Putting geography in context: A multivariate analysis
The analysis to this point identified regional effects without controlling for other possible explanatory factors. For example, Anglo Evangelicals in Los Aneles County were the oldest and best educated, begging the question, to what extent are the differences between Anglo Evangelicals in Los Angeles and those in other geographies be explained by age and educational differences as opposed to county of residence? To investigate that possibility I ran a series of logistic regressions on the individual items that comprise the behavior, belief, and social attitudes indexes. I use the individual items rather than the index in the multivariate analysis for two reasons. The first is missing values. If a respondent refused to answer a specific item, no score was calculated for the entire index. This particularly impacted the RBI. The second reason is a possible differential item contribution to the index; it might be that some of the items in the RBI had a disproportionate impact on the overall score. For example, residence in a particular geography might have a greater impact on attending a prayer group but not on speaking in tongues. Education might be associated with believing that the Bible is the word of God,
CHART 17.9 Heaven and Hell Index for Anglo Evangelicals.
but not reading the Bible literally. Age might be associated with attitudes toward abortion, but not with acceptance of evolution.
To examine the relative impact of the predictor variables on each behavior, belief, and attitude, I used a series of logistic regressions. Like the more common ordinary least squares (OLS or “multiple”) regression, logistic regression identifies the impact of each predictor variable while simultaneously controlling for all the other variables in the equation. The difference is that OLS regression predicts a “continuous” dependent variable with three or more values while logistic regression predicts a dichotomous (only two values) dependent variable. Almost all the items in the indexes analyzed above were continuous (and thus more suitable for OLS regression), so why use logistic regression, which limits the dependent variable to only two values? The first reason is the concentration of Anglo Evangelical index scores in a relatively small range at the more intensive end of all possible scores. In Chart 17.8, for example, the mean scores for Anglo Evangelicals fall between 6.6 and 7.2 out of a possible range of between 4 and 8. For the logistic regression, I chose dichotomous cut-offs reflecting the more intense responses. For example, there were five possible responses for attending prayer groups ranging from never through weekly, but 42% of the Anglo Evangelicals reported going weekly so I used attending weekly vs. less often as the cut-off for the logistic regression. There were seven possible responses for individual prayer ranging from never to several times per day. For the logistic regression I used at least daily as the cut-off because so few Anglo Evangelicals prayed less often than once a week.
The logistic regressions summarized in Tables 17.7-17.10 use four predictor variables: age, education, type of county (urban, suburban, rural), and geography (Census region and California sub-region). The cells in these tables show the Exp (B) coefficient for each category of the predictor variables with each dependent variable. The Exp(B), also known as the “odds ratio,” compares the impact of each category of a predictor variable as compared with a reference category that is not
CHART 17.10 Three Social Attitudes Index for Anglo Evangelicals.
included in the table. Each predictor variable has two or more categories with an associated odds ratio for each category' as compared with a reference category'. The reference category for education is high school or less. Looking at the first column in Table 17.7, for example, this means that a respondent with some college education is 1.472 times as likely as to attend a prayer group at least weekly as compared with a respondent with high school or less. The reference category' for age is the silent generation; for urban and suburban residence the reference category is rural residence. For geography, the reference category' is residence in the South. An Exp (B) of less than 1.0 means that a particular factor reduces the likelihood of a particular occurrence. It is comparable to a negative correlation coefficient in an OLS regression. For example, the Exp(B) for Los Angeles in the first column of Table 17.7 is 0.384. This means that a respondent in Los Angeles
Evolution is the best explanation for the origins of human life on earth
CHART 17.11 Attitude toward evolution by geography for Anglo Evangelicals.
Religion causes more problems in society than it solves 1=Strongly Agree, 4=Strongly Disagree
CHART 17.12 Attitude toward religion by geography for Anglo Evangelicals.
County was only 0.384 times as likely as a respondent in the South to attend a prayer group at least weekly.
The discussion of the logistic regressions is divided into the same divisions as the Indexes presented earlier. Behaviors are shown in Table 17.7, God and Bible in Table 17.8, Social Attitudes in Table 17.9, with Heaven, Hell, attitude toward religion, and evolution presented in Table 17.10. Within each table, I focus on the relative impact of Los Angeles/SoCal/California on each individual item, but I also comment on the implications of the other variables for Warner’s “New Paradigm” and Berger’s “Plausibility' Structure” hypotheses.