I have used LSD test for multiple comparisons of male and female but I feel LSD unfit for comparison (Need a rationale to use LSD if appropriate). Here I need some scholarly/ professional answers from gentle researchers. What would be used other than LSD? Is it HSD or any other? Please give your opinions and suggestions and provide a rationale for using a particular post-hoc test.
This discussion resolve a solution and it would make a baseline for other researchers when they encounter the same problem.
#ANOVA #MultipleComparisons #StatisticalAnalysis #SPSS #t-test #teachingLearning
It's not clear from your description what kind of model you're using. But the best approach for any reasonably complex model is to use the comparisons of marginal means supplied by the software you're using. This is likely called lsmeans or emmeans.
If there is no interaction between gender (male/female) and time (pre/post), then I would use the F-test from the ANOVA (main effect means) to compare male to female (assuming you have just those two levels). If you do have a significant interaction, you need to compare the simple effect means (male pre to female pre, etc). In these cases, I use the guidelines below as a general rule of thumb (to protest myself against making a Type I error):
- If you have a subset of preplanned comparisons between treatment groups to a control group only, use Dunnett's test.
- If you are performing all pairwise comparisons between simple effect means, use Tukey's HSD test.
- If you have few, preplanned orthogonal contrasts, use an F test.
- If you have few, preplanned non-orthogonal contrasts, use Bonferroni's test.
- If you have few, unplanned non-orthogonal contrasts, use Scheffe's test.
Dear Ashley Franklin its good to correlate analysis with different parameters. It increases my knowledge.
But though analysis, I have found significant difference between male and female using F test in ANOVA table. In addition, I did compare all pair-wise data i.e. pre-test male with pre-test female, post-test male with post-test female. Would I use Tukey's HSD or LSD? (which one is the best)?
And thank you for sharing though provoking information.
In that case, I would use Tukey's HSD. LSD is like running a bunch of independent t-tests and the Type I error rate gets inflated very quickly. LSD should pretty much never be used in my opinion
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA