I am conducting a causal-comparative study for my dissertation, and am under quality review. The reviewer told me that I should prepare for post hoc testing by conducting independent t tests a priori in addition to my G*Power analysis since my sample is small. Can someone explain to me why she has recommended this? I'm struggling to understand and respond in my research proposal.
Thank you in advance.
Tiffany Hastings
Hello Tiffany,
Unfortunately, there's not quite enough information to give a helpful answer. Perhaps you could elaborate:
What is/are the research question(s) you're trying to answer?
What sort of analysis had you planned for this study?
How many and what kind(s) of variables are involved?
What sample size do you propose to use?
Good luck with your proposal and dissertation research!
I will use MANOVA to complete the analysis. I am seeking the difference in employee commitment (affective, continuance and normative commitment) between groups of employees (above-average, average, below-average, past- and non-participation) participating in various philanthropic activities (payroll giving, one-off fundraising events, and volunteerism). So, I have 3 groups of employees per activity type, and evaluating each of these among each of the 3 types of employee commitment - 5 groups, 3 independent variables, 3 dependent variables. G*Power states that with these components for MANOVA, I need 100 survey participants (20 per group). The only thing I can figure is that she's assuming I may have to do some paired comparisons in post hoc analysis because I'll have some MANOVA results to further decipher. But that would consist of a series of one-way ANOVAs to drill down further. Would independent t tests be appropriate for determining adequate sample size for that part of the analysis? Or is this somehow to meet the adequate sample size assumption for MANOVA?
Hello Tiffany,
You have 5 employee groups (that's one IV). You have a second IV, type of activity (3 levels): Is this a repeated measures (within-subjects) or between-subjects variable? Then you have three related DV measures (employee commitment scores). Is that correct?
If so, then the G*Power analysis should involve three different comparisons: (a) Are the 5 groups comparable on commitment scores (across activity); (b) Are the 3 activity types equal with respect to commitment scores (across group); and (c) Is there a group x activity interaction with respect to commitment scores?
Of these three, use the largest "requisite" N as your guide, then add a bit to allow for attrition, failure to follow protocol, not completing the survey form(s), or other reasons for having incomplete data.
Given that (and I presume you've done this), I see no need for worrying about sample size for independent t-tests, either in lieu of the manova or as a follow-up test. Ordinarily, traditional pairwise post hoc tests are only completed in the presence of a significant omnibus test result (for you, that's the three manova hypothesis tests). All other things equal, a univariate test tends to be more powerful than a multivariate test (though the multivariate test can take account of the profile differences as well as ordinary differences in group means), so the reasoning for the t-test power determination eludes me.
As a follow-up, you'd likely be better served to try discriminant analysis to discover more clearly the ways in which groups are different across the set of measures instead of just running a bunch of (univariate) anovas followed by some sort of protected post hoc, where indicated.
Good luck with your study.
- Badges
- Science topic
- Similar topics
- Medical Science Nursing
- Dissertations