How to test for association between proportions with multiple variables? I have tested 100 bacterial isolates for resistance to 6 antibiotics (tetracycline, erythromycin, kanamycin, streptomycin, gentamicin, chloramphenicol). The result for any antibiotic could be either sensitive (1) or resistant (0). 
How can I determine if there is any significant association between resistance to various antibiotics (I can frequently find isolates resistant to three or more antibiotics but how to know if this is significant)? I could use Chi-square to compare between any two antibiotics but how to make multiple comparisons and will this make any difference except saving time. 
Scientists in the papers used Chi-square with Bonferroni adjustment. I am using Statistica for Windows software.
thanks
 A: The first decision you have you have to make is what to control for? Is it the probability of making any false discovery (FWE control) or the proportion of false discoveries (FDR control). 
Bonferroni will control for both, but is very conservative. At 100 tests, you might lose much power. 
I would have a look at Benjamini-Hochberg procedure for FDR control under mild dependence assumptions or the more general (and less powerful) Benjamini-Yekutieli procedure. I would look into Holm procedure for FWE control. 
I am not familiar with Statistica, but luckily, all the mentioned procedures, take p-values as  inputs and are easily computable "manually". If you are familiar with R, have a look at the p.adjust function. 
A: You could approach this in various ways - I assume that each isolate is tested with all 6 antibiotics - so you have 100 subjects with a response of 6 0's or 1's.
For an exploratory technique, you could try a cluster analysis or apriori analysis - this would show which collections of antibiotics (if any) have a similar reactivity pattern. 
Another way of getting at this would be to try a factor analysis on the data, with non-orthogonal rotation (like oblimin) - and see what that gives you.
I'm not sure what you mean by "significant": if some isolates are resistant to 3 antibiotics, then they are resistant - possibly rare, but why would one discount the information? Is there a probability of error in your testing procedure? Otherwise, what you sees is what you gets.
Warning: Don't do a log-linear analysis, because your observations are dependent - the subject is the isolate, not the 0/1 observation.
Here's a trick you might try


*

*Calculate the proportion resistant for each antibiotic

*For each sample, calculate the probability of that pattern of resistance, assuming independence: $p_1\times p_2 \ldots$ for each antibiotic

*Do a lot of bootstrap samples (with replacement) from you data. Observe the proportions for each possible string of 0's and 1's.

*Sort your combinations by expected probability and plot against observed probability.

*Throw a straight line y=x across your graph

*Marvel and compare.


This should illustrate the presence of an association
