I want to identify a suitable cut-off for a test with discrete rating values from 0-8. SPSS has given me cut-off points in 0.5 intervals. Realistically, the test itself can't be rated in intervals of 0.5. I've attached a picture below. How does one go about choosing a discrete cut-off, and if I pick one between 2.5 and 3.5 (ie 3) how could I calculate sensitivity and 1-specificity?
What is the context? Are you trying to rule in or rule out? Medical students often learn the mnemonics SPin and SNout to remember that you need good specificity when ruling in and good sensitivity when ruling out.
When you do decide where you want to make the cut, you can always compute a new dichotomous variable and then cross-tabulate it with the gold standard (disease) variable to get the row and column percentages you want. E.g., if you decide that a test is positive if the test value is 3 or more:
COMPUTE TestPos = TestValue GE 3.
FORMATS TestPos (F1).
CROSSTABS TestPos BY GoldStd /CELLS = COUNT ROW COL.
Two of the column percentages in that table will show sensitivity & specificity; and two of the row percentages will show PV+ and PV-. Beware that with the 0/1 coding, positive tests will be in the second row, not the first row. If you want positive tests in the first row, re-code TestPos accordingly:
RECODE TestPos (0=2) (ELSE=COPY).
VALUE LABELS TestPos 1 "T+" 2 "T-".
CROSSTABS TestPos BY GoldStd /CELLS = COUNT ROW COL.
HTH.
Bruce Weaver Ah, thank you. I understand. Really appreciate the help!
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS