A cross-country study of skills and unemployment flows

Damir Stijepic

doi:10.1186/s12651-021-00289-x

Stijepic, Damir

2021-03-27 00:00:00

Using an international survey that directly assesses the cognitive skills of the adult population, I study the relation between skills and unemployment flows across 37 countries. Depending on the specifically assessed domain, I docu- ment that skills have an unconditional correlation with the log-risk-ratio of exiting to entering unemployment of 0.65–0.68 across the advanced and skill-abundant countries in the sample. The relation is remarkably robust and it is unlikely to be due to reverse causality. I do not find evidence that this positive relation extends to the seven relatively less advanced and less skill-abundant countries in the sample: Peru, Ecuador, Indonesia, Mexico, Chile, Turkey and Kazakhstan. Keywords: Gross worker flows, Unemployment, Skills, Education, Human capital, International comparisons, Survey of Adult Skills, PIAAC JEL: J20, J24, J60, J64, I20 Furthermore, while an important role of formal educa- 1 Introduction tion is to add to the productivity of the students through At least since the human-investment revolution in eco- the formation of skills, some scholars stress the role of nomic thought in the 1960s (Bowman 1966), human education as a signaling device for the productive capaci- capital is regarded as a key factor in production. Most ties of applicants or as a rationing device for high-status activities in modern knowledge economies require a cer- jobs (Spence 1973; Collins 1979). tain set of skills, supposedly making the acquisition of Achievement tests as a measure of human capital are the relevant skills a prerequisite for a successful partici- gaining popularity due to improvements in testing tech- pation in the labor market. In the present paper, I assess niques and a broader availability. My analysis is based on the empirical content of this hypothesis by investigating the Survey of Adult Skills of the Programme for the Inter- to what extent the skills of a country’s labor force foster national Assessment of Adult Competencies (PIAAC), employment and prevent unemployment. which is an international survey that directly assesses Human capital is a complex construct. Since the semi- the cognitive skills of the adult population in a growing nal contributions to the human-capital literature by number of countries. Therefore, I have highly interna - Becker (1964) and by Mincer (1974), years of school- tionally comparable data on key skills in addition to the ing has remained the predominant measure of human traditional measures of human capital, e.g., years spent in capital. However, conceptual and qualitative differences education. in educational systems make international compari- Across the 30 advanced and skill-abundant countries sons challenging. The contribution of an additional year in the sample, I document that skills have a pronounced of schooling to the skills of the students in one coun- unconditional correlation with the log-risk-ratio of exit- try may very well differ from that in another country. ing to entering unemployment irrespective of the specific domain: 0.65 for literacy and 0.68 for numeracy. In con- *Correspondence: mail@damir.stijepic.com trast, formal education as measured by years of schooling Johannes Gutenberg University, Mainz, Germany © The Author(s) 2021. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. 9 Page 2 of 30 D. Stijepic has a modest unconditional correlation with the log- make use of the PIAAC data in order to explain inter- risk-ratio of 0.26. In the multivariate analysis with vari- national differences in macroeconomic variables. Spe - ous country characteristics, cognitive skills remain a key cifically, they show that differences in physical capital source of the international differences in unemployment together with a multidimensional measure of human flows. The instrumental-variable estimates do not sug - capital account for 42% of the variance across countries gest that reverse causality leads to a first-order upward in the gross domestic product per capita, stressing the bias. I do not find evidence that the positive relation role of cognitive skills. My findings suggest that a poten - between skills and the log-risk-ratio of exiting to enter- tially quantitatively important channel through which ing unemployment extends to the seven relatively less cognitive skills affect international differences in income advanced and less skill-abundant countries in the sam- is the capacity utilization of the factor labor, i.e., the lack ple: Peru, Ecuador, Indonesia, Mexico, Chile, Turkey and of labor underutilization in the form of unemployment. Kazakhstan. Furthermore, the employment effects of skills are of This paper is structured as follows. In Sect. 2, I review interest beyond the direct labor-income effects. A large the related literature. I introduce the data and the econo- and prominent body of literature documents the impact metric model in Sect. 3. This paper’s main findings on of displacement and employment breaks on various life international differences in skills and unemployment outcomes including divorce, criminality, mental health flows are in Sect. 4. The individual-level analysis of the and physical health (e.g., Eliason 2012; Fougère et al. relation between skills and the risks of entering and exit- 2009; Kuhn et al. 2009; Sullivan and von Wachter 2009). ing unemployment is in Sect. 5. Section 6 draws some Another related strand of the literature studies how a conclusions. Further data details and auxiliary results are country’s unemployment rate is affected by its institutions in the Appendix. including social security, employment protection, mini- mum wage, unionization and product market regulation 2 Related literature (see, e.g., Nickell and Layard 1999; Belot and van Ours A large and influential body of literature studies how 2004; Arpaia and Mourre 2012; Boeri et al. 2012; Launov skills contribute to an individual’s success in life. Heck- and Wälde 2016). I complement the literature by docu- man and Kautz (2012) review the evidence on how menting the close relation between average skill levels and school grades, the performance in achievement and IQ unemployment flows across countries. The individual-level tests, personality traits and attitudes relate to success in evidence suggests that a more skilled person is more likely life. Hanushek and Woessmann (2011) discuss the advan- to become and to stay employed. However, the skills of a tages and challenges of the cross-country comparative country’s labor force may also enhance institutions that approach that makes use of international achievement prevent inefficient unemployment or generate employ - tests in order to analyze the determinants and effects ment by fostering economic growth. Indeed, the empiri- of cognitive skills. Large international comparisons of cal evidence suggests a strong positive impact of cognitive the effects of skills among the adult population typically skills on macroeconomic growth (see, e.g., Hanushek and focus on the wage premium (e.g., Hanushek et al. 2015; Kimko 2000; Hanushek and Woessmann 2008, 2012). Fur- Leuven et al. 2004). Unemployment is only taken into thermore, the literature stresses the importance of institu- account insofar as it induces a selection bias in the wage tions for long-run growth (see, e.g., Acemoglu et al. 2001, regressions. A recent exception is Stijepic (2020a), who 2005). The question of whether unemployment is poten - documents that an individual’s cognitive skills are posi- tially a cause or a consequence of weak economic growth tively associated with the probability of being employed is outside the present paper’s scope. in all of the 32 studied countries. By exploiting the available data on unemployment The cited contributions to the literature analyze the duration, I obtain estimates of the flow rates into and out individual-level effects of skills. In contrast, I construct of unemployment and, hence, the extent of labor reallo- aggregate data in order to study the aggregate-level cation. A related strand of the literature studies the link effects of skills. Hidalgo-Cabrillana et al. (2017) similarly between the flexibility or sclerosis of labor markets and the level of unemployment (e.g., Blanchard and Sum- mers 1986; Bertola and Rogerson 1997; Blanchard and Portugal 2001). The literature addresses in particular the See also Iversen and Strøm (2020). The social returns can be quite different from the private returns to human capital. For instance, entrepreneurs may choose to acquire further In related work, I study the impact of skills on job mobility (Stijepic 2020b), skills in order to be more likely to succeed, reducing their probability of the trends and cycles in job mobility (Stijepic 2021), and the impact of dif- becoming unemployed. Provided that entrepreneurs create new job oppor- ferences between employers on labor-market outcomes (Stijepic 2016, 2017, tunities for others if they succeed, there are positive employment externali- 2019). ties. A cross‑country study of skills and unemployment flows Page 3 of 30 9 role of institutions, which determine, e.g., the degree of the same weight to each country in the pooled interna- employment protection. tional sample. The sample selection is as follows. First, I only study 3 Data and econometric model survey participants who report to be unemployed or to The following empirical analysis is based on the Sur - be engaged in paid work since transitions into and out vey of Adult Skills of the Programme for the Interna- of the labor force or non-profit activities may be par - tional Assessment of Adult Competencies (PIAAC). The tially affected by non-market considerations. Second, PIAAC is a large-scale initiative of the Organization for only respondents ages 25–54 may enter the final sample. Economic Cooperation and Development (OECD), pro- Therefore, I abstract from the peculiarities of the early viding internationally comparable data on key skills of and late stages of working life. The final sample encom - the adult population in the countries surveyed. During passes 117,183 individuals with 1512–13,691 observa- the first round in 2011–2012, 24 countries participated in tions per country. the data collection; of these, the following 23 are covered in my analysis: Austria, Belgium (specifically Flanders), 3.1 The ins and outs of unemployment Canada, Cyprus, Czechia, Denmark, Estonia, Finland, Similar to Shimer (2012), I make use of the number of France, Germany, Ireland, Italy, Japan, the Republic of employed, unemployed and short-term unemployed Korea, the Netherlands, Norway, Poland, Russia (exclud- workers in order to identify the entry rate into and the ing the Moscow municipal area), Slovakia, Spain, Swe- exit rate from unemployment. I classify survey partici- den, the United Kingdom (specifically England and pants as employed if they are engaged in paid work or Northern Ireland) and the United States. Another nine if they are temporarily away from a job or business to countries participated in the second round in 2014–2015: which they plan to return. In particular, a person who Chile, Greece, Indonesia (specifically Jakarta), Israel, is working for a family business without pay is not Lithuania, New Zealand, Singapore, Slovenia and Turkey. employed according to this classification. However, The third round in 2017 covers five countries addition - a person on parental leave is classified as employed. ally: Ecuador, Hungary, Kazakhstan, Mexico and Peru. I Survey participants are unemployed if they are not do report results for Russia. However, other studies (e.g., engaged in paid or unpaid work, if they are looking for Hanushek et al. 2015) do not use the data for Russia given paid work, and if they are able to start a new job within that, among other things, any statistics are potentially two weeks. I prefer the classification based on observed biased by the omission of the capital region. behavior, rather than the subjective self-assessed clas- The Survey of Adult Skills measures key cognitive skills sification which is more prone to cultural influences. that are essential for participation in the labor mar- However, I note that this definition of unemployment ket and in society. In contrast to IQ tests, the PIAAC excludes some people, such as discouraged workers who achievement tests measure general knowledge that can want to work but are not looking for jobs because they be acquired in schools and through life experiences. do not expect to succeed in their search efforts. Finally, The cognitive assessment is supplemented with a ques - I categorize respondents as short-term unemployed if tionnaire that collects a wide variety of background they had paid work in the preceding twelve months. All information including demographic, social, educational other unemployed survey participants are long-term and economic variables. In each country, a representa- unemployed. tive sample of adults ages 16–65 is interviewed at home. Figure 1 displays the unemployment rate by country. The standard survey mode is to answer questions on a The unemployment rate is 6.4% in the pooled international computer, but a pencil-and-paper interview option also sample, ranging from 2.2% in Belgium to 20.3% in Greece. exists for respondents who are not computer literate. In the international sample, there is an approximately The countries use different sampling schemes in select - equal share of short-term and long-term unemployed indi- ing their samples, but the samples are all aligned to viduals, i.e., 3.0% and 3.3%, respectively. The short-term known population counts with post-sampling weight- unemployment rate ranges from 1.3 in Belgium to 8.7 in ings. I employ these weights in all calculations, giving Spain. The long-term unemployment rate ranges from 0.9 in Belgium to 14.6% in Greece. I note that the statis- tics reflect the situation in 2011–2012, 2014–2015 or 2017 depending on the round of data collection. Since many See the OECD (2016) technical report for further information on the PIAAC. 