Timeline for How to conduct test for normality before conducting an ANOVA?
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 30, 2017 at 22:06 | comment | added | Frank Harrell | I assume you mean multiple comparison procedures. For this I use those tests' generalization: the proportional odds ordinal logistic model. With that model you can form any contrasts you need on the log odds cumulative probability scale. | |
| Aug 30, 2017 at 0:37 | comment | added | Anonymous | @FrankHarrell Kruskal-Wallis Test has an asymptotic efficiency of 95.5% compared to ANOVA, similarly for Mann Whitney U post hoc test compared to t-test. Which post hoc test do you guys suggest to use then? | |
| Aug 29, 2017 at 22:44 | comment | added | Frank Harrell | I don't have my finger on it but not hard to simulate. Clearest example in when researcher bias causes interpretation of test of normality to be influenced by significance of the first p-value computed. | |
| Aug 29, 2017 at 19:52 | comment | added | Michael Webb | @FrankHarrell, could you point me to any papers or resources that talk about how checking the data to find out how to model the data can distort statistical inference? This seems like a common practice but I have been wondering about this. | |
| Aug 29, 2017 at 15:21 | comment | added | Frank Harrell | But note that nonparametric methods perform excellently under normality too. | |
| Aug 29, 2017 at 15:00 | comment | added | Anonymous | Thanks a lot guys! I'm still dealing with the controversial issue as to whether likert's scale gives ordinal or interval data. | |
| Aug 29, 2017 at 13:53 | comment | added | Sal Mangiafico | Yes, combo of samples for each DV, but you want to look at the residuals from the analysis, not the original data. You should also take to heart @Frank Harrell 's comment. Especially if your dependent variables aren't really continuous variables, a nonparametric test may be more appropriate (Kruskal-Wallis perhaps). For what you're doing if you're not real comfortable looking at the assumptions of anova, probably K-W with a Dunn test post-hoc will be easy, familiar to readers, and will serve you well. | |
| Aug 29, 2017 at 13:35 | comment | added | Anonymous | Our sample size is n=20, still too sensitive to non-normality? | |
| Aug 29, 2017 at 13:32 | comment | added | Anonymous | @FrankHarrell Use Kruskal-Wallis Test instead?(the non-parametric equivalent of ANOVA) | |
| Aug 29, 2017 at 13:32 | comment | added | Anonymous | So visual methods of a combination of 3 populations for each dependent variable? Sorry but our school only taught us until Pearson Correlation Coefficient, all of these were my part of my personal research. | |
| Aug 29, 2017 at 12:47 | comment | added | Frank Harrell | A bigger problem is not having a power of 1.0 so missing non-normality when n is small. It is better practice to just use nonparametric tests if you are not somewhat certain of normal residuals. Checking the data to find out how to model the data can distort statistical inference. | |
| Aug 29, 2017 at 12:45 | history | answered | Sal Mangiafico | CC BY-SA 3.0 |