Beyond ‘Imagined Meritocracy’: Distinguishing the Relative Power of Education and Skills in Intergenerational Inequality

Abstract

Scholars argue the dominant discourse of meritocracy legitimises intergenerational inequality and the winner–loser divide. However, is our society really meritocratic? If yes, the relative power of educational qualifications per se should be smaller than that of skills/abilities in the labour market. Using the standardised data in the United States, structural equation modelling shows (1) the contribution of family background to educational attainment is as large as that to skills acquisition; but (2) the economic return to education is substantially larger than that to skills; and consequently (3) the role of education outweighs that of skills in forming social stratification. This suggests that contemporary USA is a typical credential society, where credentialism prevails over skills-based meritocracy. Nonetheless, people may misbelieve the society is meritocratic – imagined meritocracy – by conflating the levels/influences of education and skills. It is essential to distinguish these two traits and understand the credential/meritocratic nature of our society.

Keywords

credentialism education inequality meritocracy OED triangle OESD quadrangle PIAAC skills stratification structural equation modelling

Introduction

In his recent monograph, Sandel (2020) argues that meritocracy has been dominating the United States and legitimising intergenerational inequality and the divide between winners who can, and losers who cannot, enjoy favourable labour market outcomes. While such criticism against the dominance of meritocracy has long been shared among social scientists since Young described the dystopia (Mijs, 2021; Young, 1958), Sandel has effectively applied this oft-used concept to contemporary US society and displayed a new moral framework with an emphasis on restoring the dignity of work rather than aiming to empower everyone to excel in meritocratic competition.

Although he succeeds in philosophically delineating the illegitimacy of ‘winners take all’ and ‘self-responsibility of losers’ under the influential discourse of meritocracy, there is one strong and perhaps unconscious assumption behind his argument and subsequent discussions, which needs careful examination from the empirical perspective: the USA is a meritocratic society. Indeed, using various words of politicians and unethical behaviours of celebrities in conjunction with some statistics of job/income inequalities, Sandel (2020) explains meritocracy is now excessively at work. However, the fact that socio-economic assets are distributed based on individuals’ educational attainment and people internalise the meritocratic doctrine does not necessarily mean the society actually operates meritocratically (Andersen et al., 2021).

One common perspective is that individuals’ socio-economic background affects their destinations directly and/or indirectly via educational attainment, such that the advantaged are more likely than the disadvantaged to obtain favourable rewards (Breen, 2003; Goldthorpe, 2003; Witteveen and Attewell, 2020). In addition, extending earlier discussions on the meritocratic power of college education (i.e. the strength of the origin–destination link becomes smaller among tertiary educated or u-shaped once postgraduates are also considered) (Breen, 2010; Hout, 1984, 1988; Torche, 2011), recent research re-examines the function of higher education as an equaliser with close attention to its selectivity (Fiel, 2020; Torche, 2018; Zhou, 2019, 2022). However, regardless of whether (1) education and abilities/skills are underpinned by origins and (2) college education operates as an equaliser or a selector/stratifier, one cannot reject the presence of meritocracy insofar as educational qualifications precisely reflect one’s abilities/skills, which should be the determinant of labour market outcomes in a meritocratic society. Nonetheless, recent evidence suggests (a) there is a discrepancy between the level of education and that of skills, (b) these two dimensions have distinct roles in the rewards allocation process and, importantly, (c) educational credentials per se, rather than skills, exert a dominant power over economic success (Araki, 2020, 2022; Araki and Kariya, 2022). Should this be the case for the United States, it is not really meritocracy but credentialism that prevails in the society.

Indeed, the tendency that the affluent attempt/manage to get their children into elite universities implies that the United States is a credential society (Collins, 1979) where the possession of high educational qualifications is crucial for future success (Hout, 2011; Posselt and Grodsky, 2017), although its importance relative to skills is elusive. Consider, as a hypothetical case, two types of people: (a) a new graduate from a prestigious college with low skills (who achieved it thanks to his/her family background) and (b) a highly skilled person without a college degree (due to the absence of financial resources for his/her study). In a meritocratic world, it should be (b), not (a), who is rewarded on condition that one’s education and skills are distinguishable and evaluable in the labour market. In contrast, under the influence of credentialism, (a) can enjoy more economic returns than (b) despite their twisted skills levels. Although meritocracy and credentialism are not always the trade-off and the scope of education/skills sometimes overlaps, one may assume that the advantaged aim to let their children have a high degree regardless of their skills, arguably because they observe how an educational credential per se is utilised as a sorting tool even when its holders are less skilled.

Nevertheless, this potential structure has been inadequately investigated in an empirical manner. This gap does not necessarily undermine the philosophical merit of Sandel’s argument that the inequalities in socio-economic outcomes should not be justified by a certain doctrine, whether it is meritocracy or credentialism (or something else), and the dignity of work should be renewed. Yet, it is imperative to scrutinise which mechanism actually operates in designing our future. The current study therefore analyses the linkage between family background and labour market outcomes with particular attention to the relative roles of education and skills as mediators.

Data and Methods

In analysing intergenerational inequality, scholars often use the so-called OED triangle consisting of origins, education and destinations (Breen and Müller, 2020; Bukodi and Goldthorpe, 2018; Karlson and Birkelund, 2022; Pfeffer and Hertel, 2015). Extending OED, this article employs the OESD quadrangle, incorporating another mediator S (skills) (see Figure 1). In the sense that the role of skills is taken into account, the current analysis shares a similar concern with some pioneering studies (Karlson and Birkelund, 2019; Sullivan et al., 2018, 2020). However, one distinction here is to include individuals’ skills assessed concurrently with their labour market outcomes, rather than abilities in childhood. While the latter is superior to examine how skills mediate the association between family background and educational attainment, adult skills are the key to understanding the relative power of E and S in the rewards allocation process and their intergenerational transmission (Checchi and van de Werfhorst, 2018; Kerckhoff et al., 2001).

Figure 1.

OESD quadrangle.

