Why are cognitive test scores of Spanish adults so low? The role of schooling and socioeconomic background

ABSTRACT

We explore the cognitive skill gap between the adult population in Spain and in the rest of European Union countries using the Program for the International Assessment of Adult Competencies. We find that differences in schooling account for about a third of the average difference in cognitive test scores, whereas differences in socioeconomic background explain about one fourth of the average score gap. While cognitive skill gaps are increasing across the distribution of test scores, differences in educational stocks and socioeconomic factors explain a larger fraction of the gap at the bottom than at the top of the skill distribution.

KEYWORDS:

• Human capital
• cognitive skills
• PIAAC
• Oaxaca–Blinder decomposition

JEL CODES:

• J24
• J10

Acknowledgement

We would like to thank Olympia Bover, Aitor Lacuesta, Enrique Moral-Benito, Roberto Ramos, Ernesto Villanueva, the editor, two anonymous referees and seminar participants at the Bank of Spain for helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Because human capital is often regarded as a latent, possibly multidimensional, object, a central issue in the literature concerns its definition and measurement. Internationally comparable standardized examinations are usually thought to provide a better proxy of human capital than quantitative measures of schooling (e.g., years of schooling or enrolment rates). Compared to the latter, international test scores capture skills acquired outside school, embed not only the quantity but also the quality of the education system, and allow exploiting cross-country variation in skills at each level of education (Hanushek and Woessmann 2008; Hanushek et al. 2015).

2 The theoretical growth literature suggests that higher human capital levels can foster economic growth by increasing the productivity of the labor force (Mankiw, Romer, and Weil 1992), by facilitating the development of new technologies (Romer 1990) or the diffusion of existing ones (Benhabib and Spiegel 2005).

3 Hanushek and Woessmann (2011) and Woessmann (2016) provide extensive reviews of this literature.

4 A third, optional, domain aimed to assess problem-solving skills, but, as some countries (i.e., Cyprus, France, Italy and Spain) did not administer the corresponding module, we do not consider it in the following analysis.

5 On average, about 78% of respondents took the computer-based test and 22% the paper-based test. A field test, conducted prior to the data collection, suggests that the mode of assessment had no impact on respondent's performance on the test. Moreover, after controlling for several socioeconomic characteristics, there is no evidence that the test scores of respondents who took the paper-based assessment differ systematically from those of respondents who took the computer-based assessment (OECD 2013).

6 In PIAAC, skills are a latent variable that is estimated using item-response-theory models (see OECD 2013 for details). PIAAC provides 10 plausible values, instead of only one individual score, for each respondent and each skill domain. Throughout our empirical analysis, we use estimation techniques using the 10 plausible values in order to get unbiased estimates of the statistics of interest.

7 Schwerdt, Wiederhold, and Murray (2020) replicate previous analysis on IALS data by Coulombe, Tremblay, and Marchand (2004) and Coulombe and Tremblay (2006) using an extended sample of 33 countries which took the PIAAC assessment either in the first (August 2011-March 2012) or in the second (April 2014-May 2015) round. In contrast, we only use PIAAC data for the 24 countries assessed in the first round. In Appendix A we show that higher average literacy and numeracy scores are strongly associated with higher real GDP growth per capita even in the sample of countries used in our paper. The same result holds in the subsample of EU countries.

8 This group includes the following 16 countries: Austria, Belgium, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Ireland, Italy, Netherlands, Poland, Slovak Republic, Sweden, and United Kingdom.

9 In PIAAC, the number of years of schooling is associated with the highest level of education attained. All reported national categories in the achieved level of education are converted into the nominal years of schooling needed to achieve that particular level of education (see OECD 2019b for more details). Cross-country differences in average years of schooling in the PIAAC data are in line with those of other international datasets (see Table B5 in the Appendix). Indeed, the correlation between the cohort-specific average years of schooling in PIAAC and in the well-known Barro-Lee Educational Attainment Dataset (Barro and Lee 2013) is 0.8 for the full sample of countries and 0.83 for the sample of EU countries.

10 Note that, when we condition on the education levels, the average conditional gaps in Table 2 are lower than the average unconditional gaps for all three education groups. Obviously, this is due to the different distribution of educational attainments between Spain and the group of other EU countries (see Table 1).

11 As it is well-known, the decomposition in equation (2) is not unique as one could write an alternative decomposition in which the difference in observed characteristics is evaluated using the estimated coefficient for the EU countries, and Spain is the reference group to assess the contribution of the differences in the ‘returns’ of the observed characteristics. The results of this alternative decomposition are shown in Table C2 in the Appendix and they are very similar to those reported in Table 3 in the paper.

12 For example, Woessmann (2007) suggests that a large fraction of the cross-country variation in the cognitive skills of students can be ascribed to differences in the degree of school autonomy, accountability and competition. Insofar as cognitive skills in adulthood accumulate throughout the life cycle (Cunha et al. 2006; Cunha and Heckman 2007), differences in the quality of the education system can be important factors also in explaining differences in adult skills.

13 Table B1 in Appendix B shows descriptive statistics of these variables for Spain and other EU countries.

14 The inclusion of noncognitive skills within a cognitive skill production function is justified by previous work from Cunha and Heckman (2008) suggesting that noncognitive skills foster the development of cognitive skills, whereas the opposite is not true.

15 See Appendix D for a more detailed discussion about the construction of these measures.

16 See Campbell et al. (1980), Uslaner (2000), Hedengren and Stratmann (2012), Hitt, Trivitt, and Cheng (2016), Cheng, Zamarro, and Orriens (2016), Zamarro et al. (2016), Anghel and Balart (2017) and Grotlüschen (2017).

17 In contrast, in Chernozhukov, Fernández-Val, and Melly (2013) detailed decompositions are path dependent, i.e., they depend on the order in which the decomposition is performed. It is also worth noting that, whereas Chernozhukov, Fernández-Val, and Melly (2013) computes an exact decomposition of the counterfactual effect on the unconditional quantile, the methodology of Firpo, Fortin, and Lemieux (2009) holds only as a first order approximation. See Fortin, Lemieux, and Firpo (2011) for an in-depth discussion of these decomposition methods.

18 For expositional reasons, we group together the contribution of all demographic factors and of the parental education dummies. Table C3 in the Appendix shows the decomposition results for the literacy test.

19 The results for literacy are similar and they are available in the Appendix (see Table C5 and Figures C1, C2 and C3).

20 In order to ease readability, in the results for the Oaxaca-Blinder decompositions presented in this section we aggregate the contributions of the single items within the following three groups: years of schooling, parental education and demographic factors.

21 In this analysis, we dropped years of schooling from the cognitive skill production function because, after conditioning on the achieved education level, there is no variation in schooling years within each education subgroup. Therefore, in this exercise we assess the role of socioeconomic background in explaining the skill gap for each education group.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.