of Religious Conversion1

A. W. Hogan2

This version: 03/15/09

Abstract

In this essay I test religious capital models by investigating empirically the causes of religious conversion in a large cross section of American youths. I attempt to avoid the endogeneity bias that is prevalent in previous efforts to estimate religiosity in order to identify causal effects. I find support for rational choice theories of religious behavior and religious capital accumulation. Religious capital is shown to be significant in determining the likelihood of changing as well as the likelihood of changing back to the faith in which one was raised. Furthermore, I argue that this study shows evidence of bargaining with religious organizations - as shown by the significant and positive effect on the likelihood of conversion from being diagnosed with an incurable disease.

Keywords: , conversion, health JEL Classification: D10, I19, Z12

1 Thanks to Charles L. Baum II, John M. Nunley, and Mark Owens for assistance and comments. All remaining errors are my own. 2 Middle Tennessee State University, Department of Economics and Finance, 1301 E. Main St., Murfreesboro, TN 37132. e-mail: [email protected] 1 Introduction Since Gary Becker paved the way with “New Home Economics,” microeconomic theory has been applied to many different fields traditionally thought to be outside of the realm of economics. Though held by many to be the epitome of irrationality,3 even the religious behavior of households can be explored through the lens of . This paper joins a branch of the research into the economics of religion that studies the demand for religious goods and services empirically.4 To that end

I attempt to estimate religiosity and identify what influences it.

An individual's religiosity must be inferred from preferences revealed through that person's choices and actions. Previous studies have looked at the frequency of religious service attendance, contributions made to religious organizations (e.g., (Azzi and Ehrenberg 1975), (Durkin and Greeley

1991), (Iannaccone 1990, 1994, 1997b)), total time spent on religious activities (Neuman 1986), or frequency of prayer (Garza and Neuman 2004) as a proxy for religiosity. Instead, I use whether a person has changed religions (including to or from a “no religion” option). 5

In addition to examining conversion as revealed religiosity, this paper expands the literature by focusing on causal relationships between the independent and dependent variables. It is common in papers estimating religiosity to include measures of religious activity, opinions on faith, and family structure on the right hand side of the estimated equation. But how often somebody goes to church, whether they believe particular doctrines, and their choice of a spouse could all be caused by religious conversion. There is no denying that one's marriage or belief in the afterlife are correlated with religiosity, but marriage frequently leads to religious conversion.6 I leave out endogenous variables in

3 See (Stark and Finke 2000) for an excellent discussion of the popularity of the secularization theory (and its flaws). 4 See (Iannaccone 1998) for a survey of the economics of religion. 5 (Kluegel 1980) shows the importance of including the no-religion option and trends in religious conversion. 6 More information on the relationship between marriage and religion can be found in, for example, (Becker et al. 1977) and (Lehrer and Chiswick 1993) .

2 order to focus on purely causal effects to conversion.7

The first investments in a person's religious capital are made by their parents. Parents decide what faith, if any, their children will be raised in and how to educate them to these beliefs. Parental characteristics present us with an opportunity to proxy for how much investment in religious capital a person likely received.

Religious capital theory suggests that for more investment in religious capital the benefit to participating in that religion, and thus the cost of leaving it, increase. Therefore we can predict that those with larger investments in religious capital will be less likely to leave their religion. With data on who switches religions and suitable proxies for religious capital investments, I test this claim. I find that more investment from parents in their children's religious capital decreases the likelihood that they will leave the faith in which they were raised.

Those rational choice theories of religion that view the purpose of participating in religion as a good consumed after the actor's lifetime (e.g., (Azzi and Ehrenberg 1975)) immediately suggest the importance of the actor's expected lifetime. Those nearing the end of their life will be more demand more religious consumption. Using a number of health variables I attempt to estimate the effect of changes in the expected lifetime on religious conversion. By estimating the effect of being diagnosed with a number of difference diseases I show that being afflicted with cancer or a psychiatric disorder greatly increase the likelihood of converting religions.

This paper is organized as follows. The next section describes the data source and how some of the variables were constructed. Section 3 describes the estimation techniques and how the final models were chosen. Section 4 discusses the results obtained from estimation and their interpretation. Section 5 concludes and summarizes the study.

7 For details on endogeneity bias see (Greene 2003).

3 2 Data The data used for this paper come from the 1979 National Longitudinal Survey of Youth (NLSY).

