Royle method for multiple counts

[tracicastellon]

HI All,

I have a couple of questions regarding the Royle method for multiple counts (Biometrics).

First, is it possible to analyze count data in the form of decimal measures. Because our sample areas differ in different habitats (due to differences in visibility), we would like to analyze counts per unit area. Hence, many of our counts are <1.

Second, when I run the analyses, most of the output estimates have abbreviations (see below). Does anyone know what these different abbreviations stand for?

Estimated parameter estimate std.err 95% confidence interval
-------------------------- -------- ------- ------------------------
Detection probability (c) = -1.#IND 1.#QNB 1.#QNB - 1.#QNB
Avg. abundance/sample unit(lambda) = 1.#R 1.#R 1.#R - 1.#R


Thanks bunches for your help…

Traci Castellon

[jhines]

PRESENCE can handle observations < 1. Just add a column in the input form and when you save the pao file, tell PRESENCE that the last column is a frequency count (click 'yes' when asked).

The output you listed indicates a problem with the data or model. They are indicators of a computational problem (eg., divide by zero, log(neg),...) and the estimate could not be computed.

Jim Hines

[darryl]

Jim,
I think Traci is saying that her actual count data can have decimal places, not the frequencies...
Darryl

[ganghis]

Why not model the raw counts with area as a covariate (i.e., controlling for area within the model)? That might be your best bet for this type of analysis.

[darryl]

PS by counts these are of unique individuals, right? So a count of 5 is definitely 5 different individuals, not 1 individual counted 5 times (or something in between).

[tracicastellon]

darryl,

that's correct. the count data are have decimal places because the raw counts have been transformed to counts per unit area. and, yes, the counts consist of unique individuals.

is it possible to analyze the data this way? i created a small data set with decimals and it ran fine, so i assumed it would work...

thanks for your help,
traci

[darryl]

It may run, but whether it's giving a sensible answer is another thing. Jim Hines would be the one to check with about what PRESENCE in really doing with your decimals, or there was Paul Conn's suggestion of using the raw counts with Area as a covariate.

[tracicastellon]

Thanks for all your suggestions.

In answer to the suggestion about using sample area as a covariate… that won’t work for us because our goal is to assess the influence of stand age on abundance, but sample area is strongly negatively correlated with stand age because tall vegetation reduces visibility. So, using sample area as a covariate would tell us how abundance varies with sample area, but it would be confounded with the variable (stand age) that we’re really interested in. That’s why we’re trying to analyze counts per unit area, in an effort to control for differences in sample area among stands of different ages.

So, back to Jim Hines statement that PRESENCE will accept decimal counts, but we need to use the last column as frequencies… I’m not sure exactly how I’m supposed to do that. Do I add an empty column (i.e., will PRESENCE calculate the frequencies), or do I add up frequencies in excel to create a frequency column before I paste the data into PRESENCE? And, what exactly do you mean by “frequencies?” Is it the sum of all the counts (in decimals) across all sample occasions for a site?

Thanks again,
Traci

[tracicastellon]

Another question… Can the Royle method in PRESENCE handle overdispersed data? Is it possible to specify a Negative Binomial instead of a Poisson distribution?

Much thanks,
Traci

[darryl]

Including area as a covariate is doing a very similar thing to standardising the data directly. If area and stand age are highly correlated you're going to have a very hard time trying to definitively separate the 2 effects regardless of which approach you take.

You can only use a Poisson distribution in PRESENCE as far as I'm aware, but you can include covariates on the rate parameter which will account for some overdispersion.

Frequencies is the number of sites with the exact same history, so in your case, the exact same series of counts. You include that number as an additional column after the final survey occasion. Those sites also have to have the exact same covariate values.

Darryl