Hi, The Proc Severity procedure allows fitting of a 'Burr' distribution but the CDF function does not allow the 'Burr' distribution as an input. For instance I tried this code but does not work: Cum_Pct_Burr = CDF("Burr", Y, theta,alpha,gamma); Is there a way to generate a cdf for Burr distribution? thanks, A
... View more
Yes, you are right, in the example the second and third predicted scores should be 20 and 23 respectively. Is there any way to solve this problem using optimization? thanks.
... View more
Please find my response inline in blue font below: Are the allowed adjustments -2 to 2 part of the input? No they are not. The adjustments are a function of 1) Score 1 and 2) Score 2 values. The only inputs are Score 1 and Score 2, and the Target output is the desired output after the adjustment is applied to Score 1. Is score 2 always a nonnegative integer? Yes that's true In your example solution, each range is nonempty and increasing with respect to adjustment. Are those requirements, or is the solver allowed to return empty or nonincreasing ranges? Yes that's a strict requirement, the range value is increasing both horizontally and vertically. Thanks!
... View more
Thank you Rob for your reply. Following is more detail on my data: Score_1 | Score_2 | Predicted Output (based on mapping table) | Target Output 21 55 23 22 21 20 18 20 22 58 24 24 22 40 22 23 23 69 25 24 Mapping table (that needs optimization): Adjustment to Score 1 -2 | -1 | 0 | 1 | 2 Score1 / Score2 Range-> 21 0-18 19-30 31-40 41-50 51+ 22 0-20 21-34 35-48 49-58 59+ 23 0-24 25-40 41-52 53-62 63+ I am looking to calibrate the above score2 ranges in the mapping table so the sum(abs(target-predicted output)) is minimized. Hope this clarifies the optimization problem. Please let me know if further clarity is required. Thanks!
... View more
Hi, I am looking to optimize a score mapping table but am not sure if there is a way in SAS. Following is the problem set up: Score 1 is a score calculated by regression model. Score 2 is provided to us. And the predicted score is calculated based on score 1 and the value of score 2. Below is the mapping table: Score 1 Score 2 adjustment +1 0 -1 21 | Range 1 | Range 2 | Range 3| 22 | Range 4 | Range 5 | Range 6| 23 | Range 7 | Range 8 | Range 9| To illustrate, if Score 1 from the model is 21 and score 2 is within Range 1 then predicted score is 22 (21+1). The goal is to minimize absolute(Predicted score - Actual score). Would appreciate any help on this. Thanks!
... View more
Thanks for your reply. To my knowledge Proc GLM does not allow functionality for adding constraints so it does not solve my problem. My model framework is: Model: y = var1 var2 var3 Constraints: Weight of var1 should be at least 20%, i.e. var1/(var1+var2+var3) >0.2 Weight of var2 should be at least 30% i.e. var2/(var1+var2+var3) >0.3 I can use the Restrict statement in Proc Model to fit models using these constraints, however, this procedure does not output the AIC, BIC and other fit statistics. Proc Reg also does not have the functionality of fitting non-linear or inequality constraints. So is there any other procedure I can use to fit a model and get the goodness of fit statistics if its not available with Proc model. Thanks.
... View more
Hi, I want to look at the goodness of fit statistics (AIC, BIC etc.) for a linear regression model I have built with Proc Model. I could not use Proc Reg for this model as I had certain constraints (like weight of variable_1 > 20%) that proc reg can not handle. So I am looking to get the fit statistics using Proc Model now. Thanks for helping out. A
... View more