Binary Factors in SPSS 11 Multinomial Logistics

[Bill]

Hi All,

I have a number of binary explanatories (dummy variables) in some
logistical (Nominal Regression pick) regressions, and I get unexpected
displays such as the following parameter estimates. The variable
EMPLOYEE was entered as a "factor".

[EMPLOYEE=0] .082
[EMPLOYEE=1] 0(b)

This is a simple binary explanatory variable- 1 for Employee, 0 for
dependent.
My questions:
Why do the displays show both entries (neither the binary logit nor the
linear regression models do this)?
Is there a way to suppress this?
Does it affect results?

Thanks,
Bill

[Bruce Weaver]

Bill wrote:

Footnote b, which you did not report, will indicate that EMPLOYEE=1 is
the reference category. I.e., the odds ratio is for EMPLOYEE=0 relative
to EMPLOYEE = 1.

Note that for binary variables that are coded 0/1, you can simply treat
them as you would continuous predictors. You will get the one line of
output you want, for one thing; and the odds ratio will be associated
with a one unit increase in the predictor--i.e., a change from 0 to 1.
So 1 will be the reference category, in other words.

HTH.

--
Bruce Weaver
bwe...@lakeheadu.ca
www.angelfire.com/wv/bwhomedir

[Bill]

Hi Bruce:

Just to be clear- I need to follow up:
1) Footnote b) actually says "b  This parameter is set to zero because it is redundant.". I take that as basically the same as "EMPLOYEE=1 is
the reference category" Correct?
2) So I can enter these independent binary variables as "Covariates" rather than "factors".Correct?
3) Q 2 in effect answers why when I use the binary logistic pick, (which in SPSS 11 does not ask the input to be distinguished between factors and covariates), I do not get this sort of output.

I think I got it.
Please let me know if I have misinterpreted your response.
Thanks,
Bill

[Bruce Weaver]

Bill wrote:

> Hi Bruce:
>
> Just to be clear- I need to follow up:
> 1) Footnote b) actually says "b This parameter is set to zero because
> it is redundant.". I take that as basically the same as "EMPLOYEE=1 is

> the reference category" /Correct?/

Correct.

> 2) So I can enter these independent binary variables as "Covariates"

> rather than "factors"./Correct?/

Correct. For the benefit of others, in SPSS lingo, *covariate* means
scaled variable, and *factor* means nominal/categorical variable.

> 3) Q 2 in effect answers why when I use the binary logistic pick, (which
> in SPSS 11 does not ask the input to be distinguished between factors
> and covariates), I do not get this sort of output.

Right. In the binary logistic regression procedure, you have to
specifically tell SPSS to treat something as a categorical variable,
because the default is to treat it as a scaled variable.

Cheers,
Bruce

[Bill]

Bruce:
I need to get straight with my 1s and 0s- using both dependent and independent
binary variables.

In the Linear Regression, I typically use 1 to signify the event happened, 0
it did not.
For explanatory variables in the LPM, again I use 1- e.g., for "Male", 1 =
yes, 0 = no (female).
In the multinomial logistic, as we discussed below, the last category
numerically is the reference group for both the LHS and RHS variables.
Therefore results are in effect "switched", in that coefficients tell the
odds, etc of going from the 1 group to the 0.
To be consistent between LPM and logistic regressions, I need to change my 1s
to 0s and 0s to 1s for the dependent variable. Do I have this correct?

For the RHS variables, if I enter them as covariates as we discussed below, do
I again need to switch the 0s and 1s? What if it is an array of dummy (shift)
variables- e.g., days of the week, where 6 days are entered with a 1 or 0 and
the 7th day is left out? Do these need to be reprogrammed from the 6 by 6
matrix I use for the LPM?

Thanks,
Bill
Bruce Weaver wrote:

[Bruce Weaver]

Bill wrote:

> Bruce:
> I need to get straight with my 1s and 0s- using both dependent and independent
> binary variables.
>
> In the Linear Regression, I typically use 1 to signify the event happened, 0
> it did not.
> For explanatory variables in the LPM, again I use 1- e.g., for "Male", 1 =
> yes, 0 = no (female).
> In the multinomial logistic, as we discussed below, the last category
> numerically is the reference group for both the LHS and RHS variables.
> Therefore results are in effect "switched", in that coefficients tell the
> odds, etc of going from the 1 group to the 0.
> To be consistent between LPM and logistic regressions, I need to change my 1s
> to 0s and 0s to 1s for the dependent variable. Do I have this correct?

It has been a while since I used NOMREG, but that sounds right.

>
> For the RHS variables, if I enter them as covariates as we discussed below, do
> I again need to switch the 0s and 1s?

No. If you enter a 0/1 binary variable as a covariate, the odds ratio
will be associated with a one-unit increase in the covariate. There is
only one one-unit increase: 0 to 1. So the odds ratio reflects what
happens as the covariate changes from 0 to 1. I.e., 1 is the referent.

> What if it is an array of dummy (shift)
> variables- e.g., days of the week, where 6 days are entered with a 1 or 0 and
> the 7th day is left out? Do these need to be reprogrammed from the 6 by 6
> matrix I use for the LPM?

Why not just enter it as a categorical variable? If you don't like the
last level as your referent, recode it into a new variable which has the
referent you want as the last category. E.g., to swap categories 1 and
5, and leave the others as they are:

RECODE oldvar
(1=5)
(5=1)
(else=copy) into NEWVAR.
exe.

Note that in binary logistic regression, you can specify any category
you want as the referent, at least for Deviation, Simple, or Indicator
contrasts. For example, if CATVAR is a categorical variable with k
levels, you can substitute any number from 1 to k where you see X on the
/CONTRAST line below:

LOGISTIC REGRESSION VAR=status
/METHOD=ENTER CATVAR
/CONTRAST (CATVAR)=Indicator(X)
/CRITERIA PIN(.05) POUT(.10) ITERATE(20) CUT(.5) .

Whatever value of X you use will be the reference category.

HTH.