The NLSY questioned nearly thirteen thousand people between 14-22 years of age in 1979 and revisited these same people for further questioning every year or two after the initial survey.8 The variables used here are shown in Table 1. Those variables that may require addition explanation are discussed through out the rest of this section.

The NLSY asks each respondent which religion they were raised in, and subsequently asks their current religion in 1979, 1982, and 2000. The religion categories studied here are Baptist, Episcopalian,

Lutheran, Methodist, Roman Catholic, Jewish, Protestant, Other and None. “Protestant,” in this case, is mutually exclusive of the other protestant denominations enumerated above, and indicates other protestant sects as well as non-denominational Christians. These categories are the same for the variables addressing the respondent's spouse's religion. Table 2 shows the religious make up of the

NLSY population, as well as the number of converts to and from each religious category.

The main dependent variable analyzed here is religious conversion. This variable, Change, is defined as a 1 if the respondent reports being of a different religion in 2000 than the religion in which they were raised, and a 0 otherwise. Table 3 shows the conversions that are being captured by the

Change variable. This matrix shows the sum of converts for each possible change.

Additionally I examine the persistent long term effects of religious capital by looking at those who change religions but eventually return to the religion in which they were raised. The Change Back variable is a 1 if the respondent reports changing their religion before 2000, but also reports being the same religion in 2000 as they were raised. A change matrix like the one described above can been seen for Change Back in Table 4.

8 More information on the NLSY data can be found at http://www.bls.gov/nls/nlsy79.htm

4 The variable M is a dummy indicator set to 1 if the respondent is married to somebody who was raised in the same religion as they were, and a zero otherwise. This unusual specification is weakly endogenous insofar as a spouse's traits will only be observed for those who are married, and getting married could cause religious conversion.

All of the health variables (H) are indicators set to 1 if the respondent has been diagnosed by 2000 with a particular ailment (arthritis, diabetes, stroke, cancer or a psychiatric disorder)9, and a zero otherwise. While the author strives not to dismiss supernatural factors surrounding religious conversion, it is assumed that changing religions could not cause these diseases.

The religious attendance variables (A) are vectors of dummy variables for each choice the respondent was offered in answering the question “In the past year, about how often have you attended religious service?” The choices offered as a response were: 1) Not at all, 2) infrequently or once per month, 3) 2-3 times per month, 4) once per week, or 5) more than once per week. In the raw data there are six choices, but there is no statistical difference in the models shown here between those who answered “infrequently” and those who answered “once per month,” - so those answers have been collapsed here.

3 Estimation Whether a person changes religious affiliation is estimated using a probit procedure.10 How sensitive the data are to this estimation technique can be examined in Table 6, which shows the same model estimated with probit and linear probability models.

Table 5 shows results for different sets of independent variables in the estimation of Change. Both

Wald tests to drop sets of variables as well as Bayesian and Akaike information criteria support model

9 The model also included an indicator for diagnosis of heart disease, but it predicted perfectly and was dropped from the probit model. 10 For further information on probit estimation see (Greene 2003).

5 #1. This specification includes the variable for the religion in which the respondent's spouse was raised

(M), which, as previously discussed, could be endogenous. With this in mind I will focus on model #2, which is shown in equation (1). That model is used in subsequent estimation of Change (i.e., Tables 6 and 7).

C= 01 X  2 K 3 H (1)

This equation shows Change as a function of general demographic variables (X), religious capital variables (K), and health variables (H). Included in the demographic variables are Male, Age, Age2,

Black, White, region dummies, and a dummy for which religion in which the respondent was raised.

The religious capital variables (K) include Intact Family, Father Foreign Born, Father High School and Father College. The health variables are all those dummy variables for whether the respondent had been diagnosed with a given disease, shown in Table 1.

Another way to examine converts is with the variable Change Back, which indicates a person who has changed religion from that in which they were raised but later changes back to their original faith.

Change Back is estimated according to equation (2). The estimation results for Change and Change

Back can be compared in Table 7.

C B=01 X 2 K 3 H (2)

As a further exploration of religious conversion, religious participation (A) is estimated as a multinomial logit, as a supplemental model, with the same explanatory variables used to estimate

Change. While the attendance variable is ordered (from no attendance, increasing through to attend more than once per week), a Brant test shows that the assumption of proportional odds would not hold with these data. Without the proportional odds assumption the multinomial logit model is used instead

6 of an ordered logit.

