Does Higher Education Cause Religious Decline?: A Longitudinal Analysis of the Within- and Between-Person Effects of Higher Education on Religiosity

Abstract

Although there is ample empirical evidence of the associations between higher education and various aspects of religiosity, the causal mechanisms producing these associations remain unclear. I use four waves of longitudinal data, with respondents ranging in age from 13 to 29, to model the within- and between-person effects of higher education on several measures of religiosity. The results show that earning a bachelor's degree is associated with within-person declines in some but not all measured aspects of religiosity, which partially supports the argument that higher education causes religious decline. The results also suggest that those predisposed to attending religious services self-select into higher education, that relatively religious youth in general self-select into nonelite colleges, and that those with low levels of religious belief self-select into elite universities. These findings further understanding of the associations between social class and religion, particularly the causal effects of higher education.

Keywords::

• education
• religion
• longitudinal analysis

ACKNOWLEDGMENTS

This research uses data from the National Study of Youth and Religion, a research project designed by Christian Smith, of the Department of Sociology at the University of Notre Dame, and generously funded by the Lilly Endowment Inc. of Indianapolis, IN.

NOTES

Notes

¹ The full sample for wave 1 consists of 3,370 respondents. The 80 respondents (in wave 1) who comprise the Jewish oversample are deleted from each wave because the Jewish oversample is not a random sample and those cases cannot be weighted along with the other cases. The sample sizes reported above do not include the Jewish oversample.

² In wave 4, four respondents reported being 31 or 32 years of age, which is outside of the range of possible ages given respondents' ages in wave 1. Consequently, age for these four respondents was recoded to 29, the oldest possible age.

³ Because of the ordinal nature of most of the variables, the factormat option in Stata was used to perform an exploratory factor analysis on the polychoric correlation matrix. All measures loaded at .41 or higher on a single factor (.63 or higher for all measures other than support for proselytization [.56] and opposition to religious subjectivism [.41]). No other factor had multiple variables reaching the basic threshold of .40 (Costello and Osborne 2005).

⁴ Frequency of religious service attendance is a seven-category measure ranging from “never” to “more than once a week.”

⁵ Frequency of prayer is a seven-category measure ranging from “never” to “many times a day.”

⁶ Belief in afterlife, angels, demons, and miracles each coded as follows: (1) not at all, (2) maybe, and (3) definitely. Belief in God is coded (1) no, (2) unsure/do not know, and (3) yes. Respondents who believe in God were asked “Do you believe that there will come a judgment day when God will reward some and punish others?” In wave 4, an “unsure” response option was added. Those who answered no, unsure/do not know, or do not believe in God are coded zero and those who answered yes are coded one.

⁷ Importance of faith is based on the survey question “How important or unimportant is religious faith in shaping how you live your daily life?” The five-category variable ranges from “not at all important” to “extremely important.” In the first wave respondents were asked (1) if they had ever made a personal commitment to live their life for God and (2) if they experienced a definite answer to prayer or specific guidance from God. In later waves, respondents were asked if they had done or experienced these things in the time since the previous survey. In wave 4, a “maybe” response option was added. For both commitment to God and having a prayer answered/guidance from God, those who responded no or maybe are coded zero and those who answered yes are coded one. Finally, respondents were asked “How distant or close do you feel to God most of the time?” The six-category variable ranges from “extremely distant” to “extremely close.” In wave 4, a “does not apply” response category was added. Respondents who do not believe in God, and were thus not asked the question, and those who responded “does not apply” are coded as extremely distant.

⁸ Exclusivist religious perspectives are coded as follows: (1) there is very little truth in any religion, (2) many religions may be true, and (3) only one religion is true. Respondents with a religion were asked, “In the past year, how many doubts, if any, have you had about whether your religious beliefs are true?” Response options are, many doubts, some doubts, a few doubts, and no doubts. In wave 4, a “does not apply” response category was added. Does not apply and those with no religion are coded as having many doubts. Support for proselytization contrasts those who say it is “okay for religious people to try to convert other people to their faith” with those who believe “everyone [should] leave everyone else alone.” Last, opposition to religious subjectivism (Hart 1987) is based on disagreement with the following statement: “Some people think that it is okay to pick and choose their religious beliefs without having to accept the teachings of their religious faith as a whole. Do you agree or disagree that this is okay?”

