Topic
IC4NOTICE: developerWorks Community will be offline May 29-30, 2015 while we upgrade to the latest version of IBM Connections. For more information, read our upgrade FAQ.
9 replies Latest Post - ‏2012-10-31T01:10:46Z by SystemAdmin
SystemAdmin
SystemAdmin
396 Posts
ACCEPTED ANSWER

Pinned topic Empty output when running heterogeneous correlations?

‏2012-10-27T20:41:57Z |
I have run a heterogeneous correlation on a set of binary variables, and SPSS will run the analysis, but the output is empty. That is, except on the diagonal, all cells only contain an "."
I have also tried to run it with other variables, but it does not change the result.

Help, anyone?
Updated on 2012-10-31T01:10:46Z at 2012-10-31T01:10:46Z by SystemAdmin
  • SystemAdmin
    SystemAdmin
    396 Posts
    ACCEPTED ANSWER

    Re: Empty output when running heterogeneous correlations?

    ‏2012-10-28T02:20:20Z  in response to SystemAdmin
    What types of correlations is it trying to produce - that is, what is in the correlation types table?

    Can you post an example dataset and syntax? I haven't seen this behavior, but if the variables are too far from a normal distribution, the procedure can fail. In that case, though, you should get a message to that effect in the output and no tables.
    • SystemAdmin
      SystemAdmin
      396 Posts
      ACCEPTED ANSWER

      Re: Empty output when running heterogeneous correlations?

      ‏2012-10-28T20:06:23Z  in response to SystemAdmin
      It produces polychoric correlations.

      The data set is very large (N > 200000) - a series of Eurobarometers. Could this be the case?
      And the variables are not normally distributed, but I get no error message either.

      This is the matrix I get

      Pearson, Polyserial, and Polychoric Correlationsa
      Variables Statistics Variables
      v1001 v1002 v1003 v1004 v1005 v1006 v1007 v1008 v1010
      v1001 Correlation 1.000 . . . . . . . .
      Std. Error .000 . . . . . . . .
      v1002 Correlation . 1.000 . . . . . . .
      Std. Error . .000 . . . . . . .
      v1003 Correlation . . 1.000 . . . . . .
      Std. Error . . .000 . . . . . .
      v1004 Correlation . . . 1.000 . . . . .
      Std. Error . . . .000 . . . . .
      v1005 Correlation . . . . 1.000 . . . .
      Std. Error . . . . .000 . . . .
      v1006 Correlation . . . . . 1.000 . . .
      Std. Error . . . . . .000 . . .
      v1007 Correlation . . . . . . 1.000 . .
      Std. Error . . . . . . .000 . .
      v1008 Correlation . . . . . . . 1.000 .
      Std. Error . . . . . . . .000 .
      v1010 Correlation . . . . . . . . 1.000
      Std. Error . . . . . . . . .000
      Correlations computed by R Hetcor package
      a N = 233672

      The syntax is

      DATASET ACTIVATE DataSet2.
      SPSSINC HETCOR v1001 v1002 v1003 v1004 v1005 v1006 v1007 v1008 v1010
      /OPTIONS ESTIMATOR=TWOSTEP STDERR=TRUE
      MISSING=LISTWISE N=TRUE TYPE=TRUE
      /SAVE.

      Any help is appreciated :)
      • SystemAdmin
        SystemAdmin
        396 Posts
        ACCEPTED ANSWER

        Re: Empty output when running heterogeneous correlations?

        ‏2012-10-28T20:50:26Z  in response to SystemAdmin
        It could well be that the underlying R package can't handle that data. Try taking, say, just 10000 cases and see what happens. I have never tried a problem that large with this procedure.
        • SystemAdmin
          SystemAdmin
          396 Posts
          ACCEPTED ANSWER

          Re: Empty output when running heterogeneous correlations?

          ‏2012-10-28T21:11:59Z  in response to SystemAdmin
          Well, I tried with 500,000 cases of random Bernoulli data and got appropriate results. That makes me thing that there may be something about the magnitude of the values in your data that is throwing the package off. If so, first creating z scores (Descriptives can do this for you) might help.
          • SystemAdmin
            SystemAdmin
            396 Posts
            ACCEPTED ANSWER

            Re: Empty output when running heterogeneous correlations?

            ‏2012-10-28T21:42:19Z  in response to SystemAdmin
            Tried creating z-scores, and running with those, but it still doesn't work...
            • SystemAdmin
              SystemAdmin
              396 Posts
              ACCEPTED ANSWER

              Re: Empty output when running heterogeneous correlations?

              ‏2012-10-28T21:45:16Z  in response to SystemAdmin
              I tried with a different data set (940 cases) as well, but still no result. Could there be something that I have not installed, or that I haven't installed it properly? But wouldn't I get an error message then?
              • SystemAdmin
                SystemAdmin
                396 Posts
                ACCEPTED ANSWER

                Re: Empty output when running heterogeneous correlations?

                ‏2012-10-28T23:53:37Z  in response to SystemAdmin
                This is behavior that I haven't seen before. If something were not installed, you should see no output and an error message. You haven't said what version of Statistics (and R) you are using. If you can send me the smaller dataset, I can see whether the problem reproduces on my systems. You can send it to peck at us.ibm.com.

                -Jon
                • SystemAdmin
                  SystemAdmin
                  396 Posts
                  ACCEPTED ANSWER

                  Re: Empty output when running heterogeneous correlations?

                  ‏2012-10-30T22:07:38Z  in response to SystemAdmin
                  I have sent you an e-mail. Thanks for your help.

                  Guri
                  • SystemAdmin
                    SystemAdmin
                    396 Posts
                    ACCEPTED ANSWER

                    Re: Empty output when running heterogeneous correlations?

                    ‏2012-10-31T01:10:46Z  in response to SystemAdmin
                    I ran your data through the procedure trying different versions of Statistics, and everything worked. I don't have ready access to a Mac, but I'm planning on getting someone to try it there as well.
                    Regards,
                    Jon