Stats novice needs your help..::.. please..::..

[beka]

Hi

I am trying to compare the means of some values between three groups and have used ANOVA, which has shown a significant difference (0.00). I have been told that I also should report the Levene's statistic (0.026). While I can see that there is a difference between two of the groups and the final one (that is, two of the means are very similar and the other is obviously different) I think I am also meant to do a post-hoc test to show which one this is?? Is that correct? I have done a Kruskal Wallis where the mean rank (differences?) are 51.00, 22.19 and 22.27. I have written for this that there is strong evidence to suggest a difference in ranks - but I am not even entirely sure what this means, and whether the Kruskal Wallis is the right test to use? Does the significance in the Levene's statistic invalidate the fact that there is a statistical significance when using ANOVA?

You may have guessed that I am a novice dabbling in statistics because of necessity so any help would be greatly appreciated, and also, when reporting KW do I need to report the values, and which test I used (if this is the right one?)

Thanks in advance for any help, please let me know if there is anything I can clarify

Beka

[Philyuko]

Beka

As far as I know, the Levene's statistic (here, 0.026) is not great news for you. Levene tests whether your groups are appropriate (i.e., NOT different) so here you want Levene to be GREATER than .05 (I know, it's weird).

But, here's the good news. As an applied stats kinda guy (rather than a student of ...) I would argue (as I always do) that ANOVA is a pretty robust test and if you have a sig result (with your F and your P and your eta squared) and those results are in the right direction that you don't need to worry about Levene's. It's not like ALL your groups were out and all were out by A LOT! What I would say is ... if you DIDN'T get a result then LEVENE'S might explain why you DIDN'T get one ... rather than Levene's being a reason to ditch your positive result.

Now - if you have a really ODD number of cases in each group, like 12 in one and 300 in another then, well, then, well, that's not great again. Or if you have a low number in each group (like less than 20) then, well, then, well, that's not great again ...

science is art Beka, science is an art ...

[beka]

Thanks! Some of my sample sizes are on the smaller side. Ok, so one is less than 20 but there's not a lot I can do about that and is the nature of the field that I'm in...