Home Economics Economic Insights on Higher Education Policy in Ireland: Evidence from a Public System
Data and Methods
As noted in Flannery and O’Donoghue (2016), the main data requirements for calculating the net private and fiscal returns as specified in Equations (4) and (5) are a detailed micro-level dataset and an associated tax-benefit microsimulation model.3 The data for our analysis comes from the Irish component of EU-SILC. This is a cross-sectional and longitudinal micro dataset containing income, social, demographic and labour market variables at the individual and household levels. The data has been collected on an annual basis since 2003 with the estimates in this analysis using the information from the 2014 wave. The data is collected from a representative population sample from across Ireland and is weighted to reflect independent population estimates and to correct for possible attrition. The data is collected on an annual basis with the 2014 wave having over 12,000 observations, over 9000 of which are aged over 16 years.
This data specifically includes information on an individual’s highest level of education attained across six categories, namely primary education, lower secondary, upper secondary, post Leaving Certificate, third level non-degree and third level degree or above. Flannery and O’Donoghue (2016) provide a helpful step-by-step guide to estimating Equations (4) and (5) and an adapted version of these steps is outlined here:
The private (Eg) and public (Eg) costs of education are also required to calculate our private and fiscal returns to third level education. To facilitate this we use expenditure per student at tertiary level education from HEA (2014). To separate the burden of this cost across private/public contributions we multiply by the public/private share as outlined in HEA (2014).4 The annual private and public cost figures are then multiplied by 3.55 to obtain the costs in changing education levels from upper secondary to tertiary.
The indirect costs of education for the private returns (p_es x Yn) is measured using the cross-sectional weighted averages of earnings (Yn) of those aged 18-22 years with upper secondary as their highest level of education attained, in work and not in education. To obtain our finalised foregone earnings measure, this is then multiplied by an employment probability (p_es), calculated as the probability of being employed when aged 18-22 years and having upper secondary as one’s highest level of education attained.
For the indirect costs relating to the public returns to education, a similar methodology is used. However, it is the foregone taxes, benefits and social contributions that are needed. To this end, the tax and social contribution rules to the level of foregone earnings calculated above are applied and used in Equation (5). The foregone benefit term bYn is specified as the average benefit received from those in work reduced by the average benefit received by individuals while in education and in work. This completes the terms required to calculate each of the cost elements of the fiscal and net private returns to education.
The estimation of the non-pecuniary returns to education follows a more simplified approach. The data used comes from the Irish module of the European Social Survey (ESS) for 2014. Much like the SILC data, the ESS is cross-sectional microdata. However, unlike the SILC dataset it contains detailed information on a variety of subjective well-being measures, such as indices of happiness and health. It also collects information on education, demographic and income variables at the individual level. The data has been collected on a bi-annual basis since 2002 and samples just over 2000 (2390 for 2014) representative individuals in Ireland for each wave.
The subjective indicator of happiness6 within the ESS is segmented into 11 categories (0-10), ranging from extremely unhappy (0) to extremely happy (10). To explore the possible correlation between level of education and self-reported happiness, we estimate an ordered probit model with the 11 indicators of happiness as the dependent variable. This is regressed against highest level of education attained, with other factors such as income group, gender, age and parental education level included as control variables.
The indicator of health7 is broken into five categories (1-5) within the ESS, ranging from very bad (1) to very good (5). However, few people indicated that their health status is within the bottom two categories—only 2.7% of the sample cumulatively. Therefore, for our analysis we follow the approach of Oreopoulos and Salvanes (2011) and make the distinction between only those that indicate very good health and those that do not indicate that they are in very good health. We then use a binary probit model to estimate the correlation between whether an individual indicates they are in very good health and level of education. Other explanatory variables include income group, gender, age, parental education level and a measure of body mass index based upon self-reported height and weight measurements within the ESS.
|< Prev||CONTENTS||Next >|