Inequality of Educational Opportunity and Time-Varying Circumstances: Longitudinal Evidence from Peru

Abstract

This paper examines the extent to which characteristics that are beyond the control of children affect their educational outcomes. This is a matter of particular interest because the distribution of educational opportunities will shape future outcomes in other realms. While time-invariant circumstances have already been examined in the inequality of opportunity (IOp) literature, the role of time-varying circumstances has not yet been addressed. For the first time, this paper provides both lower and upper-bound estimates of IOp on learning achievement and assesses the impact of time-varying circumstances on upper-bound measures. It exploits a very rich and unusual longitudinal data set, the Young Lives Study, focusing on a cohort of children that has been followed for fifteen years, surveyed for the first time when they were around a year old. The results suggest that educational IOp is sizable and time-varying circumstances do not have a major impact on upper-bound measures using panel data.

Keywords:

• Educational economics
• human capital
• inequality of opportunity
• learning achievement

Acknowledgements

The author is grateful to Rémi Bazillier, Javier Herrera, Marta Menéndez, Josselin Thuilliez, Gustavo Yamada, as well as the participants of the 2018–2019 Development Economics seminar at Paris 1 University. They provided helpful ideas and comments during the elaboration of this paper. Special thanks to Alejandra Miranda for her insightful clarifications concerning the database, Henry Höschler for his careful revision of a previous draft, and three anonymous referees for their valuable comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data used in this study come from Young Lives, an international study on the changing nature of child poverty in Ethiopia, India, Peru, and Vietnam (http://www.younglives.org.uk). The data sets are publicly archived and available to download from the UK Data Archive. The code that supports the findings of this study is available from the author, upon reasonable request.

Notes

1 Recent reviews include Cueto and Felipe (2018), Cueto, Miranda, and Vásquez (2016), Guadalupe, León, Rodríguez, and Vargas (2017), and Ñopo and Kitmang (2017).

2 It is noteworthy that the source of inequality matters from an ethical point of view. Indeed, most ‘would agree that effects of circumstances on persons’ well-being that are beyond the control of individuals should be rectified, while at least some differential outcomes due to choice are not compensable at the bar of justice’ (Roemer & Trannoy, 2015, p. 294).

3 As a matter of fact, the General Education Law N^o 28044 (2003) states that education is a fundamental right and a free-cost public service when provided by the state, which ensures the right to an integral, high-quality, and universal education for every person.

4 That is, the fact that the full set of circumstances is not observed in the data.

5 Let us define ‘time-varying circumstances’ as those circumstances that have a high probability to change over the lifetime of a human being.

6 Some seminal philosophical works are those of Arneson (1989, 1990), Cohen (1989), and Dworkin (1981a, 1981b).

7 cf. Fleurbaey and Peragine (2013) and Ramos and van de Gaer (2016) for a compelling discussion.

8 The surveys were carried out in 2002, 2006, 2009, 2013, and 2016. Additionally, a school survey was carried out in 2010 for a subsample of 572 children from the younger cohort distributed across 132 primary schools. For more details, cf. Appendix A.

9 Indeed, the YLS sample ‘covers the full diversity of children in Peru in a wide variety of attributes and experiences. Therefore while not suited for simple monitoring of child outcome indicators, the Young Lives sample is an appropriate and valuable instrument for analyzing causal relations and modeling child welfare and its longitudinal dynamics in Peru’ (Escobal & Flores, 2008, p. iv).

10 However, it is noteworthy that the richest five percent of districts were excluded from sampling due to the goal of oversampling poor areas. As a consequence, our IOp estimates will likely be downward biased.

11 It is important to mention that in this paper the outcomes are measured as z-values of the raw scores. Indeed, the Rasch scores are not available for rounds 4 and 5 of the Peruvian sample. Regardless, rounds 2 and 3 show that both the raw and Rasch scores are strongly correlated ($\rho >0.95$).

12 Similar technical notes for rounds 4 and 5 are forthcoming.

13 In fact, the only index that respects the axioms of anonymity, normalization, population replication, scale invariance, subgroup decomposability, path-independent decomposability, and the Pigou-Dalton transfer principle, is the mean log deviation $\mathit{MLD}=\frac{1}{N}\sum _{i}ln\frac{{\mu }_y}{y_i}$ (Foster & Shneyerov, 2000).

14 The log transformation of income or earnings is a common practice since it is usually more normally distributed than the original variables, which are generally highly right-skewed. In contrast, the test score and normal distributions usually look alike. Moreover, test scores are typically ‘constructed from the raw results by means of Item Response Theory (IRT) models, which attempt to account for ‘test parameters’, so as to better infer true learning. This process generates an arbitrary metric for test scores, which are then typically standardized to some arbitrary mean and standard deviation’ (Ferreira & Gignoux, 2014, p. 212).

15 In addition, some basic descriptive statistics are provided in Appendix S.II in the Supplementary Materials.

16 The counterfactual distribution is interpreted, in the ex-ante approach, as the opportunity set for individuals belonging to the same type.

17 In any case, Brunori et al. (2019) argue that their method is ‘preferable when the intent is to compare the level of IOp in two populations’ (p. 645), which is not the case here.

18 As in section 4.1, the original logs are omitted for the dependent variable.

19 As noted by Ferreira and Gignoux (2014, p. 231), ‘the mean log deviation is not ordinally invariant in the standardization to which test scores are submitted’ in the context of item response theory (IRT) models. As a consequence, the MLD is not suitable for the present study.

20 For instance, using a Peruvian subsample of the Young Lives Study, Cueto et al. (2017) found that ‘students’ socioeconomic status at age 1 and maternal education were positively associated with their teachers’ PCK [pedagogical content knowledge] by the time students were enrolled in fourth grade’ (Cueto et al., 2017, p. 329).

21 This notion can be seen as ‘the non-arbitrary and morally significant line between childhood and adulthood and that children are not responsible for their preferences in the way that adults are deemed to be’ (Arneson, 1990, p. 179).

22 It is worth mentioning that when and how autonomy is evinced seem to be culture specific. For example, ‘autonomy, individuality, and personal freedom are strong cultural values in the United States and in most Western industrialized societies. Within this tradition, parents generally socialize their children to make their own decisions, parents will expect young adolescents to begin to demonstrate autonomy and take on additional responsibility, adolescents will have increasing desires for individual rights and responsibility, and older adolescents will move out of the home of origin and attempt to make it on their own. However, these expectations and desires vary among cultures within and outside of the United States’ (Zimmer-Gembeck & Collins, 2007, p. 193).

23 According to Ramos and van de Gaer (2017), among the three main measurement criteria—ex-ante/ex-post, direct/indirect, and parametric/non-parametric—it seems that the former choice is the most relevant since it substantially influences IOp orderings.

24 It may also be noticed that time-varying effort variables are needed to compute the upper bounds of IOp within the Niehues and Peichl (2014) framework; otherwise, the upper bounds would be 100 per cent.

25 Observable circumstances from the data set do not show high correlations with these variables. Certainly, some unobserved circumstances, such as parents’ attitudes may influence them. They are not perfect measures of exerted effort but they do reflect it to some extent.

26 Unfortunately, I was not given access to the YLS database with the child’s school identifier, but it does exist.