I have four techniques to compare, done side by side on 5 adjacent surfaces (but not the exact same surface as the test "consumes" the surface) in 7 different buildings, all buildings similar. Each test use different metrics but produces a continuous variable outcome, of unknown distribution, e.g., not necessarily normal, and a "below detectable limit" or "0" option (number of residual spores via microscopy, culturable residual spores via swab, fungal cell constituents via fluorescence, and airborne fungal spores via microscopy). The fungal cell constituent test was performed twice for each location (the others were done only once) in immediately adjacent locations. For each location, the test is a pass/fail outcome. I wish to compare the tests to see if they correspond for the 35 locations tested. How should I structure the statistical analysis? Is there any alternative to logistic regression? Would logistic regression be a poor choice?
Can you state your null hypothesis? I am trying to figure out what null you are testing from this sentence: I wish to compare the tests to see if they correspond for the 35 locations tested. But I can think of at least two different hypotheses, and they would suggest different approaches.
Dear Brad
If I understand your hypothesis, you want to compare four techniques ( treatments ) and your data not normally distributed, and the result of each test is pass or fail , and you have 35 locations ( replications ), I think you can use chi-square test to compare each two tests, you have observed ( pass number ) and ( fail number ) , the expected number will be equal ( half the total for each one technique, because your null hypothesis is equality and the alternative hypothesis is not equality ).
Good Luck
If you have some "ground truth", you could use epidemiological criteria (sensitivity/specificity).
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis