ROC Curve Questions w/ SPSS
Maria
(BMI) as the test variable (continuous) and number of cardiovascular
risk factors (dichot var) as state var. I have several questions I am
trying to resolve and would greatly appreciate any feedback you may
have.
1. On the output, what does this mean: "the test result variable has
at least one tie between the positive actual state group and the
negative actual state group. Statistics may be biased."?
2. Another statement on the output reads: "The smallest cutoff value
is the minimum observed test value minus 1, and the largest cutoff
value is the maxiumuy observed test value plus 1. All other cutoff
values are the averages of two consecutive ordered observed test
values". I tried working this out against my data values and it
doesn't make sense?? Also, is there a way to request what cutoff
values are presented (for example if I want to see the coordinate of
the curve for BMI of just rounded numbers such as 14.0, 15.0 etc., can
I do this?
3. Finally, as I am using NHANES data, when I turn on the weighting
variable, I get an error message: "Warnings The WEIGHT variable
contains at least one fractional value which is inappropriate for the
ROC procedure.
This command is not executed."
Thank you for your attention!
Maria
Richard Goy
> I am using SPSS ver 11.5 to produce ROC curves, using body mass index
> (BMI) as the test variable (continuous) and number of cardiovascular
> risk factors (dichot var) as state var. I have several questions I am
> trying to resolve and would greatly appreciate any feedback you may
> have.
>
> 1. On the output, what does this mean: "the test result variable has
> at least one tie between the positive actual state group and the
> negative actual state group. Statistics may be biased."?
To estimate the AUC (constructing trapeziods) two cordindates share
the same point. see http://gim.unmc.edu/dxtests/roc3.htm and their ref
or any math text for complete explantion.
> 2. Another statement on the output reads: "The smallest cutoff value
> is the minimum observed test value minus 1, and the largest cutoff
> value is the maxiumuy observed test value plus 1. All other cutoff
> values are the averages of two consecutive ordered observed test
> values". I tried working this out against my data values and it
> doesn't make sense??
> Also, is there a way to request what cutoff
> values are presented (for example if I want to see the coordinate of
> the curve for BMI of just rounded numbers such as 14.0, 15.0 etc., can
> I do this?
This is a huge problem with the SPSS output. try looking at SAS and
you will likely change. It is extremely difficult to determind
cut-points from the SPSS output. Perhap someone could enlighten us one
why SPSS breaks the scale into midpoints. editing the table the round
off the results give incorrect results