summed (reliability coefficient = .85). This strategy yields an index that ranges
from 6 to 30 with higher scores reflecting greater psychological distress.
Economic variable. Economic difficulties are measured using the income
scales employed by the GSS for the survey years. The scales are then rescaled
to percentages. Drawing from a variety of sources, the decision was made to
use the 13 to 14 percentile as the cut point for those falling below the poverty
line. In addition, an interaction term to represent the urban poor is also
included in the models.
Control Variables. Recent studies demonstrate that attendance at religious
services varies by a number of sociodemographic and background factors.
Dummy variables representing southern residence (South = 1, U.S. Census
designation), married respondents (married = 1), households with children
under the age of eighteen living at home, gender (female = 1), and African
American respondents are used as control variables, along with age (in years),
educational attainment (in years) and community size (six-point ordinal mea-
sure, with more urban areas receiving higher scores).
Analytical Strategy and Results
Multiple regression is used to analyze attendance at religious services since
the scale is not significantly skewed and meets the basic criteria for this type
of analysis. Two models for each time period are used to examine the main
effects of the independent variables, with and without controls, on attendance
at religious services.
Results
Table 1 shows the means and standard deviations for all variables in each of
the three time periods. For the dummy variables or dichotomous variables,
the mean reflects the proportion of respondents on that item that are coded
one. Tables two, three, and four display the bivariate correlations for all vari-
ables in the analysis.
Table 5 shows the models for the three time periods. Model I for the 1978
time period displays the coefficients for the main effects and the interaction
term for urban poverty. The model is statistically significant but explains a
very small proportion of the variance. Living in poverty has no significant
effect on attendance but does show a positive coefficient. The model does
show that the urban poor are less likely to attend. The coefficient of psycho-
348 • David Gay, Warren S. Goldstein, and Anna Campbell Buck