An Economic Analysis of Higher Education Participation in Ireland

Progression to Higher Education

In this sub-section we consider in detail three microeconometric analyses of progression to third-level education in Ireland: Denny (2014), Cullinan et al. (2013) and Flannery and Cullinan (2014). The first of these papers drew on pooled School Leavers Survey (SLS) data from 1994 to 1998 inclusive. These years were chosen as they bracket the abolition of university fees for undergraduates in 1996,4 which was the focus of the paper. The SLS was based on a stratified random sample of those leaving the second-level system, with respondents interviewed between 20 and 26 months after leaving school. The survey collected a wide range of individual, school, income, social, demographic, education and labour market related information (see Byrne et al. 2008 for further details). For example, it included details of the current education and/or labour market activities of respondents and thus allowed for identification of those school leavers in the sample who made the transition to higher education (or not). It was also possible to identify which HEI an individual chose to study at (if they did).

The analysis consisted of a series of probit models where the dependent variable was whether a student progressed to university (or not). The focus was on the socioeconomic background (as measured by the father’s occupation) and the second-level educational attainment of respondents, although some demographic controls were also included. Table 2.3 is based on Table 3 of Denny (2014)5 and reports marginal effects. Note that there are other ways of measuring socioeconomic background with this data. While there is no information on family income, we observe mothers’ occupational group and the educational level of both parents.

Table 2.3 Probit models of attending university in Ireland

Model (1)

Model (2)

Model (3)

Points/100

0.055*** (7.07)

0.055*** (7.04)

No. of honours

0.044*** (9.87)

0.044*** (9.88)

No. of fails

-0.032* (2.38)

-0.032* (2.34)

Father professional

0.305*** (8.03)

0.041 (1.85)

0.023 (0.80)

Father other white collar

0.114*** (6.67)

0.0068 (0.63)

0.016 (0.97)

Father skilled manual

-0.0104 (0.56)

-0.0081 (0.68)

0.007 (0.39)

Father unemployed 'Free fees' x Father professional

-0.074*** (3.79)

-0.0066 (0.41)

-0.021 (1.15) 0.031 (0.74)

'Free fees' x Father other white collar

-0.015 (0.79)

'Free fees' x Father skilled manual

-0.025 (1.28)

'Free fees' x Father unemployed

0.0467 (0.91)

Father disabled

-0.074* (2.21)

-0.009 (0.31)

-0.009 (0.33)

Mother disabled

-0.065 (1.42)

-0.040* (2.04)

-0.039 (1.92)

Parent dead

-0.026 (1.03)

0.034 (1.43)

0.034 (1.410)

Age

-0.074*** (11.44)

-0.019*** (4.49)

-0.019*** (4.47)

Urban

0.183*** (10.61)

0.047*** (4.14)

0.047*** (4.12)

Woman

0.057*** (5.09)

-0.0017 (0.25)

-0.001 (0.25)

Pseudo R2

0.138

0.481

0.482

Notes: n = 4983. *p < 0.05, **p < 0.01, ***p < 0.001. Absolute t statistics in parentheses. Year and region dummies not shown. Estimation is by probit and marginal effects are shown Source: Adapted from Denny (2014)

The simple formulation used in this case was to facilitate comparisons of the socioeconomic gradient across specifications.

The strong socioeconomic gradient in progression was illustrated by the estimated parameters for fathers’ occupational grouping. Relative to the omitted category (semi- and un-skilled manual), the children of professional fathers were approximately 31% more likely to progress to university. For children of ‘other white collar’ fathers the difference was smaller, at about 11%. It is interesting that there were no statistically significant differences between the blue collar/manual groups. Given that there were likely to be substantial differences in income amongst manual workers, this argues against a simple income-based explanation of differences in progression. On the other hand, the children of fathers who were unemployed (and hence have lower income) were about 7% less likely to progress to university. One other interesting finding is that having a father who is disabled also had a significant negative effect on the probability of going to university. This is a reminder that there are other forms of disadvantage than the socioeconomic variety.