The data are derived from the Organisation for Economic Co-operation and Development (OECD) Programme for the International Assessment of Adult Competencies (PIAAC), administered in the United States in 2017. While the absence of reliable skills data among adults has been a common challenge in this line of empirical research, PIAAC provides standardised scores of literacy and numeracy in conjunction with the respondents’ education, family background, labour market outcomes and socio-demographic attributes. In addition to this wide coverage of variables, one distinctive advantage of PIAAC is its international comparability. Although this article focuses on the United States, comparative research can be developed by applying the conceptual/analytic framework that follows to other PIAAC participating countries and comparing the findings cross-nationally.

It is important to note that ‘skills’ assessed by PIAAC are cognitive dimensions. Other traits (e.g. noncognitive and occupation-specific skills), which have been detected as the key to socio-economic success (Deming, 2017; Heckman et al., 2006; OECD, 2015), are not directly incorporated. Nonetheless, evidence shows that cognitive abilities operate as the foundation for other types of skills and longer-term outcomes (Autor, 2014; OECD, 2019a). Although one cannot completely reject the possibility that E measures used in the current analysis represent not only credentials per se but also some types of merits, the skills level in PIAAC can therefore be used as a comprehensive, if not perfect, metric for an individual’s ability to analyse its relative role in mediating the origin–destination linkage.

As regards family background (O), oft-used parental occupations and family income are not included in the PIAAC 2017 data. Instead, three variables are available: mother’s education, father’s education and the number of books at home (NBH). While there is a near-consensus on the importance of incorporating maternal and paternal education in this line of research (e.g. Birkelund et al., 2022; Chmielewski, 2019), two opposite views are provided on the validity of NBH as a metric for origins. Some argue NBH is a proper predictor of economic outcomes (Hanushek and Woessmann, 2011), whereas others indicate its endogeneity problem and suggest it should be excluded from empirical models (Engzell, 2021).

The current article thus adopts two strategies. First, the three measures are used to create a composite indicator of origin by merging the standardised scores of (1) a dummy for mother’s education (1 is assigned to the respondents whose mother is tertiary educated, 0 otherwise), (2) a dummy for father’s education (1 is assigned to the respondents whose father is tertiary educated, 0 otherwise) and (3) NBH (a six-point scale: 10 or less; 11 to 25; 26 to 100; 101 to 200; 201 to 500; more than 500). Second, origin is quantified by another composite measure calculated with only parental education (i.e. the sum of standardised scores of two dummies for maternal and paternal education). As the results are identical between these two approaches, the first analysis with NBH is described in the main manuscript and the second one without NBH is shown in the Appendix (see online Supplementary Tables A1 and A2).

Destination (D) is assessed by the deciles of monthly income including bonuses. Confounders include age, gender and race/ethnicity as provided in the PIAAC public use dataset. To mitigate the risk of reverse causation (i.e. labour market outcomes improve education and skills) and to better understand the up-to-date situation surrounding recent college graduates in comparison with their less educated counterparts, the respondents aged 25 to 34 are selected for analyses (N = 425). Given the limited number of samples, the bootstrap method (1000 replications) is also employed to see the robustness. Note that the findings and implication that follow are consistent even when changing the target age groups (e.g. those aged 25 to 44, respondents in their 30s and all respondents with age interactions).

Using these variables and the OESD framework (Figure 1), linear structural equation modelling (SEM) with only continuous measures and generalised SEM (GSEM) with categorical indicators are both performed. For SEM, education and skills are quantified by years of schooling and the mean score of literacy and numeracy (both of which are assessed with 0–500 raw points), respectively. The correlation coefficient (r) for these two measures is 0.591, indicating that the levels of education and skills are generally related but by no means identical (hence incorporating both dimensions in the same model is sensible). While income deciles are also treated as a continuous scale for SEM (Model 1), another nonlinear specification (GSEM) is employed by using income as an ordinal scale with continuous education/skills measures (Model 2) for a robustness check.

For another set of GSEM to directly examine the power of possessing a college degree, educational attainment is defined by a dummy variable concerning whether a respondent has attained tertiary education (i.e. International Standard Classification of Education (ISCED) 2011 Level 5 and above). Likewise, drawing on recent studies on the distinction between high educational credentials and high skills (Araki, 2020; Araki and Kariya, 2022), skills are specified by another dummy variable: whether or not the mean score of literacy and numeracy is 326 and above, the threshold for high skills set by the OECD (OECD, 2019b). As with Models 1 and 2, the outcome variable is used as both a continuous measure (Model 3) and an ordinal one (Model 4). See Table 1 for descriptive statistics.

Table 1.

Descriptive statistics.

Variables	Mean/percentage	SD	Min	Max
Income deciles	5.41	2.51	1	10
Years of schooling	14.23	2.64	6	21
Skills	271.06	49.71	131.62	400.85
Family background	0.35	2.38	−2.69	5.36
Tertiary degree (dummy)	58.4%
High skills (dummy)	13.4%
Age 25 to 30 (dummy)	46.4%
Age 30 to 34 (dummy)	53.6%
Women (dummy)	48.9%
White (dummy)	65.2%
Hispanic (dummy)	12.7%
Black (dummy)	12.0%
Asian and Pacific Islander (dummy)	6.4%
Other races/ethnicities	3.8%

Notes: N = 425. ‘Family background’ is a composite indicator of standardised scores for mother’s education, father’s education and the number of books in the home. Age is quantified by two dummies because its exact value is not provided in the PIAAC public use data for the United States.

The basic model (SEM) is describable with vectors and parameters in Figure 1 as follows:

D = β O + δ_{1} E + δ_{2} S + τ C

where C signifies confounders including age, gender and race/ethnicity. Defining E and S as the outcome of O with C and substituting these equations into the above, D can be defined as follows.