A=01 C2 X 2 K 3 H 4 M (3)

This model is used in order to analyze the effect of conversion on religious attendance.

4 Results This section discusses the results from estimating the equations described in the previous section.

The first subsection discusses results in the religious capital theoretical framework. The second subsection discusses how a person's health will affect the probability that they will convert religion and theories of bargaining with religious organizations. Finally, I present results from a supplemental model of religious attendance, focusing on the effect of religious conversion on religious attendance.

4.1 Religious Capital To capture some measure of religious capital I look primarily to parental characteristics, shown in the variables for whether each person in the data was raised in an intact family by their biological parents, whether their father was born outside the U.S., and their father's education. Additionally, despite a potential endogeneity problem, I look to whether the respondent is married to somebody who was raised in the same faith they were.

If being born abroad means a lower opportunity cost to investing in a child's religious capital (that is, cultural differences raise the opportunity cost of other activities after moving to the U.S.), then with more time to spend with their children, foreign-born parents are able to invest more in their child's religious upbringing. The religious capital model thus predicts the negative and significant effect of

Father Foreign Born on Change (see Table 7). Similarly, it is of no surprise that having been raised in an Intact Family has a negative and significant effect on a person's propensity to convert religions, having likely received more investment specific to their parents' religion.

7 Additionally, this study looks at the education of the respondent's father. The education levels of the respondent's father are significant and positive. Having a father who finishes college leaves one less likely, however, to change religions than if their father only completed High School. Note that this effect on Change and Change Back are opposites. A father's education, it would seem, installs an independence streak that encourages finding a new faith and discourages coming back to the original religion.

Iannaccone has written extensively on the predictions that the religious capital model make about religious intermarriage, and shows where being the same religion as one's spouse is highly (positively) correlated with religiosity (1990, 1998). This study considers only whether the respondent's spouse was raised in the same religion as they were (M), since marriage or divorce could be caused by religious conversion, but which religion in which somebody was raised could not. The opportunity for a married couple to enjoy higher returns to their religious participation from their joint capital, and the increased opportunity cost of switching religions, shows a significant and negative effect on Change in Table 5.

4.2 Health While being diagnosed with arthritis, diabetes, or a stroke do not significantly affect the probability of changing religions (see Table 5), diagnosis of either a psychiatric disorder or cancer both have a significant and positive effect (see Tables 5 and 7). The effect of a cancer diagnosis is the largest single effect observed in the model. The difference between cancer and psychiatric disorders and the other diseases tested here is that cancer and psychiatric disorders are incurable and not well understood.

Stark and Finke (2000) explain that people rationally employ supernatural means, that people do not use them to “rid their gardens of weeds or repair fences” but for problems without known or easy solutions. They go on to posit a formal proposition: “Humans will not have recourse to the supernatural when a cheaper or more efficient alternative is known and available.” (Stark and Finke 2000) Without

8 scientific understanding or medical recourse, the opportunity cost of pursuing assistance from the divine and from religious organizations is lower for cancer and psychiatric disorders than it is for ailments for which there are known causes and cures..

Note that being diagnosed with cancer, in addition to positively affecting Change, has a negative effect on Change Back. Also, of all the health variables, only being diagnosed with a stroke is significant in determining religious attendance. It would seem that contracting a frightening illness, more than simply increasing demand for religious activity, lowers returns for one's current religion, encouraging conversion.

Stark and Finke (2000), in their theory of the micro foundations of religion, propose that a person's confidence in their religion will “increase to the degree that miracles are credited to the religion.” They go on to define miracles as “desirable effects believed to be caused by the intervention of a god or gods in worldly affairs.” (Stark and Finke 2000) I abstain, in this paper, from arguing one way or the other as to the existence of miracles, noting that it is enough to point out that there are many people who do believe in miracles. (Garza and Neuman 2004)

Though Stark and Finke consider only positive influence through miracles, I propose that the opposite of their theory holds as well. That is, a person's confidence in their religion will decrease to the degree that they experience undesirable effects attributed to the supernatural – explaining why being diagnosed with an alien disease may cause religious conversion.

4.3 Religious Attendance As an additional way to analyze religious conversion I estimate religious attendance and include

Change as an independent variable. The marginal effects of Change on religious attendance are shown in Table 8. Generally speaking, those who convert will likely attend religious services more frequently.

