[CT Birds] population estimates (long .... and probably boring)

Chris Elphick elphick at sbcglobal.net
Tue Oct 4 08:17:11 EDT 2011


A few responses to your question ... a bit long-winded I'm afraid ....

Before I start though, as a long-term advocate of breeding bird atlases, I have to second Jaime's comments.  Most birding is pretty awful (sorry!) for assessing things like population size and trend because people tend to go to a limited number of sites; focus on really good sites (which gives a rosy impression of things); and if sites start to deteriorate, gradually shift to other locations, meaning that the birding stays good.  Atlases are no panacea, but they at least get people into places that usually do not get birded and so provide a better sense of the big picture than does regular birding.

Turning more directly to your question, like many others I was very surprised by some of the projected population estimates in the State of the Birds.  Before it was published, I was asked to review the article in question and suggested that those numbers should not be included without also providing two critical pieces of information - full details of the methods used to generate them and estimates of the uncertainty associated with them (i.e., the margin of error, or "confidence/credible intervals" to use the statistical terms).  Both of these would be standard items in a formal scientific study.  Because State of the Birds is not a scientific treatise, such details are generally not included in its articles.  This is probably a good thing, as they clutter up the text and make it less accessible to anyone without statistical training.  But these numbers struck me as sufficiently surprising that I felt the added information was necessary,
 especially as there were no scientific journal articles to back them up.

Subsequently, I have heard (from the most reliable of authorities) that cerulean warblers are probably more numerous in CT than most birders think, at least in large forest blocks in the southeast portion of the state, and that numbers have increased in recent years (I'm not sure if the same is true for Acadians, but I don't think so).  Even with that knowledge, though, I still find the estimates in the State of the Birds surprising and would not want to base conservation policy on them without some verification.

I still do not know the exact methods used to generate the numbers, but the general approach that is taken in situations like this is as follows:

Step 1. Sample the population using a standardized method at randomly selected locations.  Point counts are the most commonly used method, but there are others.  During these counts all birds seen and heard within a fixed time period (usually a few minutes) are recorded.  

Step 2. Estimate how many individuals were missed at the sample points.  This may sound strange, but there are a variety of mathematical methods available to do this - a very common one (though probably not a very good one for forest birds) involves estimating the distance to each bird detected during the count.  I won't go into how you then use that information statistically, but would be happy to follow up with an explanation if anyone is interested.  (Note that entire books, and dozens - probably hundreds - of scientific papers, have been written on these methods, which are also used for counting lots of things other than birds.)

Step 3. Extrapolate that information on bird numbers in the sampled areas (e.g., the point count locations) to the entire area of interest.  This step will only work if the points truly were randomly located such that everywhere in the area of interest was equally likely to be visited.  It also only works if you have good estimates of how much habitat there actually is.

So, that's the basic theory.  It sounds simple, but in practice converting point counts to population estimates is fraught with difficulty (as one of the PhD students in my lab will attest!!).  All of the different methods have assumptions and ensuring that those assumptions are met is often difficult.   If they are not met then there are often ways to correct for them, but that requires more math and often increases the margin of error.  Perhaps the biggest problem is that small errors at the first step can multiply (often hugely) as you work through the process - and especially if you end up extrapolating over large areas in step 3.  

For forest birds, things are especially tricky.  For example the methods that use distance estimates to determine how many birds were detected and how many were missed really depend on good estimates of those distances.  Several studies, however, have shown that even highly trained observers are often terrible at estimating distances unless they actually see the bird.  Most forest point counts rely on birds that are only heard, so this is a huge problem.  These methods also run into problems if there is double counting, which is more likely if you have mobile animals (e.g., birds) especially those you can't see moving (e.g., because they're in the forest canopy).  The standard way to deal with this is to keep the counts short, and even 5 minutes can be too long (in saltmarsh sparrows, double-counting starts to become a problem after ~4 minutes - and these birds are much easier to track than birds in a forest would be).  Yet another problem is that
 studies show that the more species you are tracking the more mistakes you make.  Even going from 6 species to 8 species has been shown to create detectable increases in the errors.  Forest bird point counts are often used to track dozens of species simultaneously.

All of this may sound as though it is impossible to use these methods in forests.  That is not the case, but it is not simple and both field and analysis methods have to be chosen very carefully.  In terms of the State of the Birds estimates, it is impossible to judge their accuracy without knowing exactly how the data were collected, how they were analysed, what assumptions were tested, etc. etc.  If this is not done right then it would be very easy to produce extrapolations that are off by an order or magnitude.  

Even if the basic estimates are good, the margins of error can be large so must be calculated so that one knows how much faith to put in an estimate.  For example, I once did an analysis of this type and came up with a confidence interval that span six orders of magnitude (needless to say, I concluded that the estimate was pretty useless) - that was for a shorebird, where the tendency to form big flocks can really magnify the uncertainty, but it illustrates the problem.

So, for Roy and the two of you still reading (I'm an optimist I know) - sorry to tie up so much ctbirds space.  Unfortunately, Paul's question is both very good and not at all easy to answer.


Chris Elphick

Storrs, CT

elphick at sbcglobal.net

More information about the CTBirds mailing list