These results are cast in a very different light by those in Model (2) which adds measures of attainment in the Leaving Certificate, specifically the total number of points scored, the number of ‘honours’6 achieved and the number of papers failed. All of these had the effect on the outcome that one would expect, but what is more important is its effect on the socioeconomic gradient which essentially disappears: fathers’ occupational group and unemployment status were no longer statistically significant and the coefficients were much smaller. Interestingly, the effects of being female and of having a disabled father also became statistically insignificant. In short, second-level attainment helped explain much of the socioeconomic gradient, amongst other things. The paper also estimated (for those who were going to university) an ordered probit model of the prestige of the university attended (using the Shanghai rankings). Because of space constraints we do not show the results here (see Table 5 of Denny 2014). What is notable is that the socioeconomic background effects remained even after conditioning on Leaving Certificate results. This may partly reflect subject mix (e.g. the lower-ranked universities did not have a medical school), but conceivably young people’s aspirations and self-confidence are also affected by their upbringing.

One of the objectives of Denny (2014) was to assess the effect, if any, of the abolition of university fees (commonly known as the ‘free fees’ reform) given the government’s stated objective that it would reduce inequalities in accessing third-level education. The paper tested this by interacting the socioeconomic background variables with a dummy variable indicating the post reform period (after 1995)—see Model (3) of Table 2.3. One is unable to reject the hypothesis that there was no change in these coefficients (p = 0.26), that is, the socioeconomic gradient was unchanged after the reform. This is hardly surprising for two reasons. First, low-income students would generally not have been paying fees as they would have been in receipt of the means-tested higher education grant. Thus, effectively, fees were abolished for better-off students only. Second, the results in Model (2) show that it is secondary school attainment (Leaving Certificate results) that largely drives the socioeconomic gradient in access and this would have been unaffected by the reform.

Cullinan et al. (2013) also employed the SLS but used more recent data, the 2007 wave. The paper used a broader concept of participation, as opposed to solely university participation. A binary logit model was estimated with a dependent variable taking a value of one if an individual participates in higher education and a value of zero otherwise. Table 2.4 presents a slight variant of the results in their Table 3 with controls for a range of individual, spatial and school-level factors. Specifically, these included the socioeconomic background of the young person, second- level attainment and a measure of teacher engagement.7 The latter variable was constructed using principal components analysis from responses to a series of questions within the survey asking students to rate the competencies of their teachers in their last year of upper secondary education. These questions included the ability of the teacher to keep order in class and the availability of teachers to talk to the student. Socioeconomic background is based on father’s occupation but is specified somewhat differently to Denny (2014): the omitted category is ‘higher or lower professionals’ with one dummy variable for ‘other white collar and skilled manual’ and a second for ‘semi- and unskilled’. For some consistency with Table 2.3, the results are presented with and without a control for attainment in the Leaving Certificate.

The results are broadly similar to those using the older SLS data in Table 2.3 in terms of the effect of social class; including upper secondary attainment significantly dampens the effect that socioeconomic background may have on progression to higher education. They show that children of ‘other white collar and skilled manual’ fathers are no longer less likely to participate in higher education compared to children of those in the ‘higher or lower professionals’ social class. While the results indicate that those in the lowest social class grouping (semi- and un-skilled) still have a 7% lower probability of participating in higher education relative to children of ‘professional’ fathers, the effect is more than halved when Leaving Certificate attainment is included.

Table 2.4 Binary logit models of higher education participation

Variable

ME

z

ME

z

Social Class II

-0.101***

(3.52)

-0.035

(1.46)

Social Class III

-0.172***

(4.7)

-0.071**

(2.46)

CAO Points

-

-

0.001***

(17.23)

Distance to Nearest HEI

-0.000

(0.3)

-0.001

(1.45)

Midlands Region

-0.053

(0.81)

-0.099*

(1.75)

Western Region

-0.001

(0.02)

0.007

(0.12)

Dublin Region

0.009

(0.16)

-0.009

(0.19)

Mid-East Region

-0.089

(1.5)

-0.084*

(1.96)

Mid-West Region

0.023

(0.41)

-0.012

(0.23)

South-East Region

0.000

(0.01)

-0.030

(0.76)

South-West Region

0.079

(1.43)

0.002

(0.04)

Youth Employment Rate

-1.884**

(2.16)

-0.904

(1.2)

Gender

-0.022

(0.37)

-0.001

(0.03)

Grinds

0.133***

(5.21)

0.049**

(2.24)

Teacher Engagement

0.041***

(5.33)

0.017**

(2.53)

Enrolment mix is female only

0.034

(0.66)

-0.008

(0.17)

Enrolment mix is male only

-0.001

(0.02)

-0.010

(0.28)

Church of Ireland sponsored

-0.157*

(1.79)

-0.081

(0.89)

school

Interdenominational sponsored

-0.103***

(2.82)