D = β O + δ_{1} (γ_{1} O + υ_{1} C) + δ_{2} (γ_{2} O + υ_{2} C) + τ C = (β + γ_{1} δ_{1} + γ_{2} δ_{2}) O + (τ + δ_{1} υ_{1} + δ_{2} υ_{2}) C

The primary focus is on the relative magnitude of γ₁δ₁ and γ₂δ₂ (mediation by education and skills) as well as their decomposition (γ₁, γ₂, δ₁ and δ₂). One may hypothesise that credentialism is at work more than meritocracy when γ₁δ₁ is significantly larger than γ₂δ₂, and vice versa, as the mechanism behind intergenerational inequality. Unlike SEM, the coefficients in nonlinear models are not interpretable in a substantive manner (Breen et al., 2013, 2018). However, the indirect effects via education and skills (and hence total effects) in GSEM can be estimated as the product of coefficients, and the difference in magnitudes of multiple paths is testable by the bootstrap method in consideration of their standard errors as with SEM (Albert et al., 2016; Bartus, 2017; Preacher, 2015; Rijnhart et al., 2017). That is, by multiplying the estimated strength of two links (i.e. OE and ED; OS and SD, respectively), the indirect relationships between origin and destination mediated by E and S can be defined. Adding up these two indirect pathways and the direct OD association, one may also calculate the total effect.

Given the potential heterogeneities in associations between O, E, S and D across socio-demographic attributes (e.g. gender and race/ethnicity) (Brand, 2010; Breen and Jonsson, 2007; Kerckhoff et al., 2001; Lu and Li, 2021), the models are finalised by incorporating interaction terms between O/E/S and C (see Table 2 Notes for more details about analytic models and Table 3 for the full model).

Table 2.

Key results of structural equation modelling of OESD with bootstrapping.

Path	Model 1 (SEM: linear estimation for D with continuous E and S)					Model 2 (GSEM: nonlinear estimation for D with continuous E and S)
	B	Bootstrapped				B	Bootstrapped
	B	SE	95%CI (Lower–Upper)		P	B	SE	95%CI (Lower–Upper)		P
Direct (β)	0.054	0.105	−0.152	0.261	0.607	0.055	0.095	−0.131	0.241	0.565
Indirect via E (γ₁δ₁)	0.136	0.047	0.045	0.227	0.003	0.115	0.039	0.038	0.191	0.003
Indirect via S (γ₂δ₂)	−0.007	0.028	−0.062	0.048	0.805	0.001	0.025	−0.047	0.049	0.964
Decomposition
OE (γ₁)	0.339	0.085	0.173	0.506	<0.001	0.339	0.085	0.173	0.506	<0.001
OS (γ₂)	5.552	1.647	2.324	8.781	0.001	5.552	1.647	2.324	8.781	0.001
ED (δ₁)	0.401	0.092	0.221	0.581	<0.001	0.338	0.079	0.184	0.492	<0.001
SD (δ₂)	−0.001	0.005	−0.011	0.008	0.796	0.000	0.004	−0.008	0.008	0.962
Controls	Yes					Yes
AIC/BIC	40496.34/40719.20					8067.97/8319.20
Path	Model 3 (GSEM: linear estimation for D with binary E and S)					Model 4 (GSEM: nonlinear estimation for D with binary E and S)
	B	Bootstrapped				B	Bootstrapped
	B	SE	95%CI (Lower–Upper)		P	B	SE	95%CI (Lower–Upper)		P
Direct (β)	0.052	0.112	−0.168	0.271	0.644	0.058	0.096	−0.131	0.246	0.550
Indirect via E (γ₁δ₁)	0.545	0.270	0.016	1.075	0.044	0.479	0.232	0.025	0.934	0.039
Indirect via S (γ₂δ₂)	0.004	0.144	−0.277	0.286	0.976	0.004	0.130	−0.251	0.259	0.974
Decomposition
OE (γ₁)	0.418	0.115	0.193	0.643	<0.001	0.418	0.115	0.194	0.643	<0.001
OS (γ₂)	0.191	0.122	−0.048	0.429	0.117	0.191	0.122	−0.048	0.430	0.118
ED (δ₁)	1.304	0.466	0.391	2.216	0.005	1.145	0.406	0.349	1.942	0.005
SD (δ₂)	0.023	0.580	−1.115	1.160	0.969	0.023	0.520	−0.996	1.041	0.965
Controls	Yes					Yes
AIC/BIC	2725.90/2936.61					2658.38/2897.45

Notes: Overall R² is 0.426. The number of observations is 425 for all models. AIC and BIC stand for the Akaike information criterion and the Bayesian information criterion, respectively. In Models 1 and 2, education (E) and skills (S) are continuous measures (i.e. years of schooling and the mean score of literacy and numeracy), for which linear estimation is applied, whereas D is estimated by linear (Model 1) and ordered logit (Model 2) links with O, E, S and controls. In Models 3 and 4, E and S are binomial variables (i.e. dummies for tertiary educated and highly skilled with the mean score of literacy and numeracy being 326 and above, respectively), for which a binary logit model is used, alongside the linear (Model 3) and ordered (Model 4) estimations for D as with Models 1 and 2. Controls include age, gender and race/ethnicity. The indirect effects, standard errors (SE) and confidence intervals (CI) are calculated by the bootstrap method (1000 replications), using the lincom (Model 1), nlcom (Models 2–4) and bootstrap (all models) commands in Stata. Cut points for nonlinear estimation for D (i.e. ordered logit for income deciles) in Model 2 are 1.313, 2.454, 3.183, 3.823, 4.575, 5.271, 5.990, 7.054 and 8.806; those in Model 4 are −2.762, −1.668, −0.975, −0.363, 0.347, 0.996, 1.674, 2.693 and 4.412. See Table 3 for the full results.

Table 3.

Full results of structural equation modelling of OESD.

