When analyzing academic performance among Swedish pupils at lower-secondary and elementary level, there are typically four standardized procedures to evaluate them: grades, national tests, international tests, and specific tests constructed by researchers (Holmlund et al., 2019; Björklund et al., 2010). Recent research (Boman, 2021a; Boman, 2021b) shows that socioeconomic and sociodemographic variables are substantially associated with grades and national test results at the aggregated school level. Earlier studies such as Myrberg and Rosén (2009), who used structural equation modeling to analyze data from Progress in Reading Literacy Study (PIRLS), inclusive of Swedish students, have found associations between socioeconomic variables and reading achievement.

However, there is a dearth of studies on Swedish students’ Programme for International Student Assessment (PISA) performance, especially more recent surveys (i.e., 2015 and 2018), which is carried though in association with the Organization for Economic Co-operation and Development (OECD). An exception is the book Northern Lights on TIMSS and PISA (see Reimer et al., 2018), but this work does only focus on PISA 2015 from the perspective of all Nordic countries. Thus, it lacks larger cross-national data sets with more participating countries, as well as analyses on 2018. Moreover, it has less emphasis on mathematics achievement compared to science achievement and focuses mostly on socioeconomic indicators and science achievement–science interest indicators. This is because in the 2015 PISA assessment, science was the main subject whereas mathematics was the main subject in 2012. While this relationship makes sense, there is no reason to only focus on the main subject because of the high intercorrelations between the achievement scores in mathematics, reading and science (e.g., r = 0.772 to 0.888 in PISA 2015). In the current study, mathematical literacy will be highlighted because of the importance of mathematics for future educational trajectories, earnings and societal development (e.g., Rindermann, 2018) but the author admits that this selection is nevertheless somewhat arbitrary.

Therefore, in accordance with methodological suggestions underlined by Gustafsson (2007; 2008), as well as Jerrim et al. (2018) and to lesser extent Guglielmi and Brekke (2017), the following article examines Swedish students’ PISA performance in 2015 and 2018. The unique feature is the combination of one cross-sectional data set with Swedish students (n = 5458 in 2015) with two cross-national data sets for 31 countries in 2015 and 2018. This increases both the internal and external validity, as well as the reliability of such a research endeavor.

Five strands of literature will be briefly highlighted below: research on the factors associated with academic achievement at lower secondary level in Sweden, research on Sweden’s PISA performance, literature related to the nexus between PISA achievement and migrant participation, literature on factors associated with PISA performance at the cross-country level, and critical studies on PISA’s methodology. Some studies overlap, such as migration background and socioeconomic status (SES) linked with PISA performance.

Many studies related to Swedish students at Grade 8 and Grade 9 point to substantial relationships between parents’ socioeconomic status and academic achievement, whether such are based on international data like PIRLS, municipality data, or individual date on grades or national tests results are used (e.g., Myrberg and Rosén, 2009; Gustafsson and Yang Hansen, 2018; Boman, 2021a; Boman, 2021b). For example, Gustafsson and Yang Hansen (2018) examined changes in the impact of family education on student educational achievement in Sweden, from 1988 to 2014, and found that correlations increased by 0.04 units between the early 1990s and 2014, partly due to immigration. Boman (2021a) found that the SES among the working population in Swedish municipalities was correlated with school results (aggregated grades) in Grade 9 in 2019. In addition, natives outperformed non-natives and municipalities with a larger share of certified teachers were positively correlated with higher grades. More or less positive SES–academic achievement correlations have been found in both the US (Sirin, 2005), other developed countries (Tan, 2015), and developing countries (Kim et al., 2019).

In relation to teacher effects, there is a rich literature from scholars in the United States which indicates that teachers’ competence (e.g., certification, experience and so forth) has at least a moderate relationship with academic performance (e.g., Hanushek et al., 1996). In Sweden, studies are lacking but Holmlund et al. (2019) stress that the entirety of national and international evidence suggests that teachers’ competence do have a positive relationship with academic performance at the secondary levels. However, findings from correlational and regression analyses might be spurious due to that certified teachers may be more frequent in at least some low-performing schools and municipalities in Sweden (Holmlund et al., 2019). Boman (2021a; 2021b) found positive relationships between teacher certification rates and grades in all subjects at the aggregated municipality level in Sweden. In Boman (2021a), these relationships were more pronounced when high-performing municipalities were compared with low-performing counterparts. However, as both Homlund et al. (2019) and Myrberg and Rosén (2009) underline, family-related factors such as parents’ SES and IQ seem to explain most of the variance in relation to various measures of school achievement. School factors account for only a few percentages of the variability in grades (Holmlund et al., 2019). However, in regard to PISA, Hanushek et al. (2019) make a rather strong case for the impact of highly skilled teachers on students’ PISA performance. The authors underline:

As Wiklund (2018) observes, Sweden’s teachers are often blamed for the poor results in general and among low-SES segments such as second-generation and newly arrived migrants in particular. However, it is far from certain that more qualified or competence teachers – those that would for instance be congruent with how Finland’s teacher pool was constituted around 2011/2012 – do have much of an impact since Swedish teachers’ PIAAC scores are not much after their Finnish counterparts and substantially above the OECD average (Hanushek et al., 2019). PIAAC stands for Programme for the International Assessment of Adult Competencies and might be described as PISA for adults and teachers, typically based on smaller but representative samples, who take tests in numeracy and literacy (Gustafsson, 2016).

