Hi,
I know that one of the assumptions for binary logistic regression is that there is a linear relationship between any continuous independent variables and the logit transformation of the dependent variable.
I have seen online there is a Box-Tidwell test that tests this assumption but I don't think this test is available on SPSS?
My dependent variable is a dichotomous variable (0 and 1), how can I transform this variable into logit on SPSS?
Thanks in advance
You don't need to transform the binary dependent variable.
Please see the link below:
https://statistics.laerd.com/spss-tutorials/binomial-logistic-regression-using-spss-statistics.php
For the Box-Tidwell test please see pages 10 to 11 if the attached document
Let, Y be the binary response variable and X be the set of predictors. Fit the logistic regression model using the following syntax
Model:
logit[Pr(Y=1|X)] = beta0 + beta1 X
Syntax:
LOGISTIC REGRESSION VARIABLES Y
/METHOD=ENTER X
/CRITERIA=PIN(0.05) POUT(0.10) ITERATE(20) CUT(0.5).
Get the parameter estimates of (beta0, beta1) as (0.458, 0.531). Finally, compute the linear predictors as below:
COMPUTE Linear_predictor=0.458+(0.531 * X).
EXECUTE.
Using the linear predictors and X you can check the linearity.
Md. Shaddam Hossain Bagmar
I think the regression equation = 0.531 + 0.458 * X.
Please correct me if I am mistaken.
Hello Surendra:
You are right. The regression equation is linear on logit scale. The full corrected equation would be:
logit[Pr(Y=1|X=x)] = 0.531 + 0.458*x
where logit is nothing the log of odds of the event Pr(Y=1|X=x).
This can be helpful https://www.reneshbedre.com/blog/logistic-regression.html
- Badges
- Science topic
- Similar topics
- Psychology