What type of correlation analysis should I use?

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

Polis3512
Posts: 3
Joined: Thu Apr 09, 2015 2:46 am

What type of correlation analysis should I use?

Postby Polis3512 » Thu Apr 09, 2015 3:02 am

So I made a questionnaire with one part being sort of a quiz with right and wrong answers, along with "Don't Know" answers. I took all these, recoded them so that the right answers are 1 and the others are 0. After that, I used Compute to give out: right answers/total # questions. There are 13 questions so if people get 12 out of 13, then they got a .92. Anything that's .50 and below would mean they have little to no knowledge on the subjects being discussed.

One of my hypotheses is about a significant relation between the individual's political party identification and their knowledge (here we have the knowledge questions on the "quiz".

Now, the party variable is categorical. The sample is asked if they identify with Party A, Party B, Party C, Non-partisan or Other. My sample chose mostly A, B, and Non-partisan, with the other options receiving really low scores. So I imagine I would mostly have to use the first two.

Which correlation analysis should I use? In my class, we talked a lot about Pearson and Spearman correlations, along with Chi-Square and others. However, I am not sure how to look for any correlation between Party ID and their knowledge. Another professor suggested t-test if I'm looking for a significant relation between party ID (if any) and the individual's knowledge.

Thanks!
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: What type of correlation analysis should I use?

Postby GerineL » Thu Apr 09, 2015 7:29 am

This would be an anova type of question.
Anova is meant to deal with research questions in which you have a categorical predictor and a continuous outcome, which is the case in your question.

You probably don't meet the assumptions of ANOVA though, because it is meant to deal with groups that are more or less of the same size.
In your case, it sounds like some "groups" are much larger than others (let's consider people with the same political preference as being members of a group).
In that case, you should go for a non-parametric alternative for anova such as the kruskall-wallis test.
People oftentimes don't even check assumptions, so maybe you are not required to check them for your project. Also, oftentimes the differences between parametric and non-parametric tests are not that large.
however, if you want to do it absolutely correctly, you need to check assumptions etc.

For more information on ANOVA, check for instance this video:
https://www.youtube.com/watch?v=3wZkn-pSVzk
Polis3512
Posts: 3
Joined: Thu Apr 09, 2015 2:46 am

Re: What type of correlation analysis should I use?

Postby Polis3512 » Thu Apr 09, 2015 11:17 am

Well, you're right. The groups aren't the same sizes. Which is why I did a recode of the party variable. The sample size is 100. Meanwhile, Party A and Party B had 27 and 26, respectively. I had joined these together and labeled them "Partisan" with 53 people (27+26). Meanwhile, "Non-partisan" was chosen by 41 people. Party C was chosen 5 times and Other was chosen 1 time. Can't the same be done with only these two groups and do a T-Test? I could do an ANOVA with Partisans and Non-partisans but I understand is has to be 3 or more groups, unless I can have Party A (27), Party B (26) and Non-partisans (41), Although I doubt it.
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: What type of correlation analysis should I use?

Postby GerineL » Thu Apr 09, 2015 11:28 am

If you have 2 groups, go for t-test, if you have more groups, go for anova.

Whether or not you should merge several parties into one is a question that should be based on theory / hypotheses, it is not a statistical question, so that is something you should determine yourself :-)
Polis3512
Posts: 3
Joined: Thu Apr 09, 2015 2:46 am

Re: What type of correlation analysis should I use?

Postby Polis3512 » Thu Apr 09, 2015 12:02 pm

Well, I could do partisan vs. non-partisan., which is what one professor recommended. My hypothesis is that there is a statistically significant relation between party ID and the individual's knowledge on X subject. I'm guessing Sig. (2-tailed) would be the relationship I'm looking for, but correct me if I'm wrong. If the alpha is supposed to be 0.05, then getting a .141 as Sig (2-tailed) would mean the hypothesis is rejected. I will include an attachment.
You do not have the required permissions to view the files attached to this post.
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: What type of correlation analysis should I use?

Postby GerineL » Thu Apr 09, 2015 12:10 pm

This is probably a nerdy a detail, but actually with a non-significant p-value you do not reject the null-hypothesis, which means that based on the data you collected there is no conclusive evidence indicating that there is a difference.

if it is significant (i.e., P is smaller than .05), you can indicate that the difference you observed in your data is unlikely to occur in case there is a difference in the actual population. I.e., if P<.05 it is likely that there is a difference.


Whether that indicates that there is no difference, or, for instance, you did not include enough participants to be able to show the effect, you cannot conclude (to directly compare null hypothesis (the two groups have equal means, thus no difference) with the alternative hypothesis (the two groups do not have equal means, thus there is a difference) you can use bayesian statistics but I assume that goes way beyond what you intend).

Who is online

Users browsing this forum: No registered users and 3 guests

cron