Multinomial regression interpretation SPSS

Question

-- start reading from the edited part --

When running a multinomial regression the two values we are really interested in are the values 'B' and B(Exp)'.

Let's say we have (fictive numbers):

• B: .098 ; B(Exp): 1.98
• B: -.098 ; B(Exp): 1.98

The B(Exp) is the same for both examples. But in one case the B is positive in the other negative. For the first I suppose it means the odds 'increase' by 1.98, but what about the second?

The odds 'decrease' by 1.98?

------ edited from here -----
(forget the above).

Clearly I should have been more specific. I will use actual numbers now. So let's assume I have a dependant variable D with three categories. And I have 2 explanatory variables E with 3 categories and sex with 2 categories.

D:

• D1
• D2
• D3 (reference)

These are the results of parameter estimates in SPSS for the category D1:

E:

• E1: B= -.111 ; Exp(B)= .895
• E2: B= 2.264 ; Exp(B)= 9.622
• E3: (reference)

Sex:

• F: B= 1.514 ; Exp(B)= 4.546
• M: (reference)

Are the following statements correct?

• For females the odds of being in category D1 rather than in D3 'increase' by a factor 4.546 compared to males.
• For E1 the odds of being in category D1 rather than in D3 'decrease' by a factor .895 compared to E3.
• For E2 the odds of being in category D1 rather than in D3 'increase' by a factor 9.622 compared to E3.

In addition if the Exp(B) is lower than 1 it's decrease and otherwise increase? What statement would be correct for males?

Answer 1

The second one is impossible in your example , when $\beta$ is negative, your $exp(\beta)$ should be less than 1 but still bigger than zero. For example, $exp(-0.98)=0.3753111$. Which means your odds decreases by 37.5% or by 0.6246889 (relative to $1$) depending on different understanding of English. And the first one should be $exp(0.98)=2.66$ when coefficient is positve the odds should be bigger than one for multinominal logistic regression.