1 Introduction

Among the most distinctive features of social and economic change in today’s societies are rapid technological progress (‘digitization’), demographic ageing, and a generally increased pace and scope of societal change (e.g., Mills & Blossfeld, 2013; Rosa, 2013; Silbereisen & Chen, 2010). One of the obvious implications of these macrolevel trends is that an ageing population will have to navigate an increasingly complex, knowledge-based, and technology-rich world in which skills that were once needed and useful may quickly become obsolete. In such a world, lifelong education and lifelong learning become the norm (Billett, 2018; Blossfeld et al., 2020; Kilpi-Jakonen et al., 2015).

Are today’s cohorts of adults prepared for this world? Because it is hard to foresee which specific skills will be needed in the future, it is sensible to ask instead whether adults possess the prerequisites to learn and update their skills across the entire lifespan. Two such prerequisites are undoubtedly literacy (or reading competence, i.e., the ability to understand, use, and interpret written text) and numeracy (or mathematical competence, i.e., the ability to access, use, and interpret mathematical information) (OECD, 2012; PIAAC Literacy Expert Group, 2009). Without sufficient levels of these basic competencies, acquiring the types of specific skills and qualifications needed to navigate one’s life in a technology-rich knowledge society is all but impossible. To illustrate, learning a foreign language through a smartphone app, understanding how to operate a CNC milling machine, reading up on the latest regulatory changes in a commodities market, or learning how to use spreadsheet tools for sales forecasting are all skills whose acquisition and execution require dealing with what is often complex symbolic (verbal and numeric) material—that is, they require literacy and numeracy skills. Apart from being prerequisites for lifelong learning, literacy and numeracy are indispensable for labour market participation, social and political participation, or managing one’s health and finances. The robust links these basic competencies have to individual-level outcomes such as income, health, and social participation as well as to macrolevel outcomes such as economic growth testify to their importance (e.g., Hanushek et al., 2015; Lunze & Paasche-Orlow, 2014; OECD, 2016; Pullman et al., 2021).

The high and perhaps growing relevance of literacy and numeracy skills directs attention to how these competencies develop over the lifespan. People acquire literacy and numeracy primarily through schooling. That much is clear. But what happens after people leave education is far less clear. How come, for example, that there is a substantial number of adults even in economically developed countries—between 12.1% and 17.5% of the working-age adults in Germany according to various estimates—who possess only low literacy levels that allow them to read simple words and sentences, but not longer or more complex texts (Durda et al., 2020; Grotlüschen et al., 2020)? Did these adults never acquire sufficient levels of literacy in the first place, or did they lose them at some stage? What are the normative trajectories of literacy and numeracy development over the lifespan? If competencies continue to change in adulthood, does change involve only losses or also gains, and how are gains and losses distributed across different segments of the population? Moreover, what individual and contextual factors drive potential gains and losses in literacy and numeracy—can we pinpoint risk and protective factors? Such questions are interesting research topics in their own right, but they are also relevant for policymakers and practitioners who are interested in promoting literacy and numeracy as well as lifelong learning more broadly (e.g., Wolf & Jenkins, 2014).

This chapter summarizes the key insights from a research project, led by the author of this chapter, that aimed to unravel these questions. This interdisciplinary project brought together psychologists, sociologists, and economists. It leveraged the unique potential to study change over time in adults’ competencies offered by two German large-scale assessment surveys: the German National Educational Panel Study (NEPS; Blossfeld et al., 2011) and the German PIAAC Longitudinal Study (PIAAC-L; GESIS et al., 2017). These surveys’ combination of large samples with a repeated-measures design has so far been a rarity in research on competence development during adulthood.

I shall proceed as follows. First, I shall briefly review the theoretical perspectives that guided the research project. Next, I shall outline the opportunities for studying adults’ competence development offered by the repeated-measures data in NEPS and PIAAC-L while also highlighting methodological challenges and how our project addressed them. I shall then summarize the key findings on the patterns and predictors of literacy and numeracy development from the research project. I shall conclude by discussing open questions and avenues for future research on competence development. Table 4.1 lists the project’s key publications, on which this chapter will dwell.

Table 4.1 Overview of the key publications from the DFG research project ‘Stability and Change in Adult Competencies: Patterns and Predictors of Literacy and Numeracy Development’
Full size table

2 Theoretical Perspectives on Competence Development in Adulthood

The literature on competence development during adulthood is highly diverse (e.g., Desjardins & Warnke, 2012). It has approached competence development from various disciplinary perspectives including psychology, economics, sociology, and adult education. As a result, there is not even consensus on terminology, with different research traditions referring to competencies as ‘skills’ (predominantly in economics), ‘proficiencies’, or ‘(crystallized) intelligence’ (predominantly in psychology). The various disciplines that have studied competencies offer several theoretical perspectives that guided our project.

In psychology, a classic theory is Raymund B. Cattell’s (1971) investment theory, initially proposed in the early 1940s and thus decades before data allowing for stringent tests of its central claims were available. In his theory, Cattell (1971) proposed that ‘intelligence’ is comprised of two distinct aspects: ‘fluid intelligence’ (Gf), the ability to process novel information and stimuli, and ‘crystallized intelligence’ (Gc), the totality of acquired knowledge and skills.Footnote 1 Crucially, Cattell posited that Gf and Gc are subject to different influences and follow different age trajectories: he envisioned Gf to be largely innate (i.e., heritable) and dependent on biological, neural functioning. For this reason, Gf would peak early in life and then slowly decline. Contrariwise, he conceived of Gc as resulting from the investment (hence ‘investment theory’) of Gf in different subject areas and being influenced by education, experience, and culture. Accordingly, he posited that Gc would continue to increase across adulthood and decline only in old age when biological ageing negatively affects overall functioning. Cattell’s theory is an early example of a theory of lifespan development, and it strongly influenced subsequent work (Brown, 2016; Nisbett et al., 2012). For example, it inspired Phil Ackerman’s (1996) distinction between intelligence-as-process and intelligence-as-knowledge in industrial/organizational psychology and Paul Baltes’s (1993) distinction between cognitive mechanics and cognitive pragmatics in research on cognitive ageing.

These concepts do not refer directly to literacy and numeracy as assessed in NEPS and PIAAC-L. However, a glance at the definition and assessment of competencies in NEPS clarifies that these surveys conceptualize literacy and numeracy as acquired functional capacities (Rammstedt, 2012; Weinert et al., 2011). This puts them into the same category as crystallized intelligence, intelligence-as-knowledge, and cognitive pragmatics. In the updated Cattell–Horn–Carrol (CHC) model (McGrew, 2009), one of the most influential structural models of cognitive abilities, the definitions of reading and writing ability (Grw)Footnote 2 and quantitative knowledge (Gq)—broad, acquired abilities at Stratum II—correspond closely to literacy and numeracy skills. At the same time, it is clear that literacy and numeracy require the operation of fluid-type cognitive basic skills such as reasoning and processing. Thus, competencies may be a blend of acquired (Gc-type) and inherited (Gf-type) abilities. This leads us to the hypothesis that the lifespan trajectories of Gf and Gc may be a blend of the ideal-typical trajectories of Gf and Gc hypothesized by Cattell (1971) and Baltes (1993). Figure 4.1 expresses this idea, which indeed receives some support from cross-sectional age profiles of literacy and numeracy presented by Paccagnella (2016) and ourselves (Lechner et al., 2021b, Supplementary Material).

Fig. 4.1
A multiline graph plots performance versus age in years. It has concave down increasing curves for crystallized pragmatics, competencies, and fluid mechanics. The curve for crystallized pragmatics becomes stable and the rest 2 curves finally decline.

Hypothesized age trajectories for different types of cognitive abilities. (Note. The figure shows hypothetical age trajectories of ‘crystallized’ and ‘fluid’ cognitive abilities according to the theories of Cattell (1971) and Baltes (1993). The age trajectory of competencies (literacy and numeracy) appears to be a blend of both, showing a prolonged plateau from early to middle adulthood and a decline slowing only in old age according to cross-sectional data reported in Paccagnella (2016) and also Lechner et al. (2021b))

Full size image

A second theoretical perspective that hails from research on adult education is more specific to the competencies measured in studies such as NEPS and PIAAC-L: this is practice engagement theory (PET; Reder, 1994, 2009) developed by Steve Reder who was an international collaborator in our research project. PET posits that individuals’ literacy proficiencies develop as a by-product of their engagement in everyday reading and writing practices. Reciprocally, literacy proficiencies affect levels of engagement in reading and writing practices. Thus, reminiscent of investment theory’s postulate that growth in Gc results from the investment of Gf in a subject area (Ackerman, 1996; Cattell, 1971), PET assigns a pivotal role for competence development to individual engagement in literary and mathematical practices at work, in the family, or during leisure. At the same time, it highlights that engagement is itself dependent on prior proficiency and a set of structural influences such as race, gender, or socio-economic status (Reder, 2009; Reder et al., 2020; Wicht et al., 2021b). PET would predict that adults can experience both gains and losses in literacy and numeracy skills over time—depending on the extent to which they engage in attendant practices in their job, leisure, or other contexts. This central tenet of PET is compatible with investment theory or Baltes’s research on ageing, but extends it by highlighting individual differences in skill development and identifying engagement as the core driver of these differences.

The third perspective, which has some prominence in life-course sociology, is the cumulative advantage or ‘Matthew effects’ hypothesis, which states that initial advantage (and disadvantage) accentuates over time. This perspective, applied to competence development, would suggest that ‘skills beget skills’ (Cunha & Heckman, 2007): from childhood on, people with higher initial levels of competencies (e.g., intelligence) and resources (e.g., higher socio-economic status) are selected into more cognitively stimulating and demanding environments (e.g., tertiary education, complex white-collar jobs). These environments, in turn, allow them to expand their competencies further. Those at the lower end of the distribution of competencies and resources, by contrast, will lose further ground, leading to a growing competence gap.

Together, these theories suggest several hypotheses about adults’ competence development. First, through being functional capacities that individuals acquire chiefly through education and practice during schooling age, literacy and numeracy should remain fairly stable across adulthood. However, they might remain malleable in principle (‘lifelong plasticity’). Second, according to investment theory and PET, respectively basic cognitive skills (Gf) and practice engagement should rank among the strongest predictors of competence development: individual differences in Gf and the extent to which adults ‘invest’ their fluid intelligence in literacy and mathematics practices will determine whether they can maintain or even expand their competencies in adulthood. Third, sociodemographic and structural factors may be related to competence development through both selection effects (e.g., individuals with a higher SES are more likely to attend tertiary education) and socialization effects (e.g., individuals with tertiary education are more likely to have complex jobs that require reading and math). According to the ‘Matthew effect’, those with higher resources may be at an advantage in competence development. These hypotheses were among those that we hoped to put to a rigorous test in our research project.

