforce concentrating on low-level manual occupations that are relatively difficult to automate (Das and Hilgenstock, 2022; Maloney and Molina, 2019; Martins-Neto et al., 2021).

For the second condition, while the skill composition of routine employment shows that 70.49% of its employment was initially middle-skilled (see Figure A1 in the  Appendix), the low exposure to routinization makes middle-skilled employment not dominantly routine. As indicated in Table 2, the middle-skilled tertile comprises routine and manual employment, with the latter having the highest employment share of 57.12% compared to 40.89% for routine employment. This suggests that the link between the task content of occupations and occupational employment is worth further examination to determine the relevance of the RBTC hypothesis to Egypt, as is undertaken in the following subsection.

Major changes in employment composition have occurred over the period and are worth highlighting (see Table A1 in the  Appendix). On the demand side, public sector employment has considerably declined. The inability of the formal private sector to absorb the labor supply has made this decline associated with informal employment expansion. Egypt’s structural transformation has also shifted employment from tradable and relatively high-quality activities. Agricultural employment has contracted. Within industrial activities, the employment share of mining, manufacturing, and utility activities has failed, while it has grown in construction. Employment has also expanded in trade, transportation and storage, and accommodation and food services, whereas it has dropped in professional and public administration services and health and education activities. On the supply side, employment initially comprises workers with below intermediate education, despite the notable educational upgrade toward intermediate education. Additionally, the age structure of employment exhibits an inverted U shape where most workers are concentrated in the middle-aged group (30–49), of which share increases over time at the expense of the young-aged group (15–29). For gender composition, overall employment is dominated by men, with a slight increase in their share over the period.

The decomposition results of overall employment change in each occupational task category are detailed in Table 3. Results per factor reveal that routine occupations are constantly declining between and within groups, regardless of the factor based on which the decomposition is undertaken. The higher and negative within-group changes compared to between-group changes suggest a highly relevant role of the hypothesis in accounting for their decline. The relative importance of the hypothesis in accounting for employment growth of abstract and manual occupations differs depending on the decomposition factor. The hypothesis is almost entirely linked to the employment growth of abstract occupations. However, when its effect is contrasted with education, education becomes completely responsible for that growth. This could be attributed to the high average share of abstract employment among workers with above intermediate and intermediate education (82.0 and 33.1%, respectively) whose employment experienced growth. Meanwhile, this average reaches only 10.1% among workers with below intermediate education whose employment has witnessed a decline.

For manual occupations, decomposition by education, age, and gender shows that within-group changes are the core drivers of their employment growth, suggesting a higher relevance of the RBTC hypothesis. Yet, the hypothesis becomes less relevant when the decomposition is undertaken by sector, formality, and economic activity. The dominantly positive between-group changes associated with the latter factors are linked to the fact that manual occupations are dominated by private, informal, and non-tradable low-quality activities workers whose employment also expanded over the study’s period. Namely, manual employment comprises 66.9% of private sector employment and 72.8% of informal employment on average. It also constitutes a very high average employment share in many growing economic activities such as transportation and storage (85.6%), construction (83.7%), accommodation and food services (65.4%), and trade (21.9%).

Findings of the overall decomposition analysis applied to all groups’ factors (the 720 groups) suggest overall support for the RBTC hypothesis. First, within-group changes are negative for routine occupations and positive for abstract and manual ones. This means that the employment behavior of those occupations is still in line with the hypothesis’s expectations, even after accounting for all occupation-specific factors. Second, within-group changes are more important in size than between-group changes for both routine and manual occupations. While the reverse is true for abstract occupations, the difference between both components is minor. This recommends that the RBTC hypothesis potentially plays an unneglectable, if not dominant, role in accounting for employment changes in all occupational task categories over the study’s period.

Many robustness checks are conducted by applying the decomposition analysis on more detailed groupings of the existing factors to ensure that the current level of disaggregation does not hide potential between-group changes (i.e. 7 institutional employment sector groups, 21 economic activity groups, 7 education groups, and 6 age groups). The results are largely consistent with the baseline decomposition across all factors in terms of both the direction of the change and the relative importance of between-group and within-group changes. The decomposition analysis is also undertaken on additional factors (i.e. 2 location groups and 6 marital status groups) and results also indicate that within-group changes are dominant and are with expected signs for all occupational task categories.

Results of the transition probability in Table 4 shows that routine workers are less likely to keep their occupational task category unchanged. While only 30.57% of routine workers in 1998 remained in the same task in 2018, this percentage reached 83.13 and 79.53% for abstract and manual workers, respectively. In addition, routine workers principally relocate to abstract (47.02%) rather than manual occupations (22.41%). This could be connected to the prevalence of overeducation among routine employees, which increases their chance of relocating to better occupations.

