Long-Run Effects of Catholic Schooling on Wages

1 Introduction

There is a substantial body of evidence, mostly from the United States, on the positive outcomes of Catholic schooling on test scores, school completion and college enrollment compared to outcomes from attending public schools.^[1] However, academic achievement is only one measure of the success of an educational system. Other important measures may include civic engagement, law abidance, happiness and personal traits (such as conscientiousness). Evidence that Catholic schooling affects outcomes other than education is scarce. Figlio and Ludwig (2012) and Mocan and Tekin (2006) find that Catholic schooling is negatively associated with engagement in risky behaviors, while Bettinger and Slonim (2006) and Dee (2005) find it to be positively associated with charitable giving and civic participation, respectively.

In this study, we examine whether Catholic schooling improves readiness for work over and above increasing the level of education attained (direct effect). There may be several channels through which such benefits may accrue. First, because Catholic education is faith-based, it may put greater emphasis on the development of “non-cognitive” or “soft” skills that are important in explaining labor market outcomes (Brunello and Schlotter 2011; Heckman, Stixrud, and Urzua 2006). Second, because they are typically from more advantaged backgrounds, Catholic school graduates may have richer networks than public school graduates. Having broad and resource rich connections is well known to be an advantage in the labor market (Calv´o-Armengol and Jackson 2004, 2007). Third, attending a Catholic school may signal the presence of personal traits, such as a strong work ethic, that may be valued by employers. Finally, independent of the level of education, Catholic school graduates may choose fields of study that have higher labor market returns. Although data limitations mean that we are unable to test the mechanism(s) through which any differences in wage trajectories are transmitted, we present descriptive statistics that may indicate the importance of one of the channels.

Previous studies by Neal (1997) and Vella (1999) have estimated short-run direct wage effects from Catholic schooling. Both studies find small positive, albeit insignificant effects, although Neal (1997) shows that results for minority groups to be much larger and significant. In contrast, large Catholic schooling wage effects have been found in a long-term study by Kim (2011). Using a cohort of graduates from Wisconsin in 1957 (Wisconsin Longitudinal Study) and controlling for selection, Kim (2011) found Catholic school graduates earn 10 and 5% more than public school graduates at 17 and 35 years after graduating, respectively. We build on these studies in three main ways. First, we build on Kim (2011) by including controls for education level to estimate long-term direct Catholic school wage effects. Second, unlike previous studies that examine outcomes at particular points in time, we examine wage outcomes over a working life, which is important to fully appreciate any benefits of Catholic schooling. Finally, by also examining outcomes from independent private schooling, we shed light on whether any benefits relative to public schooling are unique to the Catholic system.^[2]

In this study, we estimate the long-run direct effect of Catholic schooling on hourly wages by tracking outcomes of employed males 26–53 from different schooling backgrounds. In particular, we apply the approach suggested by Wooldridge (2002, p. 267), which involves estimating a log wage fixed effects model with interactions between school type and years of labor market experience, with controls for education attainment. Under this setup, the Catholic schooling effect is measured by the interaction coefficients, which represent the part of a wage gap associated with different payoffs to labor market experience between Catholic and public school graduates, or the “experience premium” from Catholic schooling. The use of fixed effects controls for time-invariant factors, such as family background, that may affect both the choice of Catholic schooling and wage growth.

Estimation is based on 10 years of data (2001–2010) from the Household Income and Labour Dynamics Australia (HILDA) panel survey. The use of Australian data is important. The Australian school system, much like that in other English speaking countries, has no tracking of students and the primary focus is on general education. Like in the United States, there is evidence of significant Catholic school effects on education outcomes (Vella 1999; Le and Miller 2003), although studies using more recent data have found much smaller effects (Marks 2007; Cardak and Vecci 2013). A key difference is that Australia has a relatively large private school sector, including large Catholic and independent private sectors, which allows for comparison of outcomes within the private sector. In Australia, 34% of all school students are enrolled in private schools (20% in Catholic and 14% in independent) (Department of Education and Workplace Relations 2011) compared to around 10% in the United States (USDOE 2012). The relatively large private schooling sector in Australia is due, in part, to government subsidies to private schools that account for around 80% of Catholic school and 45% of independent private school net recurrent income per student (Department of Education and Workplace Relations 2011). Total expenditure per student in Catholics schools is on par with that in government schools (A$10,000 per student per year), but is less than that in independent private schools (A$12,000 per student per year) (ABS 2006).

We find evidence of an experience premium associated with Catholic schooling during prime-age – 15–25 years of employment experience. In particular, we estimate that higher payoffs to employment experience in prime-age is associated with a 12% higher hourly wage compared to public school graduates. This result is robust to a range of alternative model specifications and is indicative of a long-term direct wage effect of Catholic schooling. Importantly, we also find a significant experience premium for independent private school graduates in prime-age. The implication is that the positive direct effect of Catholic schooling on wages applies generally to private schooling.

2 Data

Analysis conducted in this study is based on 10 annual waves of the unbalanced HILDA dataset, covering the period from 2001 to 2010. HILDA is a large, nationally representative panel dataset of Australian Households that contains detailed education and labor market information, similar to the U.S. Panel Study of Income Dynamics (PSID). Following Kortt and Dollery (2012a), who estimated the impacts of religion on wages in Australia using HILDA, we restrict the sample of analysis to employed males aged 26–53, dropping those who are not employed or who are self-employed.^[3]

The high rates of employment in our sample means that omitting those in employment does not drastically reduce our sample size (employment rates in our sample are over 90% for most age groups). Nonetheless, this restriction does have a larger impact on those with relatively little labor market experience, especially public school graduates. Given that the chance of selection into employment is likely to be non-random, it is possible that our results are biased to some degree. However, when we correct for selection using Inverse Probability Weighting (Fortin, Lemieux, and Firpo 2011), our main results are much the same, which suggests that sample selection bias is not a major concern.^[4] To ensure that we are estimating long-run labor market impacts, we also restrict the sample to those with at least 5 years of employment experience since leaving full-time education for the first time. In the sensitivity analysis, we examine whether this restriction affects the results.

The main variable of interest in this study is the hourly wage. Hourly wage is constructed as the ratio of individual real weekly wages and salaries in all jobs (in Australian dollars, 2010 prices) to hours usually worked per week in all jobs.^[5] We exclude a small number of observations for which there is no weekly wage or hours of work recorded. Overall, our sample contains 3,369 individuals and 19,129 person-year observations.

The type of secondary school attended is identified in HILDA by asking respondents which of the following best describes the type of school they attended in their last year of school: government school, private Catholic, other private school and other school. For the purposes of this study, we omit the small number of respondents who chose other school type.^[6] Overall, 15% of the 19,129 observations in our sample are from people who attended Catholic schools (2,904), 10% from people who attended independent private schools (1,836) and 75% are from people who attended public schools (14,389).^[7]

The definition of school type attended in HILDA means that individuals who moved to a Catholic school in their last year of secondary school are not distinguished from those who were exclusively Catholic schooled. This makes our estimates on the effects of Catholic schooling conservative because most of those who change school types move from public schools, most commonly between primary and secondary school (around age 12).^[8]

The average log hourly wage by school type and years of employment experience from our sample are presented in Figure 1. From Figure 1, it appears that initially, people who attended either type of private school have a higher wage than people who attended a public school. Furthermore, as years of employment experience grow, especially beyond 10 years, the gap widens, before narrowing again after 25 years. Indirect effects – due to differences in educational attainment – may partly explain the wage gap by school type. In our sample, people who attended public schools are much less likely than people from private schools to have attained qualifications higher than upper secondary or equivalent – 42% compared to 58% for Catholic schools and 65% for independent private schools.^[9] However, data from splitting the log hourly wage in Figure 1 by the initial observed education level show a similarly widening wage gap between private and public school graduates with a college (bachelor) degree (see Figures 2 and 3 in Appendix). This is suggestive of a direct effect from private schooling among college graduates. In the following sections, we explore this relationship further by controlling for differences in characteristics between private and public school graduates.

