Regression analysis results
independent effects of several variables on the number of arguments a respondent has heard. In some instances, the dependent variable (that is, the number of arguments heard) takes the form of the entire scope of arguments a respondent has heard about, say, the two major presidential candidates. In other cases, the focus was just on the number of arguments heard about a specific candidate among those who do not support that candidate.
The rationale for using regression analysis is to isolate the independent effects of different variables on predicting the number of arguments heard. For example, an important issue in this report is whether internet use might result in a Kerry supporter hearing fewer arguments that support George Bush’s candidacy for reelection. Although the analytical focus is on the internet effect, it is necessary to control for other effects or, in other words, to look at the internet effect while holding everything else constant. Regression analysis does this.
In the example of people’s awareness of arguments about the two presidential candidates, there were four pro-Kerry arguments and four pro-Bush arguments. It is possible for a respondent to have heard up to eight arguments. The dependent variables are numeric ranging from 0 to 8 when focusing on all arguments, and 0 to 4 when the analysis focuses just on arguments heard in favor of a single candidate. Ordinary Least Squares regression was used in conducting the analysis.
Taking the regression models run regarding the presidential campaign as an example, the number of arguments heard were modeled as a function of the following variables:
- satisfaction with the direction of the country
- several measures of news consumption
- partisan predictors (e.g., party affiliation)
- interest in the campaign
- demographic characteristics (e.g.., age, gender, race, marital status)
- internet use
Four different measures of internet use were used in the models, namely whether one goes online at all, whether one uses a high-speed internet connection at home, whether one goes online for campaign news, and whether one goes to non-traditional Web sites for political news. Internet effects discussed reflect results of models run with one measure included in the equation, with the other three excluded.
Overall, the internet effects tended to show up for Kerry supporters who said they go online for campaign news or have gone to Web sites of non-mainstream media. This was also the case for predicting exposure to the total number of Bush and Kerry arguments for all respondents. For Bush supporters, internet effects were a bit weaker and evident among high-speed users, internet users at large (in a separate model specification), and those who had gone online for campaign news. For Bush supporters, the act of going to non-mainstream media did not predict greater exposure to arguments (either arguments for Bush or arguments for Kerry).