Hi Will, It is difficult to give "advice" without knowing a lot more about the problem at hand and that is not something for an email list but for local statistical support at your University etc., but your observations as you describe them will not be independent and hence violate the assumptions of the methods. You have three repeated measurements so potentially some temporal autocorrelation. Also the observations within transects are potentially spatially autocorrelated.
You will need to account for this in your analysis, and a simple t-test does not. If you are interested in the bird "assemblage" (not sure that makes sense for birds) then CCA/RDA in Canoco (and soon to be in R via the vegan package if I pull my finger out and finish coding) will allow you test for effect of explanatory variables in presence of non-independence due to the line transects via permutation tests. With 3 temporal observations, you might be better off ignoring this temporal autocorrelation and include a "fixed effect" for "sampling time" as a dummy variable --- you can do a partial analysis (remove effect of this dummy variable) if you want to remove the complicating factor of sampling on 3 different occaisions, or leave it in the analysis and test for difference in time. You should be able to do this as a univariate regression also in Canoco (and soon vegan) if you are interested in individual species or diversity. If you are modelling individual bird species or diversity of birds as the response via more traditional regression techniques, then you need to account for autocorrelations. This starts to look like a mixed effects analysis, where you have a random effect for transect (if you consider these to be a random sample of all possible transects you could have selected), a fixed effect for sampling time (it is difficult to estimate this as a random effect with only three observations) and then a fixed effect for forest/no forest. Then you also need to account for the spatial autocorrelation within transects via specifying a form for the covariance matrix associated with the regression model residuals. Whilst not strictly correct, as you are working with counts of species (diversity) something like a poission GLM/GAM or a negative binomial GLM/GAM would be appropriate. Something similar would be appropriate for abundances of an individual species. Putting all of this together in software is difficult. In R using the mgcv package you could do a poission analysis incorporating the mixed effects and spatial autocorrelation, via the gamm() function. As you can see, it starts to get fairly complicated very quickly once you go down the generalised linear/additive mixed effects approach and have correlations in your model residuals. If you can deal with your data in CCA/RDA in Canoco (or vegan in R) then that would be an easier route to go down... HTH G > Im posting a general question to those who know something about > ecological statistics. As part of my masters thesis, I want to compare > bird species richness and abundance of birds detected in transects in > mature forests and mature forests experiencing low-density housing > development (Ive worked hard to select sites that make this comparison > valid and I will be collecting data in a way that will allow me to assign > probabilities of detection in both habitat types). > > I have 12 forested transects and 12 residential transects. I will visit > each site three times. Im essentially searching for the appropriate > statistical analysis to compare mean number of species/mean number of > individuals/and mean diversity index values between the two habitat > groups. Im assuming that basic parametric tests such as t-test/one-way > ANOVA would be appropriate, however my thesis committee challenged me to > see if there arent better ways of comparing the data, particularly since > there will be repeated visits to each transect. > Any help/thoughts/suggestions would be most welcome. >
