WesVar Frequently Asked Questions

On this page:

Q: Does the order of the factors in the external FPC and JKN factor files matter?

A: Yes. The factor files have one record per replicate. The records must be in the same order as the physical order of the replicate weights on the input file used to create the data (VAR) file. For example, suppose your replicate weights are named RWT1-RWT100, but that they were sorted in alphanumeric order: RWT1, RWT10, RWT100, RWT11, etc. The factor files must be in this order also. Replicate weights are listed on the Attach Factors screen by their physical order in the data file. Both FPC factors and JKn factors must appear in this same order on the external files.

Q: What does the following error mean in logistic regression: “Estimates are converging slowly or are diverging for full sample and replicates…”?

A: This message is issued in logistic regression when the convergence criterion for parameter estimates is not met in the specified maximum number of iterations. This could be caused by the following

  1. The convergence criterion has been changed from difference in log-likelihood in version 3 to relative difference in log-likelihood in the current version. If you were a version 3 user, you may have to manually change the convergence criterion from 0.025 to 1E-6 in Preferences and in any version 3 workbook that you import into the current version.
  2. You have few (or no) cases that have the characteristic being modeled in a cell. If your model contains a categorical independent variable, say RACE, and in replicate subsample 43 for RACE.1 there are no cases with the characteristic (dependent variable = 1), then the estimation algorithm will try to send the parameter estimate to negative infinity so that the predicted probability for RACE.1 is 0. In that case the parameter estimate will not converge but will get more and more negative at each iteration.
  3. You may need to collapse categories for one of your independent variables to eliminate the problem.

In the Regression Request/Output Control, check the boxes for Iteration History and Replicate Coefficients to get a detailed report on how the parameter estimates are changing.

Q: I can’t get WesVar to import the SAS transport files I created. How do I create the transport file so that it’s acceptable to WesVar?

A: One alternative is to use the following code to create your SAS transport files. This is code for creating a SAS transport file (with a single member) named c:\export\wesvar.xpt from the sd2 file c:\input\data.sd2:

                        options validvarname=v6;
                        libname in1 'c:\input';
                        libname out1 xport 'c:\export\wesvar.xpt';
                        proc copy in=in1 out=out1;
                        select data;

The statement “options validvarname=v6;” will cause SAS to generate unique 8-character names for variables with names longer than 8 characters. The VALIDVARNAME option is only valid using SAS Version 7 or later. It is not recognized (or needed) by earlier versions of SAS.

Q: I converted a WesVar data file to SAS XPORT format. Using WesVar, I re-imported the data into WesVar and seemed to lose all my value labels. Did I do something wrong or can value labels not be imported?

A: SAS value labels are not imported in WesVar, even if they were originally created in WesVar and exported to SAS. Use the Relabel utility.

Q: What does the following error mean: “STRATUM 1 has non-consecutive varunits for METHOD=JKn”?

A: VARUNITS must be numbered consecutively starting at 1 in each VARSTRAT. The number of VARUNITS can differ from one VARSTRAT to another.

Q: Is it possible to create weights with one PSU in some strata with JKn design?

A: WesVar does not allow the JKn method to be used unless there are at least two PSUs in every stratum. To use the JKn method, you would have to collapse any stratum with one PSU with another stratum.

Q: Can WesVar run under 64-bit Windows?

A: Yes.

Q: Can WesVar read 64-bit SAS files?

A: WesVar cannot read 64-bit SAS files. WesVar can only read 32-bit SAS files. The best way to ensure the data set is 32-bit is to specify the data set option OUTREP=WINDOWS_32 when creating the data set.