Path	Model 1 (SEM: linear estimation for D with continuous E and S)					Model 2 (GSEM: nonlinear estimation for D with continuous E and S)
Path	B	SE	Lower	Upper	P	B	SE	Lower	Upper	P
Income deciles
Family background	0.054	0.094	−0.131	0.239	0.565	0.055	0.083	−0.107	0.217	0.509
Years of schooling	0.401	0.099	0.207	0.595	<0.001	0.338	0.082	0.177	0.500	<0.001
Skills	−0.001	0.005	−0.011	0.008	0.798	0.000	0.004	−0.008	0.008	0.961
Age 30–34	−1.078	1.407	−3.836	1.680	0.444	−0.729	1.176	−3.035	1.576	0.535
Women	−2.152	1.408	−4.911	0.606	0.126	−1.663	1.190	−3.996	0.670	0.162
Hispanic	−2.534	2.102	−6.653	1.585	0.228	−1.744	1.763	−5.200	1.711	0.323
Black	0.940	2.164	−3.301	5.181	0.664	1.164	1.773	−2.310	4.638	0.511
Asian/Pacific	0.006	2.825	−5.531	5.544	0.998	−0.430	2.282	−4.903	4.043	0.850
Other races	0.388	0.543	−0.676	1.451	0.475	0.722	0.569	−0.392	1.837	0.204
Family background
x Age 30–34	−0.076	0.098	−0.268	0.116	0.439	−0.040	0.085	−0.207	0.128	0.641
x Women	−0.056	0.097	−0.247	0.135	0.567	−0.088	0.084	−0.253	0.078	0.299
x Hispanic	−0.105	0.158	−0.416	0.205	0.506	−0.073	0.140	−0.348	0.202	0.601
x Black	0.062	0.170	−0.272	0.395	0.717	0.068	0.146	−0.217	0.354	0.639
x Asian/Pacific	−0.177	0.198	−0.565	0.212	0.373	−0.151	0.189	−0.522	0.220	0.426
Years of schooling
x Age 30–34	−0.075	0.101	−0.272	0.123	0.458	−0.048	0.085	−0.214	0.118	0.569
x Women	0.101	0.102	−0.099	0.302	0.323	0.083	0.086	−0.086	0.252	0.337
x Hispanic	−0.174	0.143	−0.455	0.107	0.224	−0.160	0.125	−0.404	0.085	0.200
x Black	0.006	0.167	−0.321	0.333	0.970	−0.043	0.132	−0.302	0.216	0.745
x Asian/Pacific	−0.130	0.220	−0.561	0.302	0.555	−0.037	0.189	−0.407	0.333	0.844
Skills
x Age 30–34	0.010	0.005	0.000	0.021	0.048	0.007	0.004	−0.001	0.016	0.102
x Women	−0.002	0.005	−0.012	0.009	0.759	−0.002	0.004	−0.011	0.007	0.644
x Hispanic	0.020	0.008	0.004	0.035	0.017	0.016	0.007	0.002	0.030	0.028
x Black	−0.008	0.008	−0.023	0.008	0.341	−0.005	0.007	−0.018	0.008	0.431
x Asian/Pacific	0.005	0.011	−0.017	0.028	0.644	0.003	0.010	−0.017	0.023	0.774
Intercept	0.358	1.452	−2.487	3.203	0.805
Path	Model 1 (SEM: linear estimation for D with continuous E and S)					Model 2 (GSEM: nonlinear estimation for D with continuous E and S)
Path	B	SE	Lower	Upper	P	B	SE	Lower	Upper	P
Years of schooling
Family background	0.339	0.095	0.154	0.525	<0.001	0.339	0.095	0.154	0.525	<0.001
Age 30–34	0.004	0.231	−0.449	0.456	0.987	0.004	0.231	−0.449	0.456	0.987
Women	0.273	0.233	−0.183	0.730	0.241	0.273	0.233	−0.183	0.730	0.241
Hispanic	−1.073	0.371	−1.801	−0.345	0.004	−1.073	0.371	−1.801	−0.345	0.004
Black	−1.123	0.366	−1.840	−0.406	0.002	−1.123	0.366	−1.840	−0.406	0.002
Asian/Pacific	0.916	0.483	−0.030	1.863	0.058	0.916	0.483	−0.030	1.863	0.058
Other races	−1.541	0.600	−2.717	−0.364	0.010	−1.541	0.600	−2.717	−0.364	0.010
Family background
x Age 30–34	0.041	0.097	−0.150	0.232	0.674	0.041	0.097	−0.150	0.232	0.674
x Women	0.018	0.097	−0.173	0.209	0.851	0.018	0.097	−0.173	0.209	0.851
x Hispanic	0.053	0.156	−0.253	0.359	0.733	0.053	0.156	−0.253	0.359	0.733
x Black	0.255	0.163	−0.064	0.575	0.118	0.255	0.163	−0.064	0.575	0.118
x Asian/Pacific	0.003	0.191	−0.372	0.377	0.989	0.003	0.191	−0.372	0.377	0.989
Intercept	14.256	0.223	13.819	14.694	<0.001	14.256	0.223	13.819	14.694	<0.001
Skills
Family background	5.552	1.757	2.109	8.996	0.002	5.552	1.757	2.109	8.996	0.002
Age 30–34	−0.892	4.282	−9.285	7.501	0.835	−0.892	4.282	−9.285	7.501	0.835
Women	−2.470	4.318	−10.932	5.992	0.567	−2.470	4.318	−10.932	5.992	0.567
Hispanic	−21.365	6.884	−34.858	−7.872	0.002	−21.365	6.884	−34.858	−7.872	0.002
Black	−40.024	6.784	−53.321	−26.728	<0.001	−40.024	6.784	−53.321	−26.728	<0.001
Asian/Pacific	−16.382	8.954	−33.931	1.166	0.067	−16.382	8.954	−33.931	1.166	0.067
Other races	−32.120	11.124	−53.924	−10.317	0.004	−32.120	11.124	−53.924	−10.317	0.004
Family background
x Age 30–34	−0.885	1.806	−4.426	2.655	0.624	−0.885	1.806	−4.426	2.655	0.624
x Women	−0.318	1.807	−3.859	3.223	0.860	−0.318	1.807	−3.859	3.223	0.860
x Hispanic	6.140	2.895	0.467	11.814	0.034	6.140	2.895	0.467	11.814	0.034
x Black	4.383	3.024	−1.544	10.310	0.147	4.383	3.024	−1.544	10.310	0.147
x Asian/Pacific	6.930	3.541	−0.011	13.870	0.050	6.930	3.541	−0.011	13.870	0.050
Intercept	281.419	4.135	273.314	289.523	<0.001	281.419	4.135	273.314	289.523	<0.001
Path	Model 3 (GSEM: linear estimation for D with binary E and S)					Model 4 (GSEM: nonlinear estimation for D with binary E and S)
Path	B	SE	Lower	Upper	P	B	SE	Lower	Upper	P
Income deciles
Family background	0.052	0.097	−0.139	0.242	0.595	0.058	0.082	−0.102	0.217	0.481
Tertiary degree	1.304	0.456	0.409	2.198	0.004	1.145	0.376	0.408	1.883	0.002
High skills	0.023	0.606	−1.166	1.211	0.970	0.023	0.487	−0.932	0.977	0.963
Age 30–34	0.473	0.344	−0.201	1.147	0.169	0.339	0.282	−0.213	0.891	0.229
Women	−1.415	0.339	−2.079	−0.752	<0.001	−1.109	0.280	−1.658	−0.559	<0.001
Hispanic	−0.417	0.478	−1.354	0.520	0.383	−0.471	0.409	−1.273	0.331	0.250
Black	−0.947	0.497	−1.920	0.026	0.057	−0.683	0.399	−1.466	0.100	0.088
Asian/Pacific	−0.905	0.955	−2.776	0.966	0.343	−0.624	0.693	−1.982	0.735	0.368
Other races	0.086	0.564	−1.019	1.191	0.879	0.334	0.553	−0.750	1.419	0.546
Family background
x Age 30–34	−0.030	0.101	−0.229	0.169	0.767	−0.010	0.085	−0.176	0.155	0.901
x Women	−0.012	0.101	−0.210	0.185	0.903	−0.048	0.084	−0.213	0.116	0.564
x Hispanic	0.096	0.152	−0.201	0.394	0.526	0.033	0.128	−0.218	0.285	0.795
x Black	0.069	0.184	−0.292	0.430	0.708	0.076	0.148	−0.213	0.366	0.605
x Asian/Pacific	−0.175	0.199	−0.564	0.215	0.380	−0.174	0.178	−0.522	0.174	0.328
Tertiary degree
x Age 30–34	0.104	0.483	−0.843	1.051	0.829	0.145	0.391	−0.622	0.911	0.711
x Women	0.550	0.483	−0.397	1.497	0.255	0.324	0.393	−0.446	1.095	0.409
x Hispanic	1.083	0.711	−0.310	2.477	0.128	0.971	0.590	−0.185	2.128	0.100
x Black	−0.203	0.849	−1.866	1.460	0.811	−0.296	0.637	−1.545	0.952	0.642
x Asian/Pacific	0.597	1.160	−1.676	2.870	0.607	0.613	0.871	−1.095	2.321	0.482
High skills
x Age 30–34	1.016	0.670	−0.298	2.330	0.130	0.850	0.548	−0.224	1.924	0.121
x Women	−0.535	0.700	−1.907	0.836	0.444	−0.473	0.561	−1.573	0.627	0.400
x Hispanic	−1.708	1.442	−4.536	1.119	0.236	−1.363	1.134	−3.585	0.859	0.229
x Black	−0.308	1.705	−3.650	3.033	0.856	−0.559	1.335	−3.174	2.057	0.676
x Asian/Pacific	−0.076	1.286	−2.597	2.444	0.953	0.219	1.186	−2.107	2.544	0.854
Intercept	5.022	0.344	4.347	5.697	<0.001
Path	Model 3 (GSEM: linear estimation for D with binary E and S)					Model 4 (GSEM: nonlinear estimation for D with binary E and S)
Path	B	SE	Lower	Upper	P	B	SE	Lower	Upper	P
Tertiary degree
Family background	0.418	0.107	0.209	0.627	<0.001	0.418	0.107	0.210	0.627	<0.001
Age 30–34	−0.222	0.226	−0.665	0.222	0.327	−0.222	0.226	−0.665	0.222	0.327
Women	0.211	0.227	−0.234	0.655	0.354	0.211	0.227	−0.234	0.655	0.353
Hispanic	−0.748	0.335	−1.404	−0.091	0.026	−0.748	0.335	−1.404	−0.091	0.026
Black	−1.110	0.392	−1.878	−0.343	0.005	−1.110	0.392	−1.878	−0.343	0.005
Asian/Pacific	0.925	0.641	−0.331	2.181	0.149	0.925	0.641	−0.331	2.181	0.149
Other races	−1.334	0.596	−2.502	−0.167	0.025	−1.334	0.596	−2.502	−0.167	0.025
Family background
x Age 30–34	−0.134	0.110	−0.350	0.082	0.225	−0.134	0.110	−0.350	0.082	0.225
x Women	−0.031	0.108	−0.242	0.180	0.773	−0.031	0.108	−0.242	0.180	0.773
x Hispanic	−0.069	0.146	−0.355	0.218	0.638	−0.069	0.146	−0.355	0.218	0.638
x Black	0.366	0.214	−0.054	0.785	0.087	0.366	0.214	−0.054	0.785	0.087
x Asian/Pacific	0.262	0.324	−0.374	0.897	0.419	0.262	0.324	−0.374	0.897	0.419
Intercept	0.553	0.215	0.131	0.975	0.010	0.553	0.215	0.131	0.975	0.010
High skills
Family background	0.191	0.123	−0.050	0.431	0.120	0.191	0.123	−0.049	0.431	0.120
Age 30–34	−0.124	0.383	−0.874	0.627	0.746	−0.124	0.383	−0.874	0.627	0.746
Women	−0.688	0.405	−1.483	0.106	0.089	−0.688	0.405	−1.483	0.106	0.090
Hispanic	−0.781	0.698	−2.148	0.586	0.263	−0.781	0.697	−2.148	0.586	0.263
Black	−2.031	1.270	−4.521	0.458	0.110	−2.032	1.271	−4.523	0.459	0.110
Asian/Pacific	−0.190	0.725	−1.611	1.230	0.793	−0.190	0.725	−1.611	1.230	0.793
Other races	−13.670	575.29	−1141.22	1113.88	0.981	−13.426	509.15	−1011.33	984.48	0.979
Family background
x Age 30–34	0.065	0.135	−0.199	0.329	0.629	0.065	0.135	−0.199	0.329	0.629
x Women	0.198	0.137	−0.071	0.468	0.149	0.198	0.137	−0.071	0.468	0.149
x Hispanic	0.033	0.252	−0.462	0.527	0.897	0.032	0.252	−0.462	0.527	0.898
x Black	0.389	0.400	−0.395	1.173	0.331	0.389	0.400	−0.395	1.174	0.331
x Asian/Pacific	0.002	0.242	−0.472	0.475	0.994	0.002	0.242	−0.472	0.475	0.994
Intercept	−1.610	0.316	−2.230	−0.991	<0.001	−1.610	0.316	−2.230	−0.991	<0.001