3 NEPS and PIAAC-L: Opportunities and Challenges for the Study of Competencies

3.1 Research Designs of Prior Work

Prior research on competencies—summarized in informative reviews by Desjardins and Warnke (2012), Paccagnella (2016), and Nienkemper et al. (2021, focusing on low literacy) and expanded by recent studies (e.g., Kyröläinen & Kuperman, 2021)—offers several important insights into the age profiles and into the precursors and correlates of adults’ competencies. Much prior research is limited, however, by the research designs that fall mainly into one of two categories: large-scale cross-sectional and small-scale longitudinal studies. Cross-sectional studies are often based on large-scale international assessments such as the Programme for the International Assessment of Adult Competencies (PIAAC) or the International Adult Literacy Survey (IALS) conducted by the OECD. Studies based on these surveys have made important contributions to our understanding of the correlates (i.e., potential antecedents and consequences) of consequences, cross-national differences, and trends over time (Paccagnella, 2016). They have also enabled insights into age-related differences in literacy and numeracy skills and into the correlates of individual differences in skills. However, owing to their cross-sectional design, these studies can hardly inform us about the temporal dynamics of competence development in adulthood. These studies confound age effects with cohort (and, if they exist, period) effects, meaning that the alleged age trends they report might be biased. Cross-sectional designs cannot ascertain whether age differences reflect age-related changes or stem from pre-existing cohort differences in competencies (which may already have arisen in childhood or adolescence). They cannot provide indispensable statistics about change and stability. These statistics, for example, the change score or test–retest correlation, require repeated-measures designs to compute. Moreover, cross-sectional designs do not allow for the application of panel models such as fixed-effects (FE) regression to identify (or at least get closer to) causal effects of predictors of competence development.

Small-scale longitudinal studies are advantageous in this regard because they are truly developmental and allow competence development to be traced over time. Examples comprise the longitudinal study of adult learners (LSAL) that focuses on high-school dropouts in the US (Reder, 2009), studies on adults in basic skills programmes in the UK (Wolf & Jenkins, 2014), or studies on cognitive ageing in old age (Baltes, 1993). However, these studies, too, have limitations: because their samples are small and selective, their findings may not generalize to the population as a whole. Moreover, because they represent only specific subgroups of the population, they do not allow the study of differences in competence development across different subgroups of the population. Often, they may lack sample size and statistical power. Moreover, some of these studies did not contain as extensive assessments of competencies as those in the major international large-scale assessments such as PIAAC. Instead, many of them used ad hoc measures or established intelligence test batteries in which isolated verbal (e.g., vocabulary) or numerical abilities (e.g., number series) were only short subtests. Such assessments, though informative in their own right, do not provide the same depth of information on literacy and numeracy as PIAAC-L and NEPS.

3.2 The Unique Data Troves of NEPS and PIAAC-L

To answer the questions about the patterns and predictors of competence development posed at the outset, we need data that fulfil high requirements: large and diverse samples that allow for generalizable statements and cover relevant subgroups; a repeated-measures design that allows tracing change in competencies over time; and valid, reliable, objective, and longitudinally linked competence assessments. Until recently, such data were in short supply (Desjardins & Warnke, 2012; Reder & Bynner, 2009).

Fortunately, at least in Germany, this situation has changed with the advent of NEPS Starting Cohort 6 (SC6: Adults) and PIAAC-L. Both surveys assess literacy and numeracy in comparable ways. They offer repeated measures of adults’ literacy and numeracy spaced 3 years (in PIAAC-L) to 6 years (in NEPS) apart. Both surveys conceptualize literacy and numeracy from a functional perspective with assessment items reflecting problems and tasks encountered in everyday life (Gehrer et al., 2013). Despite some differences in the assessment approaches (for a detailed comparison of the literacy assessments in PIAAC-L and NEPS, see Durda et al., 2020), evidence from a linking study suggests high convergence between test takers’ results in the PIAAC-L and NEPS assessments (Carstensen et al., 2017). Combining these data offered the opportunity to conduct what appeared to be the most comprehensive analyses to date of change in literacy and numeracy skills during adulthood in Germany.

3.3 Challenges and Pitfalls

Thus, PIAAC-L and NEPS are unique data troves that offer unprecedented analytical opportunities for the study of adult competencies. Nonetheless, there are a number of methodological challenges that complicate the study of change in competencies (Pohl & Carstensen, 2013; Pohl et al., 2015).

3.3.1 Retest Artefacts and Regression Toward the Mean

The first challenge pertinent to our project concerned the repeated assessment of skills. In traditional designs, the same set of items are administered on different occasions. This can lead to practice effects that bias estimates of change in competencies. Retest artefacts are among the reason why cross-sectional age profiles may sometimes capture age effects on cognitive abilities more accurately than repeated-measures designs (Salthouse, 2019a). With the help of computerized testing designs (which were also adapted in recent NEPS waves), such retest artefacts can be minimized because, apart from a few ‘anchoring’ items, individuals receive a different set of items on different occasions. Still, it is possible that repeated exposure to the assessment situation alters the test results to some extent.

Another complication of analyses of competence development is a specific form of change that occurs even in the absence of identifiable external influences: regression toward the mean, a ubiquitous phenomenon particularly in two-wave data (Furby, 1973; Nesselroade et al., 1980). Regression toward the mean moves extreme values (e.g., competence scores) in the initial assessment closer to the sample mean in the reassessment. Regression toward the mean is ‘a feature, not a bug’ of two-wave data, so to speak, because for any imperfectly correlated measure, the following relation holds: E(X2|X1 = x) = corr(X1, X2) × x (Nesselroade et al., 1980).

Although measurement error is often invoked as an explanation for regression toward the mean, this equation makes it clear that the phenomenon occurs even in the absence of measurement error. The explanation is that the total set of causal influences on performance in not only the literacy/numeracy assessment (e.g., the actual competence level) but also situational factors (e.g., test motivation, fatigue, disturbances during the assessment) that lead to an extreme score at one measurement occasion are unlikely to re-occur at a second occasion, leading to less extreme scores. Whatever the reasons behind regression toward the mean, it complicates the study of competence development and makes it unlikely that ‘the competent get more competent’ over time, as a Matthew effect would imply.

3.3.2 Measurement Error

Another challenge regarding assessment, which we discuss in detail in a primer paper on the usage of data from large-scale assessments resulting from the research project (Lechner et al., 2021a), is measurement error in the target competencies. Competencies are latent variables that cannot be observed directly but only inferred indirectly from respondents’ answers to a set of test items. Any point estimate of an individual’s ability will inevitably contain measurement error. Measurement error, in turn, adds random noise that typically attenuates (i.e., biases downward) the test–retest correlation, a key indicator of stability (see below). Likewise, measurement error can attenuate regression coefficients when a pretest score of the competence is used as a predictor of a posttest score of that competence, as is the case in lagged dependent variable (LDV) models or when the competence score is used to predict some outcome (e.g., income).Footnote 3

To avoid bias from measurement error, latent-variable modelling or plausible values (PV) methodology should be used instead of individual ability point estimates (i.e., ‘test scores’; see Lechner et al., 2021a; von Davier et al., 2009; Wu, 2005). PIAAC-L has long included PVs. NEPS now also includes PVs and provides a state-of-art R tool for estimating a custom set of PVs (Scharl et al., 2020).

3.3.3 Selectivity

A third challenge is selective panel attrition. Even with the best of efforts by researchers, some respondents will inevitably drop out of a panel study, be it because they refuse to participate again, because they cannot be located, or for a variety of other reasons. NEPS and PIAAC-L are no exception in this regard. Dropout is particularly problematic if it is selective—that is, if certain individuals have a systematically higher likelihood of dropping out than others. Such selectivity can introduce bias in substantive findings. A study from the context of our research project (Martin et al., 2020) showed that this problem indeed plagues both surveys: between ALWA (the predecessor of NEPS conducted in 2007/208) and the first NEPS wave in 2010/11, and between PIAAC 2012 and the first wave of PIAAC-L in 2014, individuals with lower education and—even after controlling for education—lower literacy were more likely to refuse to participate again and drop out than those with higher education and literacy. Although this effect might not be stationary (i.e., it might not remain as strong in subsequent waves; e.g., Hammon, 2018), it will lead to a sample that is somewhat skewed toward higher skills and education. To some extent, using longitudinal weights that adjust for selective attrition and/or modern missing data methods such as multiple imputation can remedy this problem. However, neither weights nor missing data can recover cases that are lost, and they may not even remove bias if dropout is conditional on unobserved covariates.

4 How Stable or Malleable Are Literacy and Numeracy in Adulthood?

Mindful of the methodological challenges facing any inquiry based on a repeated-measures design, I shall now review the most important substantive findings of the research project on the patterns and predictors of change in competencies in adulthood. Together, the findings from these studies suggest the four basic principles of competence development in adulthood visualized in Fig. 4.2. I shall discuss each of these principles in turn.

Fig. 4.2
A set of 4 principles. They are lifelong plasticity, involving both gains and losses, dependence of competencies on basic cognitive skills, cumulative advantage, and practice makes perfect. A list of points is below each principle.

Four principles of change in literacy and numeracy during adulthood that emerged from our project

Full size image

The first fundamental question about competence development in adulthood concerns the stability over time of literacy and numeracy skills. One can only answer this question with repeated-measures data. We (Lechner et al., 2021b) utilized NEPS and PIAAC-L to investigate stability and change in literacy and numeracy across 3–6 years of adulthood. We employed two complementary measures of change: mean-level change (i.e., the change score of competencies between two points in time, ΔT1, T2) and rank-order consistency (i.e., the correlation between competencies assessed at two points in time, \( {r}_{T_1,{T}_2} \)). We inspected these measures of change in the total population and in major sociodemographic subgroups defined by age, gender, and educational attainment. The subgroup analyses aimed to identify potential social inequalities in competence development. In both surveys, we used PVs to account for measurement error in competencies and included information from background variables.

The analyses yielded several novel insights. First, mean-level change in the population was small: on average, adults experienced little change in either literacy or numeracy across the 3 years (PIAAC-L) to 6 years (NEPS) of adulthood covered by the surveys. Zooming in on potential differences across sociodemographic segments revealed that change was slightly more pronounced in some subgroups, but the extent of change was still small. Moreover, few of the subgroup differences in PIAAC-L were replicated in NEPS and vice versa. The only subgroup difference that was similar in both surveys was a tendency among young adults (18–34 years) to experience gains in literacy (and in PIAAC-L also numeracy), whereas other age groups experienced little change or showed a tendency toward losses, thereby resembling earlier findings from cross-sectional data from PIAAC (Paccagnella, 2016). As an aside, the mean-level change observed in this sample was smaller than the age differences in competencies implied by cross-sectional data from PIAAC (Paccagnella, 2016) as well as the cross-sectional age differences in competencies in PIAAC-L and NEPS Waves 1 (see Supplementary Material in Lechner et al., 2021b).