5 6 Australia is not included since its public-use file is not directly accessible Furthermore, most individuals complete their formal education by the age over the OECD website. of 25. 9 Page 4 of 30 D. Stijepic c c unemployment rate (in %) exit rate (λ ) entry rate (δ ) Belgium Japan Peru Korea Mexico Norway Hungary Singapore Austria Netherlands Chile Kazakhstan Canada Indonesia Germany Finland Israel Russia Sweden New Zealand Ecuador Czechia Denmark UK Estonia Pooled Turkey France USA Poland Cyprus Slovakia Lithuania Slovenia Ireland Italy Spain Greece 0 4 8 12 16 20 0 .3 .6 .9 1.2 1.5 0 .03 .06 .09 .12 .15 c c Fig. 1 Unemployment rate, maximum-likelihood estimates of the exit rate from unemployment, , and of the entry rate into unemployment, δ , by country. Sample restricted to survey participants ages 25–54. Sampling weights employed in all calculations, giving the same weight to each country in the pooled specification. Author’s calculations based on the Survey of Adult Skills (PIAAC) countries experienced a pronounced surge in the unem- P(s = u ) t+1 s ployment rate in the aftermath of the Great Recession, =P(s = u|∃τ ∈[t, t + 1) : s = e) t+1 τ time effects are substantial potentially. Therefore, I take (2) =P(s = u) − P(s = u ) t+1 t+1 into account fixed effects by round of data collection in the =P(s = u) − e P(s = u), t+1 t following regressions. In order to motivate the likelihood function on which respectively, where P(·) denotes the probability of the the following empirical analysis is based, I rely on a sim- respective event. ple search model. Let s denote the employment status In the steady state, the flow of employed workers into of a worker in the year t. Workers are either employed, unemployment equals the flow of unemployed workers s = e , or unemployed, s = u . Unemployed workers into employment. Hence, the steady-state probabilities become employed at a rate of > 0 and employed work- of observing an employed and an unemployed worker are ers become unemployed at a rate of δ> 0 . The prob - P(s = e) = /(δ + ) and P(s = u) = δ/(δ + ) , resp e c- abilities of observing a long-term unemployed worker, tively. Hence, by Eqs. (1) and (2), the steady-state prob- s = u , and a short-term unemployed worker, s = u , in l s ability distribution is a random sample in the year t + 1 are P(s = u ) P(s) = if s = e, 1 − e t+1 δ + δ + (3) =P(s = u|∀τ ∈[t, t + 1) : s = u) t+1 τ (1) if s = u , and e if s = u . s l =e P(s = u) and t δ + Finally, I assume the effects of the various covari - ates, denoted by x for i ∈{1, ..., n} , on the flow rates into and out of unemployment to be log-linear, i.e., A cross‑country study of skills and unemployment flows Page 5 of 30 9 n n and 0.144, respectively. This reflects the difference in the δ = exp(δ + δ x ) and = exp( + x ) . In 0 i i 0 i i i=1 i=1 shares of long-term unemployed individuals, which are order to derive the likelihood function in Eq. (3), I impose 14.6% and 9.6% in Greece and in Spain, respectively. that unemployment is at its steady-state level. I relax this assumption in the Appendix. 3.2 Covariates In a first exercise based on the function in Eq. (3), I The PIAAC measures the cognitive skills of the survey par - estimate by maximum likelihood the parameters and ticipants in three domains: numeracy, literacy and prob- δ for each country, not taking into account any covari- lem solving in technology-rich environments. However, ate effects, i.e., = δ = 0 for all i = 1, ..., n . Let these i i c c the assessment of problem-solving skills is not carried country-level estimates be denoted by and δ , where 0 0 out in all countries and among all survey participants in c is the country index. Figure 1 displays the implied exit a country. The PIAAC defines literacy as “understanding, rate from unemployment, , and the entry rate into evaluating, using and engaging with written texts to par- unemployment, δ , by country. In the pooled interna- ticipate in society, to achieve one’s goals, and to develop tional sample, the exit rate from and the entry rate into one’s knowledge and potential,” while numeracy is defined unemployment are 0.652 and 0.044, implying an average as “the ability to access, use, interpret and communicate unemployment-spell and employment-spell duration of mathematical information and ideas, in order to engage 1.5 years and 22.7 years, respectively. I note that I dis- in and manage the mathematical demands of a range of tinguish individuals who have been unemployed for less situations in adult life.” The PIAAC measures literacy and than a year and individuals who have been unemployed numeracy on a 500-point scale. In the pooled interna- for at least a year in order to identify the exit rate from tional sample, the average of the literacy score and of the unemployment. However, the variation in shorter unem- numeracy score are 267 and 265, the standard deviations ployment spells tends to imply higher transition rates. being 53 and 57, respectively. For the estimations in this Indeed, numerous studies document the negative dura- study, I standardize the scores to obtain a mean of zero tion dependence in the exit rate from unemployment for and a standard deviation of one in the pooled international some countries (see, e.g., Kaitz 1970; Elsby et al. 2013). sample in order to facilitate the interpretation. Following In the Appendix, I also exploit the variation in shorter Hanushek et al. (2017), I focus on numeracy skills, which unemployment spells in order to obtain estimates of the are most comparable across countries supposedly. transition rates. Figure 2 depicts the relation across countries between In the model, the ratio of the exit rate from to the the estimated risk ratio of exiting to entering unem- entry rate into unemployment coincides with the ratio of c c c c c ployment, /δ , and the average skill scores. Across the employed workers, /(δ + ) , to unemployed workers, c c c 37 countries in the sample, the numeracy and literacy δ /(δ + ) . Indeed, the estimation procedure constrains scores have a limited correlation with the logarithmized the ratio of the exit to the entry rate to exactly match the risk ratio of exiting to entering unemployment of 0.01. ratio of employed to unemployed workers and, hence, The associated ordinary least-squares lines explain less the unemployment rate in the sample. Under the stated than 1% of the variance across countries in the logarith- model assumptions, explaining country differences in the mized risk ratio. However, allowing for a single optimal ratio of the exit to the entry rate or country differences in break point in the intercept and slope coefficients of the the ratio of employed to unemployed workers is the same ordinary least-squares regression, the explained vari- task. In that sense, the latter interpretation of country ance increases to 48% and 44%, respectively. Figure 2 differences does not rely on specific structural assump - shows the ordinary least-squares lines with the optimal tions. However, the estimated absolute magnitudes of the break points. The regression analysis implies the same entry and exit rates are to be more narrowly interpreted optimal division of the 37 countries irrespective of the within the theoretical framework. specific skill domain: a group of seven relatively skill- Notably, some countries substantially differ in the esti - scarce economies, i.e., Indonesia, Ecuador, Peru, Mexico, mated exit and entry rates despite similar unemployment rates. For instance, both Greece and Spain face an unem- ployment rate of 18–20%. On the one hand, the exit rate and the entry rate in Greece are 0.327 and 0.083, respec- tively. On the other hand, the rates in Spain are 0.645 The respective questionnaire items relating to unemployment duration are Footnote 8 (continued) not available in the Canadian and U.S. public-use files. ployment surge in the aftermath of the Great Recession if it had adopted the Spain is among those economies of the European Union that most decid- French employment-protection legislation. edly promoted temporary-employment contracts in the past, with tempo- rary employment reaching up to one-third of salaried employees. Bentolila Throughout this paper, I use the first plausible PIAAC-score values in et al. (2012) argue that Spain could have avoided about 45% of its unem- each domain. 9 Page 6 of 30 D. Stijepic (−0.379) Correlation = 0.684 (−0.278) Correlation = 0.654 BEL BEL JPN JPN PER PER KOR KOR MEX MEX NOR NOR HUN HUN SGP SGP AUT AUT NLD NLD CHL CHL KAZ KAZ CAN CAN IDN IDN DEU FIN DEU FIN ISR ISR RUS RUS SWE SWE NZL NZL ECU CZE ECU CZE DNK DNK GBR GBR EST EST TUR TUR FRA FRA USA USA POL POL CYP CYP SVK SVK LTU LTU SVN SVN IRL IRL ITA ITA ESP ESP GRC GRC 180 200 220 240 260 280 300 320 180 200 220 240 260 280 300 320 numeracy score literacy score (−0.258) Correlation = 0.408 (−0.270) Correlation = 0.510 BEL BEL JPN JPN PER PER KOR KOR MEX MEX NOR NOR HUN SGP HUN SGP AUT NLD AUT NLD CHL CHL KAZ KAZ CAN CAN IDN IDN FIN DEU DEU FIN ISR ISR RUS RUS SWE SWE NZL NZL ECU CZE ECU CZE DNK DNK GBR GBR EST EST TUR TUR FRA FRA USA USA POL POL CYP CYP SVK SVK LTU LTU SVN SVN IRL IRL ITA ITA ESP ESP GRC GRC 10000 20000 40000 80000 −.6 −.3 0 .3 .6 GDP per capita at chained PPPs (in 2011 US−Dollars) ICT in the workplace (std) Fig. 2 Country-level relation between the displayed variables and the maximum-likelihood estimates of the ratio of the exit rate from c c unemployment to the entry rate into unemployment, /δ . Ordinary least-squares lines depicted. Lines and statistics in gray (black) are for Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey (all other countries). Sample restricted to survey participants ages 25–54. Sampling weights employed in all calculations. Author’s calculations based on the Survey of Adult Skills (PIAAC) and the Penn World Table 9.1 (Feenstra et al. 2015) Kazakhstan, Chile and Turkey, and a group of 30 rela- Therefore, the skill-scarce and skill-abundant countries tively skill-abundant economies. warrant a separate analysis. In particular, I focus on the Notably, the seven skill-scarce economies, i.e., Indo- skill-abundant countries in the main analysis since the nesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and sample of skill-scarce countries is small. Turkey, exhibit substantially lower average skill levels I assess the proximity of an economy to the technology compared to the other countries in the sample. Specifi - frontier by the average use of information and communi- cally, the average numeracy score among the other 30 cation technology (ICT) in the workplace. The measure countries is 275 with a between-country standard devia- of ICT use at work is based on how often the survey par- tion of twelve. The seven skill-scarce countries have aver - ticipants usually “use email”, “use the internet in order to age numeracy scores that are 2.1–7.5 standard deviations better understand issues related to [their] work”, “conduct below that average. Similarly, the average literacy score transactions on the internet, for example buying or sell- among the 30 skill-abundant countries is 276 with a ing products or services, or banking”, “use spreadsheet between-country standard deviation of twelve. The seven software, for example Excel”, “use a word processor, for skill-scarce countries have average literacy scores that example Word”, “use a programming language to pro- are 2.0–6.9 standard deviations below that average. All gram or write computer code”, or “participate in real-time in all, the seven skill-scarce countries have starkly lower discussion on the internet, for example online confer- skill levels and the relation of skills with the ins and outs ence, or chat groups” in their job, where the five answer of unemployment seems to be qualitatively different. categories range from “Never” to “Every day.” The scale c c c c λ /δ λ /δ 4 8 16 32 4 8 16 32 c c c c λ /δ λ /δ 4 8 16 32 4 8 16 32 A cross‑country study of skills and unemployment flows Page 7 of 30 9 for ICT use is constructed according to item-response I also compute country averages of the covariates of theory: the item parameters are estimated using the gen- interest, e.g., the numeracy score. Let the country aver- eralized partial-credit model and the person-specific ages of the respective covariates be denoted by x . In the levels of ICT use are estimated using the weighted-likeli- second step, I simultaneously regress the estimated log- c c hood method. I assign a value of zero for ICT use if a per- risks and δ on a set of country averages of the covari- 0 0 son indicates never pursuing any of the stated activities ates of interest in a seemingly unrelated regressions or not using a computer on the job. In order to facilitate (SUR) setup à la Zellner (1962). Specifically, the system the interpretation, I standardize the measure of ICT use of econometric equations is to obtain a mean of zero and a standard deviation of one c c c in the pooled international sample. Figure 2 depicts the = + x + ǫ and 0 i 0 i relation across countries between the risk ratio of exiting i=1 (4) to entering unemployment and the average use of ICT in c c c the workplace. Notably, the seven skill-scarce countries δ = δ + δ x + ǫ , 0 i 0 i δ are among the countries with the lowest average use of i=1 ICT in the workplace. c c where ǫ and ǫ denote the unexplained residuals. Making use of the Penn World Table version 9.1 (Feen- Table 1 displays estimates of the parameters and δ for i i stra et al. 2015), I measure the economic advancement of the baseline sample excluding Indonesia, Ecuador, Peru, countries by the gross domestic product (GDP) per capita Mexico, Kazakhstan, Chile and Turkey. Specifications at purchasing-power-parities (PPP) exchange rates in 2011 (1)–(5) present a first series of SUR estimates for differ - US-Dollars. Figure 2 depicts the cross-country relation ent sets of control variables. Specification (1) effectively between the risk ratio of exiting to entering unemploy- replicates the