Dealing with ties in data
Posted: Wed Nov 11, 2020 4:46 pm
Hello,
My data: participants are in two groups - treatment and control and the dependent variable of interest is the proportion of correct identification of a set of tactile stimuli (so basically ranges from 0 to 1.0). We hypothesise that the groups will perform differently on the DV due to the treatment.
Now, the problem with the data is that it is not normally distributed and that there are several ties in the data (ceiling effect with a sizeable proportion of participants from both groups scoring 1.0 or 0.9375). I understand that ties in the data are not ideal even for non-parametric tests like Mann-Whitney. So, I had a couple of questions:
1) How does the Mann-Whitney test in JAMOVI deal with ties? Does it use midranks, or randomisation (ref: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6057651/)?
2) Is it advisable to add random noise to jitter the data and then run the Mann-Whitney? What about if this is done multiple times and the average test statistic from all the iterations are calculated?
3) What ways would you recommend to deal with ties (particularly those at ceiling) in your data?
Thank you for your help!
Matthew
My data: participants are in two groups - treatment and control and the dependent variable of interest is the proportion of correct identification of a set of tactile stimuli (so basically ranges from 0 to 1.0). We hypothesise that the groups will perform differently on the DV due to the treatment.
Now, the problem with the data is that it is not normally distributed and that there are several ties in the data (ceiling effect with a sizeable proportion of participants from both groups scoring 1.0 or 0.9375). I understand that ties in the data are not ideal even for non-parametric tests like Mann-Whitney. So, I had a couple of questions:
1) How does the Mann-Whitney test in JAMOVI deal with ties? Does it use midranks, or randomisation (ref: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6057651/)?
2) Is it advisable to add random noise to jitter the data and then run the Mann-Whitney? What about if this is done multiple times and the average test statistic from all the iterations are calculated?
3) What ways would you recommend to deal with ties (particularly those at ceiling) in your data?
Thank you for your help!
Matthew