Would anyone happen to know how the percentages are calculated in SPSS for the predicted and observed categories? Is it something you can do by hand or does SPSS use some kind of internal calculation? The type of output I'm referring to is the screenshot below. Huge thanks in advance for any help!
Hello Louise Rowland. Here is an example that shows how the classification table is generated. I hope this helps.
* Use the example show here:
https://stats.oarc.ucla.edu/spss/output/logistic-regression/
.
NEW FILE.
DATASET CLOSE ALL.
SPSSINC GETURI DATA
URI="https://stats.idre.ucla.edu/wp-content/uploads/2016/02/hsb2.sav"
FILETYPE=SAV DATASET=hsb2
.
* You may see an error message after the SPSSINC GETURI DATA
* command, but it appears not to matter.
COMPUTE honcomp = (write ge 60).
LOGISTIC REGRESSION honcomp with read science ses
/CATEGORICAL ses
/SAVE PRED(predprob).
* The default CUT() value is 0.5. See:
https://www.ibm.com/docs/en/spss-statistics/28.0.0?topic=regression-criteria-subcommand-logistic-command
.
COMPUTE predhc = predprob GE 0.5.
FORMATS honcomp predhc (F1).
CROSSTABS honcomp BY predhc.
* Compare this crosstabulation to the Block 1 Classification table.
Thinking about the percent correct classification, it might be useful to look at the definitions of the Count pseudo r-square and the Adjusted Count pseudo r-square here:
https://stats.oarc.ucla.edu/other/mult-pkg/faq/general/faq-what-are-pseudo-r-squareds/
Note that while the model presented by OP correctly predicts 61% of the observations, it does no better than a model that is just RiskTaking = 1.
Hello. Just a side comment. I have read somewhere that a classification table is not the best way of assessing the "performance" of a binary LR model, as it relies just on cut-off probability value. The ROC curve and related AUC are considered to do a better job.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression