I have 76 variables and sample size is 32. The skewness and Kurtosis are under acceptable range but the Kolmogorov-Smirnov normality test shows significant value.
My intention was to do a regression analysis, but now I am confused.
The data are collected in ration scale.
Should I normalize my data using statistical tools?
or, Is there any statistical test for predictions with non-normal data?
Deep -
1) You mean you have 76 independent variables and n = 32, so your degrees of freedom is negative? That stops you there. You have too many independent variables, if that is what you mean.
2) As for normality, the data do not need to be "normal" at all. It might be nice if the estimated residuals (or, considering weighted least squares, which may be much better, the random factors of the estimated residuals) were approximately normally distributed, but the data need not be. Even simple linear regression is possible if x and y are both skewed.
Cheers - Jim
Prof. Knaub is right as far as he goes. There might be some hope yet however. Let n be the number of observations and p the number of IVs. In biology there has been great interest in biostatistics to be able to build predictive regression models when p>n. This is often done using lasso variable selection. For details see the link:
https://www.google.com/search?rlz=1C1CHBF_enUS847US847&ei=CC4IXejZJoSctAX4lJKwBA&q=predictive+regression+models+when+P%3En+using+lasso&oq=predictive+regression+models+when+P%3En+using+lasso&gs_l=psy-ab.12...47776.55622..63422...1.0..0.200.2150.0j12j1......0....1..gws-wiz.......0i71j33i22i29i30j33i299j33i160j33i10i160.x6ZOSzF4Fz0
In addition, an example but with p
James R Knaub & David Eugene Booth Thank you both for your valuable suggestions. Regards
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Biomolecules
- Peptides