February 2008 Poll - Discussion
Type & Party
Results from the February 2008 Poll were analyzed to explore whether or not the findings from the November 2007 poll would be replicated. Due to the smaller sample size, however, we adopted a more exploratory approach.
First, using multiple crosstab analyses, we examined the relationship between party self-identification (Democratic, Republican, & Independent) and variables of interest: political orientation, type preferences, age, sex, and APT membership. For these analyses, age was dichotomized into two groups: those 30 years of age and younger vs. those over 30 years of age. Our rationale for dichotomizing age is the consistent finding in political psychology research that party identification in young people is strongly influenced by parental party identfication. Futher, political ideology also may be influenced during these years, but may remain more flexible until one reaches one's late twenties. After age 30, research tends to show that political ideology becomes more stable.
Results from the multiple crosstab analyses indicated the following variables were significantly associated with party self-identification (listed in order of level of statistical significance):
Results for the S-N preferences were near-significant (p = .08). Preferences for E-I and J-P were not significant.
To explore more complex relationships, we used nominal regression (a technique that is similar to conducting many crosstabs at the same time). Due to the limited samples size, we used a step-wise process. This process included party self-identification as the dependent variable in each analysis and also included political orientation as an independent variable in each analysis. We then conducted separate analyses for each of the variables found to be significantly related to party identification in the univarite crosstab analyses. We also explored whether certain type interactions (SN by TP, SN by JP, and TF by JP) might produce significant results. Thus for example we conducted an analysis where party identfication was the dependent variable and the independent variables included political orientation and T-F preferences. (Note: We dichotomized political orientation into Liberal-Other given the skew in the data on this variable. Thus, the comparison is between self-identified Liberals and those who self-identified as Moderate or Conservative. Futher, we did not include APT membership given that this variable globally reflects some type differences as well as age differences. Rather, we used the more specific variables individually.)
Results from this analysis suggested only age contributed signficantly beyond political orientation. Near-signficant results, however, were found for T-F preferences (p = .07) and the interaction between preferences for T-F and J-P (p = .09).
Building on these results we conducted two other analyses by adding the T-F preferences and the interaction between T-F and J-P preferences. Results from these two analyses suggest a signficant result for the model with party identification as the dependent variable and the following as indepent variables: political orienation, age, and T-F preferences. Results indicated that sample members who were younger, not liberal, and expressed a preference for Thinking were more likely to self-identify as Republican than Democrat. With respect to the comparison between Independents and Democrats, only the political orienation score was signficant: Self-identified liberals were less likely to be independents.
Given that the February 2008 sample was somewhat different than the November 2007, we checked whether or not the results obtained in the February sample could be replicated in the November sample. We thus conducted a nominal regression with party identification as the dependent variable and the following as independent variables: political orientation, age, and T-F preferences. Results indicated only the political orientation variable was significantly associated with party identification in the November sample. This failure to replicate the full model suggests results may be sample specific and need further study.