2
$\begingroup$

I have data collected for 20 subjects under two conditions for each of the subject. When I want to compare the means of the two conditions, is it absolutely necessary that I use the paired T-test? My understanding is that in paired T-test it is easier to establish significance than the independent T-test because the former is affected by between subject variabilities as well. Under this argument, I believe that using independent T-test and establishing significance is stronger and valid. Am I correct?

Along the same line of argument, can I use one-way ANOVA instead of repeated measures ANOVA even when data has been collected for all subjects under all the conditions? I am tempted to check one way ANOVA instead of repeated measures ANOVA because one way ANOVA lets me compare multiple dependent variables at once where repeated measures ANOVA requires me to make subject X condition tables for each of those variables (in SPSS).

Improve this question
$\endgroup$

1 Answer 1

Reset to default
3
$\begingroup$

First, deliberately using an inefficient, pessimistically biased analysis as a sort of handicap cannot possibly be considered good scientific practice. If you want to be stringent, then lower your alpha level on your hypothesis test.

Second, while your conjecture is true more often than not, it's not true in all cases. My answer HERE explains that when the intraclass correlation is negative, then the between-subject or "unpaired" test is actually more likely to reject the null.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thanks for the insight. I was inclined to use the one-way ANOVA because I have a large number of (about 150) dependent variables to check significance and in SPSS repeated measures design requires me to arrange data in separate tables for each of these variables. Could you help me out pointing out a strategy to overcome this? $\endgroup$
    – Mohamed Athif
    Jan 22, 2019 at 19:27
  • 1
    $\begingroup$ @Athif You'll probably want to to do it programmatically in some scripting language. R is probably the best option. Python would also be good. I'm not that familiar with SPSS, but I do know that it contains a sort of scripting language under the hood that most people do not directly use. But using the SPSS scripting language directly could possibly be another option, depending on what kind of data transformation facilities SPSS comes with (I'm not sure). $\endgroup$
    – Jake Westfall
    Jan 22, 2019 at 19:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.