I am interesting the parametric test in my research. My dependent variable is continuous and sample size is 300. so what can i to do? I am request to all researcher which test is more preferred on my sample even both test are possible in SPSS.
Dear Govinder,
I have been advised that the Shapiro-Wilk test is generally more sensitive for sample sizes up to one or two thousand. I would therefore recommend looking at the Shapiro-Wilk test first then, if necessary, looking at the Kolmogorov-Smirnov test as a backup.
Both tests also have the tendency to be too sensitive for the purpose of selecting a parametric test when the sample size is larger than one or two hundred. I have been advised that in these circumstances it is wise to also look at other visual representations of normality, such as histograms with fitted normal curves.
There are also robust exceptions for many parametric tests which mean that their results are often valid even when normality tests suggest that the data is not normally distributed. These exceptions depend of the individual tests and are generally based on simulation studies.
Dear Bernardo dos Santos,
Thank you for provide the link but price of publication is more expensive for learners researcher. I am from Nepal. I cannot purchased it. But article is very useful for me. so, it can possible for read in other link
Best regards
Govinda prasad Dhungana
Dear Govinda,
If you could send a test email to [email protected], I could share the article with you.
Regards,
Anupam
you can share at [email protected] and we should apply the shapiro will test for normality ...as well as QQ plot
For small sample sizes the Shapiro wilk perform much better than any other normality test, it is commonly acnowledged https://en.wikipedia.org/wiki/Shapiro%E2%80%93Wilk_test. I attach the performances of some tests applied to an exponential distribution with parameter =1 (study with 5000 replicates). When it comes to large sample size those tests are too sensitive and I would suggest visual inspection of the histogram or the qq-plot. I hope it helps...
Dear Cristian Ricci
Thank you kindly response my quarries. It suppor to me in further analysis in applications field in real data
The Shapiro-Wilk Test is more appropriate for small sample sizes (< 50 samples), but can also handle sample sizes as large as 2000.
The normality tests are sensitive to sample sizes. I personally recommend Kolmogorov Smirnoff for sample sizes above 30 and Shapiro Wilk for sample sizes below 30.
I iteach my students to first study the scatter plot of the data, followed by a QQ plot. Exploratory data analysis is the first step. Tukey’s text on EDA explains why.
Johnson & Wichern provide a table with critical values fir the correlation test between data quantiles and normal quantiles to check the QQ plot. Then the Shapiro Wills Test is also performed. If the tests and plots do not suggest normality, either a Box Cox transformation is done or a suitable nonparametric Test is used.
The Shapiro Wilke Test is recommended overall for better theoretical properties.
Some times skewness and kurtosis value between -2 and +2 is accepted in social science.
Hi Govinda, yes given that your sample size is 300, the Kolmogorov-Smirnov test would be most appropriate. If the p value is >0.05 then you can reject the null hypothesis, that the data is not normally distributed, and proceed with parametric testing. The Shapiro-Wilk test is appropriate for sample sizes less than 50.
you guys interested in this field may read and cite this work :P where the performances of common normality fitting tests for small samples are explained Article Determining sample size adequacy for animal model studies in...
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size