How Culture Shapes Community: Bible Belief, Theological Unity, and a Sense of Belonging in Religious Congregations

Abstract

Feeling that you belong in a group is an important and powerful need. The ability to foster a sense of belonging can also determine whether groups survive. Organizational features of groups cultivate feelings of belonging, yet prior research fails to investigate the idea that belief systems also play a major role. Using multilevel data (U.S. Congregational Life Survey, http://www.uscongregations.org), this study finds that church members' traditional beliefs, group-level belief unity, and their interaction associate positively with members' sense of belonging. In fact, belief unity can be thought of as a “sacred canopy” under which the relationship between traditional beliefs and feelings of belonging thrives.

ACKNOWLEDGMENTS

The author would like to thank Joseph Baker, Phillip Connor, Paul Froese, Kevin Dougherty, Jerry Park, Brandon Vaidyanathan, and Andrew Whitehead for comments on an early draft of this article. The author would also like to thank Aaron Franzen, Jenna Griebel, Carson Mencken, and Charles Tolbert for productive dialogue. An earlier version of this manuscript was presented at the Society for the Scientific Study of Religion 2010 annual meeting in Baltimore, Maryland.

NOTES

Notes

¹ Unlike Sewell (2005), I do not collapse the term “culture” into structure but rather label social environments and networks as “structure” and symbolic features or mental schemas as “culture.” Using structure and culture rather than schemas and resources lends itself to clearer prose and follows the approach of recent work (Ridgeway 2006; Vaisey 2007).

² Here, I follow Ryle and Robinson's (2006) use of “modernists” rather than Hunter's “progressives” to avoid confusion with the contemporary usage of “progressive.”

³ See Blau and Schwartz (1984) for more on the consequences of network heterogeneity.

⁴ The USCLS employs a national sample of 1,214 congregations obtained by the University of Chicago–based National Opinion Research Center. Of these, 34 percent completed a congregational survey. Researchers have found the USCLS data to be demographically comparable with those of the 1998 National Congregations Study (Hill and Olson 2009). Were my goal to yield point estimates of congregational belonging, this lower than ideal response rate would be problematic. Although a higher response rate would be desirable, research has indicated that the precision of parameter estimates using sample data is minimally related to response rates (Singer 2006; American Association for Public Opinion Research 2008). The present study is focused on parameter estimates, not point estimates. The USCLS provides a weight that adjusts for the size of a congregation. Inferences can be drawn for the population of American congregations using this weight in the USCLS data. The procedures used in constructing this weight directly follow those utilized in McPherson's (1982) voluntary association research and subsequently employed in congregational research by Chaves et al. (1999). The substantive results of this study do not change with or without the use of this weight. A formal individual-level response rate of attendees within each congregation is not available since the USCLS did not gather information on how many persons were in attendance on the day of the survey administration. However, I calculated a pseudo–response rate by comparing the number of attendee surveys completed in a congregation with the average worship attendance reported on the profile for that congregation. The mean individual-level pseudo–response rate was 62.2 percent. I suspect that average worship attendance would be higher because it would be based on all Sundays (including Easter and Christmas) when attendance is much higher than average, and so 62.2 percent is potentially a conservative estimate of average individual-level response. Given that these data are based on self-completion surveys, it is not impossible that selection bias potentially poses a more serious challenge in this survey than other phone or personal interview surveys.

⁵ See Woolever and Bruce (2002) for more information on USCLS methods and sampling procedures.

⁶ About 11 percent of respondents did not provide household income information, by far, the item with the largest amount of missing information. The mean percent missing for the remaining variables is 1.3 percent (several items had no missing data: age, race, whether child lives in the respondent's residence, and small-group participation). I exclude cases that initially had missing values on the dependent variable (see Von Hippel 2007).

⁷ Details concerning the number of respondents in congregations are as follows: mean = 267, median = 121, minimum = 3, maximum = 2,728.

⁸ Alternate coding strategies do not substantively change the findings of this study. To examine whether findings are dependent on this particular coding of the outcome variable, I conducted analyses using two other coding strategies. First, I coded category 1 or 2 as equal to 1. Second, I coded all “do not know/not applicable” responses as missing and reran models. In none of these models did results change in any meaningful ways. The study's focal findings do not appear dependent on these response variable-coding strategies.

⁹ Using the inverse of the standard deviation for Bible belief is appropriate for two reasons. First, Bible belief is an ordinal variable, and the standard deviation is a more appropriate measure of variation among respondents in a group on an ordinal variable. The entropy index or Herfindahl index is most applicable to measuring variation in nominal variables. Second, using standard deviation to measure diversity is in keeping with its recurrent use in organizational literature (see McPherson 1983).

¹⁰ Social interaction with friends in a house of worship is associated with religious belonging in prior studies (Stroope 2011a). Because this variable may have an important impact on sense of belonging, I reran models using an alternate dummy-system coding of this variable (no close friends, 18,744, 16 percent; some friends, 21,339, 18 percent; some close friends, 60,866, 52 percent; most close friends, 16,125, 14 percent). The reference category was rotated three times so that each friend dummy was the reference once. The effects fall out consistently in a linear fashion. For example, all categories are less than category 4. Category 3 is positive compared with category 2 but negative compared with category 4. Category 2 is negative compared category 3 and category 4 but positive compared with category 1. Also, the focal results of the study did not change with these alternate codings of the friends variable.

¹¹ I also estimated models with additional control variables at the congregation level. Congregational mean income, mean educational attainment, and mean service attendance were not statistically significant as control variables and did not meaningfully alter the study's focal findings.

¹² There are 3,663 respondents in 17 black Protestant congregations.

¹³ There are 301 respondents in 5 congregations of the Jewish tradition.

¹⁴ Diagnostic analyses did not uncover problems of multicollinearity (bivariate Pearson's correlations are all less than .60).

¹⁵ This unconditional model shows that there is statistically significant (p < .001) congregation-level variance in the average log odds of belonging feelings (Raudenbush and Bryk 2002).

¹⁶ Data used in the results presented are unweighted. Conducting analyses using USCLS weights did not change results in any meaningful ways.

¹⁷ When Catholic is the reference category, all traditions except Jewish have higher belonging. But the Catholic/mainline difference disappears once congregation size is controlled, likely because many Catholic parishes are large.

¹⁸ I also explored the possibility that the effect of individuals' beliefs on sense of belonging may be moderated by a variety of structural factors. Nonsignificant interactions included mean service attendance x individual Bible belief, mean length of time at church x individual Bible belief, and percent white x individual Bible belief. An interaction between church size and individual Bible belief was statistically significant, but graphing revealed that this interaction was not substantively significant.

¹⁹ See Froese (2008) for more on the sacralization of secular ideology.