Return to Construction guide / CAMSIS home
6.1 Exporting CAMSIS model results: details
6.1.1 Exporting SPSS CA results
As mentioned elsewhere, the output from an SPSS correspondence analysis procedure is relatively easy to deal with - in most cases it can be pasted to another file, Microsoft Excel worksheets being possibly the most convenient. The standard SPSS outputs also produce graphs of the score estimates and limited summary statistics on the CA interpretation which are worth saving. (The most important statistics are the singular values and percent of inertia associated with the primary dimension, since support for the meaning of the first dimensional structue is greatest if the first dimension is considerably more influential than any other).
6.1.2 Exporting lEM RC row and column scores
First, LEM produces a number of summary statistics by default in its output files which are of great interest, most notably overall fit statistics which can be used to evaluate between alternative models, and dimensional association statistics which show the relative magnitude per dimension of occupational unit associations if more than one dimensional structure has been estimated.
More detailed results from the LEM estimations can however be slightly harder to deal with. By default, the estimated row and column scores are placed in the plain text LEM output file as tab-separated entries in rows, but the rows themselves are broken every six entries and stacked into text columns. We find a working solution to get these results into a single column, which can then be pasted next to a table of occupational unit titles, is (a) to paste the output columns into Microsoft Excel, (b) use the 'data -> rows to columns' option to separate them into tab separated columns, then (c) paste those results into a plain text file on pfe and (4) use pfe's 'replace' function to replace tab characters ("\t") with line breaks ("\n").
An additional problem is that the LEM estimation using the 'square autorecoded' occupational units will usually have included a few units where there are no cases representing either men or women. These should ideally be scored with a missing data indicator, but the LEM output gives them values of "0", despite the scale values being estimated themselves around a mean of 0. The pragmatic solution is to substitute values of "0" in the LEM output with values such as -999 or some similar indicator (this can be done in the pfe file by replacing ("\n0\n") with ("\n-999\n"), though this must also be run through twice to deal with any adjacent empty cases). There is a very slim danger in doing this that an actual score of exactly 0 will be falsely treated as missing; this rarely occurs, though when it does it is usually identifiable because it will have had a non-zero number of cases contributing to it.
The resulting columns of data can now be neatly pasted into a package used for reviewing the data scores, such as Excel, or read into a statistics package such as SPSS.
6.1.3 Exporting LEM RC residual statistics
We have noted above that our evaluation of the LEM RC models relies upon an assessment of the most extreme residuals to earlier model fits. LEM's primary output file often includes a list of all male-female combinations in the dataset with their observed and estimated frequencies, but this is not the case if the number of possible combinations is relatively high. In all cases, however, it is possible (as with the last commands in the example LEM command file above), to set the estimation to export the relevant observed and expected values to output files. These can then be read into SPSS, and values indicating the standardised residuals subsequently calculated, then analysed and if necessary exported to a plain text file or for use in Excel. This example from the supplementary file shows the SPSS syntax which can be used for the purpose reading in the two LEM output files, assuming the export options as in the lEM example command file above (model1.inp).
Return to Construction guide
Last modified 14 February 2002
This document is maintained by Paul Lambert (firstname.lastname@example.org)