topic Re: Using Binomial Variables as Independents in EM Regression in SAS Data Science

Using Binomial Variables as Independents in EM Regression

euser — Wed, 09 Oct 2013 22:40:31 GMT

Hi there,

I'm working through a regression exercise in Miner and have a bunch of independent binomial variables with a binomial target. All the binomial variables have been coded 0 and 1, with 1's indicating YES. By default, does the regression node (logistic regression) in Miner calculate the estimates against the 0's, instead of the 1's? I'm finding that the signs for the binomials are opposite of what they should be, while the signs for the interval independents seem correct.

Please let me know if additional clarification is required.

Thanks in advance,

Sidney

Re: Using Binomial Variables as Independents in EM Regression

jwexler — Thu, 10 Oct 2013 20:40:20 GMT

Hi can you send a snapshot of your model? Are you talking about the target variable being coded 1/0 as backwards, or the independent variables? Target variables are automatically sorted descending to find the value of interest, so by default "1" would be the target.

Thanks!

Jonathan

Re: Using Binomial Variables as Independents in EM Regression

jwexler — Thu, 10 Oct 2013 21:06:08 GMT

One more item to add, there is an option in EM that specifies how you handle 'Input Coding', meaning there are 2 methods to create dummy variables (Deviation and GLM). Depending on the method, interpreting the coefficients would be different. Your user guide explains how to interpret. That may explain your issue.