James Cantor wrote:
> Has anyone ever used the Rockit software to test differences between
> ROCs?
>
> Are their substantial advantages/disadvantages relative to SPSS for such
> analyses?
>
I haven't used Rockit, but I use ROC Analyzer sometimes, and I think I
can give you some relevant info: while SPSS computes ROC-AUC, with its
standard error and asymptotic confidence interval (problematic when
n<200 & AUC>0.9, BTW), it does NOT compare ROC AUCS (test with a
p-value), neither in independent nor paired samples, while ROC analyzer
does (although data need to be categorized first), and Rockit, to my
knowledge, also does. SPSS performance concerning ROC analysis is, in my
opinion, rather poor.
Anyway, with SPSS, a Z test to compare two independent samples samples
ROC-AUC can be easily computed using OMS, since the formula is quite simple:
AUC1-AUC2
Z=------------------------
SQRT(SE1^2+SE2^2)
And I have also written a macro that compares related samples AUC (more
powerful than using the formula above) and computes a 95%CI for their
difference, using bootstrap. I have tested it quite extensively, and it
gives the expected results. I can send it to you if you want to give it
a try (provided you are comfortable using SPSS syntax¯os).
HTH,
Marta García-Granero
--
For miscellaneous statistical stuff, visit:
http://gjyp.nl/marta/=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
