Sorry for late reply,objective is to over sample a rear event in the model development data but using method above you may have multiple same observations in train_final data for target=1 right? ... View more

To set referential coding for all variables, under Model Options select GLM coding. I'm assuming you want it for all, not just one variable. I'll see about setting a reference level. ... View more

In general, yes, expect the same response as long as your training data is a good sample and the observation window of your model makes sense to you. The more similar the training data to your scoring data, the more similar the response rate. In other words, the more similar the distribution of each input for both the target events (response in this case) and for the non-target events (non-response), the closer your response will be to the 10% response rate of your training data. This is because you are assuming your data is a sample that can be scaled linearly and because you used training data that captures any seasonality and you expect the response rate to keep fairly steady in the future. I hope it helps, Thanks, Miguel ... View more

Repost this in the SAS Graphics section rather than Statistical. Then one of the SAS Graphing experts will be able to help. If you're open to using the SG Procedures (GCHART is outdated) then I believe there's a way to get the data behind the graph. Proc Freq or Proc Univariate should work as well ... View more

Sounds like what you want is just the number of cases with values between the first and second percentiles, the second and third percentiles and so on. By definition, this is always just 1% of the total number of (non-missing) cases in the file. It might be slightly above or below this if the data are not truly continuous or is weighted. In your decile example, the volume numbers are all around 10% of the sample size. Unless you are trying to look at the impact of ties in the data, it is not a particularly interesting statistic. ... View more

Thank You Robert... ... View more

Any rule is going to depend on the relative cost of making an error.  In some fields, an error >10% may be a good cutpoint, but in others, it may need to be substantially less.  Also, percentages are notorious for being scale dependent.  In your example, a deviation of 13 with a denominator of 588, leads to a percentage error of 2%.  However, suppose your model predicted a deviation of only 1.3 (much better), but the denominator was only 5.88.  Now the percentage error is 22%, even though the absolute prediction is a magnitude more precise. That probably doesn't answer your question of HOW to set a cutpoint in EM, but it is something to consider when the question is put forward. Steve Denham ... View more

Kanyange, this question would be best posted in the SAS Stat Community Given your needs with PROC LOGISTIC, they would be best to respond appropriately. Let me know if you need any more help. Thanks, Jonathan ... View more

EMWS1 is the name of your workspace (enterprise miner workspace 1), not a macro variable. Let's remove the "&". Before PROC REG DATA=&EMWS1.TRANS9_TRAIN; Now PROC REG DATA=EMWS1.TRANS9_TRAIN; macro EM_INTERVAL_INPUT will resolve to the name of your interval inputs as long as you have at least one interval input. Check that the Level of at least one of your variables is set to Interval. This two things should fix you. Does it work now? ... View more

If you are saying that 0.2% (1 out of 500) respondents actually respond, then I would first check to see if the non-response is random, or not. If it is not random, then there's no point in examining goodness of fit indicators, your model is worthless. If it is random, I'd still be very worried about using a model where you have such a high non-response rate, but who knows, if the sample size (not the non-response rate) is large enough, then maybe you might have something. It's hard to say without actually knowing the details of your situation and the data itself. I don't really think there is a canned answer in this case. ... View more