Sweden has experienced high rates of inflows of low-skilled migrants the 1990s onwards (Sanandaji, 2020; Ekberg, 1999). Economists such as Sanandaji (2020) and migration scholars like Vogiazides and Mondani (2019) have claimed that the socioeconomic and sociocultural integration of migrants in Sweden has largely failed, especially in the city of Malmö, leading to an increased fiscal burden, high unemployment rates and crime rates, as well as lower school results (Sanandaji, 2020; see also Skolverket, 2009; Skolverket, 2016b). In different European countries, natives do typically outperform immigrant students. Azzaloni et al. (2012) found that in Italy and Spain achievement differences in the PISA 2009 survey were associated with family background and socioeconomic status. In Italy, the effect of academic tracking was also substantial while immigrants from Latin America in Spain had an advantage in PISA reading due to their ability in Spanish. However, socioeconomic, institutional, and linguistic factors cannot fully explain the variance in both models (Azzaloni et al., 2012). Similar results were found for first- and second-generation immigrants in Switzerland when analysing data from the PISA 2000 wave. However, larger family sizes for first-generation immigrants, typically from Albania, had a negative relationship with PISA performance (Meunier, 2011).

Research on PISA achievement at the cross-national level indicates that cognitive abilities and PIAAC scores are highly correlated with PISA performance (e.g., Gustafsson, 2016; Rindermann, 2018; Hanushek et al., 2019). Swedish teachers had high average scores in literacy and numeracy (Hanushek et al., 2019), although substantially lower than Finland and Japan. Raw IQ scores in Sweden are similar to the British mean which is used as a standardized value and is set at 100 (Sonnander and Ramund, 2003).

Moreover, Cheung (2017) has found a significant relationship between resilience and mathematic literacy in the PISA 2012 survey, which implies that substantial shares of socioeconomically disadvantaged East Asian students perform better than home-advantaged students both in these countries and relative to other PISA participants. Specifically, ‘The effect sizes are that an increase of one standard deviation in mathematics self-efficacy is able to decrease the chance that an ESCS advantaged student will be academically slackening by approximately 50–75%’ (Cheung, 2017).

However, these patterns might be inapplicable to culturally distant countries like Sweden. Nonetheless, non-cognitive abilities might be included in for example multiple regression analyses as they are similar to personality traits like conscientiousness and emotional stability, which are associated with higher academic achievement and has a generic if not universal character (e.g., Poropat, 2009).

At this point, a plethora of critical academic works has accentuated the shortcomings of the OECD-related global competition (e.g., Prais, 2004; Sung, 2011; Imsen et al., 2017). Furthermore, PISA scores are, according to some researchers, not a good indicator of educational quality (Jerrim, 2015; Sjøberg, 2015). Moreover, the PISA test constitutes a costly procedure in which the benefits are by no means clear (Sjøberg, 2015). Furthermore, educational policy borrowing can be quite problematic (Steiner-Khamsi, 2014; Steiner-Khamsi, 2014; So, 2014; Sung, 2011; Takayama, 2008; Waldow et al., 2014). It is also questionable if the surveys are comparable over time due to the substantial inclusion/exclusion of different test items (Rutkowski and Rutkowski, 2016). Hence, when using PISA data, a balanced, critical yet grateful stance might be appropriate to take. If the OECD has done something positive it is to provide researchers, politicians, practitioners and laymen with more data pertinent for intranational and cross-national comparisons (Steiner-Khamsi, 2014; Sjøberg, 2015; Hanushek et al., 2019). The current article takes such a balanced approach which implies the usefulness of PISA data. Overall it contributes to the literature on Swedish students’ PISA performance and in such a way that it focuses on two waves at the cross-national level and in one assessment at the individual level.

The following research questions are addressed in the present study: What factors are associated with high achievement in PISA 2015 and 2018 among Swedish students? Are these patterns similar at the cross-country level? What other factors at the cross-country level are associated with differences in PISA math achievement?

This study set out to answer three research questions related to what factors are associated with high achievement in PISA 2015 and 2018 among Swedish students, if these patterns are similar at the cross-country level, and what other factors at the cross-country level are associated with differences in PISA math achievement.

At the within-country and individual level, native Swedish students who took the PISA test in 2015 outperformed first- and second-generation migrants. This is similar to school and municipality level studies on Swedish students such as Boman (2021a) who found that non-natives generally performed worse than natives, as well as for example Meunier (2011) in relation to PISA. Moreover, the latent SES “number of books at home” was associated with higher PISA math achievement. It does also confirm similar research on the relationship between SES and PIRLS achievement (Myrberg and Rosén, 2009; see also; Tan, 2015) and SES in relation to PISA science achievement (Reimer et al., 2018). Some of the ambition indicators (e.g., “I want to be one of the best students in my class”) were associated with a higher PISA math score whereas others were not (e.g., “I want to be the best, no matter what I do”). The latter indicator might be tied to a counterproductive level of overconfidence whereas the former is more balanced and appropriate (e.g., see Cheung, 2017 for a discussion). The same pattern was discerned for the belongingness indicators. The picture was also a bit fuzzy as regards the different indicators of parental support, which in part constitutes a proxy of SES.

At the cross-country level, GDP per capita was associated with higher PISA scores. However, ethnic homogeneity was not substantially associated with higher scores. NIQ scores, PIAAC scores and teacher salaries were also moderately correlated with PISA math scores, which corroborates earlier research such as Rindermann (2007; 2018), Hanushek et al. (2019), and Flores-Mendoza et al. (2021). This implies that a country with high cognitive ability level, as well as cognitively competent and well-paid teachers are likely to perform better in large-scale international assessments such as PISA. Thus, to merely add more school days and/or annual instruction time does not seem to be a solution to lower performance levels, ceteris paribus. When students within the same country are compared at the individual level, it is a “comparative advantage” (see Boman, 2020) to be a native and have parents with a level of higher socioeconomic status, as well as well-developed non-cognitive abilities such as a balanced degree of ambition.

This study has several limitations. Because of the cross-sectional design, the evidence is almost entirely correlational. The same goes for the relatively small number of countries included at the cross-national level (N = 31). For example, the inclusion of two high cognitive ability level and high PISA math achievement level countries such as Taiwan and China and two with lower scores on these two measures (Kazakhstan, The Philippines), leads to a higher correlation between NIQ and PISA math (e.g., r = 0.629 in PISA 2018). Similarly, the PISA math–ethnic homogeneity relationship increases to r = 0.215 when these four PISA 2018 participants are included.

Furthermore, despite the efforts made by Lynn and Becker (2019) to weight and adjust scores from various IQ samples many of these appear somewhat unaligned with the PISA data in a temporal sense, as well as with regard to the mean and median ages within those various samples (i.e., many of the samples include both small children and adults). Nonetheless, the combination of individual data at the intranational level as well as aggregated data at the cross-national level can provide a more comprehensive picture of PISA performance among specific countries such as Sweden. Focusing merely on a single level tends to neglect important factors such as SES and migration (at the intranational level) and cognitive ability and teacher variables (at the cross-national level).

Moreover, some may notice the exclusion of gender variables because it is clearly understood that girls outperform boys in PISA reading and science but typically not in mathematics (e.g., Ma, 2008). However, these differences are smaller than the native/immigrant differences, boys are typically stronger in math and the current study is concerned about math achievement, and these differences are difficult to capture in a meaningful way with aggregated cross-country data (at least with the current research design). Hence, the inclusion of gender at the within-country level but the exclusion at the between-country level would likely lead to even more discrepancies between these two levels of analysis.

As is noticeable in several tables in this article, the explained variance in PISA math is greater at the aggregated cross-country level compared to the individual student and within-country level. This might be related to the lack of appropriate non-cognitive ability variables and items and the total exclusion of cognitive ability variables and items within the frames of the student questionnaire and related student tasks. Hence, the inclusion of two brief tests, one non-cognitive test (e.g., Big Five-10, see Rammstedt and John, 2007) and one cognitive (e.g., the Kaufmann Brief Intelligence Test) may explain the variance to a greater extent than which is currently the case. Some personality researchers have stressed that adolescents and older children may have trouble to understand and asses their own personalities (e.g., see Andersen et al., 2020) to the same extent as university students. However, these few Big Five items seem not more difficult to grasp than the PISA items (c.f., Rammstedt and John, 2007, and the PISA 2015 Student Questionnaire). Furthermore, complete cross-national data on all PISA participants and not just OECD countries would lead to more comprehensive and reliable estimates of the cross-national, aggregated effect sizes (e.g., see OECD, 2018b).

Overall, the current article contributes to research on Swedish students’ PISA performance and replicates relationships that are found at the aggregated within-country level in regard to SES, teacher competence, and migration background (e.g., Boman, 2021a) but adds other variables such as cognitive and non-cognitive abilities which likewise are associated with PISA math achievement. Future research may use a similar theoretical and methodological approach but in relation to other countries or later PISA waves with regard to Sweden.