One might be tempted to conclude that 3–6 years are simply too short for substantial change to occur. A closer look at the distribution of change revealed, however, that behind the near-zero means of the change score, there were substantial shares of adults who experienced changes in literacy and numeracy in both surveys. Importantly, the distributions of the change score were approximately symmetric around zero, meaning that gains and losses were almost equally likely to occur. This suggests that looking only at mean-level change conceals a great deal of the heterogeneity in change that actually exists.

This impression was further reinforced by the rank-order consistencies \( {r}_{T_1,{T}_2} \). The correlations across 3 years in the total population were .85 and .81 for literacy and numeracy in PIAAC-L, and the correlations across 6 years in NEPS were .61 and .70 respectively (accounting for measurement error through PV methodology). Albeit substantial, these correlations are far from unity. They are notably lower than the rank-order consistencies of basic cognitive skills over often much longer periods of time reported by studies on age trajectories of basic cognitive abilities (intelligence). For example, Gow et al. (2012) conducted one of the few longitudinal studies on the stability of cognitive abilities across the life span on data from 1017 Scottish individuals from the Lothian Birth Cohort study. These authors found that general intelligence (g) measured with the Moray House Test No. 12 (MHT) at age 11 correlated at r = .67 (r = .78 after correcting for measurement error) with intelligence reassessed at age 70. Deary (2014) reported similar rank-order consistencies in data from another Scottish cohort study (one can only envy a country so blessed with data), although rank-order consistencies were lower because the age of individuals at retest increased into the eighth decade of life. Furthermore, Schalke et al. (2013) reported rank-order consistencies for general intelligence of r = .85 from age 12 to 52 in a sample of 344 individuals from Luxembourg. The rank-order consistencies of specific abilities (e.g., fluid reasoning, comprehension knowledge, visual processing) were only slightly lower. These findings underscore that the rank-order consistencies of literacy and numeracy in PIAAC-L and NEPS are relatively low in comparison to that of general intelligence over several life decades in these studies. The reasons for their relatively lower stability are not entirely clear. While it is possible that literacy and numeracy are more sensitive to lifestyle and contextual influences than basic cognitive abilities, further research is needed to exclude the possibility that methodological differences (e.g., assessment design, test length, type of tasks) or differences in motivation and effort among test-takers underlie these findings. At any rate, the far from perfect rank-order consistencies of literacy and numeracy across the relatively short periods of 3–6 years in PIAAC-L and NEPS imply that the relative position of adults in the competence distribution is subject to change over time, or, in other words, that individual differences in competencies are not fully stable over time.

Overall, then, the descriptive findings from Lechner et al. (2021b) paint a clear picture: although competencies change little on average, a sizeable share of adults do experience change in competencies over time even when accounting for measurement error in competencies through PV. This is evidenced by the distribution of the change score and the substantial but not perfect rank-order consistencies. This change comprises gains and losses in about equal measure.

5 What Factors Drive Gains and Losses in Competencies?

If competencies are not ‘set like plaster’ but can change in adulthood even over relatively short periods, and if competence gains are as likely to occur as competence losses, then the natural next question to ask is what drives gains or losses in competencies. Identifying individual and contextual factors that explain individual differences in change was the objective to which most of the remaining studies in our project were devoted.

In two studies (Gauly et al., 2020; Gauly & Lechner, 2019), we set out to unravel whether the previously reported effects of participation in continued education and training (CET) hold up in a longitudinal setting in PIAAC-L. In subsequent studies, we explored the role of a broad set of predictors (i.e., potential determinants) of competence development in NEPS (Wicht et al., 2020, 2021a). To garner further insights into the role of motivational factors, additional studies (Lechner et al., 2019b; Miyamoto et al., 2020) expanded the project’s focus to adolescence, for which NEPS SC3 (5th Graders) offers richer measures of motivational factors than SC6.

5.1 Cumulative Advantage (‘Matthew Effects’)

As discussed earlier, Matthew effects (resource amplification) are a widely discussed pattern in research on social inequality and education (Blossfeld et al., 2020; Cunha & Heckman, 2007). Our findings provide some evidence that Matthew effects occur in competence development, too. At the same time, Matthew effects are not the full story. They occur only for some sociodemographic characteristics but not for others.

5.1.1 Education, SES, and Cultural Capital

Our analyses brought to the fore several findings that are indeed reminiscent of Matthew effects. These findings concern the role of sociodemographic characteristics that indicate resourcefulness—or, sociologically speaking, endowment with economic and cultural capital. Most crucially, higher educational attainment predicted positive change in literacy (i.e., growth or slower decline) in LDV models across 6 years in NEPS (Wicht et al., 2020). LDV or ‘residual change’ models control for initial competence levels and thus predict change over the initial competence levels (e.g., Johnson, 2005). In cross-sectional analyses based on PIAAC and its predecessors, educational attainment has long been identified as the key determinant of literacy and numeracy (e.g., Desjardins, 2003; Kyröläinen & Kuperman, 2021; Paccagnella, 2016). Analyses in PIAAC-L replicated these Matthew effects of educational attainment for both literacy and numeracy (Reder et al., 2020). Given that many of the adults in the PIAAC-L and NEPS samples obtained their last educational degrees years or even decades ago, the abiding importance of educational attainment for subsequent competence development is impressive.

Above and beyond educational certificates, a higher number of books in the household—a classic indicator of cultural capital (Sieben & Lechner, 2019)—also predicted positive change in literacy (Wicht et al., 2020). Even parental socio-economic status had a small positive association with literacy development, attesting to the ‘long arm’ of socialization and perhaps heritability. Several of these ‘Matthew effects’ were sizable, with standardized regression coefficients in excess of .20.

Although we could not test this explanation conclusively with the present data, it seems reasonable to assume that these effects reflect the selection of the higher-educated and more resourceful into more stimulating environments that require the continued application of literacy and numeracy, which, in turn, fosters the development of these competencies and protects to some extent against age-related losses. Future research may be able to unpick the reciprocal influences of selection and socialization with designs that assess competencies from childhood or youth into adulthood.

5.1.2 No Matthew Effects for Competencies as Such

Whereas we found some evidence for Matthew effects for education and cultural capital, there were no Matthew effects for competencies themselves in the sense that adults with higher initial competence levels necessarily enjoy greater gains (or smaller losses) in competencies than adults with lower competencies. Quite the contrary, what we observed when splitting the initial competence levels into quartiles and analysing the distribution of change in competencies in each of these four quartiles was regression toward the mean. That is, adults from the fourth quartile—those with the highest literacy and numeracy level at the initial measurement occasion—experienced losses in these competencies on average, whereas adults from the first quartile—those with the lowest literacy and numeracy level—experienced gains on average (Lechner et al., 2021b). This might also indicate ceiling effects in the assessments.

5.1.3 No Clear Evidence of Gender Differences But Small Age Differences

Although several studies (e.g., Borgonovi et al., 2018; Paccagnella, 2016) have reported gender differences in literacy (favouring women) and numeracy (favouring men), no clear evidence of gender differences in the change in literacy and numeracy development emerged from LDV models analysing changes in competencies across 6 years in NEPS (Wicht et al., 2020) or 3 years in PIAAC-L (Reder et al., 2020). In the descriptive analyses of mean-level change and rank-order consistency in PIAAC-L and NEPS, gender differences were also small or non-existent (Lechner et al., 2021b).

Regarding age, LDV models and controlling for initial competence levels in NEPS (Wicht et al., 2020) and PIAAC-L (Reder et al., 2020) bore out with greater clarity what was only a slight tendency in the descriptive analyses of mean-level change (Lechner et al., 2021b): higher age is negatively related with change in literacy and numeracy. Younger adults (up to age 30 or 40) tend to gain, older adults (from age 50 or so onward) tend to lose competencies, although gains and losses in the repeated-measures data of PIAAC-L and NEPS are somewhat smaller than those suggested by cross-sectional age profiles (Gauly et al., 2020; Gauly & Lechner, 2019; Lechner et al., 2021b).

5.2 Fluid Cognitive Abilities

Another important factor that is known to govern adults’ competence development are basic cognitive skills of the type that Cattell (1971) denoted as fluid intelligence (gf), Ackerman (1996) as ‘intelligence-as-process’, and (Baltes, 1993) as ‘cognitive mechanics’: reasoning ability (typically measure with matrices tests) and perceptual/processing speed (typically measured with digit–symbol substitution tests).

5.2.1 Basic Cognitive Skills Predict Positive Change in Competencies

Findings from our project make it abundantly clear that individuals with higher basic cognitive skills enjoy an advantage when it comes to acquiring literacy and numeracy in adolescence and maintaining or even expanding these competencies in adulthood. In a study in NEPS SC3 (Lechner et al., 2019b), higher levels of reasoning ability as measured in 5th grade predicted faster growth in both literacy and numeracy between 7th and 9th grade in secondary school students, especially in students who also had high interest in reading and math respectively.Footnote 4 Likewise, in adults, higher processing speed and especially higher reasoning ability measured in the first NEPS wave predicted positive change in literacy over the subsequent 6 years (Wicht et al., 2020). Thus, fluid-type cognitive abilities are tightly interwoven with the more crystallized-type literacy and numeracy competencies. This should come as no surprise to anyone familiar with structural models of intelligence in which various fluid and crystallized cognitive abilities all form a positive manifold (i.e., are all substantially positively correlated) from which a general (g factor) emerges (Lang et al., 2016; McGrew, 2009). It highlights that although fluid-type and crystallized-type abilities may indeed follow different age trajectories as hypothesized by Cattell, they are not fully independent of each other (see discussion on the meaning of ‘g’ in Nisbett et al., 2012).

5.2.2 ‘General Slowing’ Might Impair Competencies

The aforementioned association between higher age and negative (relative) change in literacy that we observed in LDV models in NEPS (Wicht et al., 2020) shrank by about one third after adding reasoning ability (i.e., Gf) and processing speed to the model. This is compatible with the view that the literacy losses in older adults are partly due to ‘general slowing’. The general slowing hypothesis states that cognitive abilities that depend on neural or biological efficiency decline with age (Choi & Feng, 2015). Similar to Cattell’s view, this, in turn, may reduce the ability to learn novel things and ultimately impair even acquired (learned) abilities. As explained earlier, although literacy and numeracy are acquired abilities, acquiring and performing these competencies requires reasoning and processing speed. It is hence plausible that the well-established age-related declines in reasoning and processing speed (Deary, 2014; Salthouse, 2019a; Schalke et al., 2013) negatively affect literacy and numeracy, too. Only a few select skills—especially those that rely on long-term memory such as vocabulary—are largely exempt from age-related declines. For a more comprehensive and conclusive test of the general slowing hypothesis and its implications for competencies, repeated measures of both literacy/numeracy and basic cognitive skills would be required. This would allow the unravelling of different hypotheses about the causal relations between basic cognitive skills and competencies: (a) declines in basic cognitive skills precede and then lead to subsequent declines in competencies; (b) declines in basic cognitive skills co-occur at the same time as declines in competencies, and both are caused by general slowing. The relations among age, fluid-type abilities, and crystallized-type abilities is a longstanding matter of investigation in intelligence research (Nisbett et al., 2012; Salthouse, 2019b).

5.3 Practice Engagement

5.3.1 Engagement in Literacy and Numeracy Practice Ranks Among the Strongest Predictors of Competence Development

A third principle that governs competence development in adulthood is well captured by the adages that ‘practice makes perfect’ (Reder et al., 2020) and you ‘use it or lose it’ (Bynner & Parsons, 1998). These adages allude to the central position that practices occupy in competence development. Practices refer to the frequency and/or intensity with which individuals engage in literacy practices such as reading books, manuals, diagrams, or letters at work and during leisure; or engage in numeracy practices such as calculating prices, costs, or budgets; preparing charts or tables; or using a calculator.

In line with PET (Reder, 1994, 2009), LDV models demonstrate that practices predict positive change (i.e., growth or slower decline) over time in both literacy and numeracy and in both NEPS (Wicht et al., 2020, 2021a) and PIAAC-L (Reder et al., 2020). Indeed, in PIAAC-L, practices emerged as the strongest predictors of change in literacy and numeracy even after accounting for educational attainment and a range of other sociodemographic characteristics (Reder et al., 2020). Cross-sectional analyses (Wicht et al., 2021b) suggest that the same holds for ‘digital literacy’ or ‘ICT competence’ in NEPS and PIAAC-L, which proved to be strongly dependent on the usage of digital technologies on the job and in everyday life (no repeated assessments of this competence are available in the data yet).

That practices contribute to competencies is intuitively plausible but far from trivial. First and foremost, practice effects presuppose that competencies—which are acquired mainly through schooling during childhood and adolescence—remain plastic (i.e., malleable) throughout adulthood. Only if competencies are sufficiently malleable can engagement in practices contribute to growth or maintenance in these competencies. Findings from our previously discussed descriptive analyses on stability and change (Lechner et al., 2021b) and the LDV results reported here suggest that this is the case. This is an important finding, because it suggests that adults’ competencies are in principle amenable to policy programmes and interventions that target everyday practices.

Second, as already intimated by PET and as further elaborated in a theoretical framework presented in Wicht et al. (2021b), practices are themselves dependent upon a range of preconditions. For example, individuals with higher competencies or higher educational attainment as well as those in employment enjoy greater opportunities for engaging in literacy- and numeracy-related practices at work and leisure than those with lower competencies or lower attainment, or those who are unemployed (Reder, 1998). This suggests that practices and thereby the likelihood of competence growth are patterned and stratified by earlier selection processes that may increase inequality in competencies in the long run.

Our analyses were limited by the fact that they relied on practices as assessed on a single occasion and could not control for unobserved heterogeneity. Although we controlled for a range of possible confounders, we cannot exclude the possibility that unobserved confounders influenced both practices and competencies. The time scale on which potential practice effects operate also remains unclear. Future work relating changes in practices to changes in competencies over longer periods of time could prove insightful. Currently, however, long-running panel data comprising repeated assessments of both competencies and practices that would allow for such analyses are unavailable.

5.3.2 Job-Related Training per se Does Not Contribute to Literacy and Numeracy Growth

Despite the demonstrable importance of practices, not all ‘practices’ contribute to competence growth—especially if the indicators of practices are coarse and unspecific. This concerns, in particular, the alleged role of job-related training specifically for literacy and numeracy growth. Earlier large-scale cross-sectional and smaller-scale longitudinal studies focusing on lower-skilled or lower-educated individuals had reported a positive association between participation in job-related training and levels of literacy and numeracy (e.g., Cegolon, 2015; OECD, 2013). Some researchers and policymakers have interpreted this association as proof that job-related training can contribute to literacy and numeracy growth. Our project’s findings using the repeated-measures data of PIAAC-L tell a different story. Although we (Gauly & Lechner, 2019) indeed replicated the positive association between job-related training and change over 3 years in literacy in pooled ordinary least squares (POLS) and LDV models, this association vanished when using first-difference/fixed-effects (FE)Footnote 5 and instrumental variable (IV) estimators, which—in contrast to LDV and OLS models—control for unobserved time-invariant heterogeneity. Conversely, higher initial literacy levels predicted higher rates of participation in job-related training. We obtained essentially the same findings with numeracy and even after differentiating by training type and intensity (Gauly et al., 2020). This strongly suggests that the previously reported positive associations between competencies and participation in job-related training reflect self-selection of more competent (and highly educated) individuals into further training rather than training effects on competencies.

Despite the strong links between training and competencies in cross-sectional work, the absence of training effects should not come as a surprise. Job-related training is a broad category that comprises anything from presentation training for white-collar trainees to safety training for workers in a manufacturing plant or forklift license training for warehouse staff. These examples patently show that by far not all job-related training includes written material, let alone numerical material, and tasks of the sort that would count as ‘practising’ literacy and numeracy skills. Some training may be fully oral or rely on visual demonstration of motor skills only. Moreover, training often comprises short courses lasting only a few hours. It is unlikely that such training, designed to foster specific job-related skills, would have ‘spillover’ effects on literacy and numeracy. Of course, job-related training that involves a considerable amount of literacy- or numeracy-related tasks may well contribute to literacy and numeracy growth. For example, workplace-based literacy training and basic skills training may be effective in lifting low-literate adults, or at least in encouraging them to increase literary practices (Bynner & Parsons, 1998; Reder & Bynner, 2008; Sheehan-Holt & Smith, 2000; Wolf & Jenkins, 2014). However, such training effects cannot be detected with the relatively coarse measures of participation in (any type of) job-related training available in PIAAC-L (and NEPS, although we did not use these data for our papers on job-related training).

