occupancy with two seasons of point count data

[birdnerd]

I am using occupancy models to analyze patterns in avian point count data to determine if a forestry practice affects the rate of occupancy of treated sites by several species. Sites were visited in 2012 and 2013. About half the sites are controls, about 25% were treated pre surveys (2011 or 2012) and another 25% were treated between 2012 and 2013.

I am planning on using a multiple season (robust design) occupancy model, but I'm unsure if this is the correct choice since I'm mainly interested in the effect of treatment on occupancy, not change in occupancy between years. However, I'm not sure how else I could do this, since using a single season model would require that I create an encounter history for each year and site, artificially inflating my sample size and resulting in not independent sites.

Secondly, I'm not sure how to group the sites that were treated between 2012 and 2013: In 2012, they're untreated sites, but in 2013 they become treated sites.

Any advice on how to proceed would be welcome.

Thank you

David

[bacollier]

I am planning on using a multiple season (robust design) occupancy model, but I'm unsure if this is the correct choice since I'm mainly interested in the effect of treatment on occupancy, not change in occupancy between years. However, I'm not sure how else I could do this, since using a single season model would require that I create an encounter history for each year and site, artificially inflating my sample size and resulting in not independent sites.

RD style models typically require at least 3 primary occasions and you have 2, so I don't think RD will really work. Using a single season model is the most straightforward way to do this.

Secondly, I'm not sure how to group the sites that were treated between 2012 and 2013: In 2012, they're untreated sites, but in 2013 they become treated sites.

Here a couple options:

1) you have treated/untreated regardless of year and you group that way,

or

2) you have treated prior to 2012 surveys,you have treated prior to 2013 surveys. Additionally, you will have to tease out the effect of treated before 2012 and surveyed in 2012 and 2013--but, effects of treatment would likely occur more than one time period post treatment, it would probably be easiest to just treat the treated before 2012 then surveyed in 2013 as a independent group, but you could do it with a treatment covariate as well, in some cases a TSM model would work but you only have 1 occasion to look at so that's a bit of overkill.

[birdnerd]

Thanks for the advice.

While perusing the forum before I posted this question, I came accross this post: viewtopic.php?f=1&t=825&p=2067&hilit=point+count+data#p2067
that suggested a multi-year model, even with just two seasons worth of data. Has MARK changed since then, to no longer allow that?

I considered using a single year model, but since I surveyed the same sites two years in a row, won't that introduce a significant amount of autocorrelation, since the sites aren't independent?