Converts are less likely to attend less than once a month or not at all, and more likely to attend once a

9 month, once a week, or more than once a week. The only exception is that those who only attend less than once per month who convert are more likely to stop attending religious services. This is likely a reflection of those converts who have changed to “no religion.”

5 Conclusion Having analyzed causal effects to religious conversion I have shown further evidence supporting rational choice theories of religious behavior. This study extends previous work looking at religiosity from an economic point of view by comparing causes of conversion and previously used proxies, such as religious attendance. I find significant support for theories of religious capital (Iannaccone 1990) by showing that those who receive more investment in religious capital from their parents are less likely to convert. Furthermore, I show evidence that ties in religious capital between spouses will increase the likelihood that they will share a single faith. Finally, and perhaps most interesting of all, by examining the role health plays in determining the demand for religious services I discover that people diagnosed with alien diseases are more likely to lose faith in their religious organization.

10 6 References

Azzi, Corry and Ehrenberg, Ronald (1975), "Household Allocation of Time and Church Attendance," The Journal of Political Economy, 83 1, 27-56.

Becker, Gary S.; Landes, Elisabeth M.; Michael, Robert T. (1977), "An Economic Analysis of Marital Instability," The Journal of Political Economy, 85 6, 1141-87.

Durkin, John T. and Greeley, Andrew M. (1991), "A Model of Religious Choice Under Uncertainty: On Responding Rationally to the Nonrational," Rationality and Society, 3 2, 178-196.

Garza, Pablo Branas and Neuman, Shoshana (2003), "Analyzing Religiosity Within an Economic Framework: The case of Spainish Catholics," IZA Discussion Paper, 868, .

Greene, Willaim H. (2003), "Econometric Analysis." Prentice Hall: Upper Saddle River, NJ.

Iannaccone, Laurence R. (1990), "Religoius Practice: A Human Capital Approach," Journal for the Scientific Study of Religion, 29 3, 297-314.

Iannaccone, Laurence R. (1994), "Why Strict Churches Are Strong," American Journal of Sociology, 99 5, 1180-1211.

Iannaccone, Laurence R. (1997a), "Rational Choice," from Rational Choice Theory and Religion. Edited by Young, Lawrence A.. pp. 25-45. Routledge, New York, NY.

Iannaccone, Laurence R. (1997b), "Skewness Explained: A Rational Choice Model of Religious Giving," Journal for the Scientific Study of Religion, 36 , 141-57.

Iannaccone, Laurence R. (1998), "Introduction to the Economics of Religion," Journal of Economic Literature, 36 3, 1465-1495.

Kluegel, James R. (1980), "Denominational Mobility: Current Patterns and Recent Trends," Journal for the Scientific Study of Religion, 19 1, 26-39.

Lehrer, Evelyn L. and Chiswick, Carmel U. (1993), "Religion as a Determinant of Marital Stability," Demography, 30 3, 385-404.

Neuman, Shoshana (1986), "Religious Observance Within a Human Capital Framework: Theory and Application," Applied Economics, 18 , 1193-1202.

Rogerson, Richard; Shimer, Robert; and Wright, Randall (2005), "Search-Theoretic Models of the Labor : A Survey," Journal of Economic Literature, 43 4, 959-89.

Stark, Rodney and Finke, Roger (2000), "Acts of Faith: Explaining the Human Side of Religion." University of California Press: Berkeley, CA.

