I need to compute the sensitivity and specificity of a screening tool, and I will report the sensitivity and specificity of it. Are there any rule of thumbs that might suggest what the bare minimum acceptable level for identification accuracy used in social sciences/psychometrics is?
Also, is it OK to report sensitivity and specificity values without confidence intervals? The 2x2 table aiding in computing the sensitivity and specificity values does not provide confidence intervals. How do you compute those?
Dear Sedat,
- Sensitivity and specificity higher than .80(80%): good model
- Sensitivity and specificity between .50 and .80 (50 to 80%): acceptable model
- lower than .50: poor
Source: Marôco, J. (2011). Análise Estatística com o SPSS Statistics (5ª ed.). Lisboa: Report Number.
good luck with your research!
Hi Sedat,
The acceptability of particular sensitivity/specificity characteristics depends on the way that a screening tool is going to be used, and the amount of misses or false positives that are deemed acceptable for your purposes.
For a screening tool for depression, for example, you would focus on high sensitivity (80% plus, as a guide), since it is important not to miss people who are genuinely in need of help. However you might find specificity as low as 50% to be acceptable, as the consequences of a false positive may be that a patient is simply referred for a further evaluation that is not especially costly or inconvenient. In another area such as medicine, however, a false positive may result in highly costly and risky interventions (chemotherapy, for example), and one might be much more demanding, requiring a specificity in the high 90s.
Generally, the practicalities of costs versus benefits will decide what psychometric properties are acceptable. Published reviews in the area of assessment of what it is that your screening tool is measuring should be able to offer you some guidance in this respect.
I hope this helps.
I Guess that a sensitivty of 50% is equivalent to tossing a coin, so the test is of no use at all.
- Badges
- Science topic
- Similar topics
- Social Science