Figure 1:

Average log hourly wage by labor market experience and school type, employed males 26–53.

Figure 2:

Average log hourly wage by labor market experience for those without a college degree.

Figure 3:

Average log hourly wage by labor market experience for college graduates.

3 Empirical Strategy

To estimate the impacts of Catholic schooling on wages we estimate a typical human capital model where the log hourly wage for individual i in wave t depends on measures of education qualifications, employment experience, school type and personal attributes, including aspects of religion:

In eq. [1] Schooli represents school type; Religioni includes denomination (including no religion), frequency of attendance at religious services and importance of religion; Educationit is the highest qualification level attained at time t; Expit is years of employment experience since leaving full-time education for the first time; Xit is a vector of individual characteristics (described below); Wavet are time-trend dummies and uit is a random error term.

In this model, the effect of school type is captured by the combination of constant term γ, which reflects any initial gap in wages by school type in the reference period, and θ, a measure of how any initial gap changes with employment experience, or the “experience premium” associated with Catholic schooling. With controls for education level, γ and θ represent direct effects, or effects over and above those from differences in education attainment.

Equation [1] allows any wage effects of Catholic schooling to be realized at different points in a career. Such a flexible approach may be important, for example, in capturing the effects of non-cognitive skills that may not be realized until prime-age when opportunities to enter management usually arise. Employment experience enters the model as six categorical variables (in five-year groups, commencing with 5–10 years of employment experience and ending with 30 or more years). The chosen reference period is working for between 5 and 10 years.

In the first instance, we estimate eq. [1] by pooling the sample and estimating OLS. However, this approach does not control for bias due to endogenous sorting into school type. Such bias is present if unobserved factors, such as family traits, are correlated with both school type selection and wage outcomes. We control for the effect of unobserved time-invariant factors that affect school type selection and wages by estimating eq. [1] using individual fixed effects estimation.^[10] This approach was first suggested by Wooldridge (see 2002, 267) to measure the gender wage gap.

With fixed effects estimation, the random error term uit is broken into a time-invariant individual unobserved heterogeneity term αi and a stochastic error εit:

The impact of all time-invariant factors over the period of analysis is subsumed into the αi term, including the school type constant (γ). Therefore any direct Catholic schooling effect under fixed effects estimation is measured only through θ, the experience premium. In model [2], identification of θ is based on individuals who are observed to move between employment experience categories over the period of analysis (Imbens and Angrist 1994). Because we use 10 years of a large panel dataset, there are ample observations in each of the five-year work experience groupings to identify θ. The robustness of the results to the choice of experience groupings is tested in the sensitivity analysis section. Importantly, there is also variation in education levels within each experience and school type grouping to identify the effect of school type separately from the effect of education on wages.

A related issue is the potential endogeneity of education level in our model, which because of correlation with school choice, may bias our interaction coefficients. While fixed effects estimation controls for time invariant unobserved factors, it does not control for time-varying unobserved factors that may be correlated with education level. However, results generated from a sub-sample of 3,156 (out of 3,369) individuals who do not change their education status over the period of analysis produces consistent results. This suggests that the effect of any bias from the endogeneity of education level may be small.^[11]

A key assumption underlying the use of fixed effects is that time-varying unobserved factors (εit) are uncorrelated with both wages and school type (common unobserved time trends). We test the robustness of our results to this assumption in the sensitivity analysis.

4 Control Variables

In estimating direct effects of the school type on wages, we control for a number of observed factors that may be correlated with both the choice of school type and labor market outcomes. Descriptive statistics for all the controls are presented in Table 3 in the Appendix. Some of these factors are time-invariant and are subsumed into αi in the fixed effects estimation. For completeness, we discuss all of the controls used in the analysis, regardless of whether they change over time.