⁹ Based on rankings of “national” universities. Wave 2 of the NSYR does not include IPEDS (Integrated Postsecondary Education Data System) codes, which indicate the specific college attending and/or graduated from. Only three respondents had bachelor's degrees in wave 2, which means ranking of institutions granting the bachelor's degree is not very relevant in wave 2. Nonetheless, 416 respondents to the wave 2 survey were currently in college. This would cause a high level of missingness for a variable indicating ranking of college or university currently enrolled in. Consequently, no such variable is included in the models. Eleven respondents received degrees from more than one four-year college or university. The college rank variables reflect the highest ranked university from which these respondents earned a bachelor's degree.

¹⁰ The multilevel modeling literature (e.g., Raudenbush and Bryk 2002), and particularly the literature on using multilevel models to examine change within individuals (Singer and Willett 2003), suggests including the mean of time-varying (i.e., level-1) variables as time-invariant variables (i.e., level-2) to assure that the time-varying measures assess change within individuals. I use a dummy variable indicating ever receiving a bachelor's degree rather than the mean across waves because the substantively relevant distinction is between those who do and do not graduate from college. Using the mean instead of a dummy variable would give more weight to those who graduated in an earlier wave. Nonetheless, the results are not meaningfully different when using the mean of time-varying bachelor's degree as the time-invariant measure rather than the dummy variable employed in this article.

¹¹ The time-invariant dummy variable for college graduation and the time-invariant mean of being in college must be correlated because those who never went to college have a value of 0 on both variables, and, conversely, graduating from college requires spending time in college (although there is a large enough gap between waves 3 and 4 for respondents to attend and graduate between waves; thus, 6 percent of respondents who were never in college at the time of a survey did graduate from college). The polychoric correlation (rho) between these two variables is .692. The time-invariant measures of bachelor's degree and mean in college are thus indeed relatively highly correlated, although no more so than is generally acceptable in multiple regression (Tabachnick and Fidell 2012). Alternative analyses that drop the mean of in college show similar results. Specifically, the focal results for attendance, prayer, the belief scale, the personal religiosity scale, and the religious certainty scale are unchanged. The only meaningful changes concern the overall religiosity scale. When the mean of in college is excluded from the model of the overall religiosity scale, the time-varying (level-1) in college variable has a moderate, negative effect (p < .05), indicating within person decline associated with being in college, and the time-invariant (level-2) bachelor's degree variable has a moderate, positive effect (p < .05), which indicates differences between college graduates and those without a bachelor's degree. Results from analyses without the mean of in college available on request.

¹² Four cases are missing data on parent service attendance. For these respondents, parent service attendance is coded at the mean.

¹³ Given the marginal reliability of the religious certainty scale, separate models for each of the individual items that constitute the scale were analyzed (results not shown). The results from those models indicate that time-varying bachelors' degree has a negative effect on within-person changes in religious exclusivism and opposition to religious subjectivism, but not on support for proselytization and lack of religious doubt (results available on request).

¹⁴ For instance, the mean overall religiosity score in wave 1 is .237 for those who eventually graduate from college and .156 for those who do not (t = 2.234, p < .05).

¹⁵ This finding of relatively little between-person difference in religiosity associated with higher education also holds when focusing specifically on those who are past the traditional college age range. For instance, limiting the sample to wave 4 and examining ordinary least squares (OLS) models of religiosity shows that bachelor's degree is positively associated with religious service attendance but not significantly associated with the other measures of religiosity (models control for sex, race, region, married, cohabit, live with parents, children, and age).

¹⁶ For instance, in wave 4, the mean overall religiosity scale for graduates of nontop 100 institutions is .145 standard deviations higher than for graduates of top 51 to 100 institutions, and .703 standard deviations higher than for graduates of top 50 institutions.

¹⁷ I examined models of religiosity separately for those in the top 50 percent and the bottom 50 percent of the overall religiosity scale in wave 1. The models are identical to those in Table 2 with the exception of splitting the sample by wave 1 religiosity. Results show that time-varying bachelor's degree has a strong, negative effect on the belief scale for those who were relatively religious in wave 1 (b = −.158, p < .001) but no effect for those were less religious in wave 1. Conversely, time-varying college graduation (b = −.198, p < .001) and in college (b = −.107, p < .01) are associated with robust declines in frequency of prayer for those who were less religious in wave 1 but not for those who were more religious in wave 1. Another interesting finding from this analysis is that the negative effect of parent education on between-person differences in religiosity is limited to those who were less religious in wave 1. Results of this auxiliary analysis are available on request.