session covariates for multi site

[Mark Whun]

good day secr google group,

I am new to secr but am the only in my office with R experience to work on this project, so please forgive me if the questions basic answers. I am using secr to get density from 8 sites, before and after treatment and to see if an effect of treatment. im not sure if I am using an appropriate design for my datas though?

the 8 sites are reserves in 3 areas (areas are 600-km apart but have the same general climate/vegetation type. the areas where sampled at different times, but same for sites in areas. Each survey we had between 7-11 sample day. so it looks like this

area1- site 1 (treated) , site 2 (not treated), survey both first in march then may

area2- site 3 (treated), site 4 (not treated), survey in July and august

area3- site 5(treated), site 6 (treated), site 7(not treated), site 8 (not treated), survey in January and september

is it okay to model the areas together by put a session covariate for area or is it best to model each areas apart? I’ve used survey as a session (16 sessions) and have with session session covariates of area (area1, area2, area3), site (treated, not treated) and survey (before or after the burn). I notice that if area modelled individually, density and confidence changes so I am not sure which is the best reflection of the datas.

I also have some questions about habitat masks vs just buffer but will save for now, as this is a long question thank you

thank you, I appreciate any advice, M. Whun 

[Murray Efford]

Hello Mark

Your question raises issues of experimental design and analysis - I hope someone with that specific expertise chips in. I think you are on the right track - considering models with area, treated/not treated and pre/post treatment as session covariates (each survey a 'session'). Your question is whether to conduct one large analysis (all areas, with area as covariate) or 3 separate analyses. It seems desirable to conduct one large analysis, but if the treatment effect variates among areas then you may end up with a rather large model (including area x treatment interactions). Strictly, in experimental design terms, multiple areas gives you important information on the variance of the effect. However the gap between pre/post surveys differs quite a lot...

Murray