Important time-invariant controls are characteristics of religion, which are time invariant because they are only observed at the time individuals are first surveyed in HILDA. Past studies have shown that Catholic men in the United States and Australia earn a wage premium relative to their protestant counterparts (see Steen [2004] and Kortt and Dollery [2014], respectively). The Catholic wage premium may stem from broader networks and/or from personal traits, such as self-discipline. As a result, without controls for religion, OLS estimates of Catholic school effects are likely to be biased. In this study, we categorize religious affiliation as protestant (Anglican, Presbyterian/Reformed, Uniting Church, Lutheran, Baptist and other protestant), Catholic, non-Christian (Buddhism, Hinduism, Islam, Judaism, other non-Christian), other religion and no religion. We are able to control for religious affiliation because being Catholic in Australia is not perfectly correlated with attending a Catholic school. Indeed, around 71% of Catholic school attendees, 14% of public school attendees and 11% of independent private school graduates report being Catholic (Table 3). We also control for the importance of religion (measured on an 11-point scale where 0 means religion is the least important thing in your life and 10 is the most important thing) and religious observance. Religious observance is measured by the number of times the respondent attends a religious service, from every day through to less than once a year, or never.

Other time-invariant co-variates include the number of siblings, whether or not an individual reports being an indigenous Australian, country of birth, father’s occupation, family information at age 14 and marital status.

The rest of the controls are time varying. Most important are controls for education level, measured by the highest International Standard Classification of Education (ISCED 1997) qualification level attained at the time of interview. We group education levels into the following ISCED categories: less than ISCED 3, which is less than a secondary school qualification; ISCED 3A and 3C, which is an upper-secondary school qualification or vocational equivalent; ISCED 4B, which is a post-secondary vocational qualification; ISCED 5B is a Diploma level qualification and ISCED 5A and 6, which is a college (bachelor) degree or higher (ABS 2001).

Another important time-varying control is age. Age enters the model as six categorical variables, each spanning a five-year interval, commencing with 25–30. Controls for age are important so that our long-run estimates of schooling type effects are independent of cohort effects that might arise due to differences in the quality of education over time. While age is correlated with experience, there is considerable overlap in the age distribution across the employment experience categories, which allows us to separately identify the effects of age and experience.

Other time-varying controls that are used in both OLS and fixed effects estimation are reported disability status, full-time/part-time employment status, union membership, state of residence (state fixed effects) and time-trend dummies (time fixed effects).

5 Results

Key results from the log wage model, estimated using OLS and fixed effects are presented in Table 1; full results are presented in Table 4 in the Appendix. Standard errors, clustered at the individual level, are presented in parentheses below each of the key estimated coefficients. In models A and B, OLS results without interaction terms show no estimated difference in wage by school type. In model C, we include interaction terms into the OLS models (as described in eq. [1]). The statistically significant and positive interaction results for model C suggest that there are direct wage benefits from Catholic schooling relative to public education that are realized through an experience premium.^[12] In model D, we introduce individual fixed effects to the model C specification (as described in eq. [2]), and if anything, the positive interaction coefficients are larger.

The interaction coefficients from model D suggest no significant direct effects of Catholic schooling early in a career associated with an experience premium. This is consistent with results from Neal (1997) and Vella (1999). However, significant effects are found during, but not beyond, prime-age of a career (15–25 years experience). In particular, during prime-age, it is estimated that Catholic schooling is associated with an experience premium of around 12%. Expressed in level terms, we estimate that greater payoffs to experience for Catholic school graduates is associated with an extra A$3/h and A$4/h, respectively, for 15–20 and 20–25 years of experience. For an individual who works 40 h per week, this equates to an extra A$120 and A$160 per week, respectively, or 11% and 14% of the average weekly pay.^[13] These are reasonably large effects.

