I'm doing a content analysis of political debates (20/20 total) broadcast before a specific election (census sample). My content analysis specifically observes and codes negative appeals in terms of two groups by political affiliation. Which statistical test/analysis should be considered for the analysis?
Jean-Jacques Boshoff Boshoff,
There are a few approaches that come to mind. Just to be sure I understand your analysis needs; you have a sample size of 20 (number of debates) and 2 groups (political affiliation).
The one statistical approach I think may be best is a paired t-test; pairs being number of negative appeals from each political group in each debate. This offers a little more power than a regular t-test and helps control for effects specific to each debate on negative appeals overall (both political groups). So the results form this will basically help determine if the two group means of neg. appeals are equivalent. Be sure to know and check assumptions.
If assumptions are not met you might consider randomization/permutation or non-parametric approaches (depending on which assumptions aren't met).
You could also do a more general test, like a chi-squared test or Fisher's exact test on a contingency table. The draw backs here is that you'll need to have another measurement(s) (e.g., neutral appeals, positive appeals) and it's not as specific to the distributions of negative appeals as both t-tests. These tests will basically help you determine if the groups and proportions of appeal type are independent. Again, know and check assumptions.
Best of luck on the analysis.
Caleb
I agree with Caleb Ashton Aldridge that t-Test would be the most appropriate statistic, although I don't see how this would be a paired comparison rather than the regular t-Test.
David,
My thinking was that each debate will somewhat differ (e.g., moderator, questions, participants, audience, venue, etc.)—at least I get that sense from observations of American political debates. I suggested the paired t-test in order to account for this variation. Is my reasoning off? If so, I'm interested to know how this possible variation might be handled—I did consider randomization type t-test for this. I look forward to hearing you answer and will likely learn something.
Sincerely,
Caleb
Hi David and Caleb,
Thank you for the helpful comments! The convenient aspect of the analysis is that the 20 debates form part of a census sample, therefore the whole population is included in the analysis. The appeals were coded into negative and positive/neutral values. For this reason I initially did a normal contingency table with bar charts. Because the analysis included a comparison between categorical variables I thought that a chi-square test would be suffice? I will consider the t-test as part of the analysis.
Thanks!
Since you have the possibility of doing a chi-square analysis, that should be sufficient.
Jean-Jacques Boshoff Boshoff , Acording to my experience, SPSS is useful for contente analysis when we have a great amount of material. I think that it is your case.
- Badges
- Science topic
- Similar topics
- Political Science