Consider the following problem: in doing binary logistic regression, I
have a variable (say gender) containing several missing values. My
client does not wish to simply eliminate a case containing a missing
value as all the information for other variables would be lost. However,
it is not desirable to treat the variable as an ordinal one. Hence I wish
to treat it as nominal.
In SPSS one can choose for the categorical variables either 'simple' or
'indicator' contrasts (among others). Suppose that gender is coded 0
for male, 1 for female, and 2 for 'missing'. Would it be advisable to use
the contrast named 'simple', where each category is compared to the
reference category (which is what I am inclined to do - but see later re
choice of reference category) or the 'Indicator' contrast where
membership of each level of the category appears to be assigned a
dummy variable?
The latter would seem to fit a multiplicative model with one factor in
the product for each level of the variable that is included, with
presumably the 'baseline' being neither male, nor female, nor missing!
Which raises the question: if 'missing' were made the baseline
category for a 'simple' contrast, with male and female being possible
multipliers -would this be an adequate way of getting round the
missing value problem?
Regards
Miland Joshi (Mr.)
Department of Epidemiology and Public Health
University of Leicester
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|