bivariate scatter plot in Fig. 2, addition- ment and GDP per capita. Notably, the seven skill-scarce ally controlling for fixed effects by round of data collec - countries are among the countries with the lowest GDP tion. A one-standard-deviation increase in numeracy per capita. Furthermore, making use of the databases of skills is associated with an increase in the exit rate from the OECD and the International Labour Organization unemployment and with a decrease in the entry rate into (ILO), I also consider key institutional characteristics of unemployment by a factor of 2.2 (exp(0.777)) and by a the labor markets as explanatory factors: the minimum factor of 0.3 (exp(−1.193)) , respectively. Hence, the risk wage relative to the median wage, employment protection ratio of exiting to entering unemployment rises by a fac- as measured by the strictness of the regulations relating to tor of 7.2 (exp(0.777 + 1.193)) . The average numeracy the dismissal of workers with regular contracts (regular), score ranges from 255 in Spain to 298 in Japan. Evalu- employment protection as measured by the restriction of ated at this range, the estimates suggest an increase in the temporary work and fixed-term contracts (temporary), ratio of employed to unemployed workers by a factor of trade union density as measured by the share of employ- 4.3 (exp((1.969/57)(298 − 255))) , corresponding to a fall ees who are union members, unemployment benefits as in the unemployment rate from 18.2 to 5.0% in the case measured by the net replacement rate during the second of Spain. month of unemployment for a single person who earned The more comprehensive multivariate specifications 67 percent of the average wage (level), and unemployment (2)–(5) in Table 1 paint a similar picture. Numeracy skills benefits as measured by the difference in the net replace - remain a key determinant of the international differ - ment rates between the second month and the 14th ences in unemployment flows. Depending on the set of month of unemployment (degression). additional controls, a one-standard-deviation increase in numeracy skills is estimated to raise the risk ratio of exit- 4 Country‑level evidence ing to entering unemployment by a factor of 5.7–9.5. A In order to obtain estimates of the impact of skills on larger public sector is associated with a lower risk ratio unemployment flows at the country level, I choose a of exiting to entering unemployment. A higher minimum two-step estimation procedure. In the first step, I maxi - wage is estimated to raise employment, mainly by lower- mize the likelihood function in Eq. (3) in order to obtain ing the entry rate into unemployment. More generous— c c estimates of the parameters and δ for each country 0 0 either higher or less degressive—unemployment benefits c. I do not take into account any covariate effects at this are associated with a higher entry rate into unemploy- step, i.e., = δ = 0 for all i = 1, ..., n . Fur ther more, i i ment. The level of the unemployment benefits is posi - tively correlated with the exit rate from unemployment. I note that I do not control for composition effects, which may be relevant given the positive association between Accessed at https:// stats. oecd. org/ (OECD database) and at https:// ilost at. ilo. org/ data/ (ILO database) on December 20, 2020. 9 Page 8 of 30 D. Stijepic Table 1 Country-level estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment in a seemingly unrelated regressions setup à la Zellner (1962) SUR GMM (1) (2) (3) (4) (5) (6) (7) (log-risk of exiting unemployment) ∗∗∗ ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ Numeracy (std) 0.777 0.766 0.849 0.777 1.018 0.807 0.281 (0.297) (0.312) (0.318) (0.288) (0.296) (0.385) (0.711) ∗ ∗ Logarithmized GDP per capita (PPP) – 0.507 – – 0.614 0.540 0.356 (0.306) (0.315) (0.329) (0.413) ∗∗ ∗∗ ∗∗ Employment in public sector (share) – − 0.920 – – −2.137 −2.227 −2.451 (0.912) (1.090) (1.105) (1.238) ICT in the workplace (std) – − 0.180 – – − 0.164 − 0.012 0.365 (0.394) (0.456) (0.492) (0.679) Minimum relative to median wage – – − 0.586 – 0.397 0.168 − 0.400 (1.038) (0.945) (0.989) (1.250) Trade union density – – − 0.163 – − 0.101 − 0.066 0.020 (0.443) (0.441) (0.446) (0.499) ∗ ∗ ∗ ∗ Unemployment benefits (level) – – – 0.883 0.965 1.038 1.219 (0.490) (0.552) (0.564) (0.650) Unemployment benefits (degression) – – – 0.129 0.228 0.114 − 0.168 (0.342) (0.383) (0.408) (0.543) ∗∗ Employment protection (regular) – – – − 0.179 −0.220 − 0.189 − 0.113 (0.117) (0.111) (0.117) (0.153) Employment protection (temporary) – – – − 0.065 − 0.007 − 0.013 − 0.027 (0.072) (0.073) (0.074) (0.083) Instrument for numeracy – – – – – numeracy PISA ages 16–19 math 2 c 0.246 0.366 0.279 0.514 0.667 0.660 0.590 R ( ) Observations 30 30 29 27 27 27 27 δ (log-risk of entering unemployment) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ Numeracy (std) −1.193 −0.982 −1.321 −1.224 −1.231 −1.575 −2.144 (0.317) (0.336) (0.340) (0.382) (0.389) (0.509) (0.926) Logarithmized GDP per capita (PPP) – 0.076 – – − 0.203 − 0.323 − 0.521 (0.329) (0.414) (0.434) (0.537) Employment in public sector (share) – 0.883 – – 1.755 1.608 1.367 (0.983) (1.433) (1.460) (1.611) ∗∗ ICT in the workplace (std) – − 0.620 – – −1.205 − 0.958 − 0.550 (0.424) (0.600) (0.650) (0.883) ∗∗ ∗∗ ∗∗ Minimum relative to median wage – – −1.455 – −2.671 −3.042 −3.657 (1.112) (1.242) (1.307) (1.626) Trade union density – – 0.528 – − 0.317 − 0.260 − 0.167 (0.474) (0.579) (0.590) (0.650) ∗∗ ∗∗ ∗∗ Unemployment benefits (level) – – – 0.450 1.560 1.678 1.874 (0.651) (0.727) (0.745) (0.846) ∗∗ ∗∗ ∗∗ Unemployment benefits (degression) – – – − 0.198 −1.098 −1.283 −1.588 (0.454) (0.504) (0.539) (0.707) Employment protection (regular) – – – − 0.095 − 0.126 − 0.076 0.007 (0.156) (0.145) (0.155) (0.199) Employment protection (temporary) – – – 0.042 − 0.130 − 0.140 − 0.155 (0.095) (0.096) (0.098) (0.107) Instrument for numeracy – – – – – numeracy PISA A cross‑country study of skills and unemployment flows Page 9 of 30 9 Table 1 (continued) SUR GMM (1) (2) (3) (4) (5) (6) (7) ages 16–19 math 2 c 0.361 0.451 0.410 0.395 0.593 0.581 0.510 R (δ ) Observations 30 30 29 27 27 27 27 − δ (log-risk-ratio of exiting to entering unemployment) i i ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 1.969 1.748 2.170 2.001 2.249 2.382 2.425 (0.409) (0.399) (0.449) (0.442) (0.397) (0.513) (0.863) ∗ ∗∗ ∗ Logarithmized GDP per capita (PPP) – 0.431 – – 0.816 0.862 0.877 (0.392) (0.422) (0.438) (0.501) ∗∗∗ ∗∗∗ ∗∗ Employment in public sector (share) – −1.804 – – −3.892 −3.835 −3.817 (1.169) (1.461) (1.471) (1.502) ICT in the workplace (std) – 0.440 – – 1.041 0.946 0.915 (0.504) (0.612) (0.655) (0.824) ∗∗ ∗∗ ∗∗ Minimum relative to median wage – – 0.869 – 3.068 3.211 3.257 (1.465) (1.267) (1.317) (1.517) Trade union density – – − 0.691 – 0.216 0.194 0.187 (0.624) (0.591) (0.594) (0.606) Unemployment benefits (level) – – – 0.433 − 0.595 − 0.640 − 0.655 (0.752) (0.741) (0.751) (0.789) ∗∗∗ ∗∗ ∗∗ Unemployment benefits (degression) – – – 0.327 1.326 1.397 1.420 (0.525) (0.514) (0.543) (0.659) Employment protection (regular) – – – − 0.084 − 0.094 − 0.113 − 0.119 (0.180) (0.148) (0.156) (0.186) Employment protection (temporary) – – – − 0.107 0.123 0.127 0.128 (0.110) (0.098) (0.098) (0.100) Instrument for numeracy – – – – – numeracy PISA ages 16–19 math 2 c c 0.485 0.624 0.512 0.613 0.797 0.796 0.795 R ( − δ ) 0 0 Observations 30 30 29 27 27 27 27 Sample restricted to survey participants ages 25–54 and excluding survey participants from Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey. Fixed effects by round of data collection not displayed. Set of covariates in specifications (3) and (5)–(7) additionally includes an indicator variable for countries without minimum-wage regulations. Sampling weights employed in all calculations. Standard errors in parentheses. Statistical significance at the 10, 5, and 1% level denoted ∗ ∗∗ ∗∗∗ by , , and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC), the Penn World Table 9.1 (Feenstra et al. 2015), OECD statistics (https:// stats. oecd. org/) and ILO statistics (https:// ilost at. ilo. org/ data/) the benefits level and the entry rate into unemployment. respective averages among teenagers ages 16–19 irrespec- All in all, the correlation between the level of the unem- tive of whether they are in the labor force or not. Survey ployment benefits and the risk ratio of exiting to entering participants in this age range are, if at all, at the beginning of unemployment is not evidently negative across the stud- their careers, limiting the effect of labor-market outcomes ied countries, but a more degressive scheme tends to be on skills. In particular, the skills of this subgroup rather positively associated with employment. reflect the skills that are acquired in education during child - Different employment patterns could directly affect hood and early adulthood in a country. Specification (6) in skills over the life cycle, leading to biased estimates due to Table 1 displays the respective instrumental-variable esti- reverse causality. For instance, employment breaks might mates. The main qualitative conclusions are unaltered. In induce skill depreciation or prevent a person from acquir- specification (7), I instrument the skills of the adult popula - ing certain skills (see, e.g., Edin and Gustavsson 2008). In tion by the average math score that the students obtained in order to address this challenge to a causal interpretation the achievement tests of the Programme for International of the estimated employment effects of skills, I make use Student Assessment (PISA) in 2006. Numeracy skills of the instrumental-variable method. Specifically, I instru - ment the skills of the adult population in a country by the Accessed at https:// www. oecd. org/ pisa/ data/ on December 20, 2020. 9 Page 10 of 30 D. Stijepic Table 2 Country-level estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment in a seemingly unrelated regressions setup à la Zellner (1962) Women and men Men (1) (2) (3) (4) (5) (6) (7) (log-risk of exiting unemployment) Numeracy (std) − 0.097 0.002 0.017 0.151 − 0.110 0.072 − 0.245 (0.298) (0.170) (0.239) (0.266) (0.177) (0.246) (0.327) × Logarithmized GDP per capita (PPP, demeaned) – – 0.373 – – − 0.218 −1.006 (0.272) (0.450) (0.597) × ICT in the workplace (std) – – – 0.882 – 1.098 2.066 (0.596) (0.843) (1.061) ∗∗ ∗ × Employment in agriculture (share, demeaned) – – – – −6.251 -6.196 −7.553 (3.080) (4.087) (4.228) Logarithmized GDP per capita (PPP, demeaned) – – 0.293 – – 0.186 − 0.032 (0.216) (0.287) (0.358) ICT in the workplace (std) – – – 0.185 – − 0.374 0.036 (0.289) (0.382) (0.483) ∗∗∗ ∗∗∗ ∗∗ Employment in agriculture (share, demeaned) – – – – −9.258 −9.698 −6.506 (2.934) (3.504) (3.208) 2 c 0.294 0.005 0.080 0.082 0.217 0.273 0.272 R ( ) Observations 7 37 37 37 37 37 37 δ (log-risk of entering unemployment) ∗ ∗∗∗ ∗∗ ∗ ∗∗ ∗∗∗ Numeracy (std) 0.176 −0.299 −0.644 −0.522 −0.327 −0.572 −0.800 (0.143) (0.170) (0.213) (0.240) (0.181) (0.238) (0.268) ∗∗∗ × Logarithmized GDP per capita (PPP, demeaned) – – −0.876 – – − 0.444 − 0.249 (0.243) (0.436) (0.489) ∗∗∗ × ICT in the workplace (std) – – – −1.498 – − 0.576 -1.024 (0.537) (0.816) (0.869) ∗∗∗ × Employment in agriculture (share, demeaned) – – – – 8.143 2.454 3.534 (3.150) (3.958) (3.461) Logarithmized GDP per capita (PPP, demeaned) – – − 0.194 – – 0.080 0.218 (0.193) (0.278) (0.293) ICT in the workplace (std) – – – − 0.387 – − 0.285 − 0.158 (0.261) (0.370) (0.396) ∗∗ ∗ Employment in agriculture (share, demeaned) – – – – 6.985 3.636 4.459 (3.001) (3.393) (2.627) 2 c 0.873 0.269 0.460 0.448 0.395 0.497 0.560 R ( δ ) Observations 7 37 37 37 37 37 37 − δ (log-risk-ratio of exiting to entering unemployment) i i ∗∗ ∗ ∗∗ Numeracy (std) − 0.273 0.302 0.660 0.673 0.217 0.644 0.555 (0.319) (0.255) (0.322) (0.355) (0.246) (0.327) (0.365) ∗∗∗ × Logarithmized GDP per capita (PPP, demeaned) – – 1.249 – – 0.226 − 0.757 (0.367) (0.600) (0.666) ∗∗∗ ∗∗∗ × ICT in the workplace (std) – – – 2.379 – 1.674 3.090 (0.796) (1.123) (1.184) ∗∗∗ ∗∗ × Employment in agriculture (share, demeaned) – – – – −14.394 −8.651 −11.087 (4.283) (5.445) (4.716) Logarithmized GDP per capita (PPP, demeaned) – – 0.487 – – 0.106 − 0.249 (0.291) (0.383) (0.399) ICT in the workplace (std) – – – 0.572 – − 0.088 0.194 (0.386) (0.509) (0.539) A cross‑country study of skills and unemployment flows Page 11 of 30 9 Table 2 (continued) Women and men Men (1) (2) (3) (4) (5) (6) (7) ∗∗∗ ∗∗∗ ∗∗∗ Employment in agriculture (share, demeaned) – – – – −16.243 −13.334 −10.965 (4.079) (4.669) (3.579) 2 c c 0.299 0.148 0.362 0.374 0.422 0.507 0.444 R ( − δ ) 0 0 Observations 7 37 37 37 37 37 37 Sample restricted to survey participants ages 25–54. Specification (1) for Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey. Fixed effects by round of data collection not displayed. Sampling weights employed in all calculations. Standard errors in parentheses. Statistical significance at the 10, 5, and 1% level denoted by ∗ ∗∗ ∗∗∗ , , and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC) and the Penn World Table 9.1 (Feenstra et al. 2015) remain a key determinant of the international differences in and with the employment share of the agricultural the risk ratio of exiting to entering unemployment. sector as a measure of structural change, respectively. Specification (1) in Table 2 effectively replicates the In line with the stated hypothesis, the effect of numer- bivariate scatter plot in Fig. 2 for the seven skill-scarce acy skills on employment is positively associated with countries, additionally controlling for fixed effects by GDP, ICT and nonfarm employment. round of data collection. The regression does not sug- The more complete specification (6) in Table 2 gest a positive relation between numeracy skills and includes the interaction terms of the average numer- the risk ratio of exiting to entering unemployment. acy score with all three measures of economic devel- Indeed, the point estimate is negative yet statistically opment. Neither GDP, ICT nor nonfarm