The disaggregation of the transition probability matrix by education further illustrates this tendency. Specifically, routine workers with above intermediate and intermediate education typically move to abstract occupations (69.28 and 59.29%, respectively), while those with below intermediate education normally move to manual occupations (48.30%). Education generally helps workers retain their occupations when they are originally in high-quality ones and move to better-quality occupations when they are originally in low-quality ones.

While the overall upgrading pattern observed in Table 4 seems contradictory to the overall downgrading pattern depicted in Figure 1, this could be explained by the difference in the data type in both analyses. The transition probability using longitudinal data shows only occupational transition among experienced workers who have more likelihood of occupational upgrading. The downgrading pattern associated with using cross-sectional data could be attributed partly to new entrants to the labor market who normally join low-skilled occupations when they are first employed. Around 43.1 and 41.7% of new entrants in the age group (15–29) who were observed only in 2018 are concentrated in low-skilled and middle-skilled occupations, respectively, compared to only 15.2% are concentrated in high-skilled occupations.

To formally investigate the role of the RBTC hypothesis at a more disaggregated level, the study estimates a logistic regression of the role of task content of occupations in explaining the behavior of occupational employment. In line with the RBTC hypothesis, the results of Table 5 reveals that abstract occupations have significantly higher employment growth probability than routine occupations at 5 and 10% significance levels but only after controlling the occupation-specific characteristics (see Models 2–7 in the same table). For the final baseline model (Model 7), the average estimated marginal effects of abstract occupations suggest that holding all other variables at their observed values, abstract occupations are associated with a 36.2% point increase in employment growth probability compared to routine occupations. This could be justified by the previously highlighted movements of displaced routine employment toward abstract rather than manual occupations and their precarious, less productive, and less dynamic nature, giving them limited growth horizons.

The model also suggests significant relevance of all other demand-side factors included. High initial public employment shares significantly decrease employment growth probability due to the considerable contraction of the public sector. While formal employment also experienced a notable decline over the study’s period, predominantly formal occupations are characterized by higher employment growth probabilities than their informal counterparts. The positive influence of formal employment stems basically from predominantly formal private than formal public occupations, reflecting the critical role of the formal private sector in ensuring sustainable employment growth compared to formal public and informal sectors. This is confirmed in one of the robustness checks applied to Model (7), where the employment shares of formal public and formal private sectors are used instead of public and formal employment shares ^[2]. Occupations with higher initial employment share in growing economic activities also show higher growth potential, emphasizing the role of product demand in shaping occupational employment.

Supply-side factors, specifically education and age, are slightly significant but not with the expected signs. Despite the overall decline in the employment share of the age group (15–29), occupational growth is still more prominent in occupations with a relatively higher intimal share of young workers. This could be explained by the proposition that occupations comprised mostly of young workers are more dynamic and have greater growth potential (Autor and Dorn, 2009). For education, the increase in the initial share of employment with below intermediate education, which is declining overall, increases rather than decreases the probability of occupational growth. This implies that those occupations have become more reliant on higher-educated workers, exposing workers to skill erosion. This is evident from the evolution of manual employment shares in the total employment of workers with intermediate and workers with above intermediate education almost doubled between 1998 and 2018 (from 28.7% to 56.5% for intermediate education, and from 6.4 to 12.7% for above intermediate education).

Many diagnostic checks are conducted to assess the quality of Model 7. The Hosmer-Lemeshow and Pearson goodness-of-fit tests indicate that the model fits well. Moreover, using a 0.5 threshold for classifying predicted probabilities, 61.5% of actual values of the dependent variable are found to be correctly predicted by the model. Variance Inflation Factor (VIF) is also calculated for each predictor to check the multicollinearity problem, and results reveal no strong multicollinearity among predictors. Since our model includes continuous predictors, it is important to check whether the logit function of the outcome variable is a linear function of the predictors. This is assessed by running the link test for model specification which confirms this linearity. Robustness checks are also applied to validate the stability of our results to changes in the estimation method, sensitivity to ignoring initial employment shares as regression weights, using alternative representations of control variables, and introducing other controls to the regression. None of these checks impact the significance of the abstract task variable and most control variables, especially the employment sector, formality, activity, and education.

This paper addressed the link between occupational task content of occupations and occupational employment in Egypt over the period (1998–2018). It tested whether routine employment significantly declined compared to abstract and manual employment. Additionally, it tested the two main employment implications that are typically, but not necessarily, associated with the routine employment decline. The first implication is employment polarization, which occurs if a relatively higher share of workers work in routine occupations and are concentrated in the middle-skilled distribution. The second implication is whether the impact of technology on manual employment is indirectly driven by movements of displaced routine workers to manual occupations rather than directly from their task content per se.