An important result from the fixed effects model (Table 1, model D) is that directs effects of schooling are also found for independent private schools graduates during prime-age. However, the effects are not as precisely estimated as those for Catholic schools, possibly because there are fewer observations in the data and because of greater variation in the practices of independent private schools. Despite the imprecision, the magnitude of the estimated interaction coefficients, especially during prime-age, is similar to those estimated for Catholic school graduates.^[14]

Using F-tests, we find no evidence that the experience premium for independent private school graduates is any different to that for Catholic school graduates at any of the work experience points. However, the F-tests are based on a relatively small number of observations for independent school graduates, which reduces the power of the tests.^[15]

The experience premium associated with private schooling in prime-age may be best explained by differences in the likelihood of being promoted into management jobs, which conceivably occurs around this time in a career. Private school graduates may be more often promoted to management roles if they have higher non-cognitive skills or richer networks. While we cannot test which, if either, of these explanations is correct, differences in non-cognitive skills may better explain the different payoffs to experience at this time. This is supported by information on the “Big-Five” personality traits available in wave 5 of HILDA that show male Catholic school graduates 26–53 to be, on average, statistically more agreeable, more conscientious and more openness to experience than public school graduates. Past studies have found levels of conscientiousness (Nyhus and Pons 2005; Almlund et al. 2011; Heineck 2011) and openness (Mueller and Plug 2006) to be positively linked with higher earnings and the chance of being a manager (Zhao and Seibert 2006). If any experience premium was due to better networks, then arguably, we should observe a positive experience premium early in a career when school graduates still maintain close ties with their school peers. Differences in non-cognitive skills fits better with the delayed timing of wage effects because they often take time to be realized in the labor market.

6 Sensitivity Analysis

As discussed above, there are two key assumptions that underpin our results. First, that there is adequate movements between the employment experience categories to allow for identification of the differential effects of employment experience by school type using fixed effects. Second, that there are no differences in unobserved post-school time trends by school type that would explain the differences in payoffs to experience. The validity of these assumptions is examined below.

7 Conclusion

This is the first study to estimate wage benefits of Catholic schooling over a working life, independent of those transmitted through higher education attainment (direct effect). We find a positive direct wage effect from Catholic education emerges during prime-age of a working life, between 15 and 25 years of labor market experience. It is estimated that higher payoffs to experience for Catholic school graduates is associated with around a 12% higher wage during prime age, compared to public school graduates. This result underlines the importance of taking a longer-term view when estimating direct wage effects from schooling programs. Previous studies that have attempted to estimate direct wage rage effects by Neal (1997) and Vella (1999) found no significant short-run effects of Catholic schooling. Interestingly, we find that these benefits are only apparent for college graduates.

Another important contribution of this paper is in being able to put labor market outcomes from Catholic education into perspective. We show that independent school graduates also earn an experience premium compared to public school graduates. We find no evidence that the experience premium for Catholic and independent school graduates is different, although this conclusion is tempered by the low power of the F-tests. The implication of these results is that any direct wage benefits relative to public schooling pertains to private schooling and not just Catholic education.

These findings have some important policy implications. In particular, the presence of labor market benefits from private schooling, independent of education outcomes, raises the issue of whether current methods of measuring school performance are sufficient. While education outcomes are linked to post-school outcomes (including labor market outcomes), our results suggest that there may be other important school outcomes that could be used to measure school performance. At present, because they are relatively easy to measure, there is a risk that schools are devoting too many resources to improving academic outcomes, potentially at the expense of developing other aspects of human capital that are important to a young person’s working life. However, to develop alternative measures that are effective in improving long-run outcomes, a better understanding is needed of the channels through which wage effects are transmitted. Given that we find that the experience premium occurs in prime-age and is restricted to college graduates, a possible explanation is that the estimated wage benefits of private schooling are associated with differences in non-cognitive abilities that are important for promotion into management roles. This hypothesis is also supported by evidence showing that Catholic school graduates have higher levels of conscientiousness and openness. Other explanations, such as better networks, may be realized early-on in a career and may not depend to the same extent on higher education qualifications. Future research to further explore and understand these potential mechanisms will be informative.