employment insignificant at conventional levels. In specification has a statistically significant impact on the employ- (2), I reestimate the relation on the full sample of 37 ment effect of skills conditional on the other factors. countries. The relation between skills and the risk All in all, I cannot readily differentiate between the ratio remains statistically insignificant at conventional three factors in the present setup. Indeed, (logarith- levels. All in all, the documented positive relation mized) GDP per capita, ICT use in the workplace and between skills and employment among the skill-abun- relative nonfarm employment have high pairwise cor- dant countries does not seem to extend to the seven relations of 0.6–0.8 in the sample. However, specifi- skill-scarce countries in the sample. cation (7) on the sample of male survey participants Why is the positive relation between skills and favors the more direct measures of technology; in employment limited to the skill-abundant countries? particular, ICT use in the workplace. I note that the A potential explanation is the distance to the tech- seven skill-scarce countries in the sample, i.e., Indo- nology frontier. Specifically, the hypothesis is that nesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and the skill-abundant countries are all at the same skill- Turkey, differ along various dimensions, e.g., infor- intensive technological frontier, whereas the skill- mal economy (see, e.g., Medina and Schneider 2018), scarce countries have not yet caught up, employing from the other 30 skill-abundant countries in the sam- technologies that are less skill-intensive. Indeed, Fig. 2 ple. However, the sample is not sufficiently large for a illustrates that the seven skill-scarce countries are comprehensive study. among the countries with the lowest GDP per capita and the lowest average use of ICT in the workplace. In 5 Individual‑level evidence order to explore the hypothesis further, I add interac- In order to obtain estimates of the employment effects tion terms between measures of a country’s skills and of skills at the individual level, I maximize the likeli- economic development to the set of controls. Spe- hood function in Eq. (3) on the pooled international cifically, specifications (3)–(5) in Table 2 additionally sample, directly taking into account the contribu- include the interaction term of the average numeracy tion of individual characteristics, x . I assume the score with GDP per capita as a measure of productive effects of the various covariates on the exit rate from capacity, with the average use of ICT in the workplace and the entry rate into unemployment to be log-lin- as a measure of the prevalence of new technologies, ear. Table 3 displays the individual-level estimates. I allow for country-level fixed effects in order to con- trol for differences in labor markets across countries. (Potential) experience is equal to age minus six minus In the regression of the numeracy score of the adult population on all other years of schooling, i.e., the typical number of years covariates, the numeracy score of teenagers (the PISA math score) has a F-sta- tistic of 19.7 (3.5). 9 Page 12 of 30 D. Stijepic Table 3 Individual-level maximum-likelihood estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment (1) (2) (3) (4) (5) (6) (7) (log-risk of exiting unemployment) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 0.166 0.174 0.084 0.162 0.165 0.164 0.163 (0.029) (0.032) (0.073) (0.030) (0.030) (0.029) (0.031) × Logarithmized GDP per capita (PPP, demeaned) – – – 0.043 – – 0.028 (0.061) (0.096) × ICT in the workplace (std) – – – – 0.061 – 0.032 (0.120) (0.192) × Employment in agriculture (share, demeaned) – – – – – − 0.352 0.054 (0.838) (0.970) Experience (decades) − 0.002 − 0.010 0.144 − 0.003 − 0.004 − 0.002 − 0.004 (0.082) (0.075) (0.361) (0.082) (0.082) (0.082) (0.082) − 0.017 − 0.013 − 0.068 − 0.016 − 0.016 − 0.017 − 0.016 Experience (decades) (0.020) (0.018) (0.103) (0.020) (0.020) (0.020) (0.020) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Female −0.276 −0.232 −0.662 −0.276 −0.276 −0.276 −0.276 (0.043) (0.041) (0.116) (0.043) (0.043) (0.043) (0.043) Log-likelihood −30,221 −26,470 −3,691 −30,203 −30,201 −30,205 −30,198 Countries 37 30 7 37 37 37 37 Observations 115,998 97,414 18,584 115,998 115,998 115,998 115,998 δ (log-risk of entering unemployment) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) −0.335 −0.356 − 0.171 −0.324 −0.338 −0.331 −0.332 (0.042) (0.044) (0.119) (0.040) (0.038) (0.038) (0.039) × Logarithmized GDP per capita (PPP, demeaned) – – – −0.170 – – − 0.042 (0.087) (0.142) × ICT in the workplace (std) – – – – − 0.271 – − 0.145 (0.171) (0.267) × Employment in agriculture (share, demeaned) – – – – – 2.230 1.029 (1.153) (1.392) ∗∗∗ ∗∗∗ ∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Experience (decades) −0.453 −0.399 −0.719 −0.449 −0.450 −0.449 −0.449 (0.101) (0.094) (0.424) (0.101) (0.101) (0.101) (0.101) ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ 0.059 0.051 0.082 0.058 0.058 0.058 0.058 Experience (decades) (0.025) (0.024) (0.112) (0.025) (0.025) (0.025) (0.025) ∗∗ ∗ ∗∗ ∗∗ ∗∗ ∗∗ Female −0.135 −0.137 − 0.117 −0.136 −0.135 −0.135 −0.136 (0.068) (0.072) (0.195) (0.068) (0.068) (0.068) (0.068) Log-likelihood −30,221 −26,470 −3,691 −30,203 −30,201 −30,205 −30,198 Countries 37 30 7 37 37 37 37 Observations 115,998 97,414 18,584 115,998 115,998 115,998 115,998 − δ (log-risk-ratio of exiting to entering unemployment) i i ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 0.501 0.530 0.255 0.486 0.503 0.495 0.495 (0.039) (0.040) (0.062) (0.036) (0.034) (0.034) (0.034) ∗∗ × Logarithmized GDP per capita (PPP, demeaned) – – – 0.213 – – 0.070 (0.086) (0.156) ∗∗ × ICT in the workplace (std) – – – – 0.332 – 0.177 (0.149) (0.189) ∗∗ × Employment in agriculture (share, demeaned) – – – – – −2.582 − 0.975 (1.085) (1.620) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Experience (decades) 0.451 0.389 0.863 0.446 0.447 0.447 0.445 (0.069) (0.071) (0.188) (0.069) (0.069) (0.070) (0.069) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ −0.075 −0.063 −0.150 −0.074 −0.074 −0.074 −0.074 Experience (decades) A cross‑country study of skills and unemployment flows Page 13 of 30 9 Table 3 (continued) (1) (2) (3) (4) (5) (6) (7) (0.016) (0.018) (0.040) (0.016) (0.016) (0.016) (0.016) ∗∗ ∗∗∗ ∗∗ ∗∗ ∗∗ ∗∗ Female −0.141 − 0.094 −0.546 −0.140 −0.140 −0.140 −0.140 (0.065) (0.068) (0.173) (0.065) (0.065) (0.065) (0.065) Log-likelihood −30,221 −26,470 −3,691 −30,203 −30,201 −30,205 −30,198 Countries 37 30 7 37 37 37 37 Observations 115,998 97,414 18,584 115,998 115,998 115,998 115,998 Sample restricted to survey participants ages 25–54. Specification (2) and specification (3) exclude survey participants from Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey, and from all other countries, respectively. Fixed effects by country not displayed. Sampling weights employed in all calculations, giving the same weight to each country. Robust standard errors in parentheses, adjusted for clustering at the country level. Statistical significance at the 10, 5, and 1% level * ** *** denoted by , and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC) and the Penn World Table 9.1 (Feenstra et al. 2015) that are associated with the person’s highest level of with disposable resources as a prerequisite for the education. trade-off between shorter search spells and better Specification (1) in Table 3 displays the individual-level jobs. Notably, Bick et al. (2018) document that average estimates for the pooled international sample including hours worked per adult are substantially higher along all 37 countries. Numeracy skills are estimated to increase the extensive and intensive margin in low-income the exit rate from unemployment and to lower the entry countries than in high-income countries. rate into unemployment—well in line with the country- In order to further explore the international differ- level estimates. Specifically, a one-standard-deviation ences in the employment effects of skills, I extend the increase in numeracy skills is associated with an increase set of controls to include interaction terms between in the exit rate from unemployment and with a decrease the numeracy score and measures of economic devel- in the entry rate into unemployment by a factor of 1.2 opment in specifications (4)–(7) of Table 3. In line (exp(0.165)) and by a factor of 0.7 (exp(−0.335)) , resp e c- with the country-level estimates, I document that the tively. Hence, the risk ratio of exiting to entering unem- effect of numeracy skills on the risk ratio of exiting to ployment rises by a factor of 1.7 (exp(0.165 + 0.335)). entering unemployment is positively associated with In specification (2) and in specification (3) of GDP per capita, with average ICT use in the work- Table 3, I estimate the transition parameters for the 30 place and with nonfarm employment across countries. skill-abundant countries and for the seven skill-scarce However, I cannot readily differentiate between the countries separately. Numeracy skills are estimated to three factors in the present setup. have a substantially smaller impact on the risk ratio of exiting to entering unemployment in the skill-scarce 6 Conclusion countries than in the skill-abundant countries. In light I construct aggregate data from the PIAAC sur- of the present paper’s scope, it is tempting to conclude vey data in order to study the aggregate-level effects that skills are less important for a successful partici- of skills, complementing the large and influential pation in the labor market in the seven skill-scarce body of literature that studies how skills contrib- countries. However, there are other competing expla- utes to an individual’s success in life. In general, the nations, between which I cannot readily differentiate. social returns can be quite different from the private My definition of unemployment may poorly reflect returns to skills. My analysis is unique in the sense the situation in the skill-scarce countries. Further- that it exploits highly internationally comparable data more, the most destituted individuals may face sub- from a large set of countries in order to explore the sistence and borrowing constraints that make longer effects of directly assessed skills on the ins and outs of unemployment spells unfeasible even in the case of unemployment. expected positive returns. Skills may be associated Across the 30 advanced and skill-abundant coun- tries in the sample, I document that skills have a pronounced unconditional correlation with the log- risk-ratio of exiting to entering unemployment irre- This variable is not available in the German public-use file. I compute the years of schooling in the German sample based on the International Standard spective of the specific domain: 0.65 for literacy Classification of Education (ISCED) and the mapping of the UNESCO Insti - and 0.68 for numeracy. The average numeracy score tute for Statistics (UIS); accessed at http:// uis. unesco. org/ en/ isced- mappi ngs/ ranges from 255 in Spain to 298 in Japan. Evaluated on December 20, 2020. 9 Page 14 of 30 D. Stijepic at this range, the estimates suggest an increase in the the third question in the affirmative and indicate hav- ratio of employed to unemployed workers by a factor ing been engaged in at least one of the eight activities of 4.3, corresponding to a fall in the unemployment stated in the second question. rate from 18.2 to 5.0% in the case of Spain. The distinction of short-term and long-term unem- The relation between skills and unemployment flows ployed survey participants is based on the following is remarkably robust across the 30 advanced countries two questions: “Have your ever had paid work? Please in the sample. Instrumental-variable estimates reject include self-employment.” (C_Q08a) and “During the hypothesis that the relation is exclusively driven the last 12 months, that is since MonthYear, did you by reverse causality, i.e., labor-market conditions have any paid work? Please include self-employment.” affecting the skills of the labor force. Strictly speaking, (C_Q08b). I classify respondents as long-term unem- I cannot firmly establish causality. Nevertheless, the ployed if they indicate never having had paid work or key determinants of the differences across advanced at least not having had paid work during the preced- countries in unemployment flows are skills or at least ing twelve months. If the latter question is answered factors that are closely related to skills. I do not find in the affirmative, I categorize the person as short- evidence that this relation between skills and unem- term unemployed. ployment flows extends to less advanced economies. I note that most questions on which this classi- fication of survey participants is based are used to determine the question routing. For instance, the interviewers obtain the following instruction for ques- Appendix: Data and auxiliary results tion C_Q01a: “The question is crucial for the routing. Further data details and summary statistics are Don’t knows and refusals are to be minimised. Please in Appendix 1 and the auxiliary results are in probe for an answer.” Appendix 2. Table 4 displays summary statistics and the maxi- mum-likelihood estimates of the unconditional tran- Appendix 1: Data sition rates for the pooled international sample and Survey participants are classified as employed if they by country. I also make use of the publicly available answer either of the two questions in the affirma- annual time series on the number of employed, unem- tive: “In the last week, did you do any PAID work for ployed and long-term unemployed workers from the at least one hour, either as an employee or as self- OECD database and from the ILO database. Figure 3 employed?” (C_Q01a) or “Last week, were you away juxtaposes labor statistics based on my PIAAC sam- from a job or business that you plan to return to?” ple with the respective statistics from the OECD (C_Q01b) and ILO databases. The ratio of employed to