I want to analyse the differences in male and female responses to a questionnaire assessing perceptions of gender equality. The questionnaire uses likert scale data only.
Which statistical test do I use?
- One way ANOVA?
-Chi-Squared
- Independant T-Test?
Are you analyzing the questions individually, or summing up responses to several questions?
I’m still unsure.
Both?
But most likely summing up responses.
Thank you!!!
Usually if people are analyzing individual Likert-type questions, they will treat the responses as ordinal, and so use, say, a Mann-Whitney test rather than a t test. But if they are adding up several questions, they will treat the results as interval data, and use e.g. anova or t test as long as the data meet the assumptions of the analysis. (There's nothing universal in approaching Likert item and scale data.).
Otherwise, traditional nonparametric tests (Mann-Whitney, Kruskal-Wallis) are fine. But understand what they are testing. They are not usually tests of the median.
The one-way anova and t test will give the same results when there are two groups. With the caveat that some software will default to not assuming homoscedasitic groups. That is, they might default to Welch's t test but not Welch's anova. Be sure you understand the assumptions of these tests if you use them.
Usually you will not want to use a chi-square test of association because it doesn't recognize the ordered nature of the response categories. That is, it sees that SA is different from A is different from N, but doesn't know that SA > A > N. If you wish to arrange your data in a contingency table with one variable treated as ordinal, then you could use the Cochran-Armitage test.
From your explanations above, make use of of independent t-test since you have two groups (Male and Female). The dependent variable must be measured at the interval level continuously. Code the data for gender nominally
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude