Trust and White Ethnic Diversity in Small Town Iowa

ABSTRACT

Trust is a highly valued social good and scholars have long been interested in what shapes who and how we trust. One line of inquiry within this area has focused on racial/ethnic diversity. In this study, we use the 2004 Iowa Community Survey to focus on European ancestry as a form of white ethnic diversity. We find that white ethnic diversity consistently decreases trust, across several indicators. Our study expands scholars’ understanding of the ethnic landscape of rural America and what such divisions in the United States mean for trust.

KEYWORDS:

• Trust
• racial/ethnic diversity
• Iowa Community Survey

Acknowledgments

For helpful comments on a previous draft, we thank Pamela Paxton and Stephanie Brockmann. We thank Amanda Harper for her excellent research assistance.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. In supplemental analyses, we explored different iterations of the diversity measure. First, we expanded the diversity equation to include the top 5 and then the top 10 largest groups. Second, we dropped the four communities (Ainsworth, Columbus Junction, Denison, Radcliffe) where Latino enters the ethnic diversity equation as one of the three largest groups. Third, we further addressed the influence of Latino by keeping all of the communities in the sample and adopting a conservative definition of white ethnicity that included only European, Canadian, and Australian/New Zealand ancestries, thereby omitting Latino from being included in the diversity equation. Last, we explored the influence of allowing an identity of “U.S.A.,” and then the categories of unclassified/other ethnicity to be included in the diversity equation calculation. Results with all of these various operationalizations were equivalent; therefore, we decided to use the Rice and Steele approach of using the three largest ethnic groups (including Latinos) for comparability.

2. We examined several operationalizations of these self-reported ancestries. Ultimately, we chose the most conservative approach, that is, we created variables that captured when respondents identified only as British, German, or Irish, though they could have chosen as many ancestries as they wanted in the ICS survey. In supplemental analyses, we then relaxed our operationalization to include respondents who identified as one of these three ancestries and any number of other ancestries. Results were equivalent; therefore, we decided to adopt the more conservative measure as this was more parsimonious.

3. In supplemental analyses, we removed these individuals and limited our sample to European-only identified individuals. Results were equivalent to those presented below.

4. Categories of formal education include: less than 9th grade; 9th to 12th grade, no diploma; high school diploma or GED; some college, no degree; associates degree; bachelor’s degree; and graduate/professional degree.

5. We recoded income from an eight-category scale to a continuous variable using the middle value of each category. For example, we recoded the category of $10,000–19,999 to $15,000.

6. Additional analyses using multilevel ordinal regression produced the same pattern of results for white ethnic diversity and similar patterns of results across the rest of the variables used in the analyses. We continue with linear regression because we could then present a figure created from predicted values using the original scale of the outcome variable. We found this provided the most straightforward interpretation of our findings.

7. We used the default “between/within” method in SAS (ddfm = bw in the model statement) for the calculation of the degrees of freedom that are used in the tests of statistical significance for the parameter estimates. The random effect was specified using the “intercept” and unstructured (type = un) covariance matrix options within the random statement. The subject option (subject = unique community identification number) specified the multilevel structure of the data (see Singer 1998).

8. Missing information was quite rare in the ICS. In our analyses, there was no missing information for the community-level measures because we used U.S. Census data. For individual-level variables, missing data ranged from a high of 9.3% (N = 871) for income to a low of .2% (N = 17) for respondents’ education. In supplemental analysis, we compared results generated both with and without multiple imputation and found similar results.

9. Practically, we used the multiple imputation procedure in SAS (Proc MI) to create the multiply-imputed datasets for each outcome variable. This procedure filled in missing values to create a complete, filled-in dataset. From these complete datasets (with no missing information) we deleted observations that had missing information on a particular outcome variable (that we flagged prior to imputation), in line with von Hippel’s MID procedure. After conducting our analyses with the mixed linear model procedure in SAS (Proc MIXED) for each dataset, we combined the results from the 10 regression equations (using SAS Proc MIANALYZE).

10. In supplementary analyses, we re-ran our models on two subsamples that were large enough to support analysis: German-identified (N = 2,382) and other European-identified respondents (N = 5,539). These models did not include the individual-level ancestry variables, but were otherwise equivalent. Our negative and statistically significant pattern for white ethnic diversity held for all but one model (new residents) for German-identified respondents and for all models with other European-identified respondents.

11. Readers may wonder about changes in the nonwhite (Latino included) and foreign-born populations throughout the 1990s in Iowa. For the entire state, the nonwhite population grew by 3.3% between the 1990 and 2000 Censuses. As such, 7.4% of Iowa’s 2000 population was nonwhite (Latino included). But this growth is not reflected in small town Iowa. In supplementary analyses, we created additional variables (means in parentheses) that represented the overall nonwhite population difference in our sample between 1990 and 2000 (53.86 persons), the difference in the percent nonwhite between 1990 and 2000 (1.94%), percent nonwhite in 2000 (3.06%), the percent foreign-born in 2000 (1.55%), the change in the percent foreign-born between 1990 and 2000 (.83%), and the change in the percent Latino between 1990 and 2000 (1.84%). We did not calculate percent change for some of the categories because a number of communities did not have any nonwhite residents, for example, in 1990. The 1990–2000 difference in the percent nonwhite and the percent nonwhite in 2000 are strongly correlated with the percent Latino, which we already control for, at .88 and .93, respectively. Nevertheless, we examined the potential influence of all of the above variables on trust by including them one-by-one in separate models and found that there were few statistically significant relationship. No other results were affected by including these variables.

Additional information

Notes on contributors

Matthew A. Painter

Matthew A. Painter II is Associate Professor of Sociology faculty at the University of Wyoming. He received the Ph.D. in sociology from The Ohio State University in 2010. His primary research interests involve immigration, social stratification, race/ethnicity, methodology and statistics, and wealth inequality.

Chloe Flagg

Chloe Flagg is a Major Gift Officer within the University of Wyoming Foundation. She received her M.A in sociology from the University of Wyoming in 2017. Her thesis on white ethnic diversity and community attachment was published in Rural Sociology.