I am using Mahalanobis Distance for outliers but based on the steps given I can only insert one DV into the DV box. Unfortunately, I have 4 DVs. After I have done all the steps for MD, Probability and the outlier, I have found out that each DV showed the same outliers. Does this same outliers results tell me that I need to delete the same cases from each DV?
Hello Shafeeka,
What software are you using? Of course, the Mahalanobis distance (D^2) is computed based on the set of relevant variables, not just one at a time.
In SPSS, the way to get D^2 is to run the linear regression subprogram, put all four variables in as INDEPENDENT variables, then name something else as the DV. In the dialog box for regression, click the "Save..." button, then check the box for Mahalanobis distance in the subsequent box. This will cause the D^2 values to be added to your data worksheet. (Ignore the actual regression output itself.)
If you're using R, here's a link that explains how to obtain D^2 values:
https://www.rdocumentation.org/packages/stats/versions/3.6.1/topics/mahalanobis
Good luck with your work.
Take a look at the attached paper. You might find it useful. This is very easy to do in a stat package. I have code in some of them. Best, David Booth
David Morse hello Sir, I highly appreciate your reply. I am using SPSS for the Mahalanobis Distance Test. So technically what i did is that I compute the sum of all IVs & DVs. I then put one DV into the DV BOX and repeat this same step with the other 3 DVs and i put all the INDEPENDENT VARIABLES into the IV BOX provided. Therefore, I have 4 sets of MD from different DV but with the same outliers results.
Hence, from reading your comment, I understand that I need to create a dummy for the DV box, and put all DVs and IVs into the independent box provided? Is this correct?
Plus, i also have mediators and moderators. Should i include them as well in the IV Box?
David Eugene Booth thank you very much Sir, will refer to it
Hello Shafeeka,
1. So far as computing Mahalanobis distance goes, yes; put all 4 measures in SPSS as independent variables, and anything (even random numbers, though there has to be _some_ variation in the ersatz DV in order for the program to run) as a DV in the regression subprogram to elicit the computed D^2 values.
2. You didn't mention mediator or moderator variables previously, so obviously nobody commented on them. So now, we go back to the beginning: what is/are your research question(s)? It's starting to sound like a SEM model framework to me. As regards moderators, it depends on the nature of the variable; many of these are categorical, representing subgroups. If so, it would likely make more sense to check for outliers separately by group. Mediators likely should be included in assessment of data for outliers (or, outright liars). The answer for one specific model might not be the same as the answer for a different model.
Good luck with your work!
Hello Akmal,
Many simply refer the Mahalanobis distance to a chi-square distribution with k degrees of freedom, where k is the number of variables involved in the computation of the D-squared values. Give this previous thread a look : https://www.researchgate.net/post/Anyone_have_mahalanobis_distance_critical_value_calculator_or_excel
Here's a more formal look at the question (as related to using D-squared as a possible indicator of a case as an outlier, from some presumed distribution):
Penny, K. I. (1996). Appropriate critical values when testing for a single multivariate outlier by using the Mahalanobis distance. Journal of the Royal Statistical Society, Series C (Applied Statistics), 45, 73-81. doi:10.2307/29986224
Good luck with your work.
Shafeeka fadlikh Zamri it is really good question..
David Morse thanks a lot for the valuable contribution, it is really helpful
- Badges
- Science topic