Statistical Computing Tips

Question: I have a large number of SPSS listing files (output from a statistical analysis) that I am interested in extracting information from for subsequent analysis. For instance, the SPSS regression procedure (REGRESSION) does not allow me to save various regression summaries to a data file (e.g. MSE, R squared, beta coefficients, F values, p values, etc.). I understand that I can use the MATRIX - END MATRIX commands to write my own statistical routines, thus enabling me more flexibility in the output of information that is calculated. However, for the sake of efficiency, I am not interested in recreating statistical procedures that have already implemented. Nor am I interested in reading information off of the listing files to type back into a data file. Is there a simple way to read these listing files back in as data and extract the information I am interested in?

Answer: SPSS provides several alternatives to the more commonly used FIXED' and FREE' data formats. The SPSS commands FILE TYPE - END FILE TYPE allow the user to define data for the following complex file types: mixed files, nested files, grouped files, and repeating groups (see the SPSS Reference Guide (1990) for further details). Here, I will demonstrate the usage of the FILE TYPE NESTED command.

Suppose I am interested in performing ordinary least squares regression of YVAR regressed onto XVAR for each of 1000 individual units where each unit has the following data structure:

	case_id       YVAR   XVAR
	    1          114    126
	    1          125     93
	    1           86     80
	    1          100     84
	    1          109    122
	    2            .      .
	    2            .      .  
	    .         etc.      .
	    .            .      .
	 1000          112    109
	 1000          115    103

This data is used as input for the SPSS procedure REGRESSION. If the SPLIT FILE command is used with the case_id' as the grouping variable before using the REGRESSION procedure, we will get regression output for each of the 1000 units. Once this listing output has been saved into a text file (for example - regress.out), we can read this text file back in as a data file:

  SET WIDTH=80.
  FILE TYPE NESTED FILE='regress.out' RECORD=1-80 (A).
  RECORD TYPE  case_id:'.
  DATA LIST / case 14-18.
  RECORD TYPE  Multiple R'.
  DATA LIST / corr 21-27.
  RECORD TYPE  XVAR'.
  DATA LIST / slope 16-23.
  RECORD TYPE  (Constant)'.
  DATA LIST / intercept 15-23.
  RECORD TYPE  Durbin-Watson'.
  DATA LIST / durbin 23-30.
  END FILE TYPE.

FORMATS case (F4.2) corr (F6.5) slope (F8.6) intercept (F8.6) durbin (F6.5).

LIST variables=case corr slope intercept durbin.

This syntax above causes the current working file to be replaced with the following variables and data cases:

	case       corr       slope   intercept  durbin
  
	1          .249       .340    72.427     2.863      
	 .         .          .         .         .
	1000       .          .         .         .
			      etc.

The SPSS command FILE TYPE NESTED defines a file in which the record types are related to each other hierarchically. In the present example, the last record type defined is the lowest level of the hierarchy; this defines a case. The RECORD subcommand defines the record-identification variable and its column location(s) in the data. Once the data of interest have been extracted, the working file can be saved for subsequent analyses.

Previous Article Next Article

If you have any problems or questions about this server, contact us as soon us as soon as possible. You can send mail to the following address: www@unt.edu