BookmarkSubscribeRSS Feed
Obsidian | Level 7

Hi Everyone, 


So I have two variables I want to denote as targets "race" and "mental illness".

I've attached it below, I preprocessed most of the interval variables into 1,0, while Ive also added a armed category and made the races as asian, black, hispanic, white, etc. 


At the moment this is how my diagram looks


I want to analyze the effect of M.I and Race on police shootings and want to carry out logistic regression and decision trees to do the prediction based on the two possible target variables.


However after looking at the regression node page on sas i saw that it doesnt accept nominal variables? Now i'm lost and confused because on certain places it does mention it accepts nominal. How should I go about this?

Should I only go with mental illness as my target variable now?


p.s the original dataset is from WAPO Fatal police shooting dbs.


Hello @b_smsha ,


You are using the wrong terminology here.

If you say: I want to analyze the effect of M.I and Race on police shootings 

, then

  • M.I. and Race are predictors / independent variables / explanatory variables / input variables / X-variables
  • and police shootings (whatever measurement scale it has) is your target variable / dependent variable / outcome variable / output variable / Y-variable.

And the Regression node in Enterprise Miner can definitely handle "CLASS" inputs.

The default coding for class inputs is DEVIATION (from-the-mean) coding aka EFFECT CODING.


I haven't used EMiner since a long time (using Model Studio in VIYA nowadays) and it is possible (I do not remember) that the Regression node in EMiner does not support multinomial target variables (i.e. categorical targets with >2 levels), I guess that's where your error message comes from. 


Kind regards,


Obsidian | Level 7

Hi yes, sas's regression node page says that but then the the "Predictive Modelling with SAS Enterprise Miner "by Kattamuri S. Sarma mentions its possible so I feel very confused. The node works but my concern is whether its producing the correct output.


Yes my target variables are Race and M.I 


while independent are variables such as age, gender, year, manner of death etc.


Hello @b_smsha ,


I cannot test it as I don't have access to Enterprise Miner now, but you can maybe watch this (old) video on YouTube:

Logistic and Multinomial logistic regression on SAS Enterprise Miner


[EDIT]: I haven't looked at the video myself and I am now also noticing it is not on the official SAS channel for YouTube.




SAS Employee

SAS Enterprise miner regression node help documentation clearly says the regression node can perform Logistic and ordinal logistic regression only. Available link functions are, logit, cumulative loglog and Probit. GLogit is not included. Therefore, generalized logistic regression using GLOGIT link function is not available in SAS EM regression node.





Available on demand!

Missed SAS Innovate Las Vegas? Watch all the action for free! View the keynotes, general sessions and 22 breakouts on demand.


Register now!

How to choose a machine learning algorithm

Use this tutorial as a handy guide to weigh the pros and cons of these commonly used machine learning algorithms.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 4 replies
  • 3 in conversation