I have performed a Pearson correlations for need for cognition scale scores (NCS) against each of the 3 sub-scales for multifactorial memory questionnaire (MMQ-Contentment, MMQ-Ability & MMQ-Strategy) however if I calculate outliers for each of these variables independently based on 1.5*IQR, 20 out of my 220 participants are removed which seems excessive and possible over-conservative.
I have tried to detect multivariate outliers in jamovi but I am really struggling with the code! Can anyone please help?
I've tried to use https://towardsdatascience.com/mahalonobis-distance-and-outlier-detection-in-r-cb9c37576d7d for guidance but don't know what the equivalent of 'airquality' would be in my data... I tried the term 'nc1' but I'm receiving the error object 'nc1' not found which I think means this should be a pre-defined variable. Below is the code I've tried to use resulting in the above error.
nc1.data = nc1[c("NCS", "`MMQ-C`")]
nc1.center = colMeans(nc1.data)
nc1.cov = cov(nc1.data)
distance = mahalanobis(nc1.data, nc1.center, nc1.cov)
cutoff = qchisq(p=0.95, df = ncol(nc1.data))
nc1.data[distance > cutoff, ]
Help much appreciated!!
Repeat the below screenshot search for many suggested packages that will be much easier to use. Google your selected package for example code with explanation. Best wishes David Booth
BTW I don't think that your formulas for calculations were correct. You might want to look at the attached paper and compare with your formulas. Best wishes David Booth
David Eugene Booth thank you so much- apologies for the delay in my reply.
I've been able to use spss to find the multivariate outliers and find only 1 with p < .001.
Given the large difference between the IQR method on the individual variables vs. the multivariate method, are you able to advise which is most appropriate for my particular scenario? My supervisor has raised the concern that as the variables are expected to covary, the multivariate method may not be appropriate but as covariance is calculated in order to determine the outliers, I thought that it was. Unfortunately, as they are unfamiliar with the multivariate method, they can't advise any further on whether I should proceed with it or revert to the outliers for each variable individually. Any help, much appreciated!
FGS just read the references I sent you. It should then become clear they know what they are doing so follow that advice. I have attached a simple introduction to the concepts. Good luck to you. D Booth