### Topics

- I agree with Dr. Tomer Czaczkes but maybe you could try logistic regression, fitting your model to each entrance of the olfatometer.
- There is a limitation with the independence between experimental units (olfactometer and insects) that depending on your hypothesis can produce a pseudoreplication problem. Please have a look to this paper that we published with other coleagues some years ago.
- Here it is...
- There is a possibility of pseudo-replication depending on how the four tire olfactometer is handled. ANOVA may be misleading. repeated-measures ANOVA using PROC MIXED would be appropriate since the insects are from similar colony (i guess) with similar four tire in the olfactometer. Since perhaps you are looking at the response of the insect to dose/concentration levels of your treatments, i suggest repeated-measures logistic regression via generalized estimating equations using GENMOD procedure.

You could check out these two papers; J. Insect Physiol. 55, 774-780;

Journal of Experimental Biology 214, 956-962

Cheers - Hello,I would use the Wilcoxon-one-sample test to evaluated whether the time spent by the insect in the test field differed significantly from the null hypothesis (total observation time / four fields, e.g. you expected 75 seconds performance for each field in a 150 second observation period). You could find this statistic here: Basic Appl. Ecol. 12: 403-412 (2011)
- I agree with Eduardo. I know his paper and it is an excellent reference on pseudoreplication. The heterogeneity G-test maybe a simple option for testing observed versus expected proportions in your olfactometer.
- Hello I did in my thesis like ........ Percentage data (Percent time spent in each arm) should be converted into proportion data and square root transformed to normalise distributions (Dytham 2003) and subject to simple ANOVA. Post-hoc Tukey HSD test should be used to determine which factor level differed significantly (Crawley 2007) While using R statistical Programming.
- Thanks to you all. However, I am actually more particular about the initial choice of each insect for the odor presented at any of the four arms of the olfactometer rather than time spent in the field.
- Exactly, i did hypothesis that you needed to test different concentration/doses of the different odour sources. And as i had my earier suggestion would suite your situation.

Cheers - If your data is related to Kruskall-Wallis test could be provided Friedman non parametric data.

Or could be a test of two-way ANOVA - You may try Chi-square test, with an assumption whether the number of the crickets going into each arm or stand still was the same and each arm received theoretically 1/5 of the total number experimented was made.
- Chi-square test goodness of fit will be very useful for your experiment
- I think it depends on the response variable. If you record the distribution of each insect at the first choice, the data are independent and I agree that Chi-square test will be appropriate. But if you take the time spent in each arm of the olfactometer then, data are correlated and paired. I have seen the use of Friedman test for the latter case (GUIMARÃES & ZUCCHI, 2004. Neotropical Entomology 33(2) 217-224). Also take care if some factor is pseudo replicated.
- Principal component analysis (PCA)
- I agree with some comments that stated using ANOVA tests whwn seeking the effect of the treatment (factor). The test (One-/Two-) ANOVA should be supported by post-hoc Tukey test when outcomes are positive.
- The functional relationship between parasitoids' behavioral responses and the different odor sources offered in the four-arm olfactometer was examined with a log linear model (a generalized linear model, GLM). (http://chemse.oxfordjournals.org/content/30/9/739.full)
- It depends on the data. The Student's t-test and ANOVA make the assumptions that 1. the data is normally distributed and 2. that these are continuous random variables. I imagine that either is true and that your data are discrete random variables (i.e. counts or frequency of occurrence). This being the case, Chi squared is the most suitable, using the Observed and Expected probabilities of the experiment.
- Be aware that you should not use a test assuming independent variable; in an olfactometer, the data collected are not completely independent. Check Turlings et al. 2004 for details on how to test response of insect in a multiple choice assay or the link provided by Monica. Both papers explain how such data should be tested.
- Either a generalized linear model or a One/Two Way ANOVA followed by a means separation test, would be suitable.
- In my opinion, I would say Compositional data analysis would be the best analysis for this type of trial as neither of the choices are independent from each
- Thanks everyone. Sorry, it took me a while to reply

## All Answers (22)