Identifying multivariate outliers using Mahalanobis distance in SPSS

[@sarahcorcoran5317]

So I have done this and identified two outliers. Do I just remove them from my data set?

[@1988sheldon]

Dear Mike Thanks for the video. I wanted to know how this procedure is applied in a longitudinal dataset with IVs and DVs measured twice.

[@varunanewatiya7133]

Sir what do we do when the p value for some obervations are 1.00000

[@moroomario4007]

Professor Mike, if I am correct we have to convert in to summated scores for each responses of the items before proceeding to multivariate normality test?

[@shabraizmalik901]

Yes

[@scienceandtechnology9228]

how calculate degree of Freedom for chi square if we have paired samples pre and post test?

[@mimnun1]

Sir, shall I consider mediating variable as independent variable to check outliers

[@mikecrowson2462]

If you are running mediation analysis, then one of the regressions models will involve a regression of the dv onto the independent variable and mediator. So, if you want to identify potential multivariate outliers in relation to that regression equation, then this would do the trick. Cheers!

[@wibowo_ha]

Ok mike.. thnk you. and I subscribed you.

[@omidmahdieh7882]

Hello Mike. Thanks for your helpful demonstration. Can items be used to calculate Mahalanobis distances? Or should I use variables? I mean composite variables.

[@bandungmee]

Hi, do you have the answer for this? I am also figuring out how to perform MD test - total of 7 IVs and 23 sub-constructs. thanks

[@chuypulido]

Hi Mike. For a large number of independent variables, how do you determine which ones you'll use for the Mahalanobis distance calculation? Most videos I've watched seem to choose these variables rather arbitrarily. Thanks!

[@mikecrowson2462]

Hi Chuy, since this approach is aimed at identifying multivariate outliers on a complete set of IV's (for instance in regression), then it makes most sense to me to include that full set of independent variables when using this method. If you select a subset of the IV's then you are only identifying multivariate outliers on those, but you aren't identifying them on the full set of IV's. So, I don't really know the reasoning behind arbitrarily selecting a subset of the variables for which you are seeking to examine for the presence of multivariate outliers. Best wishes!

[@shafeekafadlikhzamri7068]

hello sir, what if my Md is too big and my p-value shows 1.00 . is it okay or not okay?

[@JunkFoodPapas]

Hello why do I only have an option for univariant?

[@rizwanwar434]

How can I generate simulated data with outliers

[@Drago6781]

I am really interested about the outliers issues, do you know any reference that supports this method? Thank you

[@mikecrowson2462]

The chi square test is referenced in Tabachnick & Fidell (2013), Using multivariate statistics test. The dependent variable doesn't matter when generating MD. You'll get the same MD with any dv because MD's are generated only from the predictors. Not sure where I read about the graphical method but I have seen it before. It makes sense to me.

[@scorpford2277]

I can not see what you typed in the expressions box for mean and deviation. Can you please tell me what was typed.

[@mandysnyder5896]

He typed 50 and 3

[@goldparas3381]

but i have 5 dependent variables. what should i do to detect the outliers??

[@mikecrowson2462]

Include them as predictors in a regression with a random variable as DV, and ask for Mahalanobis distance (as shown in the video). Mahalanobis distance is about detecting multivariate outliers, and you can detect them using this procedure - irrespective of whether the variables are IV's or DV's in your main statistical model.

[@mandysnyder5896]

@@mikecrowson2462 Great video! Thank you for breaking down these steps. Just to clarify on the response to gold paras: The random DV does not have to go into your final or main statistical model (as you demonstrated in this video) but the DV has to be continuous because it is a linear regression (not logistic regression), correct?

[@anupaav1250]

Hi Sir, Can you please tell me how to use the SPSS when I have two groups - patients and controls? I want to use the covariance matrix from the controls alone. Any helps would be appreciated. Thanks.

[@mikecrowson2462]

Are you asking about computing Mahalanobis distance on one group? I believe that you should be able to use the Select Cases function in SPSS to select which group you want to generate the values for. Here's a tutorial I found online: ezspss.com/how-to-select-cases-in-spss/