occupancy with two seasons of data

[dajones]

I am using occupancy models to analyze avian point count data in MARK. Sites were visited in both 2006 and 2007, with three visits in June of each year. Given that I only have two years of data, I haven’t considered using multi-season models to look at change in occupancy. Rather, I am interested in conducting what is essentially a single-season analysis using the information from both years to improve estimates. I saw a previous posting with a similar situation, and it seems the suggestion was to put all visits from multiple seasons into a single encounter history, and assume that occupancy was constant across years. However, I am pretty sure that I do not meet the assumption and am afraid this will lead to substantial bias of estimates of both occupancy and detection. One alternative would be to create a unique encounter history for each site and year, which would yield two rows of data per site. I then worry that sample sizes may be inflated and confidence intervals too small because of possible lack of independence for sites used in both years. I could increase c-hat to values >1 to try to account for this, but am not sure how I would choose a value or if this is the best way to tackle this problem. I have not been able to find any other solutions in either the forum postings or other MARK documentation. Any advice would be greatly appreciated.

[darryl]

Even with only 2 years of data I'd suggest the multi-year models (called robust-design occupancy in MARK, last time I looked) are a better way of analyzing the data. While we talk about it terms of the dynamic rates of change, colonization and extinction probabilities can be considered as a way of accounting for temporal autocorrelation if that's what' you're concerned about. These models can be reparameterised so that that you estimate occupancy in each year and one of the dynamic rates if you claim you're more interested in occupancy each year. Just make sure that you check the other dynamic rate (which is now a derived parameter) is estimated to be between 0 and 1. Gary never use to enforce this constraint (though he might now), so if it's outside 0 and 1 I'd claim the results are not biologically reasonable.
Cheers
Darryl