Hi,
My model has total of five variables, Four IVs one DV and one Moderator. I am checking for the outliers and normality of the data. Please guide me if am adopting the right process. I calculate the means of all the responses and then put all the variables in Explore using SPSS and check for the Normality option.
When I remove the possible outliers using Box Plots and P-P and Q-Q plots and run the explore again, I get the outliers again. Eventually when the outliers are gone, the data still does not get normal.
Please guide.
As first you must decide if you want to remove the observations that are outlliers 'SEPARATELY FOR EACH DIMENSION' or 'GENERAL OUTLIERS' in the five dimensional space. If the latter is the case you should work in the Mahalanobis distance space from the centroid of the data cloud (the point vector having as components the avergae values of the five variables). Distance-from-the-centroid is now a single variable on which you can work in the usual way to detect eventual outliers.
On the other hand a given distribution can be non-normal not because of outliers but because it is INTRISICALLY NOT NORMAL (see attached file), notwithstanding that, contrary to the accepted folklore this situation can be managed as well by usual inferential testing strategies.
On the other hand you should deliberately use a non-parametric (or any other approach not asking for normality) for other reasons like censored data or because you do not trust in the actual numerical value of your measures but only on their rank order...that is to say that (like in every situation) A GUIDE AN BE EFFECTIVE ONLY IF YOU KNOW WHERE YOU WANT TO GO.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs