Hi, I have a data set I am working on. The data is not normally distributed (Image 5). I tried to transform with log transformation and Square root transformation; both of them did not work—any suggestions to approach this.
If you cannot find a transformation that normalises the data, you might have to use non-parametric test. Some thoughts here: https://blog.minitab.com/blog/adventures-in-statistics-2/choosing-between-a-nonparametric-test-and-a-parametric-test
Hi,
as Donald Batten suggested, a non-parametric test would work, but it depends on your data.
Check the following non-parametric test to find the one which might fit to your analysis.
https://www.spss-tutorials.com/nonparametric-tests/
Best regards
Why do you want them normally distributed? If you want transformations that will match things like skewness being low, try the Box-Cox family.
Also, to convince someone that data are not approx normal, a qq plot would be better.
Try log2 or log10 transformation. If it is not working, try nonparametric test
There are 5 critical decisions for the distribution of normality. If there are 3 of these, there is a normal distribution and parametric testing can be used.
1. The variation must be less than 30%
2. Detrended Normal QQ plot test should show random scattering
3. Skewness and kurtosis must be between -1.5 and + 1.5. Also, their standard errors of double the amount must be greater than the statistical value (numerical value)
4. Normality tests should be> 0.05
5. Histogram chart should be the normal curve
If the data is not normal after all the possible exhausting transformation methods, then one may need to resort to non-parametric statistical approaches. Although, the non-parametric approaches are not as strong or robust, but are accepted. For example, instead of one-way ANOVA, one might need to use Kruskal-Wallies test. Best wishes.