Hello all, I am having a conundrum with deciding on which simple statistical approach I should use in analyzing choice trial data. I have reviewed the literature and I see a relatively equal split between paired t-tests and chi-square analysis. I have provided an outline of the question, the experimental design, and the hypothesis being tested. I'd appreciate some input on the subject. I have already run both analysis but I'm interested in others interpretation here.
Question: Does the presence of herbicide residues affect female insect oviposition behavior? Experimental Design: A recently mated captive female butterfly is placed in a 60x60x60cm tent with two equal age host plants of the same species. One host plant was sprayed with water 3 days prior while another was sprayed with an herbicide 3 days prior (the herbicide used was a plant-specific herbicide not intended to kill the host plant). Distance between plants was maximized and their orientation in the tent was randomized. The female was left in the tent for 48 hours after which she was removed. Next we counted the number of eggs on the control (water only) and the treated (herbicide) plant. Hypothesis: If no preference exists to avoid ovipositing on herbicide exposed plants then each egg has an equal chance of ending up on the control or treated plant. That is the (# eggs on control : # of eggs on treated = 1 : 1). Results (minimal information provided here): 1) The total number of eggs laid by each female varied widely from 14 to 170+ (N=32) 2) Data is normally distributed 3) Control and Treated egg numbers have equal variance Chi-square I'll be upfront and admit this is the analysis I favor. As I see it, the expected ratio is what needs to be tested not the means. However, some have argued that Chi-square is not appropriate here as the eggs on the control and treated plants are not independent groups. If this were true then wouldn't the classic use of chi-square in testing a coin flip be invalidated? That is if each coin flip is an independent test of probability then isn't each oviposition (egg laying) event an independent test of probability. Paired T-test (Dependent T-test) Here we would be testing for difference between mean eggs numbers on control versus treatment. In my experience this is the classic test for before and after comparing relative changes in means in one individual (with multiple individual for replication). However, in this case there is not before or after. In this case the female is choosing between one of two options (control or treated) at the same time. This is further complicated by the fact that each female has a finite number of eggs she can lay in a 48 hour period and each decision to oviposit she makes will limit the potential to lay eggs on either plant in the future. Which do you think is best? One, none, or both and why? I apologize if this seems like an overly-simple problem but if I am going to teach this in the future I want to understand why I should use one statistical approach over another. Cheers, Tyler Tyler L. Hicks Ph.D. Student Washington State University Vancouver E-mail: [email protected] Web Page: http://thingswithwings.org "We were certainly uncertain. At least, I'm pretty sure I am." - Modest Mouse _________________________________________________________________ The New Busy think 9 to 5 is a cute idea. Combine multiple calendars with Hotmail. http://www.windowslive.com/campaign/thenewbusy?tile=multicalendar&ocid=PID28326::T:WLMTAGL:ON:WL:en-US:WM_HMP:042010_5