6 Avenues for Future Research

Our project was enabled by the unique large-scale two-wave competence data from PIAAC-L and NEPS. Our findings on change in competencies over a 3- to 6-year period constitute an important advancement that expands prior small-scale longitudinal and cross-sectional evidence. Nonetheless, our project still leaves many questions unanswered that call for further research—in particular, (a) how competencies develop over longer time periods and during major transitions; (b) which predictors have the strongest influences on development, and which of them are causal; and (c) how other equally important competencies beyond literacy and numeracy develop. I shall discuss these in turn.

The PIAAC-L and NEPS data available to our project comprised two occasions on which literacy and numeracy were assessed spanning 3–6 years of adulthood. Although two measurement occasions for competencies are clearly better than one, a two-wave design is not a truly longitudinal design (e.g., Ployhart & MacKenzie Jr., 2015). Future research should extend our work to include additional measurement occasions and cover longer time spans during which more change might happen. Unlike PIAAC-L, NEPS is ongoing and will continue to grow in this regard. Tracing competence development over multiple occasions would provide deeper insights into average trends (e.g., age-related declines) and individual differences in change. A particularly important question is how competencies develop across key transitions such as job entry, parenthood and parental leave, unemployment and reemployment, or retirement. It is across these transitions that we expect competencies to change most strongly and perhaps most long-lastingly.

Additional assessment waves would also greatly expand the options available for causal identification. Our in-depth analyses of the link between job-related training and competencies (Gauly & Lechner, 2019; Gauly et al., 2020) illustrate the benefit of applying different estimators (OLS, LDV, FE, IV) to scrutinize potential causal effects on competencies. However, most studies in our project were limited in their ability to eliminate unobserved heterogeneity by, for example, applying panel (e.g., FE) regressions or related models, because some key predictors of competence development were measured only once. Thus, no causal claims can be made for findings reported herein, including those regarding the most important predictors of competence change (i.e., practice engagement, basic cognitive skills, education, and others). This, however, would be crucial in order for this type of research to inform policy and practice about the most potent causal influences on competence development that may prove to be apt targets (or target groups) for interventions and other measures.

Another question that we left largely unaddressed is how competencies other than literacy and numeracy develop in adulthood. For example, digital literacy (or ‘ICT literacy’) is a competence that is increasingly on-demand in the labour market (Wicht et al. 2021b). So far, these competencies have been assessed only cross-sectionally in PIAAC-L and NEPS SC6. Given how similar our results regarding literacy and numeracy were in general, and given that digital literacy depends strongly on reading literacy, it is an open question whether digital literacy is sufficiently distinct from literacy and numeracy. Comparing trajectories and precursors of digital literacy to those of literacy and numeracy may help resolve this question. It would be even more important to extend our work to include socio-emotional or ‘non-cognitive’ competencies. Related to the increasing penetration with technology, competencies such as communicating and cooperating with others or coping with uncertainty and insecurity may also become increasingly important for leading successful and healthy lives in the twenty-first century (e.g., Allen et al., 2020; Lechner et al. 2019a). Investigating the development of adults’ socio-emotional competencies and their interplay with literacy and numeracy may prove highly insightful and indeed indispensable if researchers are to gain a more complete understanding of how the competencies that are needed to lead a successful life in current societies develop in adulthood.

7 Conclusion

The findings from our project lead to the following broad conclusions about the development of competences in adulthood. First and foremost, literacy and numeracy are not ‘set in stone’ during adulthood. Instead, these competencies remain malleable and can change even over the relatively limited periods of 3–6 years observed in PIAAC-L and (thus far) in NEPS. Gains and losses occur in almost equal shares, which leads to a mean-level change of zero on the population level that conceals the fact that many adults do experience change in competencies. Future work needs to establish the extent to which this change reflects true change in competencies as opposed to or artifacts such as regression to the mean or flucations in test-takers’ effort and motivation.

Three main groups of factors predict whether an individual is more likely to experience gains or losses in competencies over time: (1) sociodemographic characteristics that indicate resourcefulness or social advantage in which we observed a ‘Matthew effect’ for educational attainment and cultural capital (but no gender differences); (2) basic cognitive skills in which we observed that processing speed and especially reasoning ability predict change in competence, which partly explains age differences in competencies (increases in young adulthood, decreases in old age) in line with the ‘general slowing’ hypothesis; and (3) engagement in literacy and numeracy practices in which we observed that more frequent reading/math at work or during leisure robustly predicted positive change in competencies in line with the ‘practice makes perfect’ and ‘use it or lose it’ principle. These findings confirm and qualify some of the previous insights from previous small-scale longitudinal and large-scale cross-sectional research (Deary, 2014; Desjardins & Warnke, 2012; Nienkemper et al., 2021; Paccagnella, 2016).

Future work should replicate and expand our project’s findings by analysing competence development over longer periods while addressing some of the methodological challenges we have outlined, especially selectivity and potential retest artefacts such as regression toward the mean and effects of prior exposure to the tests.