Beginner Stats: which t-test should I use for these scenarios?

#1
I have been asked to run t-tests in a dataset for the following two scenarios:

1. Test whether people with a masters degree watch less television than those with a high school degree

2. Test whether women or men have had more sexual partners

Initially I ran (1) with an independent and (2) with a paired samples. This is because I thought level of education and TV-watching were unrelated, whereas gender and sexual partners might be related.

However, upon reviewing why the tests are run, I actually think that I should be running (1) with a paired samples and (2) with an independent samples. This is because (1) examines two related samples (two samples with different levels of education) and (2) examines two independent samples (women and men). Can someone let me know if I am on the right track?
 
Last edited:

rogojel

TS Contributor
#2
hi,
paired means that you can link one measurement from the first group to exactly one measurement from the second group. E.g. if you had data on the same persons from the time time they had a high school degree and later after the masters - you could use a paired test, the link being the person.

I dont see a way to link data in either case.

regatfs
 

Dason

Ambassador to the humans
#3
rogojel nailed it. One way to think about this too is to recognize that when you have paired data that like rogojel said the measurements in one group are somehow 'linked' to one in the other group. So let's say you were testing if shooting free throws 'traditionally' or 'granny style' resulted in better results. If you had 10 players and had each of them shoot 10 free throws using each style then you would have a 'traditional' and a 'granny style' measurement for each player. So we have Tim's results for traditional being 'linked' with his result for 'granny style'. You might want to do this since some players might be better than others so by pairing the results in this way you account for the variation associated between players. If instead you had ten players shoot traditionally and then ten other players shoot granny style there wouldn't be the pairing involved. Tim only shot traditionally but not granny style and it doesn't make sense to say Tim's traditional shots should be linked with Frank's granny style shots. Now the point I want to make is that you can also think about how you're collecting the data and if it makes sense to end up with different sample sizes in each group then you don't have paired data. For instance in that last scenario there is no reason I couldn't have had ten players shoot traditional and 17 players shoot granny style. If we were doing the pairing then if I get 10 players then I'm going to have ten measurements in both groups.
 
#4
Ah, I see. Thank you both - I really appreciate this!

I guess it is the notion of linkage that is/was confusing me. If I'm understanding it correctly now, it's neither whether the two variables (e.g. # of sexual partners vs. sex) are related, nor is it whether the levels within the variable are related (e.g. high school degree vs. master's degree). Rather, it's whether the samples themselves are related somehow - in the case of (1), it would be need to be the same people obtaining both the high school and then the master's degree.

To follow-up, my professor mentioned the possibility of using paired samples t-tests in matched paired designs with an experimental group and control group. Could you treat (1) in that way whereby those with the master's degree are seen as the experimental "treatment" group and the participants with the high school degree are seen as the "control"?