11 TABLE 1: VARIABLES AND SUMMARY STATISTICS Variable Mean Std.Dev. Definition Attendance=1 0.209 0.406 Attended no religious service in 2000. Attendance=2 0.168 0.374 Attended religious service less than once a month in 2000. Attendance=3 0.143 0.350 Attended religious service around once a month in 2000. Attendance=4 0.118 0.322 Attended religious service once a week in 2000. Attendance=5 0.071 0.257 Attended religious service more than once a week in 2000. Change 0.282 0.450 Respondent reports a different religious than the one in which they were raised. Change2 0.588 0.492 Respondent reports changing religious between 1982 and 2000. Respondent reports changing religion but reports the same religion in 2000 in which Change Back 0.441 0.496 they were raised. Dad-College 0.263 0.440 Respondent's father finished college. Dad-High School 0.633 0.482 Respondent's father finished high school. Region-North Central 0.146 0.353 Respondent lived in North Central U.S. Region-North East 0.098 0.298 Respondent lived in North East U.S. Region-South 0.260 0.438 Respondent lived in Southern U.S. Region-West 0.124 0.330 Respondent lived in Western U.S. Diagnosed-Cancer 0.001 0.031 Respondent has been diagnosed with Cancer. Diagnosed-Psychiatric 0.032 0.175 Respondent has been diagnosed with a psychiatric disorder. Diagnosed-Arthritis 0.043 0.204 Respondent has been diagnosed with arthritis. Diagnosed-Diabetes 0.022 0.146 Respondent has been diagnosed with diabetes. Age (79) 17.898 2.306 Age of respondent in 1979. College 0.970 0.171 Respondent finished college. Dad-Foreign 0.096 0.295 Respondent's father was born outside the U.S. Family Intact 0.676 0.468 Respondent lived with both biological parents. High School 0.997 0.055 Respondent finished high school. Male 0.480 0.500 Respondent is male. Race-Black 0.238 0.426 Respondent is black. Race-Other 0.045 0.208 Respondent is neither black nor white. Race-White 0.661 0.474 Respondent is white. M 0.189 0.392 Respondent is married to somebody who was raised in the same religion as them.

12 TABLE 2: RELIGION OF ENTIRE NLSY POPULATION AND CONVERTS Religion Population (2000) Converted From Converted To Net Gain Protestant 1261 15.74% 196 8.68% 644 28.51% 448 Baptist 2103 26.25% 555 24.57% 205 9.07% -350 Episcopalian 90 1.12% 44 1.95% 30 1.33% -14 Lutheran 360 4.49% 124 5.49% 86 3.81% -38 Methodist 384 4.79% 207 9.16% 113 5.00% -94 Presbyterian 150 1.87% 69 3.05% 62 2.74% -7 Catholic 2104 26.26% 756 33.47% 124 5.49% -632 Jewish 59 0.74% 9 0.40% 6 0.27% -3 Other 644 8.04% 160 7.08% 334 14.79% 174 None 856 10.69% 139 6.15% 655 29.00% 516 Total 8011 100.00% 2259 100.00% 2259 100.00% Notes:

13 TABLE 3: CHANGES IN RELIGION FROM RAISED TO 2000 Religion (2000) Protestant Baptist Episcopalian Lutheran Methodist Presbyterian Catholic Jewish Other None Total Religion (Raised) Protestant 0 33 3 9 14 3 22 1 27 84 196 Baptist 174 0 5 9 40 11 29 0 133 154 555 Episcopalian 5 5 0 2 3 3 2 0 6 18 44 Lutheran 32 12 0 0 7 6 24 0 13 30 124 Methodist 43 44 1 10 0 10 15 0 28 56 207 Presbyterian 16 4 1 4 8 0 8 1 4 23 69 Catholic 270 76 14 39 32 21 0 4 91 209 756 Jewish 1 0 0 0 0 0 1 0 2 5 9 Other 49 17 0 4 4 5 5 0 0 76 160 None 54 14 6 9 5 3 18 0 30 0 139 Total 644 205 30 86 113 62 124 6 334 655 2259 Notes: Those that report a different religion in 2000 than the one in which they were raised are represented here. Each number shows the total who switched from the religion of that row to the religion of that column.

14 TABLE 4: CHANGES IN RELIGION FOR THOSE WHO CHANGE BACK Temporary Religion Religion (Raised/2000) Protestant Baptist Episcopalian Lutheran Methodist Presbyterian Catholic Jewish Other None Total Protestant 0 90 3 15 31 31 47 0 14 71 302 Baptist 73 0 2 1 46 6 19 0 11 159 317 Episcopalian 3 5 0 0 1 1 2 0 1 2 15 Lutheran 9 4 0 0 3 2 4 0 1 7 30 Methodist 7 17 1 1 0 6 3 0 2 11 48 Presbyterian 5 3 0 1 2 0 5 0 1 7 24 Catholic 26 21 4 1 6 3 0 5 7 42 115 Jewish 0 0 0 0 0 0 1 0 0 3 4 Other 176 27 1 1 4 4 8 1 0 37 259 None 21 16 2 8 6 4 10 2 2 0 71 Total 320 183 13 28 99 57 99 8 39 339 1185 Notes: Those that change religion but return to the religion in which they were raised by 2000 are represented here. Each number shows the total who switched from the religion of that row to the religion of that column (and eventually back to the religion of that row).

