Hi everyone,
I have been given an assignment for uni and I am not sure whether to keep or remove outliers. After running normality assumptions, the data reported to be normally distributed, however, study of box plots i spotted a few outliers in a few of the independent variables. My main concerns are, is it still ok to run correlation and regression analysis, and should I report the outliers in the write up? I wasn't at data collection, therefore they may be errors, but it is an assignment as part of my masters degree, so the correct way to go about this would be very helpful.
Thanks
Dear Peter,
I think the right way is to check the entries. Sometimes there are incorrect data entries and it is quite simple to correct the wrong data. For example, the hight of the patients cannot be 173 m.
Another reason might be, that some patients have really patholic measurements - the measurements might be correct and must be analysed.
In some cases, the measurements may be wrong - for example because of insufficient devices. In these cases, you have to check carefully if it is possible to take the values or if is better to delete them. It you do so, you should describe it in your methods.
Good luck for your study!
Best
Tanja
Thank you Tanja.
I inspected the data and they don't seem to be mistakes, therefore, I will keep them in and also mention them in the methods.
The real issue is not so much whether outliers exist but whether they influence the results of your analysis. If you are using correlation and regression, there are specific techniques (e.g., assessing leverage) for determining the impacts of outliers.
Dear Peter,
I think it might be useful to conduct the correlation/regression analysis considering the two scenarios: with and without the outliers.
This way you can examine whether the existance of outliers in the dataset affects the conclusion/outcome of the analysis or not.
Best Regards,
Attila
- Badges
- Science topic
- Similar topics
- Psychology