Notes: The number of observations is 425 for all models. ‘Lower’ and ‘Upper’ mean the 95% confidence intervals (CI). SE, CI and the p-value are shown without bootstrapping and therefore they are slightly inconsistent with Table 2. See Table 2 for AIC/BIC for all models and cut points in Models 2 and 4. The reference for race/ethnicity is White. Some of the predictor variables show a relatively large SE and CIs, arguably because of the limited sample size. The bootstrap method therefore qualifies as a sensible approach as illustrated in Table 2.

Results

Table 2 summarises the key results of four models with bootstrapping. In Model 1 (linear estimation for D with continuous E and S), the indirect path via education (γ₁δ₁) shows a substantially positive sign (B = 0.136, 95%CI: 0.045 to 0.227, P = 0.003) whereas the direct origin–destination link (β) and the indirect route through skills (γ₂δ₂) do not demonstrate significant coefficients. The decomposition indicates this difference between the mediating roles of E and S is primarily due to the heterogeneous return to E and S, rather than their link with O. While OE (γ₁) and OS (γ₂) are both positive (B = 0.339, 95%CI: 0.173 to 0.506, P < 0.001 for OE; B = 5.552, 95%CI: 2.324 to 8.781, P = 0.001 for OS), the magnitude of SD (δ₂) is negligible unlike the ED (δ₁) association (B = 0.401, 95%CI: 0.221 to 0.581, P < 0.001 for ED; B = −0.001, 95%CI: −0.011 to 0.008, P = 0.796 for SD). Figure 2 (Panel A) also illustrates the standardised coefficients for each path of OESD to compare the relative magnitude of E and S. This confirms (1) OE and OS are almost identical, and (2) there is a significant gap between ED and SD, with the effect size of the latter being nearly zero as with OD.

Figure 2.

Strength of OESD paths. (a) Standardised coefficients for each path of OESD. (b) Predicted probability of possessing a tertiary degree and high skills by the O measure. (c) Predicted probability of income deciles by the possession of a tertiary degree and high skills.

The confounders (i.e. age, gender and race/ethnicity) and their interactions with O/E/S also display interesting variation (Table 3). In addition to oft-reported racial/ethnic inequalities (e.g. Black and Hispanic are less likely than White to obtain higher levels of education and skills), it is noteworthy that (1) the interaction between the origin measure and Hispanic shows a significantly positive coefficient for skills (B = 6.140, 95%CI: 0.467 to 11.814, P = 0.034), and (2) the one between skills and Hispanic indicates a positive sign for income (B = 0.020, 95%CI: 0.004 to 0.035, P = 0.017). This suggests that, while cognitive skills do not show a notable link with income in general, their possession does matter among Hispanics. Put differently, skills-based meritocratic rewards allocation takes place for this particular racial/ethnic group. Although the primary focus of the current article is on the overall difference in magnitude between E and S, future research must scrutinise these heterogeneities across socio-demographic attributes.

The same structure is observed when the outcome variable is treated as an ordinal scale in Model 2 (nonlinear estimation for D with continuous E and S). This includes (1) the notable contribution of family background to both E and S; (2) the substantial economic return to E without explicit rewards for S; and consequently (3) the significant mediating role of E, which is not the case for S (B = 0.115, 95%CI: 0.038 to 0.191, P = 0.003 for the OED path; B = 0.001, 95%CI: −0.047 to 0.049, P = 0.964 for the OSD path).

In Model 3 (linear estimation for D with binary E and S), both the significantly positive indirect path via E and the insignificant mediation by S are confirmed (B = 0.545, 95%CI: 0.016 to 1.075, P = 0.044 for OED; B = 0.004, 95%CI: −0.277 to 0.286, P = 0.976 for OSD). Model 4 (nonlinear estimation for D with binary E and S) also returns the identical result. Looking into the decomposition, one difference with Models 1 and 2 is that the association between family background and the possession of high skills does not show a clear sign, albeit being positive (B = 0.191, 95%CI: −0.048 to 0.429, P = 0.117 in Model 3; B = 0.191, 95%CI: −0.048 to 0.430, P = 0.118 in Model 4), whereas the significant OE and ED pathways are reconfirmed.

To illustrate these linkages, Figure 2 displays the average marginal effect of family background on educational attainment and skills acquisition (i.e. the probability of obtaining a tertiary degree and high skills by the O measure: Panel B); and that of E and S on income (i.e. the probability of reaching each income decile by the possession of a tertiary degree and high skills: Panel C). Panel B shows a clear upward trend for a tertiary degree in tandem with higher scores for family background, but this is not the case for high skills. Likewise, Panel C shows tertiary graduates are likely to attain mid–high income deciles with lower probabilities of falling into the low ranges, whereas high skills demonstrate a flatter distribution of chances across income deciles.

All these findings are corroborated by another set of robustness checks by excluding NBH from the O measure (see Tables A1 and A2).

Discussion and Conclusion

Vast evidence shows that an individual’s family background significantly determines his/her socio-economic outcomes directly or indirectly via education. Despite this socially constructed unfair structure, the discourse of meritocracy operates in justifying the divide between winners who can, and losers who cannot, enjoy favourable rewards. This happens particularly when such an unequal rewards allocation is framed as a consequence of meritocratic competition, where educational attainment is often used as a proxy for one’s merit (Mijs, 2016). Against such a backdrop, recent studies put an emphasis on renewing the dignity of work and other outcomes as a moral framework, rather than attempting to weaponise everyone with higher levels of education to compete under the meritocratic doctrine, in which even the definition of merit can be set arbitrarily (Sandel, 2020).

