I want to explore regression assumptions for my variables of interest (in SPSS).
I have already made a Kolmogorov analysis, but I'm interested to know the residuals (errors) in scores variables.
How can I analyze such data?
I want to do a moderation analysis, but Kolmogorov analysis showed me that my moderator variable is not normally distributed, also the dependent variable and predictor variable. I have a sample of 200 subjects.
Thanks for all!
I understand that you have only one variable factor, if yes, you can do analysis of variance (ANOVA) using Design Expert or Minitab and do residuals analysis.
No real variable is normally distributed. This is known and does not need to be tested. And this is not the relevant point.
What is relevant is kind and amount of "non-normality" and if this may have any considerable impact on your analysis or conclusions.Unfortunately, there is no simple or general answer to that. What one can do is to check if there is some really striging "non-normality" assuming that this may likely be problematic. Minor deviations from normality are considered unproblematic.
The problem with small data sets is usually that even striking deviation might not be recognizeable, and it also may often be the case that random variation just looks like a striking non-normality. In larger data sets it is much simpler to recognize severe non-normality and it is also less likely to accidentally find stange patterns. But formal hypothesis tests on nomality will be sensitive to whatsoever minor and irrelevant deviation from normality, what is not helpful. Especially in larger samples non-normality is even less a problem, because the inference is done on the likelihood (or the sampling distribution) which will always be a better approximation to the normal distribution, and the approximatiin gets better with increasing sample size.
Instead of doing some formal tests (that are, by the way, really illogical in this context), you should plot the residuals and look for obvious patterns. There are several diagnostic residual plots that show different aspects. Especially the normal-quantile-quantile plot (Normal-QQ plot) is a good way to see if there is any severe problem with non-normality.
You can google for residual diagnosic plots and normal-QQ plots. There are many resources explaining the possibilities, details and help you with the interpretation.
Dear Sara
When you have the linear regression dialog box, you can see a button on the right side of the box named "save". Click on it and in the residuals menu select the appropriate one. Usually for normality test i check mark "unstandarded residuals". After clicking final ok, one variable will be added to your data sheet. Now you can select this variable for normality test.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression