I have two different measures from two different groups. Both M1 (measure 1) and M2 (measure 2) from both G1 (group 1) and G2 (group 2).
Within each group, I want to find correlation between M1 and M2, that is, correlation between M1 and M2 for G1 and correlation between M1 and M2 for G2.
Between the two groups I want to compare the means of the measures, that is, means of M1 between G1 and G2 and means of M2 between G1 and G2.
My questions:
1. How exactly do I go about analyzing this?
2. Can I use the same data set (for example: 30 participants in each group G1 and G2) to do both of these analyses?
3. If yes, how does it affect my significance testing, power analysis etc.
Thank You!
You can combine the two groups into a single data set and use regression with a dummy variable (1 & 0 scoring) for group membership, and then use a t-Test for each of your continuous variables.
Hello Oviya,
To address the question in the title of your query: Yes, the same data may be used.
1. To test correlations across measures (M1, M2) separately by groups, ordinary Pearson r might suffice. For power determinations, the per-group n would apply.
2. To compare groups on measures, you'll need to first determine whether it makes more sense to do a multivariate comparison or two univariate comparisons (though it sounds as if you're more inclined towards the latter). For univariate comparisons, ordinary t-tests (as suggested by David L Morgan ) would likely work. For multivariate, one-way manova or Hotelling's T-squared (special case of one-way manova) would likely work. For power considerations, the combined N (n1 + n2) would apply.
Good luck with your work.
David L Morgan David Morse
Thank you!
I have a concern regarding significance testing as well.
If I would like to test significance at p=0.05, since I am doing two sets of analysis on the same data, would I be required to test significance for both tests at p/2, that is, p = 0.025 in order to ensure that there is indeed statistical significance?
Because, as I understood it, this is one of the major concerns with using the same data for multiple analysis or whilst testing multiple hypotheses.
Hello again Oviya,
You may certainly impose protection against an inflated risk of type I error over a set of tests by applying a more stringent significance criterion to the individual tests. Your proposed approach will certainly accomplish that goal.
Some folks distinguish instances of multiple tests that are independent of one another vs. multiple tests that are related (e.g., same error terms, for example), and you can find instances in which no adjustment is made for independent tests while adjustment is more frequently applied to dependent tests.
Good luck with your work.
You can certainly correlate M1 and M2 of either G1 or G2 separately using bivariate correlations of SPSS from SPSS data. Enter your data in two columns in SPSS data view for M1 and M2 respectively. Through Variable view, name the columns M1and M2 and run the analysis through Analyze -correlate - bivariate.
Using the same pattern of data input (M1 of G1 and M1 of G2 first, followed by M2 of G1 and M2 of G2 separately), clink Analyze - Compare means - Pairs-sample T Test to perform T Test analysis in other to compare the means of G1 and G 2 for their corresponding measures.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis