24988 - Obtain a subset of data for the FIXONE, FIXTWO, RANONE or RANTWO methods in PROC TSCSREG

Sample 24988: Obtain a subset of data for the FIXONE, FIXTWO, RANONE or RANTWO methods in PROC TSCSREG

The %TSCSDATA macro

Contents:

Purpose / History / Requirements / Usage / Details / Limitations

PURPOSE:

The %TSCSDATA macro helps identify and select those observations in a panel data set which can be used with the FIXONE, FIXTWO, RANONE or RANTWO methods of PROC TSCSREG.

HISTORY:

Version 2.1: Fixed problem of sorting wrong data set if intended data is not last-created data set. Default IN= data set is last-created data set. Turn log notes off during execution. Fixed nondisplay of observations excluded for too few cross section values.

REQUIREMENTS:

The %TSCSDATA macro requires only Base SAS^®. However, to analyze the resulting data set with PROC TSCSREG requires SAS/ETS^®.

USAGE:

Follow the instructions in the Downloads tab of this sample to save the %TSCSDATA macro definition. Replace the text within quotes in the following statement with the location of the %TSCSDATA macro definition file on your system. In your SAS program or in the SAS editor window, specify this statement to define the %TSCSDATA macro and make it available for use:

   %inc "<location of your file containing the TSCSDATA macro>";

Following this statement, you may call the %TSCSDATA macro. See the Results tab for an example.

The following arguments can be specified in the %TSCSDATA macro call. All are required except for IN=.

in=: Input data set name. If not specified, the last-created data set is used.
out=: Output data set name.
ts=: The variable in the input data set which identifies the time points.
cs=: The variable in the input data set which identifies the cross sections.
vars=: List of model variables with the response variable first followed by explanatory variables, all separated by spaces.
nreg=: Number of regressors (explanatory variables).
keep=: Name of the binary variable in the in= and out= data sets which identifies those observations to be included in the analysis. The variable should have values 0 or 1, where 1 means keep the observation for analysis by TSCSREG. Any observation in the in= data set with a zero value for the keep= variable will have a zero for the keep= variable in the OUT= data set.

DETAILS:

The FIXONE, FIXTWO, RANONE and RANTWO methods allow for unbalanced data, but they impose the following requirements on the input data:

There must be at least two time point observations (with nonmissing values for all the model variables) for each cross section.
There must be at least two cross section observations (with nonmissing values for all the model variables) for each time point.

The %TSCSDATA macro requires a binary variable in the panel data set which identifies those observations to be included for the analysis. Use the keep= argument to specify this variable. Typically, the value for this variable is 1 initially for all observations.

The macro then changes the keep= variable value to 0 for those observations with missing values for any of the model variables and for those observations which violate the requirements (1) and (2) above. When the macro is invoked, it displays a note in the log indicating if an observation is excluded from the analysis.

Try invoking the macro an additional time if the preceding invocation results in exclusion of any observation. Once the macro has run without excluding additional observations, try running PROC TSCSREG using the WHERE clause.

LIMITATIONS:

This macro may help solve the problem of violations of the above two requirements. In rare cases, depending on the particular unbalanced structure of some panel data sets, the above approach will never yield a set of observations for which PROC TSCSREG will successfully run. In these rare cases, try a simpler model which excludes explanatory variables with many missing values.

These sample files and code examples are provided by SAS Institute Inc. "as is" without warranty of any kind, either express or implied, including but not limited to the implied warranties of merchantability and fitness for a particular purpose. Recipients acknowledge and agree that SAS Institute shall not be liable for any damages whatsoever arising out of their use of this material. In addition, SAS Institute will provide no support for the materials contained herein.

EXAMPLE:

In this example, data set TSCS is generated with missing values and the %TSCSDATA macro is used to select observations for use in PROC TSCSREG.

        data tscs;
          drop _temp ;
          do state=1 to 50;
          do date=1 to 40;
            x1=normal(1);
            x2=normal(1);
            x3=normal(1);
            x4=normal(1);
            y=10*(1+x1+x2+x3+x4)+state+date+normal(1);
            _temp=(state<10)*(date>1);
            if _temp then x2=.;
            _temp=(state<50)*(date>30);
            if _temp then x3=.;
            output;
          end;
          end;
          run;


