Responding to these comments from Alan:
This is not so much a question of accuracy in data, as it is of sampling
probability for different configurations of observers.
If 60 years ago we had one observer in that 10-mile wide strip along the south
edge of Coos County, we can trust that they'd spot any Rufous Hummingbird that
passes, say, within 25 yards on either side, during whatever part of the day
that they're paying attention. That's about a 1 in 350 chance. What they
recorded may be "accurate" but it's a sample.
If now we have 15 observers spread out evenly in the same strip, not talking to
each other, that raises the odds of spotting a given hummingbird by a factor of
15. To compare their observations with the first case of a solo observer, we
should take the date of 8th detection (the median date).
I think that's what this author tried to do with the logistic regression,
though there are a lot of things left unexplained. For example, if one of those
observers spends 4 hours per day watching while the others spend just 5 or 10
minutes per day, how is that accounted for?
Factoring in use of hummingbird feeders and communication among stationary
observers brings more levels of complication. Suppose one of these observers
(let's call him James Jobinshirk just to make up a name) spends 5 hours a day
patrolling that strip. When he spots a Rufous, he alerts everyone else in the
local network, so all at once they're more focused. Is the median arrival for
that network still representative of what our lone observer in 1960 would have
found?
Not only that, but word of Mr. Jobinshirk's sighting carries on to Portland,
and people rush to hang out their feeders. Chatter carries over to Washington
and on to British Columbia. Is that a different situation from when Theed
Pearse, Esq. was watching patiently for his first hummingbird of the season in
Courtenay, B.C.? Especially now that Mr. Pearse's grandchildren might have a
feeder out for Anna's Hummingbirds all winter long?
These are just some of the difficulties in comparing these two very different
datasets. Both may be highly "accurate" but the new system is a completely
different sampling network, which furthermore is evolving rapidly from year to
year.
Joel
On Thu, 2019-03-21 at 21:08 -0700, Alan Contreras wrote:
This isn’t in my wheelhouse either, but it seems worth mentioning that early
movements of Rufous must be along one of the narrowest fronts of any migrant
land bird in North America, basically the outer coast below the Klamath
mountains and southern Coast Range. Most of the early migrants are in an area
no more than ten miles wide, using extremely predictable sites and habitats.
Wouldn’t that tend to allow fewer observers to generate data of fairly high
accuracy over time?
As Rufous does not breed in California, these early birds are probably not
doing much upslope drift in Curry County in February - they’d hit the snow
zone. So the pulse that gets to southern Coos really is the northern edge
each year and observers there are getting the best data.
In an area with low population and limited visitors, this also shows that if
the more active local observers are not using eBird, the eBird data becomes
pretty soft.
Alan Contreras
acontrer56@xxxxxxxxx
Eugene, Oregon