Although this philosophical discussion as such is valuable, one fundamental assumption has been inadequately examined in an empirical manner: meritocracy is at work in our society. In particular, regardless of the extent to which origin exerts its influence over children’s education and destination, we know little about whether benefits are distributed according to an individual’s merit. Given the observed significant association between education and socio-economic outcomes, one may simply conclude that we live in a meritocratic society at least in the rewards allocation phase. However, considering recent arguments that the level/impact of education and that of skills/abilities are inconsistent (Araki, 2020, 2022), the seemingly positive education–destination linkage does not necessarily prove the existence of meritocracy. This is because, as a hypothetical case, a highly skilled person without a college degree might be penalised socio-economically when a college graduate unaccompanied by high skills enjoys favourable returns – such a typical credential society where credentialism prevails over meritocracy.

The current study thus analyses the relationship between origin, education, skills and destination (i.e. the OESD quadrangle) using the OECD PIAAC data. SEM and GSEM reveal that (1) the strengths of OE and OS links are identical; but (2) the magnitude of SD is negligible whereas the ED association is substantial; and consequently (3) the relative role of education far outweighs that of skills in intergenerational inequality. This structure is robust regardless of model specification. In the sense that skills are not explicitly rewarded, especially as compared with educational attainment, it is logical to argue that credentialism, not meritocracy, operates in contemporary USA. Nevertheless, arguably by conflating education and skills without close attention to their discrepancy, we tend to misbelieve that our society is meritocratic and accordingly, social inequalities can be legitimised, at least at the conceptual level, based on such an ‘imagined’ discourse of meritocracy.

As argued, this empirical evidence does not undermine the philosophical merit of some arguments that the dignity should be restored because, whether it is meritocracy or credentialism that drives intergenerational inequality, their focus is on how we perceive socio-economic outcomes including the winner–loser divide. Nonetheless, from the sociological perspective, it is imperative to accurately understand the social mechanisms whereby family background exerts its influence and rewards are allocated lest we mislead scholarly and policy discussions as well as public debate. Indeed, so long as we precisely frame the United States as a credential society without considering the imagined meritocracy as a substantive social force, one cannot simply legitimise the returns to higher levels of education as proper awards for his/her merit, especially when they are backed by their origin.

To this end, future research must extend the scope of this article. First, diverse measures should be incorporated for origin (e.g. parental occupation), education (e.g. fields of study, institutional prestige and their (mis)matching), skills (e.g. noncognitive and occupational skills), destination (e.g. non-economic outcomes), confounders (e.g. nativity and genetic information) and societal-level conditions (e.g. labour market structure and social policy). Second, longitudinal data collection and analyses are necessary, preferably with the life-course perspective (Birkelund et al., 2022; Cheng et al., 2021; Hällsten and Yaish, 2021; Yaish et al., 2021). Third, comparative studies are essential to better understand the OESD structure including its cross-national similarities and differentials. The framework and findings demonstrated here would be a foundation for such elaboration.

Supplemental Material

sj-pdf-1-soc-10.1177_00380385231156093 – Supplemental material for Beyond ‘Imagined Meritocracy’: Distinguishing the Relative Power of Education and Skills in Intergenerational Inequality

Supplemental material, sj-pdf-1-soc-10.1177_00380385231156093 for Beyond ‘Imagined Meritocracy’: Distinguishing the Relative Power of Education and Skills in Intergenerational Inequality by Satoshi Araki in Sociology

Footnotes

Acknowledgements

I would like to thank Takehiko Kariya, Richard Breen, Jan O Jonsson, Herman van de Werfhorst, Yuki Honda, Hideki Hirota and Mai Araki for their invaluable comments. The computations were performed using research computing facilities offered by Information Technology Services, the University of Hong Kong.

Funding

The author disclosed receipt of the following financial support for the research, authorship and/or publication of this article: this research was supported by the Seed Fund for Basic Research, the University of Hong Kong [104006736].

ORCID iD

Satoshi Araki

Supplemental material

Supplemental material for this article is available online.

Satoshi Araki is an Assistant Professor in the Department of Sociology, the University of Hong Kong. His research centres on social stratification and inequality, education and work, well-being and research methods. His work has appeared in the American Sociological Review, European Sociological Review, Higher Education, Journal of Happiness Studies, International Journal of Comparative Sociology and Journal of Applied Gerontology, among others.

References

Albert

Geng

Nelson

(2016) Causal mediation analysis with a latent mediator. Biometrical Journal 58(3): 535–548.

Andersen

Lue Kessing

Østergaard

(2021) ‘We have equal opportunities – in theory’: Lay perceptions of privilege, meritocracy and inequality in Denmark. Sociology 55(6): 1117–1134.

Araki

(2020) Educational expansion, skills diffusion, and the economic value of credentials and skills. American Sociological Review 85(1): 128–175.

Araki

(2022) Beyond the high participation systems model: Illuminating the heterogeneous patterns of higher education expansion and skills diffusion across 27 countries. Higher Education. Epub ahead of print 22 August 2022. DOI: 10.1007/s10734-022-00905-w.

Araki

Kariya

(2022) Credential inflation and decredentialization: Re-examining the mechanism of the devaluation of degrees. European Sociological Review 38(6): 904–919.

Autor

(2014) Skills, education, and the rise of earnings inequality among the ‘other 99 percent’. Science 344(6186): 843–851.

Bartus

(2017) Multilevel multiprocess modeling with gsem. Stata Journal 17(2): 442–461.

Birkelund

Karlson

Yaish

(2022) Life-cycle economic returns to educational mobility in Denmark. Sociology 56(6): 1121–1139.

Brand

(2010) Civic returns to higher education: A note on heterogeneous effects. Social Forces 89(2): 417–433.

10.

Breen

(2003) Is Northern Ireland an educational meritocracy? Sociology 37(4): 657–675.

11.

Breen

(2010) Educational expansion and social mobility in the 20th century. Social Forces 89(2): 365–388.

12.

Breen

Jonsson

(2007) Explaining change in social fluidity: Educational equalization and educational expansion in twentieth-century Sweden. American Journal of Sociology 112(6): 1775–1810.

13.

Breen

Müller

(eds) (2020) Education and Intergenerational Social Mobility in Europe and the United States. Studies in Social Inequality. Stanford, CA: Stanford University Press.

14.

Breen

Karlson

Holm

(2013) Total, direct, and indirect effects in logit and probit models. Sociological Methods and Research 42(2): 164–191.

