AutoReg

Regression is one of the most common methods to explain and predict data in statistics.
In these pages we won't explain what regression is or how it works: it is assumed that the reader already knows the subject so that he can understand (and maybe correct or improve) the code.
The program (developed in SAS language) has been created to help statisticians who are using large datasets.
The code probably won't find the "Best Model" for a given problem, but can help giving a trace to analyze data.

In these pages we'll describe how to use the procedure and how it works. A detailed description of the code is not provided, but we'll simply describe the method, and input and output information.

To use the program properly, you need to know:

•   - which kind of data you have
•   - what are the input parameters of the macro
•   - which files must be treated as input in the procedure
•   - which files the program provides as output
•   - how the procedure works
•   - the SAS code (entirely available, but with no comments)
  
  To better understand the meaning and how to use input, output and parameters, have a look at this example.
  
  

  Release Notes

  The program was developed for SAS v.8 and tested on v.9.1.3 . Since version 9.3, SAS has modified ModelFit table resulting from the genmod procedure.
  Below you can find an example of this table for a previous version (on the left) and for version 9.3 (on the right).

  Criterion DF Value ValueDF Deviance 987 974.5303 0.9874 Scaled Deviance 987 974.5303 0.9874 Pearson Chi-Square 987 992.7954 1.0059 Scaled Pearson X2 987 992.7954 1.0059 Log Likelihood _ -487.2651 _ AIC _ 1000.5303 _

  Criterion DF Value ValueDF Log Likelihood _ -1733.0628 _ Full Log Likelihood _ -1733.0628 _ AIC (smaller is better) _ 3488.1256 _ AICC (smaller is better) _ 3488.1503 _ BIC (smaller is better) _ 3568.1815 _

  
  It is not possible to extract a value of this table choosing a specific criterion because the field changes according to the SAS installation language. Thus, the procedure selects a specific value filtering by number of row: the changes described above cause the selection of incorrect information in model valuating step and, as consequence, the entire procedure doesn't work correctly.
  To overcome this problem it is possible to change the code released inserting a key field in the table of input data (lines 599-602)
  
  

  data &in._pt; set &in ;   key = _n_;   keep &ludecla key ; run;

  
  and changing all calls to genmod procedure adding the aggregate field in the options of the model statement.
  
  

  proc genmod data=&in._pt descending NAMELEN=50; model &v_risp = / dist = &distribuzione aggregate=(key) ; output out=&in._dcorr predicted=predetti stdreschi=residui lower=inf upper=sup xbeta=xbet; run;

  
  The changes described above have the purpose of forcing SAS to generate a ModelFit table like the one below.
  

  Criterion DF Value ValueDF Deviance 11E3 3466.1256 0.3243 Scaled Deviance 11E3 3466.1256 0.3243 Pearson Chi-Square 11E3 11938.3041 1.1171 Scaled Pearson X2 11E3 11938.3041 1.1171 Log Likelihood _ -1733.0628 _ Full Log Likelihood _ -1733.0628 _ AIC (smaller is better) _ 3488.1256 _ AICC (smaller is better) _ 3488.1503 _ BIC (smaller is better) _ 3568.1815 _

  
  These changes have NOT been made in the code presented in these pages: you can update and find a better method for future versions.
  
  
  

  ---

  
  

  Main index
  Programs index	Information index	Code Licenses	Photos
  Vai alla versione Italiana

  Creation date: 17 Sep 2010
  Translation date: 30 Dec 2012
  Last change: 17 May 2013
  
  Translation reviewed by Giulia Di Lallo