When two samples are related, or dependent, but the observations are not matched, are there any tests that will determine if the samples (means or otherwise) are different? I've searched extensively and have only found tests for matched samples, which is not what I need.
Asked
Active
Viewed 104 times
2
-
4Two-way comparisons with paired samples are not restricted to matching. Siblings, twins, pre/post measurement, or responses to a question asked to both wife and husband might all be tested using, e.g., a t-test for paired samples. Could you clarify what your situation actually look like? – chl Oct 12 '13 at 18:57
-
Example -- a sample is partitioned into two groups using a clustering algorithm. Those groups are then dependent. However, different subjects reside in each group and are not matched in any way. So, a paired t-test can't be used since no pairing is possible. However, the dependence between the two groups precludes using methods that require sample independence. – CaperKen Oct 12 '13 at 23:22
-
This probably shouldn't be tagged with 'paired-data' – ndoogan Oct 12 '13 at 23:50
-
@ndoogan I added it, but it's possible I misunderstood the situation. I will remove it. The OP should clarify the nature of the dependence. – Glen_b Oct 13 '13 at 00:13
-
Some related posts with interesting answers: https://stats.stackexchange.com/questions/25941/t-test-for-partially-paired-and-partially-unpaired-data?noredirect=1&lq=1 and https://stats.stackexchange.com/questions/370953/analysing-pre-and-post-intervention-study-with-anonymous-responses – kjetil b halvorsen Nov 12 '18 at 17:09