Hello Van. YMMV, but IMO, it is rather "silly and pointless" to rely on statistical tests of normality when deciding whether a parametric test or procedure is defensible. This short conference presentation I gave several years ago explains why I think that:
- https://www.researchgate.net/publication/299497976_Silly_or_Pointless_Things_People_Do_When_Analyzing_Data_1_Testing_for_Normality_as_a_Precursor_to_a_t-test
Furthermore, I suspect that you are not the first person to ever study the two DVs you are examining. And if you are not, what is known about the population distributions of those variables from previous studies? The point here is that relying exclusively on very small samples to make inferences about population shapes does not make if lots is already known about those population shapes. This is captured nicely in the final paragraph of this BMJ Statistics Note by Martin Bland:
The aversion to parametric methods for small samples may arise from the inability to assess the distribution shape when there are so few observations. How can we tell whether data follow a normal distribution if we have only a few observations? The answer is that we have not only the data to be analysed, but usually also experience of other sets of measurements of the same thing. In addition, general experience tells us that body size measurements are usually approximately normal, as are the logarithms of many blood concentrations and the square roots of counts.
I hope this helps.
------------------------------
Bruce Weaver
------------------------------
Original Message:
Sent: Sat August 24, 2024 01:16 PM
From: Van Filopov
Subject: Test for normality of distribution (Explore) SPSS
Please advice. I have three groups - 1 experimental, 2 experimental and control (n=8 animals in each). I am testing the level of ALT and GLO.
I will use one-way ANOVA and the Shapiro-Wilk normality test for a small sample. For ALT, there was a normal distribution in 2 experimental and control group, but not in 1 experimental group. For GLO, there was a normal distribution in 1 and 2 experimental groups, but not in the control group.
If one or both groups show a normal distribution and the other non-normal - what is the conclusion - is the distribution normal or not for the respective dependent variable?The conclusion is that the test for normality of distribution found that some of the variables have a different distribution. To run a One-way ANOVA I need all three groups, and in fact it turns out that I can only use a fraction of them (with a normal distribution). My question is - what about the group that is non-normally distributed? There is no way I can run a non-parametric test (Kruskal-Wallis) with just her and a parametric test with the others, because therefore I will get two differences in p-values.
In this case, can I assume that all groups show a non-normal distribution and switch to a non-parametric Kruskal-Wallis test? what should i do Here is the result table:
------------------------------
Van Filopov
------------------------------