Posted by
Roberts, Michael on
Jan 05, 2007; 7:25pm
URL: http://spssx-discussion.165.s1.nabble.com/Significant-difference-Means-tp1072801p1072807.html
Game, Set, and Match!!!
mike
-----Original Message-----
From: SPSSX(r) Discussion [mailto:
[hidden email]] On Behalf Of Marta Garcka-Granero
Sent: Friday, January 05, 2007 1:50 PM
To:
[hidden email]
Subject: Re: Significant difference - Means
Hi Samir
I've seem to have missed all the fun while I was out...
After reading your original posting plus all the replies you got (mainly those from Jan and Richard), I think the topic should be closed about the correct statistical method you should use: 2 samples independent t-test. Anyway, I'm going to add a small explanation. I think you got a wrong idea about "dependent samples" concept, and that might the root of the problem accepting Jan's advice:
You said:
SO> I think that I can not use Independent samples T-test since I have
SO> dependent variables. Those 500 students are the part of 1000
SO> respondents.
This is not correct. You can talk about dependent samples only if you have MATCHED (paired) data. For instance:
Subject Before After
1 25 27
2 18 19
3 23 23
4 15 21
5 31 32
. . .
. . .
. . .
The above example is the most usual pairing of data (each subject is measured twice, three times... over time, or data on left eye is compared to right eye...). You can also have naturally matched data
(siblings) or artificially paired data, like in matched case-control studies (1:1 or 1:2, 1:3 matched subjects on key features).
On the other hand, you talk about independent samples when the subjects are different, and no effort has been made to match them (on age, gender... or other attributes).
You also said
:
SO> I do not want to test students vs. non-students but students vs. all
SO> (non-students + students).
As you can see, the situation you are describing falls in neither category, and you can't use any of the above approaches to answer that question. Besides, the question could be considered statistically wrong (you can't test part of a sample against the whole sample).
On the other hand, if you have SOLID theoretical reasons to consider the 1000 students as a finite population, and not a sample drawn from a bigger (technically infinite) population formed by all possible students you could have studied, then you could use the 1000 students mean as a parameter (not as a sample statistic) and run a one-sample t-test, but not the one provided by SPSS (it assumes infinite
populations) but with a finite-population correction term (right now I don't have the formula at hand, but Googling "finite smple correction"
showed a lot of promising links, like this one at Steve Simon's great web site:
http://www.childrens-mercy.org/stats/size/population.asp or this one:
http://myphliputil.pearsoncmg.com/student/bp_berenson_bbs_9/section7_3.pdf).
I insist, you must have solid reasons for considering your 1000 students as a limited population (there won't ever be more of them or something like that?) and run a one-sample t-test with finite population correction. Personally, I'd stick to the independent-samples t-test (500 vs 500 students).
--
Regards,
Dr. Marta Garcia-Granero,PhD mailto:
[hidden email]
Statistician
---
"It is unwise to use a statistical procedure whose use one does not understand. SPSS syntax guide cannot supply this knowledge, and it is certainly no substitute for the basic understanding of statistics and statistical thinking that is essential for the wise choice of methods and the correct interpretation of their results".
(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)