-0.042

(1.38)

school

Other sponsored school Number of observations

-0.134**

  • (1.96)
  • 858

0.119**

(1.98)

Notes: The models are binary logit models with clustered standard errors and sample weights and the table reports the average marginal effects (MEs). The base category for the regional dummies is the Border region of Ireland. The base category for the school sponsorship dummies is a Catholic-sponsored school. The base category for the school enrolment mix dummies is a mixed enrolment. Absolute values of z statistics are presented in parentheses. *** Denotes significant at 1%, ** denotes significant at 5%, and * denotes significant at 10%

Source: Adapted from Cullinan et al. (2013)

In other findings, Cullinan et al. (2013) also showed that having taken extra private tuition (grinds) outside of normal class hours results in a higher probability of participating in higher education, while the gender mix of a student’s school was not an important determining factor. There were also no statistically significant differences in progression between Catholic, Church of Ireland and interdenominational schools, once second-level attainment, spatial and socioeconomic factors were con?trolled for. However, the results did show that positive teacher engagement had a statistically significant association with higher education participation.

In a subsequent study, Flannery and Cullinan (2014) also used the 2007 SLS and considered at what type of HEI students chose to study at if they progressed to third level. Specifically, they defined two binary outcomes. The first was whether students attended a university or a non-university institution and the second was whether they did honours or non-honours degrees (National Framework of Qualifications [NFQ] level 8 versus NFQ level 7). Since these decisions are unlikely to be independent, conditioning on covariates, they used a bivariate probit model. Table 2.5 shows the results from their Table 3. As with the previous studies, they controlled for Leaving Certificate attainment, in the form of points, and socioeconomic background. The socioeconomic background effects were as one would expect; for example, those in the lowest category were 26% less likely to go to a university than a nonuniversity and they were 29% less likely to do an honours degree than a non-honours one. The results also highlighted significant gender effects, with males less likely to progress to university compared to females. It was also shown that having taken extra private tuition outside of normal class hours resulted in a higher probability of undertaking an honours degree. There was also no significant relationship between attending a Catholic sponsored or DEIS-designated second-level school and variation in HEI type.

What is striking here is that these effects were conditional on the students’ CAO points. Since the outcomes are different from those of Denny’s (2014) (and the specification of the model somewhat different), the results are not necessarily inconsistent. What we see is that for those going to third-level, socioeconomic background mattered even conditional on points. In general, universities are more prestigious than other third-level institutions and, likewise, honours degrees are more prestigious than non-honours. So the results are, to some extent, in line with those reported in Denny (2014) and discussed briefly above, where we conjectured that higher aspirations or greater self-confidence or some other non-cognitive skill by those from better-off backgrounds may play a role.

Table 2.5 Bivariate probit model of university participation and degree type

Variable

University

Honours degree

ME

z

ME

z

Minimum Distance to University

-0.001

(0.41)

-0.0005

(0.18)

Minimum Distance to Non-University

-0.006

(1.14)

-0.004

(0.72)

Midlands

0.53*

(2.27)

0.41

(1.35)

West

0.04

(0.16)

0.09

(0.33)

Dublin

-0.15

(0.51)

-0.08

(0.30)

Mid-East

-0.41

(1.40)

0.28

(0.83)

Mid-West

-0.15

(0.51)

0.24

(0.74)

South-East

-0.19

(0.76)

0.26

(1.08)

South-West

-0.01

(0.07)

0.12

(0.44)

Gender

0.29***

(2.68)

0.11

(0.98)

CAO Points

0.009***

(10.83)

0.007***

(12.50)

Grinds

0.05

(0.56)

0.20*

(1.70)

Social Class II

-0.24**

(1.96)

-0.22**

(2.07)

Social Class III

-0.26*

(1.64)

-0.29*

(1.69)

Deis

-0.02

(0.11)

-0.13

(0.79)

Sponsorship

0.19

(1.56)

-0.05

(0.43)

Wald x2

452.11

P

0.57*** (47.2)

Number of observations

761

Notes: The model is a bivariate probit model with clustered standard errors and sample weights and the table reports the average marginal effects (MEs). The base category for the regional dummies is the Border region of Ireland. Absolute values of z statistics are presented in parentheses. *** Denotes significant at 1%, ** denotes significant at 5% and * denotes significant at 10%

Source: Adapted from Flannery and Cullinan (2014)

 
Source
< Prev   CONTENTS   Source   Next >