Close Home Forum Sign up / Log in

npar 1 k-s test accepable for testing normailty?

A

======= Date Modified 14 Jul 2011 20:13:56 =======

hi im doing the stats for my msc disso at the moment, testing the data to see if it meets the parametric assuptions i looked through my notes and realised that there are 2 ways to use the Kolmogorov-Smirnov test.
1) Analyze-Nonparametric Tests-1-Sample K-S… place h variables in the Test Variable List… box, click OK
2) Analze - Descriptive Statistics, then Explore, plots button. Under descriptives click on the Histogram. Click on Normality plots with test.

i get differnce answers for both: 1) non-sig 2) sig. i want to use it to test normality ideally using 1)
i have got 6 variables (interval)

sample size is 185 and even thou ive got 2 groups (gender) i want to test it as a whole sample so i would assume 1) NPAR method is acceptable or am i wrong??

hope this makes sense

The K-S test is dodgy at the best of times, out of about 40 students, I get about 35 each year with sig K-S tests. I'd report it, but then say after you inspected histograms, skewness and kurtosis values were deemed to be acceptable (reference here) and therefore the data was treated using parametric analyses (if you actually think the histograms do look normal).

A

Quote From sneaks:

The K-S test is dodgy at the best of times, out of about 40 students, I get about 35 each year with sig K-S tests. I'd report it, but then say after you inspected histograms, skewness and kurtosis values were deemed to be acceptable (reference here) and therefore the data was treated using parametric analyses (if you actually think the histograms do look normal).


thanks for that quick reply, what about the fact that i use One-Sample Kolmogorov-Smirnov Test to test the normality which is non-signifcant
and is it acceptable to use this method even though i have 2 groups/ 6 varibles within the data but test it as a one sample?

check the normality within each group. Although if you have 6 variables you may want to consider a MANOVA, if they are all dependant variables, or at least make sure you make a boneferroni correction to your p value.

18831