Is knowledge capital theory degenerate? PIAAC, PISA, and economic growth

ABSTRACT

Extending recent analyses using PISA data, the current study utilises the OECD’s Survey of Adult Skills (PIAAC) to test the central claims of knowledge capital theory. PIAAC has a distinct advantage over PISA in that it more directly tests levels of purported ‘knowledge capital’ across an entire national workforce, rather than offering approximations based on the performance of 15-year-old students. Findings from our two original studies reported herein further refute the tight linkage between cognitive levels and GDP growth per capita envisaged by proponents of knowledge capital, most notably the OECD and World Bank. These results suggest that knowledge capital theory is now degenerate. If scholars are willing to extend this reporting of results refuting knowledge capital theory, this will likely accelerate the theory’s loss of momentum in the coming years.

KEYWORDS:

• Human capital theory
• OECD
• international large-scale assessments
• adult skills

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed here.

Notes

1. Knowledge Capital Theory is, in this sense, an extreme version of Human Capital Theory. While Human Capital Theory assumes that adults’ overall skill levels, including cognitive levels, affect national economic growth, Knowledge Capital Theory assumes that only adults’ cognitive levels determine national economic growth and, moreover, that adults’ cognitive levels are being accurately assessed by PISA and similar international learning assessments. The extreme simplicity of the Knowledge Capital Theory, which seems to wilfully overlook long-standing debates in that domain of research, is perhaps one reason why a recently published comprehensive review of Human Capital Theory does not even mention the work of Hanushek and colleagues (Tan 2014). While it is true that Tan (2014) was published before the publication of H&W’s major book (Hanushek and Woessmann 2015a), all the widely cited lead up studies (e.g, Hanushek and Kimko 2000; Hanushek and Woessmann 2008, 2012) and major reports written by H&W for the World Bank (Hanushek and Woessmann 2007) would have been available. This underscores just how marginal – in an extreme sense – Knowledge Capital theory remains in the wider work on Human Capital. We note too that this focus on knowledge occurs against the larger backdrop of the idea of a ‘Global Knowledge Economy’, a major World Bank policy thrust since the 1990s (see Scheicher quote below).

2. To clarify: the piece by Ramirez et al. (2006) was published before the publication of H&W’s major works. In the paper, Ramirez et al. (2006) challenge Hanushek and colleagues’ earlier work (Hanushek and Kimko 2000), which turns out to be the first step of the larger trajectory that culminates in the H&W work.

3. We failed to replicate H&W figure of 73%. One possible reason could be that although we used the same GDP per capita dataset as that used by Hanushek and Woessmann (2015a) we used an updated version. Despite the difference in the percentage between H&W’s and Komatsu and Rappleye (2017) studies, these studies are qualitatively the same in that the variation in GDP per capita growth among countries was primarily explained by the variation in test scores.

4. In the debates over knowledge capital theory to date, we can already find an example of isolation: Ramirez et al.’s (2006) critique of Hanushek and Kimko (2000), which was an earlier study by Hanushek which paved the way to the more elaborated H&W’s work of 2007–2015. Ramirez et al. (2006) reported the relationships of test scores for a given period with economic growth in subsequent periods were not statistically significant. In response to this critique, Hanushek and Woessmann (2015a, 50) re-analysed their data and again confirmed a statistically significant relationship of students’ test scores with economic growth in a subsequent period. The difference in results between Ramirez et al. (2006) and Hanushek and Woessmann (2015a) was most probably derived from the differences in methods and data. But faced with the different results and Hanushek’s response, scholars such as Sahlgren (2014) assumed that the difference between Hanushek and Woessmann (2012) and Ramirez et al. (2006) was caused by the fact that Ramirez et al. (2006) used older data and analysed fewer countries and ultimately decided to ignore the counterexample. That is, and in the language of the philosophy of science, the findings by Ramirez et al. (2006) were isolated.

5. The periods for Round 1 and 2 were, respectively, 2008–2013 and 2012–2016, but tests for these rounds were respectively performed during 2011–2012 and 2014–2015. The OECD is currently implementing PIAAC Round 3 for additional six countries: Ecuador, Hungary, Kazakhstan, Mexico, Peru, and Romania.

6. More countries are planned for inclusion, particularly low-income countries admitted via the PISA for Development (PISA-D) assessment (OECD 2017b). The goal is to have all countries worldwide involved in PISA by 2030 (see Auld, Rappleye, and Morris 2019).

7. This does not mean we agree with the methodology utilised by H&W. Our point is to examine whether or not H&W’s results hold when using exactly the same methodology and more relevant data (i.e., PIAAC data instead of PISA data).

8. At the current moment, we are unable to extend the period for calculating GDP per capita growth to, say, 1995–2035. We must leave it to future work to examine whether or not our results change when using GDP per capita growth data for a longer period.

9. We did not include other control variables than GDP per capita. This does not mean we assume other variables are unimportant for determining GDP per capita growth. We did so merely to replicate the procedures of H&W to examine the validity of the assertion made by H&W.

10. According to normal probability plots, we found no apparent outliers among the data used in Figure 3.

11. This corroborated with the fact that we found no apparent outliers among the data used in Figure 3 according to normal probability plots.

12. In this sense, ILSAs can be viewed as a ‘special apparatus’ (Kuhn 1962), one specifically designed for the ‘anticipated function’ of measuring the assumed connection with greater precision. But the special apparatus ultimately produces a novelty/anomaly that undercuts the core assumptions of the entire pursuit and thus the ‘traditional pursuit prepares the way for its own change’ (65).