A generalized approach taking the whole temperature profile into account is to use a sinusoidal regression describing the reference condition to predict the temperature of the treatment condition, also described by a sinusoidal regression, then analyze differences between the regressions using a repeated measures analysis. The objects being studied need to be paired in some way to do this, but the gls function in the nlme package, R, can be used to correctly estimate the regression error if autocorrelation is present, which can then be modeled with an AR term.
JE Janisch -----Original Message----- From: Ecological Society of America: grants, jobs, news [mailto:ECOLOG-L@LISTSERV.UMD.EDU] On Behalf Of Christopher Brown Sent: Wednesday, February 06, 2013 13:16 To: ECOLOG-L@LISTSERV.UMD.EDU Subject: [ECOLOG-L] Statistical Question on Temperature Profiles Ecologgers, I have a master's student who is examining thermal preferences of two species of scorpions in the Sky Islands of southeastern Arizona. She has gathered some field temperature data as part of her thesis, but we are unsure how best to analyze the data (or perhaps more specifically, what data to analyze). I've given some details below, if you have some insight for us! The short version of the experiment: these scorpions are found under rocks during the day, and we have determined thermal profiles for 15 rocks under which scorpions were found and 15 rocks under which scorpions were not found. For both sets of rocks, we measured length and width and selected a range of sizes based on binning the rocks into three categories (small, intermediate, and large) and then choosing 5 rocks in each size range. Each rock had an iButton placed under it, and temperatures were recorded every 30 minutes for 48 hours. Her basic question is then, do the thermal characteristics of chosen rocks differ from the thermal characteristics of non-chosen rocks? Our problem is, what data should we use? Our first though is at a simple level: we could calculate mean temps for the two rock categories and compare them with a t-test, and/or we could compare variances or ranges (max-min) with a t-test to determine if variability differs between rocks. We've found a couple of different variations of this kind of analysis in the literature, but we'd like to know if this is the best (or "best") way to analyze the data, or are there more sophisticated techniques that involve analysis of the whole profile? If we do use a fairly simple analysis based on some type of summary variable, what is the best summary variable to use (mean? Variance? Range? Something else?) and the best analysis to do? If anyone has any experience in analyzing this type of data and has some suggestions, we'd be happy to hear from you! Thanks, CAB *********************************** Chris Brown Associate Professor Dept. of Biology, Box 5063 Tennessee Tech University Cookeville, TN 38505 email: cabr...@tntech.edu website: iweb.tntech.edu/cabrown