The international survey of adult skills, PIAAC, records large differences in numeracy and literacy skills between immigrants and non-immigrants. We examine how these differences relate to the countries’ average skills and skill rankings. Immigrants are defined by country of birth or in terms of languages spoken. For almost all countries, the differences in average skills between non-immigrants and the country’s entire population are significant but small. Regarding skill rankings significant differences are found only for Sweden and these are found to be sensitive to the treatment of individuals that could not conduct the skill tests due to language difficulties.
Theory is of limited guidance when it comes to the relation between immigration and adult skills. Whether a country’s human capital benefits or suffers from immigration depends on many immigrant-specific and host country specific factors, as well as time- specific conditions.
This paper suggests that simple numerical measures are likely to be of limited guidance, too. For example, the large differences in average literacy skills between immigrants and natives, to the natives’ advantage, reported in OECD (2013a) for many of the countries participating in PIAAC, might be taken to be informative about the consequences of immigration for the skills of a country’s entire population. However, our analysis makes it clear that these differences are not sufficient statistics for how countries’ average scores are influenced by immigration. And their connection to the countries’ relative positions in the international skill distribution is even more elusive.
Our analysis also shows that when it comes to assessing changes in country rankings it is very important to account for the statistical uncertainties involved. Moreover, we have demonstrated that standard confidence intervals underestimate the rank uncertainties, compared to an appropriate method for their computation that has been proposed by Leckie and Goldstein (2011).
For almost all countries participating in PIAAC, we find that average numeracy and literacy skills are statistically significantly higher when immigrants are excluded compared to when the country’s entire population is considered. However, the differences are not very substantial. The largest difference, found for Sweden, is 10 score points, from 279 to 289, or 3.5 percent. These numbers are obtained when immigrants are defined as foreign-born individuals. When immigrants are defined as individuals not speaking (any of) the language(s) of assessment the differences become somewhat smaller. Again, however, the Swedish difference is the largest: 7 points or 2.7 percent.
Country rankings based on non-immigrants only are shown to be very similar to the country rankings based on the countries’ entire populations. With respect to numeracy skills, only one country – Sweden – obtains a significantly better ranking when the immigrants are left out. This holds under both of the immigrant definitions considered. In each case Sweden’s rank confidence interval is changed from rank [3,6] to a rank interval including only the top rank. However, a sensitivity analysis shows that the Swedish results are partly dependent upon the fact that Sweden, unlike almost all other countries, assigned imputed (low) scores to all individuals that did not take the skill assessments tests due to language difficulties. Under the alternative – frequently employed by most of the participating countries – of simply disregarding these individuals in the computation of average scores, Sweden’s significant rank differences between non-immigrants and the entire population are either markedly reduced or vanish altogether, depending on the definition of immigrants employed.
With respect to literacy skills the country rankings based on non-immigrants only are even more similar to the corresponding rankings based on the countries’ entire populations than for numeracy skills. France is the only country obtaining a better ranking when the countries’ non-immigrants are considered, instead of their entire populations. Moreover, the improvement is marginal: the rank confidence interval changes from rank  to rank [17,18] and occurs only under the immigrant definition based on country of birth (D2). These results are not sensitive to the treatment of language-related non-response in Sweden.
A one-sentence conclusion from our results is the following: If you are not satisfied with your country’s performance in PIAAC, don’t blame the immigrants!
With respect to future waves of PIAAC, there are two lessons from this paper. First, when the results are presented country rank uncertainties should be computed according to the method suggested by Leckie and Goldstein, as in this paper. Given the paramount interest in country skill rankings and our finding that the standard confidence intervals employed in OECD (2013a) underestimate the uncertainty in the rankings, this would be an important improvement. Secondly, efforts should be taken to streamline the treatment of language-related non-response across countries. This would increase comparability across countries with respects to results as well as with respect to resources devoted to the survey.
Chosen excerpts by Job Market Monitor. Read the whole story at Relations between immigration and adult skills: findings based on PIIAC – IFAU