PACES '04
7/28/04 Press Release

Survey and Analysis Methodology

Survey Methodology

The PACES (Public Agendas and Citizen Engagement Survey) has surveyed 760 adults, aged 18 and up, by telephone since February 18, 2004. The last interviews included in the analysis in the press release were completed on July 25, 2004. The survey sample is an approximate random sample of American adults. The survey has included an oversampling of young Americans age 18-26, and the results in the press release were based on analyses which reweighted the interview cases to compensate for the youth oversampling. Other than age, no weighting was applied.

The number of interviews conducted between February 15 and May 7 was 546, and the number conducted between May 8 and July 25 was 210.

Survey work was conducted by UC Berkeley’s Survey Research Center.

Analytical Methodology

The importance of assessments of the economy was tested with an ordinary least squares regression equation run on all the cases described below. The dependent variable was preference for president (vote choice if election were held now, without mentioning Ralph Nader explicitly). The 3 key independent variables included partisan identification, a summary index of policy-related opinions, and a variable representing assessments of the US economy. The variable for economic assessments was the average of opinion on 2 questions: (1) how close the US was to a strong economy; and (2) how much of a problem the US economy would be over the next year. The model also included 3 product terms which were the interaction of each of those three key independent variables with a dummy variable for survey date (0= Feb 15- May 7; 1= May 8 – July 25). The date dummy variable was also included in the model as a main effect. Various sociodemographic characteristics served as control variables in the model. Because of case loss from missing data on some variables, the total number of cases in the regression was 614 weighted cases. Also, the cutoff date for the interviews that could be included in this analysis was July 20.

The results of this regression with product terms yielded a substantial interaction term for economic assessments with date of interview, with a 2-tailed p-value <.05, meaning that it passed the normal threshold for statistical significance (or “margin of error”).

The change in the importance of traditional family values as a predictor of vote preference emerged as statistically significant when 3 regressions were run, separately for 3 date periods: Before Iraq Bad News = February 18 – March 31; During Iraq Bad News Peak = April 1 – May 11; Post Iraq Bad News Peak = May 12 to July 20 (the cutoff date for this particular analysis). The dependent variable was, as above, presidential vote preference. The key independent variable included a variable representing family values, which was an average of opinion on abortion, homosexuality and gay rights, issues of church and state, women’s rights, sex in the media, and moral decline. The control variables were partisan identification, economic assessments and other opinions on numerous other issue topics, such as the environment, Iraq, etc. Family values emerged as a substantially larger predictor of vote preference in the last of the 3 time periods, compared to the previous 2.

For other questions about survey or analytical methodology, please email Douglas Strand at: dstrand@csm.berkeley.edu.