paired samples or independent samples t test?

[aznologist480]

You see, I'm trying to help out a friend of mine in the Education
department with his dissertation. There was a study conducted in 1996
with school educators in various counties here in MS. From the study
different educator tasks or jobs were given a relevance score, which is a
composite of 2 other scores. This year, my friend resurveyed the Mississippi counties' schools. His immeidate intention is to see if the relevance scores have significantly changed or not.

Given that he wants to test if the scores have changed significantly in
the past 10 years, I thought that t-tests would be appropriate. However,
I can't decide whether the paired samples t test or the independent
samples t test is appropriate. In my opinion, paired samples t tests
should be conducted on the relevance scores. However, not all of the same
respondents from 1996 responded in the 2006 survey. So the 2006 sample is
not the same as the 1996 sample. This is understandable since after 10
years some school educators may have moved on to another job and new
educators would have been employed.

Here is where I have a problem(s). My friend says he's assuming that the
2 samples from 1996 and 2006 are independent and "equally representative".
This is the reason why I am debating whether or not to use pair samples
or independent samples t test. Again, he says they are independent
samples, which I can't totally agree with because some of the respondents
participated in both the 1996 and 2006 surveys.

On the other hand, his study question seems more in line with the
rationale of the paired samples t test. The 1996 and 2006 samples may not
be exactly the same, but as I understand it a matched pairs design can be
used. In this case, he could not test the same subject twice, but I think
that the 2 samples are very similar in that they are school educators in
MS counties. Therefore, I think that a paired samples t test would
be appropriate.


So, finally could any of you give me some advice as how to approach
this problem? If you have any questions or need me to clarify anything,
please, ask.

Thanks
aznologist480

[waltbydefault]

perform the setup and do the math for both - measure against the normal curve then look carefully at the way the writup was done the first time. I manage by exception and if there's no major changes in either - count it as a no change - If there's a better than fifteen percent change in both count it as positive change - if it's worse on both, count is as worse. The issue gets down to whether you have a situation where the results are mixed. Personally I'd call that a null - call your advisors and see whether you're challenged on it. Don't be afraid to call your fac advisors fast when you need guidance. That's what they're there for.

very best to you

vr

walt