I have longitudinal data over several weeks (17). For the outcome measured, some of the weeks have noramlly distributed data (n=10/17 weeks) for the remaining 7 weeks the data are not normally distributed. How should I best describe the results? As per their distributeion ( 10 weesk as mead and 7 weeks as median) or all as mean or all median?
Thanks
hi Dawn Cooper
What kind of measurement scales are you using? It may be that a metric scale (continuous variable) is not normally distributed; or it may happen that your data comes from ordered categorical scales (Likert scale type).....the situation in either case is different. You must also take into account other issues: What is your sample size? What degree of deviation is there from normality? If your sample is large, the variables are continuous and the deviation from normality is not large, you could estimate the mean for all periods without problem. If your scales are ordinal (categorical ordered), I suggest you use the median in all periods
However, some authors suggest that in the latter case ordinal variables can be treated as continuous under certain circumstances.
https://www.statisticssolutions.com/can-an-ordinal-likert-scale-be-a-continuous-variable/
Article Resolving the 50-year Debate Around using and Misusing Likert Scales
Of course if sample size is large and the variables are continuous and the deviation from normality is not very high we can use the mean for all periods .But if the data is ordinal ,categorical or nominal. We can use the median.
Hi. You may find my response here: https://www.researchgate.net/post/Is_Repeated_Measure_Analysis_for_a_Non-parametric_data_possible
interesting. Best wishes, David Booth
If I may ask you what is the objective of the study. If you need to include all the 17 weeks into the study and analyze, then carry our a KS & Shapiro Wilk's Test, Check for Kurtosis/Skewness check if sig. value is above 0.05 or not, provided sample size is large enough. But if you need to split the entire study into two seperate intervals for the sake of objective of the study. Then go for seperate normality test for each intervals and then conclude whether they would be parametic or non-parametric.