unem- In order to identify unemployed individuals, I rely ployed workers is highly correlated across the dif- on three survey questions: (i) “In the 4 weeks end- ferent data sources. However, the discrepancies are ing last Sunday, were you looking for paid work at particularly large for Belgium and for Japan. I note any time?” (C_Q02a), (ii) “In the four weeks end- that the PIAAC data is for the Flemish region exclu- ing last Sunday, did you do any of these things...”: sively, whereas the OECD data and ILO data are for “get in contact with a public employment office to the entire country. The OECD, ILO and PIAAC ratio find work?” (C_Q04a) or “get in contact with a pri- of employed to unemployed workers in Japan are 22, vate agency (temporary work agency, firm special- 22 and 44, respectively. Notably, the PIAAC statistics ising in recruitment, etc.) to find work?” (C_Q04b) based on observed behavior substantially differ from or “apply to employers directly?” (C_Q04c) or “ask those based on the self-reported classification in the among friends, relatives, unions, etc. to find work?” case of Japan. Specifically, the ratio of employed to (C_Q04d) or “place or answer job advertisements?” unemployed workers based on the self-assessed clas- (C_Q04e) or “take a recruitment test or examination sification in the PIAAC is 23, i.e., it is approximately or undergo an interview?” (C_Q04g) or “look for land, one-half of the ratio that is based on observed behav- premises or equipment for work?” (C_Q04h) or “apply ior. The ratio of short-term to long-term unemployed for permits, licences or financial resources for work?” workers is less highly correlated across the different (C_Q04i), and (iii) “If a job had been available in the data sources. According to the OECD database and week ending last Sunday, would you have been able to the ILO database, the ratio of short-term to long-term start within 2 weeks?” (C_Q05). I classify survey par- ticipants as unemployed if they answer the first and A cross‑country study of skills and unemployment flows Page 15 of 30 9 Table 4 Summary statistics and maximum-likelihood estimates of transition rates for survey participants ages 25–54 Observations Sample shares (in %) Transition rates Sample averages c c c c Employed Unemployed Women δ /δ Education Numeracy Literacy Experience Overall Short Long Austria 2,876 96.004 3.996 2.181 1.816 49.138 0.789 0.033 24.024 12.403 281.352 274.695 20.573 (0.365) (0.365) (0.272) (0.249) (0.932) (0.102) (0.005) (2.287) (0.048) (0.902) (0.808) (0.178) Belgium 2,704 97.830 2.170 1.312 0.858 46.204 0.928 0.021 45.088 13.175 290.997 284.524 19.889 (0.280) (0.280) (0.219) (0.177) (0.959) (0.161) (0.004) (5.951) (0.049) (0.946) (0.876) (0.183) Canada 13,691 95.635 4.365 3.005 1.361 47.376 1.166 0.053 21.908 13.929 274.473 281.439 17.477 (0.175) (0.175) (0.146) (0.099) (0.427) (0.061) (0.004) (0.916) (0.021) (0.460) (0.423) (0.081) Chile 2,537 95.791 4.209 3.223 0.987 44.454 1.451 0.064 22.756 12.236 216.693 226.573 17.101 (0.399) (0.399) (0.351) (0.196) (0.987) (0.175) (0.010) (2.250) (0.066) (1.127) (0.983) (0.220) Cyprus 2,364 91.622 8.378 4.275 4.103 48.592 0.714 0.065 10.936 13.296 271.129 273.940 18.017 (0.570) (0.570) (0.416) (0.408) (1.028) (0.073) (0.008) (0.812) (0.060) (0.944) (0.831) (0.217) Czechia 2,749 94.128 5.872 2.905 2.967 44.570 0.683 0.043 16.029 13.512 280.767 278.454 18.805 (0.448) (0.448) (0.320) (0.324) (0.948) (0.078) (0.006) (1.300) (0.049) (0.809) (0.765) (0.186) Germany 2,990 95.012 4.956 2.229 2.726 46.502 0.598 0.031 19.047 14.161 280.229 276.114 18.845 (0.398) (0.397) (0.270) (0.298) (0.912) (0.074) (0.005) (1.600) (0.053) (0.937) (0.847) (0.172) Denmark 3,387 94.035 5.965 3.621 2.344 47.890 0.934 0.059 15.765 13.314 288.379 279.258 15.391 (0.407) (0.407) (0.321) (0.260) (0.858) (0.087) (0.007) (1.144) (0.044) (0.853) (0.790) (0.174) Ecuador 2,487 94.132 5.868 2.235 3.633 43.012 0.479 0.030 16.042 12.979 193.912 199.781 15.998 (0.471) (0.471) (0.296) (0.375) (0.993) (0.065) (0.005) (1.369) (0.085) (1.072) (0.997) (0.229) Spain 3,202 81.792 18.208 8.653 9.555 45.207 0.645 0.144 4.492 12.133 255.483 260.008 19.276 (0.682) (0.682) (0.497) (0.520) (0.880) (0.039) (0.011) (0.206) (0.061) (0.848) (0.818) (0.180) Estonia 3,996 93.667 6.333 2.886 3.447 50.346 0.608 0.041 14.790 12.647 278.385 279.169 17.230 (0.385) (0.385) (0.265) (0.289) (0.791) (0.058) (0.005) (0.961) (0.042) (0.699) (0.693) (0.157) Finland 2,792 95.003 4.997 2.763 2.235 48.011 0.805 0.042 19.011 13.316 295.043 300.431 15.153 (0.412) (0.412) (0.310) (0.280) (0.946) (0.094) (0.006) (1.651) (0.052) (0.934) (0.899) (0.190) France 3,564 92.295 7.705 3.977 3.729 47.921 0.726 0.061 11.978 12.048 263.411 268.603 18.933 (0.447) (0.447) (0.327) (0.317) (0.837) (0.062) (0.006) (0.752) (0.057) (0.910) (0.790) (0.172) UK 4,594 93.789 6.195 2.592 3.604 46.065 0.542 0.036 15.100 13.289 270.283 280.434 15.165 (0.356) (0.356) (0.234) (0.275) (0.735) (0.050) (0.004) (0.923) (0.035) (0.798) (0.705) (0.154) Greece 2,454 79.702 20.298 5.663 14.635 42.498 0.327 0.083 3.927 12.596 258.311 254.586 19.396 (0.812) (0.812) (0.467) (0.714) (0.998) (0.028) (0.008) (0.197) (0.067) (0.963) (0.940) (0.210) Hungary 3,213 96.618 3.382 1.819 1.563 47.478 0.772 0.027 28.564 12.437 281.336 271.992 17.750 (0.319) (0.319) (0.236) (0.219) (0.881) (0.103) (0.004) (2.788) (0.051) (0.892) (0.758) (0.174) Indonesia 2,709 95.535 4.465 2.647 1.818 27.192 0.899 0.042 21.396 11.594 210.514 203.743 17.983 9 Page 16 of 30 D. Stijepic Table 4 (continued) Observations Sample shares (in %) Transition rates Sample averages c c c c Employed Unemployed Women δ /δ Education Numeracy Literacy Experience Overall Short Long (0.397) (0.397) (0.309) (0.257) (0.855) (0.110) (0.006) (1.990) (0.069) (1.042) (0.985) (0.183) Ireland 3,150 88.490 11.510 3.934 7.576 46.537 0.418 0.054 7.688 15.505 265.100 273.765 15.845 (0.569) (0.569) (0.346) (0.472) (0.889) (0.038) (0.006) (0.429) (0.051) (0.909) (0.815) (0.189) Israel 2,580 94.816 5.184 3.111 2.074 48.018 0.916 0.050 18.289 13.581 264.057 265.250 14.818 (0.437) (0.437) (0.342) (0.281) (0.984) (0.106) (0.007) (1.624) (0.049) (1.163) (1.015) (0.199) Italy 2,536 86.869 13.131 5.228 7.903 42.262 0.508 0.077 6.616 11.513 256.673 256.142 20.992 (0.671) (0.671) (0.442) (0.536) (0.981) (0.045) (0.008) (0.389) (0.076) (0.971) (0.877) (0.200) Japan 2,647 97.766 2.234 1.786 0.449 42.467 1.605 0.037 43.755 13.563 297.723 305.496 19.618 (0.287) (0.287) (0.257) (0.130) (0.961) (0.259) (0.008) (5.754) (0.045) (0.807) (0.693) (0.172) Kazakhstan 2,987 95.724 4.276 2.081 2.196 44.703 0.667 0.030 22.384 12.571 250.471 252.407 17.333 (0.370) (0.370) (0.261) (0.268) (0.910) (0.086) (0.005) (2.024) (0.041) (0.691) (0.717) (0.170) Korea 3,421 96.940 3.060 1.521 1.540 40.658 0.687 0.022 31.675 13.612 267.314 275.714 18.392 (0.295) (0.295) (0.209) (0.211) (0.840) (0.097) (0.004) (3.144) (0.048) (0.723) (0.660) (0.180) Lithuania 2,537 89.589 10.411 4.260 6.151 51.085 0.526 0.061 8.605 13.783 270.809 268.521 18.260 (0.606) (0.606) (0.401) (0.477) (0.993) (0.051) (0.007) (0.559) (0.049) (0.962) (0.832) (0.208) Mexico 2,708 96.855 3.145 1.791 1.354 40.460 0.843 0.027 30.799 11.218 215.912 223.069 18.474 (0.335) (0.335) (0.255) (0.222) (0.943) (0.125) (0.005) (3.391) (0.084) (0.941) (0.896) (0.208) Netherlands 2,691 95.959 4.041 2.420 1.621 46.608 0.913 0.038 23.745 13.884 289.711 292.602 17.446 (0.380) (0.380) (0.296) (0.244) (0.962) (0.117) (0.006) (2.324) (0.048) (0.927) (0.892) (0.197) Norway 2,773 96.702 3.298 1.615 1.683 47.405 0.673 0.023 29.320 14.733 289.090 288.179 14.952 (0.339) (0.339) (0.239) (0.244) (0.948) (0.102) (0.004) (3.118) (0.045) (1.026) (0.875) (0.181) New Zealand 3,000 94.432 5.568 2.703 2.865 49.129 0.665 0.039 16.960 14.222 280.348 287.830 17.043 (0.419) (0.419) (0.296) (0.305) (0.913) (0.075) (0.005) (1.350) (0.046) (0.966) (0.845) (0.192) Peru 3,710 97.358 2.642 1.506 1.136 43.402 0.844 0.023 36.843 15.935 185.869 195.227 18.981 (0.263) (0.263) (0.200) (0.174) (0.814) (0.116) (0.004) (3.771) (0.079) (1.056) (0.853) (0.168) Poland 3,113 91.801 8.199 2.868 5.331 46.412 0.430 0.038 11.197 13.529 265.518 271.118 16.197 (0.492) (0.492) (0.299) (0.403) (0.894) (0.046) (0.005) (0.732) (0.053) (0.885) (0.853) (0.187) Russia 1,512 94.692 5.308 3.544 1.764 47.115 1.102 0.062 17.839 13.978 272.839 277.485 17.427 (0.577) (0.577) (0.476) (0.339) (1.284) (0.158) (0.011) (2.046) (0.086) (1.011) (1.070) (0.255) Singapore 2,951 96.565 3.435 2.191 1.244 46.094 1.016 0.036 28.109 12.513 266.941 263.445 17.464 (0.335) (0.335) (0.270) (0.204) (0.918) (0.132) (0.006) (2.841) (0.056) (1.183) (1.028) (0.205) A cross‑country study of skills and unemployment flows Page 17 of 30 9 Table 4 (continued) Observations Sample shares (in %) Transition rates Sample averages c c c c Employed Unemployed Women δ /δ Education Numeracy Literacy Experience Overall Short Long Slovakia 2,809 89.979 10.021 3.878 6.143 45.779 0.489 0.055 8.979 13.709 283.999 279.566 18.518 (0.567) (0.567) (0.364) (0.453) (0.940) (0.047) (0.006) (0.564) (0.050) (0.822) (0.708) (0.185) Slovenia 2,857 88.778 11.222 2.997 8.225 46.713 0.311 0.039 7.911 10.796 265.569 261.714 17.896 (0.591) (0.591) (0.319) (0.514) (0.934) (0.034) (0.005) (0.469) (0.035) (0.975) (0.863) (0.193) Sweden 2,366 94.569 5.431 2.667 2.764 47.854 0.675 0.039 17.414 12.729 286.869 287.625 16.205 (0.466) (0.466) (0.331) (0.337) (1.027) (0.087) (0.006) (1.580) (0.049) (1.111) (1.002) (0.209) Turkey 1,934 92.624 7.376 3.695 3.682 23.050 0.695 0.055 12.557 9.363 234.901 235.302 19.209 (0.595) (0.595) (0.429) (0.428) (0.958) (0.084) (0.008) (1.092) (0.080) (1.197) (0.976) (0.242) USA 2,592 92.141 7.859 4.313 3.546 48.792 0.796 0.068 11.724 13.898 260.674 274.935 17.686 (0.529) (0.529) (0.399) (0.363) (0.982) (0.077) (0.008) (0.856) (0.062) (1.123) (0.975) (0.200) Pooled 117,183 93.645 6.354 3.044 3.310 45.496 0.652 0.044 14.735 13.195 264.972 267.413 17.660 (0.071) (0.071) (0.050) (0.052) (0.145) (0.011) (0.001) (0.176) (0.009) (0.168) (0.154) (0.030) Sampling weights employed in all calculations, giving the same weight to each country in the pooled specification. Education and experience in years. Delta-method standard errors in parentheses. Author’s calculations based on the Survey of Adult Skills (PIAAC) 9 Page 18 of 30 D. Stijepic Correlation = 0.865 Correlation = 0.845 IDN NOR NOR PER MEX MEX KOR KOR SGP ECU IDN HUN HUN NZL AUT NZL AUT NLD JPN NLD JPN ISR ISR KAZ RUS RUS DEU DEU CHL SWE SWE CHL GBR CZE GB CZE R FIN CAN FIN CAN BEL BEL DNK DNK USA USA FRA FRA ITA CY POL P ITA CY PO PL TUR TUR SVN SVN LTU LTU EST EST SVK SVK IRL IRL ESP ESP GRC GRC 5 10 20 40 5 10 20 40 PIAAC employment/unemployment PIAAC employment/unemployment Correlation = 0.347 Correlation = 0.325 KOR KOR MEX MEX CAN KAZ ISR ISR CHL CAN NZL NZL ECU SWE TUR TUR SWE FIN CYP FIN CYP AUT NOR AUT NOR DNK DNK USA USA RUS POL NLD NLD GBR GBR JPN POL RUS FRA HUN ESP FRA HUN ESP CZE CZE IDN LTU LTU DEU DEU BEL BEL JPN ITA SVN ITA SVN EST EST IRL IRL SVK SVK GRC GRC .5 1 2 4 .5 1 2 4 PIAAC short−term/long−term unemployment PIAAC short−term/long−term unemployment Fig. 3 Comparison of unemployment statistics based on the PIAAC data with unemployment statistics from the OECD and ILO databases. Ordinary least-squares lines (black) and 45-degree lines (gray) depicted. Sample restricted to survey participants ages 25–54. Sampling weights employed in all calculations. Author’s calculations based on the Survey of Adult Skills (PIAAC), OECD statistics (https:// stats. oecd. org/) and ILO statistics (https:// ilost at. ilo. org/ data/) unemployed workers is extremely high in some coun- i.e., less than high school (low), high school (medium) tries, exceeding 40 in Mexico and 200 in Korea. and more than high school (high). Table 5 displays the estimates for each subgroup separately. The effect of skills on the risk ratio of exiting to entering unem- ployment is statistically significant at the five-percent Appendix 2: Auxiliary results level in all subgroups. In a first sensitivity analysis, I reestimate specification In the baseline specification, I distinguish workers (5) of Table 1 dropping one country at a time from who have been unemployed for less than a year and the sample. Figure 4 displays the estimated effects workers who have been unemployed for at least a year of numeracy skills on the transition rates once the in order to identify the exit rate from unemployment. respective country is excluded from the sample. The However, most countries report the unemployment effect of skills on the risk ratio of exiting to entering duration in months in their public-use files. For those unemployment remains statistically significant at the countries, I additionally exploit the variation in unem- one-percent level in all estimations, indicating that ployment spells of less than a year in order to obtain the cross-country pattern is not driven by individual an alternative estimate of the exit rate. Under the con- countries. Furthermore, I reestimate specification stant inflow assumption, the probability of observing (5) of Table 1 for four age groups, i.e., 25–34, 35–44, a worker with an unemployment duration of n months 45–54 and 55–64, and for three education groups, n n+1 is exp (−µ ) − exp (−µ ) , wher e µ denotes the 12 12 OECD short−term/long−term unemployment OECD employment/unemployment .5 2 8 32 128 5 10 20 40 ILO short−term/long−term unemployment ILO employment/unemployment .5 2 8 32 128 5 10 20 40 A cross‑country study of skills and unemployment flows Page 19 of 30 9 λ −δ λ δ numeracy numeracy numeracy numeracy Belgium Japan Peru Korea Mexico Norway Hungary Singapore Austria Netherlands Chile Kazakhstan Canada Indonesia Germany Finland Israel Russia Sweden New Zealand Ecuador Czechia Denmark UK Estonia Turkey France USA Poland Cyprus Slovakia Lithuania Slovenia Ireland Italy Spain Greece 0 .6 1.2 1.8 2.4 3 0 .3 .6 .9 1.2 0 −.4 −.8 −1.2 −1.6 c c Fig. 4 Country-level estimates of the effects of numeracy skills on the risks of exiting unemployment, , and of entering unemployment, δ , based on specification (5) in