Help using Base SAS procedures

Domain Analysis using Proc Surveylogistic - problem with empty cells in categorical covariates

Reply
New Contributor
Posts: 2

Domain Analysis using Proc Surveylogistic - problem with empty cells in categorical covariates

Hi SAS Community:

I appreciate your help in advance.  Using SAS/STAT 14.1 - on SAS installation 4.1 TS Level 1M3

 

I am conducting a domain analysis using Proc Surveylogistic, with categorical covariates.  For the domain of interest, some of the categories of my categorical covariate are null (zero observations), and yet the domain analysis in surveylogistic is still producing estimates. This would seem to be a bug in SAS.

 

For instance, my covariate is insgroup (insurance group):

 

Table of gt_include_high_risk_table1 by insgroup

gt_include_high_risk_table1     insgroup

Frequency‚
Percent  ‚
Row Pct  ‚
Col Pct  ‚       .‚None    ‚Medicaid‚Medicare‚Other ‚Private ‚  Total
         ‚        ‚        ‚        ‚        ‚blic    ‚        ‚
-----------------------------------------------------------------------------------------------------
       0 ‚     74 ‚     14 ‚    260 ‚    571 ‚     33 ‚    964 ‚   1916
         ‚   2.87 ‚   0.54 ‚  10.09 ‚  22.15 ‚   1.28 ‚  37.39 ‚  74.32
         ‚   3.86 ‚   0.73 ‚  13.57 ‚  29.80 ‚   1.72 ‚  50.31 ‚
         ‚  70.48 ‚ 100.00 ‚  72.22 ‚  77.90 ‚ 100.00 ‚  72.32 ‚
-----------------------------------------------------------------------------------------------------
       1 ‚     31 ‚      0 ‚    100 ‚    162 ‚      0 ‚    369 ‚    662
         ‚   1.20 ‚   0.00 ‚   3.88 ‚   6.28 ‚   0.00 ‚  14.31 ‚  25.68
         ‚   4.68 ‚   0.00 ‚  15.11 ‚  24.47 ‚   0.00 ‚  55.74 ‚
         ‚  29.52 ‚   0.00 ‚  27.78 ‚  22.10 ‚   0.00 ‚  27.68 ‚
ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ
Total         105       14      360      733       33     1333     2578
             4.07     0.54    13.96    28.43     1.28    51.71   100.00

 You'll notice that for two groups I have zero observations in those categories - 'None' and 'Other' for the domain of interest.

 

When I run Proc surveylogistic with the domain statement, it creates the appropriate domain (same sample size - in bold) but produces estimates for the effect for the categories without any data.  How can this be?           
            
The SURVEYLOGISTIC Procedure

Domain Analysis for domain gt_include_high_risk_table1=1

                   Domain Summary

Number of Observations                          2578
Number of Obs with Nonpositive Weights             8
Number of Observations in Domain                 661
Number of Observations not in Domain            1909
Sum of Weights in Domain                   676.25300

 

...

 

                          Analysis of Maximum Likelihood Estimates

                                      Standard
Parameter                 Estimate       Error    t Value    Pr > |t|

Intercept                  -3.2715      0.2752     -11.89      <.0001
insgroup  Medicaid          1.0421      0.4668       2.23      0.0257
insgroup  Medicare          0.8652      0.3975       2.18      0.0296
insgroup  None              1.4822      0.8113       1.83      0.0678
insgroup  Other public      1.2911      0.6004       2.15      0.0316

        NOTE: The degrees of freedom for the t tests is 2465.


                  Odds Ratio Estimates and t Confidence Intervals

Effect                                   Unit     Estimate     95% Confidence Limits

insgroup Medicaid     vs Private       1.0000        2.835        1.135        7.082
insgroup Medicare     vs Private       1.0000        2.375        1.089        5.180
insgroup None         vs Private       1.0000        4.403        0.897       21.608
insgroup Other public vs Private       1.0000        3.637        1.120       11.805

      NOTE: The degrees of freedom in computing the confidence limits is 2465.

 

 

 

 

Ask a Question
Discussion stats
  • 0 replies
  • 226 views
  • 0 likes
  • 1 in conversation