I’m working with some data from an activated sludge respiration inhibition test. We have 6 dose concentrations and a control, all consisting of a single test vessel. The units of measurement are mg/L of oxygen per hour. We have calculated the percent inhibition from the control for each test vessel, in order to calculate an EC50 concentration. We are using the computer program of C.E. Stephan, if anyone is familiar with it. It calculates effect concentrations by binomial probability, moving average, or probit. Based on the results and output, the user determines which analysis is appropriate for the data. It’s really primarily designed for use with “groups”; i.e. you dose 3 groups of 10 subjects at different concentrations, enter the group size and number responding into the program, and it calculates the dose that effects 50% of the subjects. We have applied it to the respiration data by entering 100 as the “group” size for each concentration, and using the percent inhibition as the number responding. I know that’s kind of cheating, but no matter what the study design, the program uses the input data to calculate percentages for effect concentrations anyway. In fact, that’s how effect concentrations are typically calculated, using percent response. Our resident stone-thrower says that this is an inappropriate use of the analysis because we don’t have “replication”. That is, we only have one test vessel instead of multiple subjects. I’m wondering why this matters. This program, or any other, has no idea when we input the data if it’s coming from a study with “replication” or not. They calculate percentages and effect concentrations and that’s that. We could theoretically put in any “n” number for group size, and as long as we put in the right number for “responding” to get the appropriate percentage, we would get the same answer. Our expert says that we should be using non-linear regression, or graphing the data and dropping a line to the axis to obtain the 50% inhibition concentration. If this sounds like I’m being statistically dense, I apologize. As I’ve said before, I’m not statistician. Any thoughts? Thanks.