The human capital of firms and the formal training of workers

Human capital investments at work account for more than one-half of the human capital accumulated over the life cycle (Heckman et al., 1998). The contribution of work experience to human capital accumulation, acquired either through learning on the job or formal training, might be as important as the contribution of education itself (Jedwab et al., 2021). Firm-specific investments in human capital through employee training can generate rewards for firms and for workers, because it can increase firm performance by providing workers with necessary skills, increasing innovation and generally raise the level of competitiveness (Almeida and Aterido, 2015; Boikos et al., 2022). It can also update skills of workers in the fast-changing world of digitization and automation through retraining, especially in the context of an ageing workforce in advanced economies (Brunello and Wruuck, 2020). The wage returns to training can be high for workers (Konings and Vanormelingen, 2015, Almeida and Faria, 2014). In developing economy contexts, it can help compensate for low quality schooling. Estimates suggest that firm-provided training accounts for may account for 38% of cross-country wage growth differences and 12% of cross country income differences (Ma et al., 2022).

However, investments in workers by firms have been inadequate. The incidence of formal training across firms has been low, especially in the Middle East and North Africa (MENA) region. World Bank Enterprise Survey 2019/2020 data across 35 economies in the Europe and Central Asia region (ECA) and MENA region show that on average only 39 percent of medium and large firms provide formal training. For six developing MENA economies – Egypt, Jordan, Lebanon, Morocco, Tunisia, West Bank and Gaza - the figure is lower with 29 percent of percent formal medium and large businesses providing formal training. Around 20 percent of formal medium and large enterprises in the developing MENA economies (2019/2020) indicated that an inadequately educated labor force is a major or severe constraint to the operations of their business. Firms have disincentives to provide general skills training to their workers due to externalities, and also may face challenges in acquiring finance to fund costly training. European governments have responded by funneling subsidies to encourage training in firms (Brunello and Wruuck, 2020).

In this study, we add to the literature by exploring whether the existing base of human capital in the firm is an important determinant of whether firms invest in training. We capture human capital in two forms – one is the management practices of the firm and second is the share of workers with university degrees – by harnessing firm-level surveys that containing a unique module on management practices and a survey question on the share of workers that are university educated. We also account for manager experience. The data on management practices was only collected for medium and large enterprises, which defines our sample. Furthermore, several factors make the ECA and MENA regions good comparators. First, in a practical sense, the rollouts of the surveys happened exactly at the same time, allowing us to compare the private sector across the two regions in a consistent timeframe. Second, The ECA and MENA regions contain a large share of middle-income economies, making them good comparators. Third, many of the former soviet economies had a large degree of government involvement. The MENA region is characterized by an oversized role of the state. Finally, regional comparisons are useful as a lot of policy lending by international organizations are undertaken using a regional framework. Our findings show that both management practices and the share of university educated workers matter overall, but there are strong differences by region. The share of university educated workers is a strong predictor of both the incidence and intensity of training in the developing MENA region. This is also true for the ECA region, but the result is not as robust. In contrast management practices are a strong predictor of training in the ECA region but not in the MENA region. This study builds on several studies that have explored the determinants of training (Frazer, 2006; Rosholm et al, 2007; Pierre and Scarpetta, 2013; Almeida and Aterido, 2015; Liaqat and Nugent, (2015, 2016)).

Training represents investments in future productivity that come at a cost (Wolter and Ryan, 2011). Becker (1962) noted that firms receive little benefit in providing formal training if it is general as opposed to firm specific as workers may leave and general skills are transferable. However, extensions of the original model noted that firms may invest in general skills to attain informational advantages and monopsony power (Katz and Ziderman, 1990; Acemoglu and Pischke, 1998). Firms with better managerial quality may be better able to understand these advantages. Improving the managerial quality of a firm may lead to greater incidence of training. The measurement of managerial quality in accordance with Bloom et al. (2013) captures several dimensions including problem resolution, monitoring of performance indicators, production targets (ease of attainment, length of focus, and knowledge), basis of bonuses, promotion of non-managers and dismissal.

