Desktop version

Home arrow Management arrow Improving Access and Quality of Public Services in Latin America: To Govern and To Serve

Quantitative Analysis

Our goal was to compare the performance of students in Uruguay and Chile across some institutional features (those that show enough within- country variation, as well as between both countries), after controlling for those individual characteristics and school resources that clearly affect education performance.

Table 5.2 Mixed-methods approach with quantitative and qualitative techniques

Dimension

Methodology

Sources

Results

Governance

Historical

Secondary

Outline the main features of

settings

analysis

references, literature

the reforms carried on in the

review

1980s and 1990s.

Equity and

Quantitative

PISA 2009

Compare both countries in

quality (DV)

Qualitative

Interviews Chile/

terms of quality and equity of

Uruguay

education

Governance

Quantitative

PISA 2009

Estimate the impact of

factors (IV)

Qualitative

In-depth semi-

divergent institutional factors

structured

and school organization on

interviews

student performance. Understand the mechanisms

through which the different factors operate.

DV: Dependent variable; IV: Independent variable Source: Elaborated by authors

The quantitative approach consisted of estimations of an EPF using OLS at the individual (student) level. In this way we could directly link a student’s performance to his/her teaching environment, and control for individual background influences on student performance, as well as the influence of school resources and teacher characteristics, and the possible influence of some relevant institutional features.

However, many difficulties arise when trying to analyze the factors behind education performance, especially contamination of the OLS results by endogeneity and selection biases, mainly because of important unobserved (omitted) variables. Some of these problems may be overcome by using panel data, but this was not possible because:

a. PISA is not a panel database, so we cannot follow individuals through time. Therefore, we cannot compute differences in outcomes over time at the individual level.

b. We explored the possibility of following schools (not individuals), but this too is a problem as PISA sampling methods assure the sample is representative at the national and some subnational levels, but not at the school level.

Therefore, estimated coefficients do not show a causal relationship, but rather a global association between each governance factor (or group of variables that describe a governance factor) and performance. The purpose is not to establish causality, but rather to get a better description of the situation in both countries.

Regarding the multicollinearity issue between institutional factors, it is certainly a problem in our data. This is because some institutional factors tend to be jointly implemented. For example, private provision is almost always associated with school autonomy regarding allocation of resources. It should be noted that the effect on performance of an institutional factor may come not only directly by the factor itself (e.g., introducing more school autonomy regarding resource allocation), but also by the indirect effect of other variables (e.g., if more school autonomy comes along with more intensive use of achievement data for relevant decisions). In the latter case, if we introduce the two covariates (or estimate conditional effects), this could introduce an endogeneity problem since the last covariate could be affected by the first.

Taking into consideration the aforementioned limitations, the quantitative analysis was carried out in two sequential stages. First, we undertook a descriptive analysis to characterize each type of school in both countries and identified sources of variation to be exploited. We also described performance in both countries across socioeconomic groups. Second, using the PISA 2009 pooled database of Chile and Uruguay, we estimated separate regressions for each set of variables describing a governance factor, using the full sample, separate samples for public and private schools, and separate samples for different socioeconomic groups, in all cases with necessary controls. This was followed by an analysis of the difference in academic achievement between the two countries (full regression and Oaxaca decomposition).

Given the importance of school progression on performance and the large differences between Chile and Uruguay, we also estimated Probit models for school progression on the full sample and by subsamples according to quartiles of school socioeconomic status.

 
Source
< Prev   CONTENTS   Source   Next >

Related topics