Table 1, excluding the respective country from the sample. Black, dark gray and light gray indicate statistical significance at the 10, 5, and 1% level, respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC), the Penn World Table 9.1 (Feenstra et al. 2015), OECD statistics (https:// stats. oecd. org/) and ILO statistics (https:// ilost at. ilo. org/ data/) alternative exit rate. The alternative entry rate into country with only limited longitudinal information. unemployment, denoted by ρ , is determined by the However, with the OECD times series and the ILO condition that unemployment is at its steady state, time series, I can relax this assumption. For an arbi- i.e., ρ/(µ + ρ) . Fig ur e 5 displays the country pairs of trary probability of being unemployed in the year t, the baseline and alternative transition rates. All alter- the probability of being unemployed in the year t + 1 native transition rates are above the 45-degree line, is i.e., additionally exploiting the variation in shorter −(+δ) unemployment spells leads to higher estimates of the P(s = u) = 1 − e t+1 + δ (5) exit rate from unemployment and, hence, to higher −(+δ) + e P(s = u). estimates of the entry rate into unemployment. I t also reestimate specification (5) of Table 1 with the Additionally exploiting the relation in Eq. (1) or in Eq. alternative transition rates. The first specification (2), data on the number of employed, unemployed and in Table 6 shows the estimates. The main qualitative long-term unemployed workers in the current and in the implications are unaltered. previous year is then sufficient to identify the exit rate from unemployment and the entry rate into unemploy- In order to derive the likelihood function in Eq. (3), I ment. Shimer (2012) provides a more detailed exposition. impose that unemployment is at its steady-state level. I reestimate specification (5) of Table 1 with the OECD This assumption is necessary since the PIAAC data is and ILO transition rates as implied by Eqs. (1) and (5). typically based on a single cross-sectional survey per The second and third specification in Table 6 display the 9 Page 20 of 30 D. Stijepic Table 5 Country-level estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment in a seemingly unrelated regressions setup à la Zellner (1962) Age Education 25–34 35–44 45–54 55–65 Low Medium High (log-risk of exiting unemployment) ∗∗ ∗∗∗ ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 0.749 1.485 0.706 1.582 0.614 1.074 1.109 (0.340) (0.395) (0.280) (0.605) (0.447) (0.394) (0.418) ∗∗ ∗ ∗∗∗ ∗ Logarithmized GDP per capita (PPP) 0.700 0.590 0.541 1.394 − 0.096 1.077 0.618 (0.329) (0.442) (0.360) (0.779) (0.454) (0.350) (0.358) ∗∗∗ ∗ ∗∗ Employment in public sector (share) 0.114 −3.704 −1.731 0.168 −3.579 0.165 − 0.341 (1.274) (1.248) (1.007) (1.737) (1.461) (1.798) (1.035) ICT in the workplace (std) − 0.270 − 0.717 0.059 − 0.552 0.787 − 0.386 − 0.311 (0.406) (0.561) (0.434) (0.897) (0.819) (0.382) (0.404) Minimum relative to median wage − 0.266 0.392 1.126 3.060 2.132 0.094 0.267 (1.068) (1.248) (1.012) (2.154) (1.547) (1.047) (1.132) Trade union density − 0.356 0.678 − 0.781 0.440 0.305 − 0.462 0.069 (0.478) (0.571) (0.478) (0.944) (0.677) (0.611) (0.510) ∗∗∗ ∗ Unemployment benefits (level) 0.698 0.383 1.693 − 0.228 0.634 1.106 − 0.341 (0.623) (0.746) (0.613) (1.354) (0.794) (0.649) (0.727) Unemployment benefits (degression) 0.365 0.254 0.159 1.154 0.566 0.247 0.257 (0.420) (0.532) (0.414) (0.870) (0.601) (0.456) (0.484) ∗ ∗∗ ∗∗ ∗∗ Employment protection (regular) −0.212 − 0.227 −0.262 −0.533 − 0.066 −0.293 − 0.007 (0.128) (0.146) (0.116) (0.249) (0.168) (0.134) (0.138) Employment protection (temporary) − 0.014 − 0.039 0.030 0.110 0.019 − 0.013 − 0.074 (0.078) (0.095) (0.079) (0.182) (0.103) (0.082) (0.086) 2 c 0.538 0.665 0.655 0.584 0.459 0.683 0.528 R ( ) Observations 27 27 27 27 26 26 26 δ (log-risk of entering unemployment) ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) −1.918 − 0.243 −1.220 − 0.577 − 0.480 − 0.606 −1.788 (0.485) (0.500) (0.428) (0.626) (0.479) (0.497) (0.514) ∗∗∗ Logarithmized GDP per capita (PPP) − 0.319 0.496 − 0.635 − 0.969 −1.341 − 0.449 − 0.240 (0.468) (0.559) (0.551) (0.805) (0.486) (0.440) (0.441) ∗∗∗ ∗ ∗∗∗ ∗ Employment in public sector (share) 0.039 0.037 5.115 3.258 6.973 4.028 0.318 (1.817) (1.581) (1.540) (1.796) (1.564) (2.265) (1.273) ∗∗∗ ∗∗ ICT in the workplace (std) − 0.148 −2.474 − 0.141 0.084 2.212 0.296 0.157 (0.579) (0.711) (0.664) (0.927) (0.877) (0.481) (0.497) ∗∗∗ ∗ Minimum relative to median wage -1.010 -1.710 −4.373 -1.485 −2.793 − 0.958 -1.651 (1.522) (1.582) (1.548) (2.226) (1.656) (1.318) (1.392) ∗∗∗ ∗∗∗ Trade union density 0.884 0.068 −2.446 − 0.976 −2.195 − 0.916 − 0.055 (0.681) (0.724) (0.731) (0.976) (0.725) (0.769) (0.627) ∗ ∗∗∗ Unemployment benefits (level) 0.297 1.712 2.578 − 0.887 0.958 0.476 − 0.191 (0.888) (0.945) (0.938) (1.399) (0.850) (0.818) (0.894) ∗∗ ∗∗ Unemployment benefits (degression) − 0.193 −1.672 − 0.984 − 0.443 1.477 − 0.601 − 0.974 (0.599) (0.674) (0.632) (0.899) (0.644) (0.574) (0.596) ∗∗∗ ∗∗ ∗ Employment protection (regular) 0.082 − 0.260 − 0.285 − 0.410 −0.714 −0.389 0.330 (0.183) (0.185) (0.178) (0.257) (0.180) (0.169) (0.169) Employment protection (temporary) 0.014 −0.230 − 0.062 − 0.132 0.180 0.017 − 0.130 (0.112) (0.121) (0.121) (0.188) (0.110) (0.103) (0.105) 2 c 0.590 0.476 0.636 0.474 0.646 0.483 0.530 R ( δ ) Observations 27 27 27 27 26 26 26 − δ (log-risk-ratio of exiting to entering unemployment) i i A cross‑country study of skills and unemployment flows Page 21 of 30 9 Table 5 (continued) Age Education 25–34 35–44 45–54 55–65 Low Medium High ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 2.666 1.728 1.926 2.159 1.093 1.680 2.897 (0.424) (0.486) (0.442) (0.332) (0.501) (0.580) (0.464) ∗∗ ∗∗ ∗∗∗ ∗∗ ∗∗∗ ∗∗ Logarithmized GDP per capita (PPP) 1.019 0.094 1.175 2.363 1.245 1.526 0.858 (0.409) (0.544) (0.569) (0.427) (0.508) (0.514) (0.398) ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Employment in public sector (share) 0.075 −3.742 −6.846 −3.090 −10.553 -3.862 − 0.659 (1.587) (1.537) (1.593) (0.953) (1.636) (2.643) (1.149) ∗∗ ICT in the workplace (std) − 0.121 1.756 0.200 − 0.636 -1.424 − 0.682 − 0.467 (0.506) (0.691) (0.687) (0.492) (0.917) (0.561) (0.448) ∗∗∗ ∗∗∗ ∗∗∗ Minimum relative to median wage 0.744 2.101 5.500 4.544 4.925 1.052 1.918 (1.330) (1.538) (1.600) (1.182) (1.732) (1.539) (1.256) ∗∗ ∗∗ ∗∗∗ ∗∗∗ Trade union density −1.240 0.610 1.665 1.416 2.500 0.454 0.124 (0.595) (0.704) (0.756) (0.518) (0.758) (0.898) (0.566) Unemployment benefits (level) 0.401 -1.329 − 0.885 0.658 − 0.324 0.631 − 0.150 (0.775) (0.918) (0.969) (0.743) (0.889) (0.955) (0.807) ∗∗∗ ∗ ∗∗∗ ∗∗ Unemployment benefits (degression) 0.558 1.926 1.143 1.597 − 0.911 0.848 1.232 (0.523) (0.655) (0.654) (0.477) (0.673) (0.670) (0.538) ∗ ∗∗∗ ∗∗ Employment protection (regular) −0.293 0.034 0.023 − 0.123 0.649 0.095 −0.337 (0.160) (0.180) (0.184) (0.136) (0.189) (0.197) (0.153) ∗∗ Employment protection (temporary) − 0.028 0.190 0.092 0.242 − 0.161 − 0.030 0.056 (0.097) (0.117) (0.125) (0.100) (0.116) (0.120) (0.095) 2 c c 0.780 0.693 0.674 0.805 0.713 0.689 0.797 R ( − δ ) 0 0 Observations 27 27 27 27 26 26 26 Sample excluding survey participants from Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey. Fixed effects by round of data collection and minimum- wage regulation not displayed. Sampling weights employed in all calculations. Standard errors in parentheses. Statistical significance at the 10, 5, and 1% level ∗ ∗∗ ∗∗∗ denoted by , , and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC), the Penn World Table 9.1 (Feenstra et al. 2015), OECD statistics (https:// stats. oecd. org/) and ILO statistics (https:// ilost at. ilo. org/ data/) SGP ESP Correlation = 0.803 Correlation = 0.702 CHL PER MEX KOR CHL LTU IDN SGP ITA JPN GRC KAZ HUN IDN CYP TUR RUS RUS ISR BEL ISR ECU FRA MEX FIN NLD KAZ NZL ECU KOR DNK NZL AUT PER FIN NOR DNK SVK SVN TUR POL SWE CZE IRL CYP GBR SW EST E CZE HUN LTU FRA NLD GBR EST AUT DEU POL DEU ITA ESP JPN NOR SVK SVN BEL IRL GRC .4 .8 1.6 .02 .04 .08 .16 c c exit rate (λ ) entry rate (δ ) Fig. 5 Comparison of the estimated baseline transition rates with the estimated alternative transition rates. Ordinary least-squares lines (black) and 45-degree lines (gray) depicted. Sample restricted to survey participants ages 25–54. Sampling weights employed in all calculations. Author’s calculations based on the Survey of Adult Skills (PIAAC), OECD statistics (https:// stats. oecd. org/) and ILO statistics (https:// ilost at. ilo. org/ data/) alternative exit rate (μ ) .4 .8 1.6 3.2 alternative entry rate (ρ ) .02 .04 .08 .16 9 Page 22 of 30 D. Stijepic Table 6 Country-level estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment in a seemingly unrelated regressions setup à la Zellner (1962) Monthly Out of steady state unemployment Contemporaneous 2011–2017 2001–2019 PIAAC OECD ILO OECD ILO OECD ILO (log-risk of exiting unemployment) ∗∗∗ 1.022 0.336 0.128 0.183 − 0.155 − 0.194 − 0.476 Numeracy (std) (0.354) (0.560) (0.560) (0.503) (0.515) (0.456) (0.468) Logarithmized GDP per capita (PPP) 0.546 − 0.159 0.001 − 0.319 − 0.297 − 0.111 − 0.071 (0.479) (0.595) (0.595) (0.535) (0.547) (0.484) (0.497) ∗∗∗ ∗∗ ∗ ∗ ∗∗ ∗ Employment in public sector (share) −4.015 −4.296 −3.969 −3.405 -2.924 −3.617 −3.234 (1.352) (2.062) (2.062) (1.852) (1.896) (1.678) (1.722) ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ICT in the workplace (std) 1.104 2.680 2.920 2.653 3.078 2.316 2.696 (0.563) (0.863) (0.863) (0.775) (0.793) (0.702) (0.721) Minimum relative to median wage 1.777 2.900 2.740 2.368 2.302 1.457 1.501 (1.155) (1.788) (1.788) (1.606) (1.643) (1.454) (1.493) Trade union density 0.040 0.115 0.134 − 0.172 − 0.071 0.300 0.363 (0.543) (0.834) (0.834) (0.749) (0.766) (0.678) (0.696) Unemployment benefits (level) 0.185 − 0.599 − 0.726 − 0.447 − 0.697 − 0.511 − 0.751 (0.664) (1.046) (1.046) (0.939) (0.961) (0.851) (0.873) ∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗ Unemployment benefits (degression) 1.110 1.493 1.662 1.351 1.510 0.850 1.043 (0.581) (0.725) (0.725) (0.651) (0.667) (0.590) (0.606) ∗ ∗ ∗ ∗∗ ∗ ∗∗ ∗∗ Employment protection (regular) −0.273 −0.361 −0.376 −0.384 −0.359 −0.401 −0.385 (0.166) (0.209) (0.209) (0.188) (0.192) (0.170) (0.175) ∗ ∗ ∗ ∗ ∗ ∗ Employment protection (temporary) 0.152 0.240 0.252 0.193 0.217 0.191 0.217 (0.089) (0.138) (0.138) (0.124) (0.127) (0.112) (0.115) 2 c R ( ) 0.674 0.526 0.567 0.573 0.601 0.599 0.628 Observations 25 27 27 27 27 27 27 δ (log-risk of entering unemployment) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗ ∗∗ Numeracy (std) −1.237 −1.294 −1.481 −1.254 −1.570 −0.789 −1.069 (0.287) (0.400) (0.436) (0.409) (0.463) (0.446) (0.481) ∗∗ ∗ Logarithmized GDP per capita (PPP) −0.785 −0.718 − 0.736 − 0.488 − 0.514 − 0.352 − 0.349 (0.389) (0.426) (0.463) (0.434) (0.492) (0.474) (0.511) Employment in public sector (share) − 0.804 − 1.228 − 1.119 − 1.384 − 1.059 − 0.764 − 0.488 (1.097) (1.474) (1.604) (1.505) (1.705) (1.643) (1.770) ICT in the workplace (std) 0.322 0.553 0.877 0.691 1.270 0.399 0.833 (0.456) (0.617) (0.671) (0.630) (0.713) (0.688) (0.741) Minimum relative to median wage − 1.041 − 0.385 − 0.433 0.079 0.096 0.785 0.831 (0.937) (1.278) (1.390) (1.305) (1.478) (1.424) (1.535) ∗∗ ∗∗ ∗∗ ∗∗ Trade union density 0.078 1.207 1.327 1.371 1.429 0.895 0.948 (0.440) (0.596) (0.648) (0.608) (0.689) (0.664) (0.716) Unemployment benefits (level) 0.843 0.163 0.022 − 0.272 − 0.570 − 0.418 − 0.615 (0.538) (0.747) (0.813) (0.763) (0.864) (0.833) (0.898) Unemployment benefits (degression) − 0.756 − 0.090 − 0.027 0.089 0.375 0.552 0.769 (0.471) (0.518) (0.564) (0.529) (0.599) (0.578) (0.622) ∗∗ ∗ ∗ ∗∗ ∗∗ Employment protection (regular) − 0.016 −0.298 −0.268 −0.274 − 0.243 −0.378 −0.375 (0.134) (0.150) (0.163) (0.153) (0.173) (0.167) (0.180) ∗ ∗ Employment protection (temporary) 0.062 0.100 0.112 0.175 0.215 0.144 0.175 (0.072) (0.098) (0.107) (0.101) (0.114) (0.110) (0.118) 2 c 0.662 0.603 0.580 0.536 0.543 0.442 0.475 R ( δ ) 0 A cross‑country study of skills and unemployment flows Page 23 of 30 9 Table 6 (continued) Monthly Out of steady state unemployment Contemporaneous 2011–2017 2001–2019 PIAAC OECD ILO OECD ILO OECD ILO (log-risk of exiting unemployment) Observations 25 27 27 27 27 27 27 − δ (log-risk-ratio of exiting to entering unemployment) i i ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ ∗∗ Numeracy (std) 2.258 1.630 1.609 1.437 1.415 0.595 0.593 (0.399) (0.362) (0.357) (0.373) (0.376) (0.289) (0.292) ∗∗ ∗ Logarithmized GDP per capita (PPP) 1.331 0.559 0.737 0.169 0.216 0.242 0.277 (0.540) (0.385) (0.379) (0.396) (0.400) (0.308) (0.310) ∗∗ ∗∗ ∗∗ ∗∗∗ ∗∗ Employment in public sector (share) −3.211 −3.068 −2.850 − 2.021 − 1.865 −2.853 −2.746 (1.524) (1.335) (1.314) (1.373) (1.386) (1.066) (1.074) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ICT in the workplace (std) 0.782 2.127 2.043 1.962 1.808 1.917 1.863 (0.634) (0.559) (0.550) (0.575) (0.580) (0.446) (0.449) ∗∗ ∗∗∗ ∗∗∗ ∗ ∗ Minimum relative to median wage 2.818 3.285 3.173 2.289 2.206 0.672 0.670 (1.302) (1.157) (1.139) (1.191) (1.202) (0.924) (0.931) ∗∗ ∗∗ ∗∗∗ ∗∗∗ Trade union density − 0.038 −1.092 −1.193 −1.543 −1.499 − 0.595 − 0.585 (0.612) (0.540) (0.531) (0.555) (0.560) (0.431) (0.434) Unemployment benefits (level) − 0.658 − 0.762 − 0.748 − 0.176 − 0.127 − 0.093 − 0.136 (0.748) (0.677) (0.666) (0.696) (0.703) (0.540) (0.544) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ Unemployment benefits (degression) 1.866 1.583 1.690 1.262 1.135 0.298 0.273 (0.654) (0.469) (0.462) (0.483) (0.487) (0.375) (0.377) Employment protection (regular) − 0.258 − 0.063 − 0.108 − 0.111 − 0.116 − 0.023 − 0.011 (0.187) (0.135) (0.133) (0.139) (0.141) (0.108) (0.109) Employment protection (temporary) 0.091 0.140 0.140 0.018 0.002 0.047 0.042 (0.100) (0.089) (0.088) (0.092) (0.093) (0.071) (0.072) 2 c c 0.810 0.831 0.836 0.821 0.817 0.892 0.890 R ( − δ ) 0 0 Observations 25 27 27 27 27 27 27 Sample restricted to survey participants ages 25–54 and excluding survey participants from Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey. Fixed effects by round of data collection and minimum-wage regulation not displayed. Sampling weights employed in all calculations. Standard errors in parentheses. ∗ ∗∗ ∗∗∗ Statistical significance at the 10, 5, and 1 percent level denoted by , , and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC), the Penn World Table 9.1 (Feenstra et al. 2015), OECD statistics (https:// stats. oecd. org/) and ILO statistics (https:// ilost at. ilo. org/ data/) estimates. Numeracy skills continue to be a key determi- unemployment remains statistically significant at the nant of the international differences in the risk ratio of five-percent level. exiting to entering unemployment. In Table 7, I separately estimate the effects on the Another concern is the limited time span of the risks of exiting and entering unemployment of several PIAAC data. The obtained effects may exclusively additional covariates. In specification (1), a one-stand- reflect the situation in the survey year. Therefore, I ard-deviation increase in literacy skills is associated also reestimate specification (5) of Table 1 with the with an increase in the exit rate from unemployment averages of the OECD log-risks and of the ILO log- and with a decrease in the entry rate into unemploy- risks in 2011–2017, respectively. The fourth and fifth ment by a factor of 2.4 (exp(0.869)) and by a factor of specification in Table 6 display the estimates. The 0.4 (exp(−1.046)) , respectively. Hence, the risk ratio main conclusions are unaltered. In the sixth and in of exiting to entering unemployment rises by a factor the seventh specification of Table 6, I use the long- of 6.8 (exp(0.869 + 1.046)) . All in all, literacy skills and run averages of the OECD log-risks and of the ILO numeracy skills have a similar impact on the risk ratio log-risks in 2001–2019, respectively. The effect of of exiting to entering unemployment, i.e., the esti- numeracy skills on the risk ratio of exiting to entering mated effect does not crucially depend on the specific 9 Page 24 of 30 D. Stijepic Table 7 Country-level estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment in a seemingly unrelated regressions setup à la Zellner (1962) (1) (2) (3) (4) (5) (6) (7) (log-risk of exiting unemployment) ∗∗∗ Literacy (std) 0.869 – – – – – – (0.273) Education (years) – 0.069 – – – – – (0.068) ∗∗∗ Social trust (std) – – 0.490 – – – – (0.165) Experience (decades) – – – − 0.075 – – – (0.381) Female – – – – 1.678 – – (2.708) ∗∗ Government effectiveness (std) – – – – – 0.250 – (0.108) ∗∗ Logarithmized lagged GDP per capita (PPP) – – – – – – 0.324 (0.156) 2 c 0.308 0.105 0.286 0.076 0.086 0.216 0.191 R ( ) Observations 30 30 30 30 30 30 30 δ (log-risk of entering unemployment) ∗∗∗ Literacy (std) −1.046 – – – – – – (0.316) Education (years) – − 0.099 – – – – – (0.079) Social trust (std) – – − 0.339 – – – – (0.210) Experience (decades) – – – 0.417 – – – (0.439) Female – – – – 0.508 – – (3.182) ∗∗ Government effectiveness (std) – – – – – −0.302 – (0.125) Logarithmized lagged GDP per capita (PPP) – – – – – – − 0.146 (0.193) 2 c 0.311 0.107 0.135 0.087 0.061 0.214 0.077 R ( δ ) Observations 30 30 30 30 30 30 30 − δ (log-risk-ratio of exiting to entering unemployment) i i ∗∗∗ Literacy (std) 1.915 – – – – – – (0.389) Education (years) – 0.167 – – – – – (0.110) ∗∗∗ Social trust (std) – – 0.829 – – – – (0.271) Experience (decades) – – – − 0.492 – – – (0.624) Female – – – – 1.169 – – (4.502) ∗∗∗ Government effectiveness (std) – – – – – 0.552 – (0.165) Logarithmized lagged GDP per capita (PPP) – – – – – – 0.470 (0.263) c c 0.495 0.153 0.305 0.106 0.089 0.336 0.175 R ( − δ ) 0 0 Observations 30 30 30 30 30 30 30 A cross‑country study of skills and unemployment flows Page 25 of 30 9 Table 7 (continued) Sample restricted to survey participants ages 25–54 and excluding survey participants from Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey. Fixed effects by round of data collection not displayed. Sampling weights employed in all calculations. Standard errors in parentheses. Statistical significance at the 10, 5, ∗ ∗∗ ∗∗∗ and 1% level denoted by , , and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC), the Penn World Table 9.1 (Feenstra et al. 2015) and World Bank statistics (http:// info. world bank. org/ gover nance/ WGI/) domain in which the cognitive skills are assessed. In unemployment conditional on different sets of covariates. contrast, specification (2) suggests that formal educa- The effect of numeracy skills on the risk ratio of exiting to tion has only limiting explanatory power. The point entering unemployment is statistically significant at the one- estimates associated with years of education are not percent level in all specifications. statistically significant at conventional levels. I note Notably, education is estimated to have a positive and sig- that the between-country component of the vari- nificant impact on the risk ratio of exiting to entering unem - ance in years of schooling and in the numeracy score ployment at the individual level. Furthermore, education account for 10.0% and for 4.4% of the respective total has a substantially larger impact on the exit rate than on the variance. entry rate. In line with these estimates, Mincer (1991) states Social trust is the sum of the values that a person assigns that “the reduction of the incidence of unemployment is to the statements “There are only a few people you can trust found to be far more important than the reduced duration completely” and “If you are not careful, other people will of unemployment in creating the educational differentials in take advantage of you,” where the answer categories range unemployment rates” in the Panel Study of Income Dynamics. from “1–Strongly agree” to “5–Strongly disagree.” Notably, Why is education closely related to the risk ratio of exiting to numerous studies document that the level of trust explains entering unemployment at the individual level but not at the international differences in aggregate outcomes such as country level? Important factors that relate to the formation economic growth and institutions (e.g., Knack and Keefer of skills include country differences in the quality of school 1997; Zak and Knack 2001). In specification (3) of Table 7, ing or country differences in the preschool system. However, social trust is estimated to have a positive and significant some aspects of formal education that are not related to impact on the exit rate from unemployment, suggesting the formation of skills have predominantly an impact at the that an unemployed person has better job opportunities in individual level but not necessarily at the aggregate level. For a trustful environment. Government effectiveness from the instance, insofar as signaling and rationing take place within Worldwide Governance Indicators (WGI) captures percep - countries but not between countries, the two mechanisms tions of the quality of public services, the quality of the civil provide a rationale for the discrepancies between the indi- service and the degree of its independence from political vidual-level and the country-level effects of education. pressures, the quality of policy formulation and implemen- tation, and the credibility of the government’s commitment to such policies. In specification (6) of Table 7, government The measure of the readiness to learn is based on the values that a person effectiveness is estimated to be associated with a higher exit assigns to the statements “When I hear or read about new ideas, I try to relate them to real life situations to which they might apply”, “I like learning rate from unemployment and with a lower entry rate into new things”, “When I come across something new, I try to relate it to what unemployment. I already know”, “I like to get to the bottom of difficult things”, “ I like to In Table 8, I estimate the country-level effects of numer - figure out how different ideas fit together” and “If I don’t understand some- thing, I look for additional information to make it clearer”, where the five acy skills on the risks of exiting and of entering unemploy- answer categories range from “Not at all” to “To a very high extent.” The ment conditional on different sets of covariates. The effect of scale for the readiness to learn is constructed using item response theory. numeracy skills on the risk ratio of exiting to entering unem- I standardize the learning measure to obtain a mean of zero and a standard deviation of one in the pooled international sample. I distinguish three cat- ployment is statistically significant at the one-percent level in egories of parental education: neither parent has attained upper secondary all specifications. In Table 9, I estimate the individual-level education (low), at least one parent has attained upper secondary education effects of numeracy skills on the risks of exiting and entering (medium), and at least one parent has attained tertiary education (high). Stijepic (2020a) documents that the employment effect of education con - ditional on numeracy skills tends to be more pronounced in countries with higher unemployment. A possible interpretation is that education as a ration- ing device for jobs, in the meaning of Collins (1979), is particularly important if jobs are scarce or, in other words, if the unemployment rate is high. Fur- thermore, Stijepic (2020a) finds that the average numeracy skills among indi - viduals ages 16–19 have a significant impact on the unemployment margin of education conditional on numeracy skills in a country. A possible interpreta- tion is that education as a signaling device for productive capacities, in the meaning of Spence (1973), is particularly important if the quality of education Accessed at http:// info. world bank. org/ gover nance/ WGI/ (WGI project) on is high, or, in other words, if education is a good indicator of skills. December 20, 2020. 9 Page 26 of 30 D. Stijepic Table 8 Country-level estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment in a seemingly unrelated regressions setup à la Zellner (1962) (1) (2) (3) (4) (5) (6) (7) (log-risk of exiting unemployment) ∗ ∗∗ ∗ Numeracy (std) 0.375 0.566 0.701 0.493 0.564 0.526 0.371 (0.374) (0.333) (0.284) (0.383) (0.368) (0.307) (0.265) Logarithmized GDP per capita (PPP) – – – – 0.465 – − 0.438 (0.394) (0.388) Employment in public sector (share) – – – – −1.639 – −1.422 (0.980) (0.906) ∗∗ ICT in the workplace (std) – – – – − 0.071 – −0.753 (0.486) (0.384) Minimum relative to median wage – – – – – − 0.547 − 0.017 (0.966) (0.823) Unemployment benefits (level) – – – – – 0.144 0.167 (0.502) (0.437) ∗∗ Unemployment benefits (degression) – – – – – 0.592 0.841 (0.399) (0.354) ∗ ∗ Social trust (std) 0.352 – – 0.267 0.419 0.045 0.197 (0.213) (0.232) (0.244) (0.222) (0.190) ∗∗∗ Government effectiveness (std) – 0.151 – − 0.068 − 0.201 0.336 0.809 (0.118) (0.168) (0.236) (0.229) (0.242) ∗ ∗ ∗ Logarithmized lagged GDP per capita (PPP) – – 0.274 0.269 0.065 0.302 0.359 (0.144) (0.201) (0.231) (0.162) (0.198) 2 c 0.309 0.285 0.328 0.356 0.430 0.627 0.743 R ( ) Observations 30 30 30 30 30 28 28 δ (log-risk of entering unemployment) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) −1.394 −1.022 −1.176 −1.257 −1.432 −1.513 −1.544 (0.413) (0.360) (0.319) (0.421) (0.380) (0.429) (0.381) Logarithmized GDP per capita (PPP) – – – – − 0.440 – −1.063 (0.408) (0.558) Employment in public sector (share) – – – – 0.626 – 2.048 (1.014) (1.301) ∗∗ ∗∗ ICT in the workplace (std) – – – – −1.106 – −1.171 (0.503) (0.552) ∗∗ ∗∗ Minimum relative to median wage – – – – – −2.693 −2.713 (1.350) (1.183) Unemployment benefits (level) – – – – – 0.921 0.584 (0.702) (0.628) Unemployment benefits (degression) – – – – – − 0.878 − 0.415 (0.558) (0.509) Social trust (std) 0.176 – – 0.347 0.383 0.099 0.170 (0.235) (0.255) (0.252) (0.311) (0.274) Government effectiveness (std) – − 0.123 – − 0.252 0.267 − 0.145 0.454 (0.128) (0.184) (0.244) (0.320) (0.347) Logarithmized lagged GDP per capita (PPP) – – − 0.063 0.077 0.220 − 0.007 0.500 (0.162) (0.221) (0.239) (0.226) (0.284) 2 c 0.373 0.380 0.364 0.420 0.545 0.499 0.635 R ( δ ) Observations 30 30 30 30 30 28 28 − δ (log-risk-ratio of exiting to entering unemployment) i i ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 1.770 1.588 1.877 1.751 1.996 2.039 1.915 A cross‑country study of skills and unemployment flows Page 27 of 30 9 Table 8 (continued) (1) (2) (3) (4) (5) (6) (7) (0.535) (0.450) (0.395) (0.540) (0.472) (0.450) (0.424) Logarithmized GDP per capita (PPP) – – – – 0.905 – 0.626 (0.505) (0.621) ∗ ∗∗ Employment in public sector (share) – – – – −2.265 – −3.470 (1.258) (1.448) ICT in the workplace (std) – – – – 1.034 – 0.418 (0.624) (0.614) ∗∗ Minimum relative to median wage – – – – – 2.146 2.696 (1.418) (1.316) Unemployment benefits (level) – – – – – − 0.777 − 0.417 (0.737) (0.699) ∗∗ ∗∗ Unemployment benefits (degression) – – – – – 1.470 1.256 (0.586) (0.567) Social trust (std) 0.175 – – − 0.080 0.036 − 0.054 0.027 (0.305) (0.326) (0.313) (0.326) (0.304) Government effectiveness (std) – 0.273 – 0.184 − 0.467 0.480 0.355 (0.159) (0.236) (0.302) (0.336) (0.387) Logarithmized lagged GDP per capita (PPP) – – 0.337 0.192 − 0.154 0.309 − 0.141 (0.201) (0.283) (0.297) (0.238) (0.316) 2 c c 0.491 0.531 0.529 0.539 0.662 0.730 0.779 R ( − δ ) 0 0 Observations 30 30 30 30 30 28 28 Sample restricted to survey participants ages 25–54 and excluding survey participants from Indonesia, Ecuador, Peru, Mexico, Kazakhstan, Chile and Turkey. Fixed effects by round of data collection not displayed. Set of covariates in specifications (6), (7) additionally includes an indicator variable for countries without minimum- ∗ ∗∗ wage regulations. Sampling weights employed in all calculations. Standard errors in parentheses. Statistical significance at the 10, 5, and 1% level denoted by , , ∗∗∗ and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC), the Penn World Table 9.1 (Feenstra et al. 2015), OECD statistics (https:// stats. oecd. org/) and World Bank statistics (http:// info. world bank. org/ gover nance/ WGI/) Social trust is estimated to raise the risk ratio of employed person may enjoy stable employment by exiting to entering unemployment at the individual forming closer relationships with colleagues. level. The increase in the risk ratio is almost exclu- All the three human-capital measures, i.e., numeracy sively explained by the reduction in the risk of enter- skills, education and social trust, seem to play an impor- ing unemployment. In contrast, social trust tends to tant role in determining the risk ratio of exiting to enter- be predominantly associated with a higher exit rate ing unemployment at the individual level. In contrast, from unemployment at the country level. The dis- numeracy skills are the dominant factor at the country crepancies between the country-level and the indi- level. Notably, years spent in education, numeracy skills vidual-level effects of social trust potentially reflect and social trust all sizably reduce the risk of entering the different channels through which social trust unemployment at the individual level. However, only affects economic outcomes: An unemployed person’s numeracy skills have a sizable and statistically significant job prospects may depend on the trust of other peo- impact on the exit rate from unemployment. ple, in particular, the trust of employers. A trusting 9 Page 28 of 30 D. Stijepic Table 9 Individual-level maximum-likelihood estimates of the effects of the displayed variables on the risks of exiting unemployment and of entering unemployment. Sample restricted to survey participants ages 25–54. Fixed effects by country not displayed (1) (2) (3) (4) (5) (6) (7) (log-risk of exiting unemployment) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 0.174 0.168 0.165 0.162 0.154 0.163 0.125 (0.029) (0.030) (0.027) (0.029) (0.028) (0.032) (0.032) Social trust (std) 0.024 – – 0.024 – – 0.021 (0.020) (0.019) (0.022) Readiness to learn (std) – 0.021 – 0.019 – – 0.021 (0.022) (0.023) (0.024) Education (years) – – 0.007 0.006 – – 0.011 (0.009) (0.009) (0.010) Experience (decades) − 0.009 0.011 0.017 0.006 − 0.011 − 0.001 − 0.012 (0.080) (0.081) (0.083) (0.081) (0.082) (0.082) (0.083) − 0.011 − 0.016 − 0.016 − 0.012 − 0.009 − 0.017 − 0.008 Experience (decades) (0.020) (0.020) (0.020) (0.020) (0.021) (0.021) (0.022) ∗ ∗ Medium parental education – – – – 0.094 – 0.093 (0.053) (0.055) High parental education – – – – 0.040 – 0.039 (0.070) (0.073) ∗∗∗ ∗∗∗ Female – – – – – −0.275 −0.272 (0.044) (0.047) Native – – – – – 0.045 0.041 (0.074) (0.073) Log-likelihood −29,453 −30,266 −29,952 −29,170 −28,237 −30,166 −27,113 Countries 36 37 37 36 37 37 36 Observations 113,725 115,998 115,885 113,612 110,553 115,934 108,279 δ (log-risk of entering unemployment) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) −0.307 −0.335 −0.195 −0.189 −0.341 −0.316 −0.179 (0.041) (0.043) (0.046) (0.046) (0.042) (0.044) (0.051) ∗∗∗ ∗∗∗ ∗∗∗ Social trust (std) −0.172 – – −0.137 – – −0.130 (0.032) (0.034) (0.037) ∗∗ ∗ Readiness to learn (std) – 0.010 – 0.052 – – 0.049 (0.025) (0.025) (0.027) ∗∗∗ ∗∗∗ ∗∗∗ Education (years) – – −0.118 −0.115 – – −0.118 (0.013) (0.014) (0.014) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Experience (decades) −0.494 −0.456 −0.598 −0.617 −0.517 −0.463 −0.659 (0.104) (0.104) (0.106) (0.105) (0.109) (0.101) (0.102) ∗∗∗ ∗∗ ∗∗ ∗∗∗ ∗∗∗ ∗∗ ∗∗∗ 0.070 0.062 0.063 0.069 0.080 0.062 0.085 Experience (decades) (0.026) (0.025) (0.025) (0.025) (0.026) (0.025) (0.025) ∗∗ Medium parental education – – – – 0.055 – 0.141 (0.066) (0.071) High parental education – – – – − 0.038 – 0.120 (0.065) (0.063) Female – – – – – −0.131 − 0.045 (0.069) (0.073) ∗∗ ∗∗∗ Native – – – – – −0.251 −0.326 (0.101) (0.103) Log-likelihood −29,453 −30,266 −29,952 −29,170 −28,237 −30,166 −27,113 Countries 36 37 37 36 37 37 36 Observations 113,725 115,998 115,885 113,612 110,553 115,934 108,279 A cross‑country study of skills and unemployment flows Page 29 of 30 9 Table 9 (continued) (1) (2) (3) (4) (5) (6) (7) − δ (log-risk-ratio of exiting to entering unemployment) i i ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Numeracy (std) 0.480 0.503 0.360 0.351 0.495 0.479 0.305 (0.038) (0.039) (0.042) (0.042) (0.039) (0.037) (0.042) ∗∗∗ ∗∗∗ ∗∗∗ Social trust (std) 0.196 – – 0.161 – – 0.151 (0.019) (0.021) (0.021) Readiness to learn (std) – 0.011 – − 0.033 – – − 0.028 (0.026) (0.026) (0.027) ∗∗∗ ∗∗∗ ∗∗∗ Education (years) – – 0.125 0.122 – – 0.130 (0.014) (0.014) (0.014) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Experience (decades) 0.485 0.467 0.615 0.623 0.506 0.461 0.648 (0.072) (0.070) (0.065) (0.068) (0.074) (0.070) (0.068) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ −0.081 −0.078 −0.079 −0.082 −0.089 −0.079 −0.093 Experience (decades) (0.017) (0.016) (0.016) (0.016) (0.017) (0.017) (0.017) Medium parental education – – – – 0.039 – − 0.048 (0.048) (0.049) ∗ ∗ High parental education – – – – 0.078 – −0.081 (0.046) (0.047) ∗∗ ∗∗∗ Female – – – – – −0.144 −0.227 (0.064) (0.064) ∗∗∗ ∗∗∗ Native – – – – – 0.295 0.367 (0.066) (0.063) Log-likelihood −29,453 −30,266 −29,952 −29,170 −28,237 −30,166 −27,113 Countries 36 37 37 36 37 37 36 Observations 113,725 115,998 115,885 113,612 110,553 115,934 108,279 Sampling weights employed in all calculations, giving the same weight to each country. Robust standard errors in parentheses, adjusted for clustering at the country ∗ ∗∗ ∗∗∗ level. Statistical significance at the 10, 5, and 1% level denoted by , , and , respectively. Author’s calculations based on the Survey of Adult Skills (PIAAC) Acknowlegements Arpaia, A., Mourre, G.: Institutions and performance in European Labour I thank Martin Biewen, Denis Stijepic, Francis Teal, Klaus Wälde, two anony- markets: Taking a fresh look at evidence. Journal of Economic Surveys mous referees, and the participants of the Chair in Macroeconomics Research 26(1), 1–41 (2012) Day at the Johannes Gutenberg University (Mainz, 2016) and the Annual Con- Becker, G.S.: Human capital: A theoretical and empirical analysis with special ference of the European Association of Labour Economists (St. Gallen, 2017) reference to education. Columbia University Press, Columbia (1964) for helpful comments. I gratefully acknowledge the financial support provided Belot, M., van Ours, J.C.: Does the recent success of some OECD countries in by the Fritz Thyssen Foundation under the Grant No. 40.16.0.028WW and by lowering their unemployment rates lie in the clever design of their labor the German Research Foundation (DFG) under the Grant No. 433336278. The market reforms? Oxford Economic Papers 56(4), 621–642 (2004) usual disclaimer applies. Bentolila, S., Cahuc, P., Dolado, J.J., Barbanchon, T.L.: Two-tier labour markets in the Great Recession: France versus Spain. Economic Journal 122(562), F155–F187 (2012) Authors’ contributions Bertola, G., Rogerson, R.: Institutions and labor reallocation. European Eco- The author read and approved the final manuscript. nomic Review 41(6), 1147–1171 (1997) Bick, A., Fuchs-Schündeln, N., Lagakos, D.: How do hours worked vary with Received: 24 September 2020 Accepted: 4 February 2021 income? Cross-country evidence and implications. American Economic Review 108(1), 170–199 (2018) Blanchard, O., Portugal, P.: What hides behind an unemployment rate: Compar- ing Portuguese and U.S. Labor Markets. American Economic Review 91(1), 187–207 (2001) Blanchard, O.J., Summers, L.H.: Hysteresis and the European Unemployment References Problem. In: Chapters, N.B.E.R. (ed.) NBER Macroeconomics Annual 1986, Acemoglu, D., Johnson, S., Robinson, J.A.: The colonial origins of comparative vol. 1, pp. 15–90. National Bureau of Economic Research Inc (1986) development: An empirical investigation. American Economic Review Boeri, T., Conde-Ruiz, J.I., Galasso, V.: The political economy of flexicurity. Jour - 91(5), 1369–1401 (2001) nal of the European Economic Association 10(4), 684–715 (2012) Acemoglu, D., Johnson, S., Robinson, J.A.: Institutions as a Fundamental Bowman, M.J.: The human investment revolution in economic thought. Sociol- Cause of Long-Run Growth. In: Aghion, P., Durlauf, S. (eds.) Handbook of ogy of Education 39(2), 111–137 (1966) Economic Growth. Vol 1 of Handbook of Economic Growth, pp. 385–472. Collins, R.: The credential society: An historical sociology of education and Elsevier, New YorK (2005) stratification. Academic Press, New York (1979) 9 Page 30 of 30 D. Stijepic Edin, P.-A., Gustavsson, M.: Time out of work and skill depreciation. ILR Review Medina, L., Schneider, F.: Shadow Economies Around the World: What Did We 61(2), 163–180 (2008) Learn Over the Last 20 Years? IMF Working Papers 2018/017, International Eliason, M.: Lost jobs, broken marriages. Journal of Population Economics Monetary Fund (2018) 25(4), 1365–1397 (2012) Mincer, J.: Education and Unemployment. NBER Working Papers 3838, National Elsby, M.W.L., Hobijn, B., Sahin, A.: Unemployment dynamics in the OECD. The Bureau of Economic Research, Inc (1991) Review of Economics and Statistics 95(2), 530–548 (2013) Mincer, J.A.: Schooling, Experience, and Earnings. NBER Books, National Bureau Feenstra, R.C., Inklaar, R., Timmer, M.P.: The next generation of the Penn World of Economic Research Inc., London (1974) Table. American Economic Review 105(10), 3150–3182 (2015) Nickell, S., Layard, R.: Labor market institutions and economic performance. Fougère, D., Kramarz, F., Pouget, J.: Youth unemployment and crime in France. In: Ashenfelter, O., Card, D. (Eds), Handbook of Labor Economics Vol 3 of Journal of the European Economic Association 7(5), 909–938 (2009) Handbook of Labor Economics. New York: Elsevier, pp. 3029–3084 Hanushek, E., Woessmann, L.: Do better schools lead to more growth? Cogni- OECD. Technical Report of the Survey of Adult Skills (PIAAC), 2nd edn. OECD tive skills, economic outcomes, and causation. Journal of Economic Publishing, Paris (2016) Growth 17(4), 267–321 (2012) Shimer, R.: Reassessing the Ins and Outs of Unemployment. Review of Eco- Hanushek, E.A., Kimko, D.D.: Schooling, labor-force quality, and the growth of nomic Dynamics 15(2), 127–148 (2012) nations. American Economic Review 90(5), 1184–1208 (2000) Spence, M.: Job Market Signaling. The Quarterly Journal of Economics 87(3), Hanushek, E.A., Schwerdt, G., Wiederhold, S., Woessmann, L.: Returns to skills 355–374 (1973) around the world: Evidence from PIAAC. European Economic Review Stijepic, D.: Small employers, large employers and the skill premium. Econom- 73(3), 103–130 (2015) ics Bulletin 36(1), 381–387 (2016) Hanushek, E.A., Schwerdt, G., Wiederhold, S., Woessmann, L.: Coping with Stijepic, D.: Globalization, worker mobility and wage inequality. Review of change: International differences in the returns to skills. Economics Let - International Economics 25(1), 108–131 (2017) ters 153, 15–19 (2017) Stijepic, D.: The impact of the productivity dispersion across employers on the Hanushek, E.A., Woessmann, L.: The role of cognitive skills in economic devel- labor’s income share. Economics Bulletin 39(1), 73–83 (2019) opment. Journal of Economic Literature 46(3), 607–68 (2008) Stijepic, D.: Employment effects of skills around the world: Evidence from the Hanushek, E.A., Woessmann, L.: The Economics of International Differences PIAAC. International Labour Review 159(3), 307–338 (2020a) in Educational Achievement. Vol. 3 of Handbook of the Economics of Stijepic, D.: Job Mobility and Sorting: Theory and Evidence. Journal of Econom- Education. Elsevier 2, 89–200 (2011) ics and Statistics 240(1), 19–49 (2020b) Heckman, J.J., Kautz, T.: Hard evidence on soft skills. Labour Economics 19(4), Stijepic, D.: Trends and cycles in U.S. job mobility. The Manchester School 451–464 (2012) 89(2), 203–222 (2021) Hidalgo-Cabrillana, A., Kuehn, Z., Lopez-Mayan, C.: Development accounting Sullivan, D., von Wachter, T.: Job displacement and mortality: An analysis using PIAAC data. SERIEs Journal of the Spanish Economic Association using administrative data. The Quarterly Journal of Economics 124(3), 8(4), 373–399 (2017) 1265–1306 (2009) Iversen, J.M.V., Strøm, B.: Skills, employment, and labor market institutions: Zak, P.J., Knack, S.: Trust and Growth. Economic Journal 111(470), 295–321 Evidence from PIAAC. LABOUR 34(3), 277–304 (2020) (2001) Kaitz, H.B.: Analyzing the length of spells of unemployment. Monthly Labor Zellner, A.: An Efficient Method of Estimating Seemingly Unrelated Regres- Review 93(11), 11–20 (1970) sions and Tests for Aggregation Bias. Journal of the American Statistical Knack, S., Keefer, P.: Does social capital have an economic payoff? A cross- Association 57(298), 348–368 (1962) country investigation. The Quarterly Journal of Economics 112(4), 1251–1288 (1997) Publisher’s Note Kuhn, A., Lalive, R., Zweimueller, J.: The Public Health Costs of Job Loss. Journal Springer Nature remains neutral with regard to jurisdictional claims in pub- of Health Economics 28(6), 1099–1115 (2009) lished maps and institutional affiliations. Launov, A., Wälde, K.: The employment effect of reforming a public employ- ment agency. European Economic Review 84(3), 140–164 (2016) Leuven, E., Oosterbeek, H., van Ophem, H.: Explaining international differences in male skill wage differentials by differences in demand and supply of skill. Economic Journal 114(495), 466–486 (2004)

http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png

Journal for Labour Market Research Springer Journals

http://www.deepdyve.com/lp/springer-journals/a-cross-country-study-of-skills-and-unemployment-flows-BDXZf0tcKf

Loading next page...

Publisher: Springer Journals
ISSN: 1614-3485
eISSN: 2510-5027
DOI: 10.1186/s12651-021-00289-x
Publisher site: See Article on Publisher Site

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A cross-country study of skills and unemployment flows

A cross-country study of skills and unemployment flows

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A cross-country study of skills and unemployment flows

A cross-country study of skills and unemployment flows

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies