Hello,
Maybe is a bit silly question but, I have a variable (N=78) and on shapiro-wilk test the sig. is .008 which is just over (greater) than 0.05 meaning that is just about normally distributed. However, skewness is -.305 and kurtosis .237. Does this mean it is not so it's better to run non-parametric tests?
Victoria St, I can appreciate your concern. Perhaps the document I'm attaching will help, particularly the last page. I put that document together a couple of years ago when I was troubled by the same kinds of issues you raise.
In essence, check your histograms!
All the best with your research.
I agree with Robert Trevethan. The acceptable values for normal distribution for skewness is ±2 and ±7 for kurtosis. This is recommended for data having a sample size greater than 300. Stats tests like Shapiro-wilk sig. etc. are based on certain assumptions and hence have their limitations. Thus checking data skew/kurtosis values and graphically (histogram, Q/P-plot along with stats test to get more confidence.
The interpretation will depend on the nature of the data and the respective analysis
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running