15 TABLE 5: TESTING CHANGE PROBIT MODELS 1 2 3 4 Log-Likelihood Intercept Only: -4765.11 -4765.11 -4765.11 -4765.11 Log-Likelihood Full Model: -4480.56 -4616.34 -4621.9 -4648.62 Likelihood Ratio (27) 569.09 (26) 297.54 (21) 286.42 (17) 232.97 Prob > Likelihood Ratio: 0.000 0.000 0.000 0.000 McFadden's R2: 0.060 0.031 0.030 0.024 McFadden's Adj R2: 0.054 0.026 0.025 0.021 AIC: 1.126 1.159 1.159 1.165 AIC*n: 9017.12 9286.67 9287.79 9333.25 BIC: -62794.64 -62532.08 -62565.9 -62548.4 BIC': -326.4 -63.84 -97.66 -80.16

P-value to drop 0.000 0.065 0.000 0.000

X X X X X K X X X H X X M X

Intact Family -0.0492 *** -0.0563 *** -0.0575 *** Father – High School 0.0509 *** 0.0536 *** 0.0543 *** Father – College 0.0328 * 0.0360 *** 0.0359 *** Father – Foreign Born -0.0152 -0.0316 * -0.0340 * Diagnosed – Stroke -0.0184 -0.0203 Diagnosed – Diabetes 0.0361 0.0393 Diagnosed – Arthritis 0.0072 0.0151 Diagnosed – Psychiatric 0.0911 *** 0.1145 *** Diagnosed – Cancer 0.4143 *** 0.3888 *** Spouse Raised in Same Religion -0.1773 *** Notes:*, **, and *** indicate significance at the 10, 5, and 1% levels.

16 TABLE 6: COMPARE CHANGE MODELS

Probit OLS Marginal Standard Marginal Standard Effect Error Effect Error Intact Family -0.056342 0.0117 *** -0.0547 0.0112 *** Father – High School 0.053630 0.0124 *** 0.0515 0.0118 *** Father – College 0.036044 0.0135 *** 0.0337 0.0128 *** Father – Foreign Born -0.031612 0.0177 * -0.0323 0.0181 * Diagnosed – Stroke -0.020349 0.2351 -0.0025 0.2214 Diagnosed – Diabetes 0.039301 0.0847 0.0341 0.0798 Diagnosed – Arthritis 0.015060 0.0525 0.0151 0.0511 Diagnosed – Psychiatric 0.114451 0.0265 *** 0.1143 0.0239 *** Diagnosed – Cancer 0.388809 0.1439 *** 0.3802 0.1281 *** Notes: *, **, and *** indicate significance at the 10, 5, and 1% levels.

17 TABLE 7: COMPARE CHANGE AND CHANGE BACK

Change Change Back Marginal Standard Marginal Standard Effect Error Effect Error Intact Family -0.056342 0.0117 *** -0.0444 0.0119 *** Father – High School 0.053630 0.0124 *** -0.0315 0.0126 * Father – College 0.036044 0.0135 *** -0.0179 0.0136 Father – Foreign Born -0.031612 0.0177 * 0.0325 0.0203 Diagnosed – Stroke -0.020349 0.2351 -0.0824 0.3121 Diagnosed – Diabetes 0.039301 0.0847 -0.1222 0.1015 Diagnosed – Arthritis 0.015060 0.0525 -0.0607 0.0695 Diagnosed – Psychiatric 0.114451 0.0265 *** -0.1250 0.0275 *** Diagnosed – Cancer 0.388809 0.1439 *** -0.3582 0.0567 *** Notes: *, **, and *** indicate significance at the 10, 5, and 1% levels.

18 TABLE 8: MARGINAL EFFECT OF CHANGE ON RELIGIOUS ATTENDANCE

Predicted Probability of Attendance Change 0 1 Difference Attends no religious services 0.18% 24.67% 0.245 Attends religious services less than once a month 30.03% 21.54% -0.085 Attends religious services once a month 24.99% 19.09% -0.059 Attends religious services once a week 18.73% 19.95% 0.012 Attends religious services more than once a week. 8.12% 14.76% 0.066 Notes:

19