15.

Breen

Karlson

Holm

(2018) Interpreting and understanding logits, probits, and other nonlinear probability models. Annual Review of Sociology 44: 39–54.

16.

Bukodi

Goldthorpe

(2018) Social Mobility and Education in Britain: Research, Politics and Policy. Cambridge: Cambridge University Press.

17.

Checchi

van de Werfhorst

(2018) Policies, skills and earnings: How educational inequality affects earnings inequality. Socio-Economic Review 16(1): 137–160.

18.

Cheng

Brand

Zhou

, et al. (2021) Heterogeneous returns to college over the life course. Science Advances 7(51): 1–15.

19.

Chmielewski

(2019) The global increase in the socioeconomic achievement gap, 1964 to 2015. American Sociological Review 84(3): 517–544.

20.

Collins

(1979) The Credential Society: An Historical Sociology of Education and Stratification. New York, NY: Academic Press.

21.

Deming

(2017) The growing importance of social skills in the labor market. The Quarterly Journal of Economics 132(4): 1593–1640.

22.

Engzell

(2021) What do books in the home proxy for? A cautionary tale. Sociological Methods and Research 50(4): 1487–1514.

23.

Fiel

(2020) Great equalizer or great selector? Reconsidering education as a moderator of intergenerational transmissions. Sociology of Education 93(4): 353–371.

24.

Goldthorpe

(2003) The myth of education-based meritocracy: Why the theory isn’t working. New Economy 10(4): 234–239.

25.

Hällsten

Yaish

(2021) Intergenerational educational mobility and life course economic trajectories in a social democratic welfare state. European Sociological Review 38(4): 507–526.

26.

Hanushek

Woessmann

(2011) The economics of international differences in educational achievement. In: Hanushek

Machin

Woessmann

(eds) Handbook of the Economics of Education, 1st edn. Amsterdam, The Netherlands: North-Holland, 89–200.

27.

Heckman

Stixrud

Urzua

(2006) The effects of cognitive and noncognitive abilities on labor market outcomes and social behavior. Journal of Labor Economics 24(3): 411–482.

28.

Hout

(1984) Status, autonomy, and training in occupational mobility. American Journal of Sociology 89(6): 1379–1409.

29.

Hout

(1988) More universalism, less structural mobility: The American occupational structure in the 1980s. American Journal of Sociology 93(6): 1358–1400.

30.

Hout

(2011) Social and economic returns to college education in the United States. Annual Review of Sociology 38: 379–400.

31.

Karlson

Birkelund

(2019) Education as a mediator of the association between origins and destinations: The role of early skills. Research in Social Stratification and Mobility 64: 100436.

32.

Karlson

Birkelund

(2022) Family background, educational qualifications, and labour market attainment: Evidence from Danish siblings. European Sociological Review 38(6): 988–1000.

33.

Kerckhoff

Raudenbush

Glennie

(2001) Education, cognitive skill, and labor force outcomes. Sociology of Education 74(1): 1–24.

34.

(2021) Vertical education-occupation mismatch and wage inequality by race/ethnicity and nativity among highly educated US workers. Social Forces 100(2): 706–737.

35.

Mijs

JJB

(2016) The unfulfillable promise of meritocracy: Three lessons and their implications for justice in education. Social Justice Research 29(1): 14–34.

36.

Mijs

JJB

(2021) The paradox of inequality: Income inequality and belief in meritocracy go hand in hand. Socio-Economic Review 19(1): 7–35.

37.

OECD (2015) Skills for Social Progress: The Power of Social and Emotional Skills. Paris, France: OECD Publishing.

38.

OECD (2019a) Skills Matter: Additional Results from the Survey of Adult Skills. Paris, France: OECD Publishing.

39.

OECD (2019b) The Survey of Adult Skills: Reader’s Companion, 3rd edn. Paris, France: OECD Publishing.

40.

Pfeffer

Hertel

(2015) How has educational expansion shaped social mobility trends in the United States? Social Forces 94(1): 143–180.

41.

Posselt

Grodsky

(2017) Graduate education and social stratification. Annual Review of Sociology 43: 353–378.

42.

Preacher

(2015) Advances in mediation analysis: A survey and synthesis of new developments. Annual Review of Psychology 66: 825–852.

43.

Rijnhart

JJM

Valente

MacKinnon

, et al. (2017) The use of traditional and causal estimators for mediation models with a binary outcome and exposure-mediator interaction. Structural Equation Modeling 176(3): 139–148.

44.

Sandel

(2020) The Tyranny of Merit: What’s Become of the Common Good? New York, NY: Farrar, Straus and Giroux.

45.

Sullivan

Parsons

Green

, et al. (2018) The path from social origins to top jobs: Social reproduction via education. British Journal of Sociology 69(3): 776–798.

46.

Sullivan

Parsons

Ploubidis

, et al. (2020) Pathways from origins to destinations: Stability and change in the roles of cognition, private schools and educational attainment. British Journal of Sociology 71(5): 921–938.

47.

Torche

(2011) Is a college degree still the great equalizer? Intergenerational mobility across levels of schooling in the United States. American Journal of Sociology 117(3): 763–807.

48.

Torche

(2018) Intergenerational mobility at the top of the educational distribution. Sociology of Education 91(4): 266–289.

49.

Witteveen

Attewell

(2020) Reconsidering the ‘meritocratic power of a college degree’. Research in Social Stratification and Mobility 66: 100479.

50.

Yaish

Shiffer-Sebba

Gabay-Egozi

, et al. (2021) Intergenerational educational mobility and life-course income trajectories in the United States. Social Forces 100(2): 765–793.

51.

Young

(1958) The Rise of the Meritocracy. London: Thames and Hudson.

52.

Zhou

(2019) Equalization or selection? Reassessing the ‘meritocratic power’ of a college degree in intergenerational income mobility. American Sociological Review 84(3): 459–485.

53.

Zhou

(2022) Attendance, completion, and heterogeneous returns to college: A causal mediation approach. Sociological Methods and Research. Epub ahead of print 1 August 2022. DOI: 10.1177/00491241221113876.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.30 MB