I have done a survey, I have got some data , each variable has a value ranges from 1 to 5, my variables are not normal, is there any way to normalise them? the winsorising does NOT work with them.
Good morning Lara, which exact variables are not normally distributed? the response variables are not necessary to be normally distributed, in most linear model cases, they typically are not normal, the residuals from the linear model have to be normally distributed. If the residuals turn out to be non-normal, you can use Robust Regression. moreover, Non-parametric tests can be used If your samples are independent.
I suggest you determine your dependent and independent variables first. Which is your main outcome measure? If your dependent variable is not normally distribute, try opt for non-parametric tests.
Dear Professor Ghazi,
Thank you very much for your prompt and useful response. I really highly appreciate that.
Kind regards,
Lara
Hi Lara, from what you say, it looks like multinomial data. It is common that these variables are not normal and usually dependent variables are not normal due to the effect of treatment. It is residual which should be normal. So, I think you can use multinomial distribution in your model and also, I prefer not doing non-parametric test as they are not powerful and nowadays, you can use glimmix procedure with opportunity of adding distribution of your data. Hope this helps.
- Badges
- Science topic
- Similar topics
- Statistics
- Survey