The study started with an exploratory analysis of occupational employment evolution against the occupational skill level using parametric and nonparametric graphical inspection methods. It also provided a preliminary link between the occupational task content and the depicted employment pattern. This is done by tracking changes in occupational employment by occupational task category and identifying the location of routine employment across the skill distribution. Then, the study moves for more formal methods by conducting decomposition and logistic regression analyses, considering supply and demand side factors of the labor market that could shape occupational employment over time.

The initial examination of occupational employment evolution by various skill measures at both aggregate and disaggregated levels suggests that the employment structure is closer to a downgrading pattern than a polarizing one. In parallel, routine employment did experience an employment decline compared to both abstract and manual employment, indicating that the unsupportive evidence of employment polarization does not necessarily mean the irrelevance of the RBTC hypothesis. Polarization is not supported mainly because routine occupations have a relatively lower employment share compared to their abstract and manual counterparts. This does not make them dominant in the middle-skilled distribution, although most of their employment is middle-skilled.

A decomposition analysis of employment changes per task category is then applied, considering several occupational controls, to separate between-group changes (accredited to occupation-specific factors) from within-group changes (accredited to the RBTC hypothesis). Overall, decomposition results recommend that the RBTC hypothesis plays an unneglectable, if not dominant, role in accounting for employment changes. This is particularly relevant for routine occupations which are consistently and predominantly declining within groups compared to between groups. Additionally, tracking workers’ transition between occupational task categories reveals a higher likelihood of switching occupational tasks among routine workers than abstract and manual workers. They are also more likely to move to abstract than manual occupations, though the movement direction seems to be tied to their initial education level.

Finally, regression results pointed out that, unlike manual occupations, abstract occupations are significantly associated with a high employment growth probability relative to routine occupations, suggesting their greater importance in sustaining employment generation. Demand-side factors, particularly private formal employment and economic activity, have a high and positive significance in driving occupational growth over time. Supply-side factors like age and education show a relatively lower significance in accounting for occupational growth. Generally, occupations dominated by relatively younger and low-educated workers have higher growth probabilities. While the first case reflects higher flexibility and dynamism of young occupations, the second indicates a deskilling problem.

Many policy implications could be inferred from the findings. First, given that employment is declining in routine occupations and that abstract occupations have greater prospects for sustaining employment generation than manual occupations, policies need to target adjusting workers’ skills to align with labor market needs. Recent projections for Egypt reveal that abstract skills like cognitive, technology, self-efficacy, and management skills are identified as the top core skills that business companies need. Projections also show that reskilling and upskilling are highly needed toward a wide range of abstract skills like analytical and creative thinking, marketing and media, AI and big data, and technological literacy, among others (WEF, 2023). This does not necessarily imply expanding higher education, given the evident deskilling problem among university graduates. One area of focus could be to attract students to Science, Technology, Engineering, and Math (STEM) fields that have greater promise, especially since they are currently underrepresented, constituting only 16.9% of all tertiary education graduates in 2022 (UNESCO, 2024).

Second, results also propose a vital role of the formal private sector in driving occupational employment growth compared to formal public and informal sectors due to its relatively higher sensitivity to market incentives. This implies that supporting its development and growth is a cornerstone for guaranteeing sustainable expansion of high-quality jobs that are relatively more resilient to routinization, especially with the major contraction of public sector employment. International experience of countries such as Chile and Sweden already confirms that ensuring an overall business-friendly environment for the private sector is closely linked to generating sufficient employment opportunities within a short time frame (European Bank for Construction and Development, 2017).

Third, while the effect of product demand shift is crucial in accounting for occupational employment, most growing activities are typically non-tradable and characterized by outside-establishment work such as wholesale and retail trade, construction, and transportation and storage. In addition to the fact that these sectors are typically associated with precarious employment, evidence from Egypt shows that those activities have lower employment elasticity to sectoral value-added than activities like manufacturing and mining and social services which are declining in employment shares (El Ehwany and El Megharbel, 2008). In this regard, priority should be given to economic activities with high employment growth potential by evaluating the employment elasticity of different activities at a highly disaggregated level, while considering the potential impact of emerging technological trends.

The limitations of the study lie in the following:

1. The study employed standardized measures of occupational skills and task content. While such measures are sometimes useful when reliable country-specific measures are not available, it is always important to develop high-quality country-specific measures that reflect countries’ realities.

2. In the analysis of workers’ transition, the study addressed whether routine workers are more likely to switch their occupational task category, without considering whether they are also more likely to be unemployed or not as a result of technology.

3. The scope of the study is limited to addressing the employment impact of occupational task content at the extensive margin. Future research could investigate the impact on other variables like wage inequality, workers’ objective and subjective well-being, etc., or address the same relationship at the intensive margin.