Hi, Melissa. This is from one of our statisticians:
----
NOMREG (Multinomial Logistic Regression in the menus) produces McFadden's original measure. I don't have access to the full text of what I think is the paper of interest, but it seems that the measure referred to is the adjusted version of McFadden's measure. If so, then the following could be done (assuming no split files in effect to complicate things; that would require some additional syntax):
DATASET DECLARE temp.
OMS
/SELECT TABLES
/IF COMMANDS=['Nominal Regression'] SUBTYPES=['Model Fitting Information']
/DESTINATION FORMAT=SAV NUMBERED=TableNumber_
OUTFILE='temp' VIEWER=YES.
NOMREG y WITH x1 x2 x3 /PRINT=PARAMETER SUMMARY LRT CPS STEP MFI KERNEL.
OMSEND.
DATASET ACTIVATE temp.
COMPUTE M_A=(ChiSquare-2*(df+1))/LAG(@2LogLikelihood).
EXECUTE.
The NOMREG command would need to be edited to involve the correct dependent and predictors, which might include a BY with factors, etc. On the PRINT subcommand, the import things are MFI and KERNEL. Also, the COMPUTE assumes the intercept is included. If not, then the "df+1" would just be "df". After the COMPUTE, the single calculated value of M_A is the adjusted McFadden pseudo-R^2.
----
------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------
Original Message:
Sent: Thu June 16, 2022 02:59 PM
From: Melissa Fry
Subject: Pseudo R-squared and Logistic Regression
SPSS produces the Cox and Snell and Nagelkerke R-squared statistics in the Logistic Regression output. I just read a 2018 piece in Sociological Methods and Research that suggests the McFadden Horowitz would be less sensitive to sample size and distribution of the dependent variable. I have a medium sample size, but somewhat skewed binary dependent variable. Anyone know if or how I can get SPSS to run the McFadden-Horowitz.
------------------------------
Melissa Fry
------------------------------
#SPSSStatistics