Timeline for is it statistically correct to combine paired and unpaired samples together and perform tests?
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 15, 2022 at 23:20 | comment | added | Glen_b | 1. It shouldn't lose much efficiency wise if both the conditions I mentioned hold; otherwise I wouldn't do it. 2. There is a question - I should have mentioned it before - of why there were some unpaired values. e.g. If those were missing not at random, you have to carefully consider the biases involved in handling the overlapping samples. | |
| Aug 15, 2022 at 15:18 | comment | added | Najeha Mohamed | thank you so much for your input! just exactly what i was looking for. Is it okay to perform tests without the independent samples? | |
| Aug 15, 2022 at 15:17 | comment | added | Najeha Mohamed | thank you so much for your input! just exactly what i was looking for. Is it okay to perform tests without the independent samples? | |
| Aug 15, 2022 at 10:18 | comment | added | Glen_b | This situation is sometimes called (partially) overlapping samples and tests for it are called (partially) overlapping samples tests. e.g. see cran.r-project.org/web/packages/Partiallyoverlapping . The manual has references. (When the within-pair dependence is strong and the paired observations comprise most of the data, the independent observations might not tend to add a lot of information to what you get with the paired data.) ... also see tqmp.org/RegularArticles/vol18-1/p055/p055.pdf ... These should give you further references and suitable search terms. | |
| Aug 15, 2022 at 9:42 | history | asked | Najeha Mohamed | CC BY-SA 4.0 |