See Bloom et a., (2013): https://people.stanford.edu/nbloom/sites/default/files/mia.pdf, pp 21

The share of highly educated workers in a firm may have several implications for training. High education levels may signal high ability, and thus firms may be determined to retain these workers by investing in them, especially in developing economies where the supply of highly educated individuals is low. Thus, the correlation between highly educated workers and the provision of formal training may be positive. Furthermore, highly educated workers may require repeated training throughout their career as they develop different skill sets with experience. Also the cost of training may decline with higher educated workers who may have developed learning skills, thereby incentivizing firms to invest in training (Bassanini et al., 2007). However, the hiring of highly educated workers may obviate the need for formal training, especially if they are of high ability and can quickly learn on the job. Furthermore, highly educated workers may entail a larger flight risk as they have more bargaining power and are therefore more likely to move to other firms. Accordingly, firms with a larger share of educated workers may be less likely to invest in formal training. Finally, the share of highly educated workers in a firm may proxy for the level of general human capital in the economy. Assuming that in general human capital is low in an economy, firms may implement training programs to invest in high ability individuals that may not have the required education in order to compensate for the lack of education, or the presence of low-quality education. Thus, formal training may be more prevalent across firms that in economies with low human capital. Whether the portion of university education workers is positively or negatively related to formal training is an open empirical question.

The challenge of low provisions of formal training provided by firms is especially a concern for the MENA region. Liaqat and Nugent (2015) note that high youth unemployment, lengthy school-to-work transitions, and a sizeable gap between the skills firms want and young graduates possess are characteristics of the region. Thus, workers in the region are well poised to benefit from formal training. However, as mentioned earlier, the incidence of training is particularly low. Also noted is that firm-supplied training is found to be more effective than government-supplied training, therefore the lack of formal training by firms cannot be easily substituted. This study provides additional focus on the MENA region given the well-documented low prevalence of formal training in the private sector. The analysis of firm-level data over two waves for the MENA region is an important contribution of this study given the lack of readily accessible data in the region (Islam and Lederman, 2020; Ekhator-Mobayode and Hoogeveen, 2021).

This is the first time World Bank Enterprise Surveys of two waves of 6 MENA economies over roughly the same period of time is available.

Our empirical strategy is to exploit firm-level variation in formal training, management practices and worker education, while accounting for several confounding factors. We employ two samples. The first sample is a pooled cross-section of 8,470 firms across 35 economies. This includes two waves for economies the MENA and ECA region (2013 and 2019/2020) where we estimate the effect of management practices and worker education on the presence of formal training (extensive) and proportion of workers trained in manufacturing firms (intensive). Several firm-level factors are accounted for, including country fixed effects. A second sample entails a subsample of panel firms interviewed across both waves for both region. In these estimations, firm-level fixed effects are used to account for time-invariant firm-level omitted variables. An important concern is endogeneity. Although employee training could potentially lead to better management practices, this is unlikely as the training captured is only directed to employees, and furthermore even though training may affect some of the management practices, it is unlikely to affect the aggregate score. This is distinct from the emerging literature that has documented the effects of external management training that specifically targets improving management practices (McKenzie and Woodruff, 2017; Higuchi et al, 2019). More of a concern is that training might attract high ability workers who tend to be more educated. This is most likely if the presence of training programs is the key attraction for highly educated workers. However, the presence of training programs could be correlated to several other features of firms that highly educated workers find attractive. We account for these firm characteristics to the extent that the data allows us.

