Data set failed test of Normality now establishing the correct course of action I need to take regards using non parametric statistical tests?
Scenario:
Internal and external Training load data for a week
Analysis I wish to conduct:
1) Compare each day training day across the week
2) Compare the training load for each position ( defender,midfielder , forward) for each day across the week
3) compare Starters vs Non Starters
Having done some reading here is my suggestions:
1) Conduct a Kruskal wallis test followed by Mann Whitney. As the kruskal wallis will establish if there is significant difference and the Mann whitney U will establish exactly were the differences are.
2) Genuinely have,nt a clue how to do this one .
3) Mann Whitney U test
Any advice would be greatly appreciated . Thank you for your time .
Firstly, what is sample size? how you test normality? and what is kurtosis and skewness values? maybe you can perform parametric tests.
If you must use nonparametric tests, first and third research question's statistical tests seems true. For second question you must perform non parametric version od manova. In R enviroment, nmpv and coin packages are using for this. But i never used it.
I would start by telling us about the data you have collected. Different methods require specific data types, so that would help greatly. I would suggest that some sort of regression is probably what you want for most of these. I would start with this link: https://www.google.com/search?q=How+to+choose+a+type+of+regression+model&rlz=1C1CHBF_enUS874US874&oq=How+to+choose+a+type+of+regression+model&aqs=chrome..69i57.29346j0j1&sourceid=chrome&ie=UTF-8
Also note that if your data doesn't fulfill the requirements of a particular method there are often ways to fix that, data transformations, etc. We can discuss that once you know what you want to do. Please describe what you mean by "data set failed test of normality". I would not immediately go to nonparametrics based on what you described. Please plot your data in the most reasonable manner. This is usually the first step in a data analysis. Then come back with your question. Best, David Booth
Hi Hakan Koğar & David Eugene Booth ,
Nice to meet you and thank you for taking the time to provide guidance and advice, its very much appreciated.
Data Description :
20 athletes monitored over a weeks training ( 7 days in total).
External load = ( daily GPS metrics )
Internal Load= ( daily session RPE & training load units)
Analysis wish to conduct:
1) Describe each training day ( means) and the potential differences between days.
2) Describe and identify any potential differences between the different playing positions on each day.
3) athletes fall into starters and non starters and I want to identify any potential differences across each day during the training week.
My initial approach:
Test of Normality:
Data set would have less than 2000 data points , so I should use Shaipro Wilks but all my variables have p values less the.05 , therefore the data is not normally distributed. Am i correct in this assumption?
Kolmogorov- Smirnov indicates all the variables bar one have a p value under 0.05.
This has lead me to think that I may need to run non - parametric tests. Has this painted a better picture .
Normality is the least of you worries! Being a bit more serious, you have repeated measurements of the same individuals so that you have to take that structure into account otherwise the standard errors are likely to be incorrect due to the inherent dependence (within person similarity) in your design. In general and in my experience, such non-independence trumps non-normality with continuous measures unless there are serious outliers present.
Have a read of this straightforward and short piece for inspiration about what you can do.
http://www.bristol.ac.uk/media-library/sites/cmm/migrated/documents/repeated-measures.pdf
and for lots more see
http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#multipleresp
I agree with Dr. Kelvyn Jones. You dont have to use non-parametric statistics with 2000 data points. Test normality not with statistical techniques like kolmogorov-smirnov. Skewness and kurtosis values is better choice in big data sets. If you have valuse betwen -1 and +1, there are no worries about normality. I am sure that normality is not a problem here.
I also agree with Dr. Jones but you might want to look a recent paper that compares both types of approach. I am still not sure of your data but I will attach the paper in case it will be useful for you. Best wishes again, D. Booth
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training