        * Add variable to indicate which observations are included
          in the subset for PROC TSCSREG.  Initialize with the value
          1.0.
        *;
        data tscs;
          set tscs;
          _keep=1.0;
          run;


        title "DATA TSCS generate an error causing TSCSREG to fail";
        proc tscsreg data=tscs;
          where _keep=1;
          id state date;
          model y=x1-x4/rantwo fixtwo;
          run;

        * First invocation of %TSCSDATA macro results in exclusion of
          additional observations, as noted in the SAS log.  %TSCSDATA
          should be run again until no observations are excluded.
        *;
        %inc "<location of your file containing the TSCSDATA macro>";
        %tscsdata(in=tscs, 
                  out=tscs, 
                  ts=date,
                  cs=state,
                  vars=y x1 x2 x3 x4,
                  nreg=4,
                  keep=_keep)

        * Now the second invocation of %TSCSDATA.  It results in no
          additional observations being excluded.  So we can try running
          PROC TSCSREG again, using the where clause to identify which
          observations are included/excluded.
        *;
        %tscsdata(in=tscs, 
                  out=tscs,
                  ts=date,
                  cs=state,
                  vars=y x1 x2 x3 x4,
                  nreg=4,
                  keep=_keep)

        title "TSCSREG will now run";
        proc tscsreg data=tscs;
          where _keep=1;
          id state date;
          model y=x1-x4/rantwo fixtwo;
          run;

RESULTS:

Following is what is displayed in the SAS Log by the first PROC TSCSREG step and the two executions of the %TSCSDATA macro. The results of the final PROC TSCSREG step are not shown.

          proc tscsreg data=tscs;
            where _keep=1;
            id state date;
            model y=x1-x4/rantwo fixtwo;
            run;

ERROR: Not enough observations with non-missing model variables for model statement  in
       cross section state=1.

          %tscsdata(in=tscs,
                    out=tscs,
                    ts=date,
                    cs=state,
                    vars=y x1 x2 x3 x4,
                    nreg=4,
                    keep=_keep)
Observations excluded due to missing model variable values:
   Obs=2    state=1    date=2
   Obs=3    state=1    date=3
   Obs=4    state=1    date=4
   Obs=5    state=1    date=5
   Obs=6    state=1    date=6
   Obs=7    state=1    date=7
   Obs=8    state=1    date=8
   Obs=9    state=1    date=9
    ... listing continues ...
   Obs=1959    state=49    date=39
   Obs=1960    state=49    date=40
Number excluded due to missing values: 751
Observations excluded due to too few time points for the cross section:
   Obs=1    state=1    date=1
   Obs=41    state=2    date=1
   Obs=81    state=3    date=1
   Obs=121    state=4    date=1
   Obs=161    state=5    date=1
   Obs=201    state=6    date=1
   Obs=241    state=7    date=1
   Obs=281    state=8    date=1
   Obs=321    state=9    date=1
Number excluded due to too few time series values: 9
Observations excluded due to too few cross section values for a time point:
   Obs=1550    state=50    date=31
   Obs=1600    state=50    date=32
   Obs=1650    state=50    date=33
   Obs=1700    state=50    date=34
   Obs=1750    state=50    date=35
   Obs=1800    state=50    date=36
   Obs=1850    state=50    date=37
   Obs=1900    state=50    date=38
   Obs=1950    state=50    date=39
   Obs=2000    state=50    date=40
Number excluded due to too few cross section values: 10

          %tscsdata(in=tscs,
                    out=tscs,
                    ts=date,
                    cs=state,
                    vars=y x1 x2 x3 x4,
                    nreg=4,
                    keep=_keep)
Observations excluded due to missing model variable values:
Number excluded due to missing values: 0
Observations excluded due to too few time points for the cross section:
Number excluded due to too few time series values: 0
Observations excluded due to too few cross section values for a time point:
Number excluded due to too few cross section values: 0

Type:	Sample
Topic:	SAS Reference ==> Procedures ==> TSCSREG

Date Modified:	2016-06-01 15:21:14
Date Created:	2005-01-13 15:02:53

Product Family	Product	Host	SAS Release
			Starting	Ending
SAS System	SAS/ETS	All	n/a	n/a

Support

Sample 24988: Obtain a subset of data for the FIXONE, FIXTWO, RANONE or RANTWO methods in PROC TSCSREG

The %TSCSDATA macro

Operating System and Release Information