I have two groups each group has before and after readings (continuous data). I compared between the means using paired t-test in each group separately. and then compute the difference in each group and compared using independent t-test. Is this procedure correct? is there any other way to compare between the groups in a whole?
Hello Hadeed,
Options:
1. A one-between (groups: 2 levels), one-within (test occasion: 2 levels) ANOVA. The test occasion represents a repeated measures dimension. This design allows you to test for: (a) main effect of group (across the two occasions); (b) main effect of occasion (across groups); and (c) a group by occasion interaction. This last test will tell you if one group showed a greater pre-post change than another.
2. Use the pretest scores as a covariate, and run a one-way ancova. The result will indicate whether groups, when adjusted for pretest differences, had equal adjusted/estimated/predicted scores on posttest.
3. As you and Ajit suggest, run pairs of dependent t-tests, and pairs of independent t-tests. This last approach is least satisfactory to me, as you're running four tests, with altogether different error terms for each one.
Of course, each of these options requires making assumptions about the nature of the scores, as evidenced by the behavior of model residuals (normality, homogeneity--for the between groups tests, and independence).
Good luck with your work.
David Morse Thank you so much that was really helpful. I used both options (1&2) and both of them gave the same results. However, I prefer option 2 as it gives the estimated means of post-readings after adjustment for the pre-readings.
My question here: can I add another factor (for example degree of gingival inflammation) as a random factor to adjust for in the analysis?
Hello again Hadeel,
Yes, you may add other covariates to the model.
Good luck with your work.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis