Excluding the constant is equal to forcing it to be equal to zero. The
model is ln(P/1-P)=b0+b1x1+..., ln(P/1-P)=0+b1x1+... is equal to
ln(P/1-P)=b1x1+...
As the constant gives you the probabilitiy of Y=1 when all the
independent variables are zero (P(Y=1)=exp(b0)/1+exp(b0)), you force
this probability to be .5 by excluding the constant. I can't see a good
reason for doing this, because if it's .5, the model will estimate it as
.5, an if it's not, the model isn't appropriate when the constant is
excluded.
HTH
Christian
Steve Barron wrote:
> Hi there
>
> When performing logistic regression via the menus there is a box under
> Options that you can untick to "exclude the constant". This action
> generates the sub-command
> /ORIGIN
>
> The sub-command to exclude the constant from ANOVA is
> /INTERCEPT=EXCLUDE
>
> Is the logistic regression sub-command excluding the constant or just
> forcing it to be equal to 0? And is there a difference?
>
> I'm thinking (and i might be very wrong) that my logistic regression
> coefficients shouldn't change if i include or exclude the constant term,
> but i get very different values by ticking and unticking the "Include
> constant in model" box. The method i am using is "Enter".
>
> Many thanks
> Steve Barron
>
>
Locked