Our findings indicate that, accounting for a broad range of country and firm-specific controls, formal training is positively related to worker education for the ECA and MENA region, although the finding for the latter is more robust. Good management practices are related to formal training for the ECA region but not the MENA region. Presence of corruption and lack of formal record keeping seem to be potential reasons for the ineffectiveness of management practices in the MENA region. The negative effect of corruption on management practices is consistent with findings of Athanasouli and Goujard (2015). The relationship between worker education and training has been documented before, although we look specifically at university education while other studies look at secondary education or the occupation level to determine skill levels. Furthermore, we validate the findings using panel estimations. Our study also builds and confirms several findings in the literature on formal training including the positive correlation with firm size and quality certification (Almeida and Aterido, 2015; Liaqat and Nugent, 2015). Our study provides several policy implications. To the extent that there is a causal mechanism running from university education to training, increasing university education of workers in MENA can incentivize firms to invest in more training, potentially updating skills that are more robust to automation and digitization and developing soft skills. However, given the documented high unemployment among university graduates in the region, this may not be enough and may need to be coupled with reforms that strengthen the private sector and improve the business environment. Furthermore, if the prevalence of training is to adapt and update skills of educated workers towards work in the private sector, then the policy implication is not only be to increase university education, but also to ensure it serves the private sector as well. Finally, firms with better management practices tend to provide more training in ECA but not in MENA. Improving the business environment in terms of reducing corruption and increasing formal record keeping may improve the effectiveness of management practices in the MENA region. These findings are important for the MENA context given the low provision of training in the formal private sector.

In summary our study makes several contributions to the literature. First it explores the role of management practices and education of workers on formal training both at the intensive and extensive margin. Second it utilizes a panel dataset to account for several firm-level characteristics to validate the findings. And finally, it updates several studies in the literature by employing recent firm-level that includes both economies across Europe, Central Asia, and the MENA region. The rest of the paper is structured as follows. Section 2 describes the data. Section 3 provides the empirical specification and identification strategy, while section 4 provides the results. Section 5 presents the robustness checks and section 6 concludes.

The main source of firm-level data is the World Bank's Enterprise Surveys (ES). This includes two samples of medium and large enterprises.

Medium and large enterprises are defined as having 20 or more employees. Do note that the choice of medium and large enterprises is due to the inclusion of management practices measures that were only collected for medium and large firms. There are some small firms that are in the sample for two reasons. One is that some of these firms were identified as medium or large firms in the sample frame, and thus were administered the management practices module. Second is that some of these small firms grew to become medium firms between the 2013 and 2019/2020 waves, and thus they are retained to maintain the panel component. We retain these firms in the sample given that they are very likely to have attributes similar to medium and large firms as identified in the sample frames.

The key outcome variable is the presence of formal training. This is derived from the survey question: Over fiscal year [Insert last complete fiscal year], did this establishment have formal training programs for its permanent, full-time workers? Formal training is defined as training that has a structured and defined curriculum. It may include classroom work, seminars, lectures, workshops, and audio-visual presentations and demonstrations. However, it excludes training to familiarize workers with equipment and machinery on the shop floor, training aimed at familiarizing workers with the establishment's standard operation procedures, or employee orientation at the beginning of a worker's tenure. A second variable used is the share of workers in the firm that received formal training. This question is only asked for manufacturing firms. Figure 1 presents the incidence rates of formal training across medium and large firms in the sample at the country level. Figure 2 provides the share of workers in medium and large manufacturing firms that received training across countries. Figures 5 and 6 repeat the same for the two waves – 2013 and 2019/2020 for the MENA and ECA regions respectively. A few fairly consistent patterns are apparent. Training in both the intensive and extensive margin seems to be low in the MENA economies in comparison to other economies in ECA. Egypt and West Bank and Gaza are in the bottom four economies of the sample in terms of incidence of training. Morocco is the only economy in the top half of the sample of economies. With regards to intensity of training (only manufacturing firms), Egypt and West Bank and Gaza are the worst performing, with no economy in MENA in the top half of the sample. Second, over time the finding is mixed. On average, the share of firms offering formal training in the MENA region has increased, although the intensity of training has marginally decreased (Figure 5). In ECA, slight declines in both the incidence and intensity of formal training are observed (Figure 6).

Do note that for the share of workers that received formal training, zero is imputed for firms that did not offer formal training. However, if we restrict the sample to just firms that offer formal training, we see the same trends for the ECA and MENA regions over time.

Figure 1

Figure 2

Figure 3

The key explanatory variable is the quality of management practices, consistent with the methodology implemented by Bloom et al., (2013). This consists of eight components: (i) Problem resolution (ii) Number of performance indicators measured (iii) Level of ease or difficulty to achieve production or service provision targets (iv) Knowledge of production or service provision targets (v) Basis of manager bonuses (vi) Length of focus of production targets (vii) Promotion of non-managers and (viii) Dismissal of underperforming managers. The scoring for each component is provided in Table A2. The management practices module is only implemented for medium and large firms. Apart from Tunisia, MENA economies are in the bottom half of the sample with regards to management scores (Figure 4). Across both the MENA and ECA regions, average management scores have mostly declined between 2013 and 2019 (Figures 5 and 6).

Figure 4

Figure 5

Figure 6

A second variable of interest that was only captured for these surveys is What percentage or how many of this establishment's permanent full-time employees employed at the end of fiscal year [Insert last complete fiscal year] had a university degree? This question is typically not asked in the standard ES module, although the literature has used proxies. These include defining the skill of the worker in terms of occupation or the share of workers that have received secondary education (Almeida and Aterido, 2015; Liaqat and Nugent, 2015). The share of workers with university education employed in formal medium and large private firms in the MENA region (32 percent) is higher than ECA (27 percent). With the exception of Tunisia, MENA economies are in the top half of the sample with regards to share of workers employed with a university education (Figure 3).

The share of workers with university education in medium and large firms can be higher than the share of the population in general with a university degree. Take Egypt for example. Around 43 percent of all workers in formal firms (largely manufacturing and services) have a university education. In general, about 19 percent of the population has university education, according to the 2018 Egypt Labor Market Panel Survey (ELMPS).

Several control variables are also employed obtained from the Enterprise Surveys. These include firm size, age, outward orientation, quality certification, access to finance, informality, and perceptions of education quality and labor regulations. Summary statistics for all the variables are provided for the whole sample (Table 1), MENA cross-section sample (2013–2019/2020) in Table A3, the ECA cross-section sample (2013–2019/2020) in Table A4, the MENA panel sample (Table A5) and the ECA panel sample (Table A6). Table A7 provides the sample composition.

The data reveals some similarities and differences between firms that provide training and those that do not across both regions (Table A1). Firms that provide training in both MENA and ECA tend to be larger, younger, exporters, foreign-owned, and digitally connected as proxied by owning a website. In MENA, firms that offer formal training are more likely to be in the service sectors, while in ECA there does not seem to be noticeable differences. Across both regions, training firms are not more likely to be run by women than non-training mangers. However, firms that provide training are more likely to have a female owner in the MENA region, with no noticeable differences in the ECA region.

The following equation is estimated for the cross-section sample. (1) Traini=β0+β1MGMTi+β2Sizei+β3Agei+β4ManagerExpi+β5UniEdui+β6Manfi + βzZi+δ1Dc+δ2Dt+εi \begin{array}{*{20}{l}}{Trai{n_i} = {\beta _0} + {\beta _1}MGM{T_i} + {\beta _2}Siz{e_i} + {\beta _3}Ag{e_i} + {\beta _4}ManagerEx{p_i} + {\beta _5}UniEd{u_i} + {\beta _6}Man{f_i}}\\{\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; + {\beta _z}{Z_i} + {\delta _1}{D_c} + {\delta _2}{D_t} + {\varepsilon _i}}\end{array}

Where Train is either (i) whether or not a firm offers formal training or (ii) the share of workers receiving formal training only in manufacturing firms. The variable UniEdu is the share of workers with a university degree. MGMT is the average management practices score.

We also alternatively used Z-scores instead of the average scores for management practices. The results are largely unchanged. For the cross-sectional sample, the coefficient and statistical significance for management practices is the same for both MENA and ECA subsamples. For the panel estimations there a slight improvement in the statistical significance of management practice variable for the ECA subsample. The results are the same for the panel MENA subsample.

