SPSS ROC analysis produces average values of consecutive cutoffs, which provides other sensitivity and specificity values than by using the plain formulas for these values. I would like to understand why SPSS produces these average values and I am wondering if other researchers describe these average values in stead of the plain formula values, because it will have consequences for the chosen cutoff and the comparability of different research outcomes? I am working on an article describing the cutoff at the PQ-16, a screener used to detect psychosis in a two step screening procedure with the CAARMS interview.
Hello Yvonne de Jong. I gather you are concerned by the fact that when the test values are whole numbers, the cut-offs shown in the ROC output are mid-way between two scores--e.g., the cut-off between 7 and 8 is 7.5. Is that what you're talking about? If so, I don't think it is having any adverse effect on the Sensitivity and 1-Specificity values that are shown in that same output. Here is an example you can try. If it does not address your concern, please try to use the same bankloan.sav dataset (which all SPSS users have access to) to illustrate the nature of your problem. HTH.
* Use the bankloan.sav data file that comes with SPSS.
* Modify the path on the next line as needed.
GET FILE='C:\SPSSdata\bankloan.sav'.
* Use age as the test variable and default as the condition/disease.
ROC ANALYSIS age BY default (1)
/MISSING USERMISSING=EXCLUDE
/CRITERIA CUTOFF=INCLUDE TESTPOS=SMALL DISTRIBUTION=FREE CI=95
/DESIGN PAIR=FALSE
/PLOT CURVE=ROC(REFERENCE) MODELQUALITY=FALSE
/PRINT SE=TRUE COORDINATES=ROC CLASSIFIER=FALSE.
* Let's take the cutpoint 34.5 as an example.
* The ROC output says that Age
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS