The general computing convention for 0/1 is that 0 is False, 1 is True;
your data has that reversed. This will be vastly confusing to any
professional who looks at your data or results.
- In order to minimize confusion among the people who gave you
the data (if that matters), instead of merely reversing your 0/1 scores
(by subtracting each from 1), it could be wiser to RECODE the 0s as 2s.
That way, they can KNOW that a value of 1.5 does not arise from some
use of 0/1. If I simply reversed the scores, I would also insist on changing
the original variable names, if anyone else had ever seen or used the data.
A two-group prediction is done with simple discriminant function or
with Logistic regression. You might consider an interaction term or two,
especially if there are any strong main effects observed.
However, your file has only 88 fatalities -- that places an implicit limit on
how many variables you can safely use in one model of about 9... at most.
(If you know what you are doing, you can try more. But if you knew that,
you would not be asking what you are asking.)
--
Rich Ulrich
> Date: Thu, 4 Apr 2013 07:01:30 -0700
> From:
[hidden email]> Subject: how to develop probit or multilogit models for accident data
> To:
[hidden email]>
> i had given the following coding
> severity sv Rear end head on swipe lanewid rural drvage female alcohol
> 1 1 0 1 1 7 1 21 1 1
> 0 1 1 1 0 7 1 20 1 1
> 1 1 0 1 1 7 1 27 1 1
>
> as final_coded.xls
> <http://spssx-discussion.1045642.n5.nabble.com/file/n5719262/final_coded.xls>
>
> severity single vehicle rear end head on swipe rural female alcohol
> fatal=0 yes=0 yes=0 yes=0 yes=0 yes=0 yes=0 yes=0
> non fatal=1 no=1 no=1 no=1 no=1 no=1 no=1 no=1
>
> how to develop model for different types of collision based on severity
>
> pl. help me
>
> ...
Locked