We utilize the same specification for the Panel sample as presented in equation (2) below, by replacing country fixed effects with firm-fixed effects (D_i). Standard errors are clustered at the country level for the panel estimations. (2) Trainit=β0+β1MGMTit+β2Sizeit+β3Ageit+β4ManagerExpit+β5UniEduit + β6Manfit+βzZit+γ1Di+γ2Dt+∈it \begin{array}{*{20}{l}}{Trai{n_{it}} = {\beta _0} + {\beta _1}MGM{T_{it}} + {\beta _2}Siz{e_{it}} + {\beta _3}Ag{e_{it}} + {\beta _4}ManagerEx{p_{it}} + {\beta _5}UniEd{u_{it}}}\\{\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; + {\beta _6}Man{f_{it}} + {\beta _z}{Z_{it}} + {\gamma _1}{D_i} + {\gamma _2}{D_t} + { \in _{it}}}\end{array}

The rationale for several of the control variables is based on the literature. Firm characteristics such as size, age, access to finance, and outward orientation – both in terms of exporter status and foreign ownership - (Almeida and Aterido, 2015). Other covariates that have been accounted of in previous estimations include quality certification and website ownership. Perceptions of labor regulations have been used to proxy for labor regulation stringency while senior management time spent in dealing with requirements of government regulations and average number of visits or required meetings with tax officials have been used to proxy for enforcement (Liaqat and Nugent, 2015). The perceptions of whether the inadequately educated workforce is a major or severe constraint to operations has been used to account for perceptions of education quality of the workforce at large (Liaqat and Nugent, 2015). We also include the following additional control variables: whether the top manager of the firm is a woman, the share of temporary workers, whether the firm purchased fixed assets, and finally whether the firm competes with informal firms.

There are a number of challenges with the empirical estimations. An important concern is simultaneity bias. One possibility is that training could lead to better management practices. This would imply that our estimates are biased upwards for the effect of management practices on training. This may be unlikely as the training captured is only directed to employees, and furthermore even though training may affect some of the management practices, it is unlikely to affect the aggregate score. It is also possible that training could attract high ability workers who tend to be more educated. This is only likely if the presence of training programs is the key attraction for highly educated workers. However, the presence of training programs could be correlated to several other features of firms that highly educated workers find attractive, and we account for these firm characteristics to the extent the data allows us. For the panel samples we account for firm-level fixed effects that capture time invariant firm-level omitted variables.

Table 2 presents the base results for three pooled cross-section samples: (i) All firms, (ii) the MENA sample and (ii) the ECA sample, including all firms surveyed in 2013 and 2019. For the overall sample we find that both management practices and the proportion of university educated workers are positively related to the presence of formal training (Table 2, column 1). The coefficients for both variables are statistically significant at the 1 percent level.

We also explored the results by splitting into the 2013 and 2019 waves. For the overall sample including MENA and ECA, management practices and share of university educated workers is positively related to the incidence of formal training, regardless of whether it the 2013 wave of the 2019 wave. For the share of workers trained in manufacturing firms, the results largely stand for the 2019 wave but are statistically insignificant for the 2013 wave. For MENA the findings for university education are largely driven by the 2019 sample. While the findings for ECA is mostly driven for both the 2013 and 2019 samples for management practices, the coefficient for university educated workers is statistically insignificant for the outcome variable of share of workers trained in manufacturing firms for the 2013 wave. Results are available from the authors upon request.

In columns 4,5 and 6 of Table 2 we replicate the same estimations as in columns 1,2 and 3 using the share of workers that received formal training as the outcome variable. Note that this information is only available for manufacturing firms.

We also ran the estimations for the incidence of training for the manufacturing firms alone. The sign and statistical significance of the coefficient for management practices is the same for both MENA and ECA manufacturing firms and the whole sample (manufacturing plus services). However, for the ECA sample of manufacturing firms, there is no statistically significant relationship between proportion of university educated workers and the incidence of formal training. For the MENA subsample, the sign and statistical significance of the coefficient of the proportion of university educated workers are retained. These results are available upon request.

In Table 3 we explore whether different types of management practices matter. We run the estimations replacing the overall score with the 8 subcomponents. For the overall sample (column 1) as well as the ECA sample (column 3), four subcomponents of management practices have statistically significant coefficients – (i) Number of production or service provision performance indicators monitored, (ii) Personnel's knowledge of production or service provision targets, (iii) Basis for promoting non-mangers and (iv) When underperforming managers were dismissed or reassigned. However, for the MENA subsample, only the coefficient for the number of production or service provision performance indicators monitored is statistically significant (column 2). The results are starker when the outcome variable is the share of workers that received formal training (manufacturing firms only). None of the management practices subcomponents have statistically significant coefficients for the MENA sample. For the overall as well as ECA samples, the number of production or service provision performance indicators monitored score and Personnel's knowledge of production or service provision targets score have positive coefficients that are statistically significant at the 1 percent level. These results confirm the fact that management practices, regardless of types, are far less a contributing factor to the use of formal training by firms in MENA than the rest of the sample.

In Table 4, we turn to the MENA panel estimations.

We also conducted panel estimations for the whole sample (as opposed to the regional breakdown). The coefficient of the management score is positive and statistically significant for the intensity but not the incidence of training. The coefficient for the proportion of university educated students is statistically insignificant. These results are available upon request.

In Table 5, we turn to the ECA panel estimation results. In column 1 we present the findings on whether or not the firm provides formal training. In column 2 we present the findings with the share of workers with formal training (manufacturing firms only). The coefficient for the overall management score is positive and statistically significant at the 10 percent level for the incidence of formal training and 1 percent for the intensity of formal training. The magnitudes are larger for the panel sample than the cross-sectional sample. The coefficient for the share of university educated workers is positive but statistically insignificant for the incidence of formal training. However, the coefficient for the share of university educated workers is negative and statistically significant at the 5 percent level. These findings run counter to the pooled cross-sectional results for the ECA sample. In columns 3 and 4 of Table 5, we explore whether the subcomponents of management practices matter. For the incidence of formal training, two of management score coefficients are positive and statistically significant – (i) the action when problem in the production/service provision arose score, and the (ii) Personnel's knowledge of production or service provision targets score. However, the basis for promoting non-managers score is negative and statistically significant at the 10 percent level. For the incidence of training, 3 scores have a positive and statistically significant coefficient – (i) Personnel's knowledge of production or service provision targets score, (ii) Focus of production targets score, and (iii) When underperforming managers were dismissed or reassigned score. However, the basis for promoting non-managers score is negative and statistically significant at the 10 percent level. In terms of comparisons across panel and cross-section ECA samples, only the management practices subcomponent on personnel's knowledge of production or service provision has a statistically significant coefficient across both samples. Overall, the findings show some heterogeneity across the components of the management score in the ECA sample, but the main finding that firms with better management practices seem to have higher incidence and intensity of formal training remains.

In this section we consider as robustness checks a number of additional variables that could be correlated with the prevalence of formal training or have been found to by the literature to be of importance. These include the skill level of production workers in manufacturing firms, the gender composition of workers, and labor regulations and enforcement. In Table 7 we include a variable that captures the skill level of the worker mostly based on the occupation. This is to account for the possibility that the portion of university educated workers may simply be capturing skilled workers, and also to check if our results stand after accounting for a more commonly used measure in the literature. In the survey, highly skilled workers were defined as those who were professionals and tasks required extensive theoretical and technical knowledge. The question was only asked of manufacturing firms. In Table 7 we provide estimations for the MENA cross section sample (columns 1 and 2), MENA panel sample (columns 3 and 4) and the ECA cross section sample (columns 5 and 6), and the ECA panel sample (columns 7 and 8), alternating between the prevalence of formal training and the share of workers that received formal training as outcome variables. For each of the estimations, we account for the share of high -skilled production workers. As shown in Table 7, for both the MENA cross-section and panel samples, the coefficient for the share of university educated workers is positive and statistically significant at the 1 percent level even after accounting for the proportion of skilled production workers. There is no statistically significant relationship between either the proportion of university educated workers or the proportion of highly skilled production workers with the prevalence of formal training for the ECA cross section and panel samples. Consistent with the base line estimates, the coefficient for the share of university educated workers is negative for the ECA panel sample when the outcome variable is the share of workers receiving formal training. The coefficient for management practices is positive and statistically significant at the 1 percent level across all the ECA samples. In summary, the relationship between university educated workers and training holds for the MENA region even after accounting for skilled workers.

In Table 8 we replicate the same estimations presented in Table 7 by substituting skilled workers with the proportion of women workers in the firm. The coefficient for the proportion of university educated workers remains statistically significant at the 1 percent level for the MENA cross-section sample after accounting for the proportion of women workers (Table 8 columns 1 and 2). This is true regardless of whether the outcome variable is the incidence or the intensity of formal training. The coefficient for the proportion women workers is positive but statistically insignificant for the MENA cross-section sample. However, the results change somewhat for the MENA panel sample (Table 8, columns 3 and 4). The positive coefficient of the proportion of women gains statistical significance at the 1 percent level and 5 percent level for the incidence and intensity of formal training, respectively. The statistical significance for the coefficient of the share of university educated workers drops to 10 percent with the incidence of formal training outcome variable. For the intensity of training outcome variable, the coefficient of the proportion of university educated workers is statistically significant at the 5 percent level. Thus, at least for the MENA panel sample, accounting for the share of women workers, lowers the statistical significance of the proportion of university educated workers. Part of the reason could be that in the developing MENA economies, a significant share of women workers is university educated. For the ECA cross-sectional sample the coefficient for the share of women workers is negative and statistically insignificant for incidence of training, but statistically significant at the 5 percent level for the intensity of training (Table 8, columns 5 and 6). This is a sharp difference from the MENA sample where the relationship between women workers and formal training is positive. This finding is stronger for the ECA panel subsample where the coefficient of the share of women workers is negative and statistically significant for both the incidence and intensity of training. The coefficient of the management practices variable is positive and statistically significant at the 1 percent level for the ECA cross-section sample. For the ECA panel sample, the coefficient for the management practices score is positive but only statistically significant for the intensity of training outcome variable.

In Table 9, we explore the relationship between labor regulations and perceptions of the workforce in the economy using cell averages consistent with Liaqat and Nugent (2015). A cell is defined by the location (within each country as defined by the strata), size and sector of the firm. Each cell average excludes the responding firm's answer. The labor regulations cell average is the average of the responses that state that labor regulations are major or severe obstacle to the running of the business. The labor regulations cell average may capture the stringency of labor regulations, with more stringent regulations restricting worker mobility and therefore incentivizing firms to invest in workers. The cell average of the perceptions of education of the workforce could proxy for the general perception of the education of labor, or the difficulty of hiring educated workers. The proportion of time spent dealing with government regulations and the average number of visits from tax officials can proxy for regulation enforcement. The findings in Table 9 largely show statistically insignificant coefficients for the proportion of time spent dealing with government regulations and the average number of visits from tax officials can proxy for regulation enforcement. The coefficient for the cell average of labor regulations is positive and only statistically significant (10 percent level) for the MENA cross-section when the outcome variable is the incidence of formal training (column 1). The coefficient is not statistically significant for the MENA panel estimations regardless of the training outcome variable. Similarly, the inadequately educated workforce cell average is negative and only statistically significant (at the 5 percent level) for the developing MENA cross-section sample alone when the outcome is the intensity of formal training. While not robust, both findings do hint at the possibility that in MENA, labor regulation stringency may incentivize firms to invest while negative perceptions of an inadequately educated work force may discourage investment in workers through training. Regardless, the coefficient for the share of university educated workers is positive and statistically significant at the 1 percent level for both the MENA cross sectional and panel estimations.

For the ECA samples, the coefficient of the inadequately educated workforce cell average is statistically insignificant for both the cross-section and panel estimations. The labor regulations cell average variable is positive but statistically insignificant with regards to the incidence of formal training for the ECA cross section sample. However, the coefficient is statistically significant at the 10 percent level for the training intensity outcome variable. For the ECA panel estimations, the coefficients for the labor regulations cell average are positive and statistically significant at the 1 percent level. The coefficient for the management practices is positive and statistically significant at least at the 5 percent level for both the ECA cross sectional and panel estimations. To summarize, while there is some hint that labor regulations may be positively correlated with formal training, and the low perceptions of the education of the workforce may reduce training, the findings are not robust.

An additional robustness check was conducted to see if corruption has any effect on formal training (as found by Boikos et al., 2022). For the pooled estimation, there is barely any statistical significance between bribery incidence and formal training (the exception being a negative coefficient for bribery incidence with a 10 percent significance for the ECA sample alone). For the panel sample, the coefficient for bribery incidence flips depending on whether the outcome variable is this incidence or intensity of formal training for the MENA panel sample. There is no statistically significant relationship between bribery incidence and formal training for the ECA panel sample. Thus, for these samples of analysis, we do not uncover a robust relationship between bribery incidence and formal training.

These findings are not reported but available upon request.

Furthermore, we explored the following control variables to account for product complexity and competition as a robustness check. Innovation measures are used to proxy for product complexity. This includes whether the firm spent on R&D, introduced a new product/service, or introduced a new process. In addition, ISIC 2-digit sector fixed effects are used to capture the nature of the product. Note that the adoption of ISO certification, a potential measure for product complexity, is included as a control in base estimations. Furthermore, firms are asked to provide the number of competitors they face in the main market, and this is accounted for in this robustness check. The results show that the main results are unchanged (Table 10). Management practices has a positive and statistically significant coefficient for ECA but not for MENA, while the share of the university educated workforce is positive correlated with presence of formal training for both regions.

This study explored the relationship between human capital in the firm in two forms – management practices and the share of university educated workers – and the incidence and intensity of formal training. It provided insights for the MENA region where training provisions by firms has been known to be low. The study also harnessed firm-level panel data from the developing MENA region to validate the findings. The findings show that the share of university educated workers is a robust predictor of formal training in MENA, while management practices is a robust predictor of formal training in ECA

The study provides some important policy implications. Increasing university education in the workforce in the MENA region may incentivize firms to invest more in their workers. This has positive implications of improving the workforce by retraining and updating skills. However, the high unemployment among the university graduates documented and the prevalence of queuing for public sectors jobs (Gatti et al., 2013) in the region suggests that this is not sufficient and would need to be coupled with reforms that strengthen the private sector and improve the business environment. Furthermore, if the prevalence of training is to adapt and update skills of educated workers towards work in the private sector, then the policy implication is not only be to increase university education, but also to ensure it serves the private sector as well. Furthermore, unlike the MENA region, in ECA, better management practices are likely to lead to greater incidence and intensity of formal training. Further analysis indicates that corruption and the lack of formal record keeping by business in MENA may be the reason why management practices are ineffective. For a subset of firms that did not experience corruption or have formal records, good management practices are positively correlated with the provision of formal training. Policies to improve the business environment and increase digitization may reap benefits by improving the provision of formal training in the private sector.

The study does have a number of limitations. Despite the robustness checks, and the use of panel estimations, we cannot completely rule out the possibility of simultaneity bias between our key variables – management practices and share of university educated workers – and our outcome variables of incidence and intensity of formal training. There is room for experimental research to validate these findings. Regardless, the study does leverage new data and point to interesting directions for future research. For one, it would be interesting to theoretically explore why certain types of management practices are more likely to lead to training, and what types of training are more likely. Second, it may be worth investigating if policies that led to increases in university education had corresponding effects on firm behavior as the pool of workers available to them became more educated and possibly more valuable. Finally, studies on informal training could complement the findings of this study.