Birds Seem to Be Moving North. Part of It Is More Birdwatchers.

When a bird is first recorded in a northern place, two things can be true. The bird moved there, or an observer finally went there. A range map cannot tell these apart unless it also shows the effort.

My thesis going in was this: a meaningful share of reported northward bird shifts reflects more observers in new places, and the shift shrinks when analysts correct for effort. After reading what I could reach, I keep the first half and trim the second. The best effort-aware test I found shows both outcomes in one dataset. One group of birds kept its shift. Another group's shift was weakened by effort. I found no published share that covers all birds, and I will not invent one.

The question

How much of an apparent northward range shift is a change in where and how often people look? I want a number. What I have is a set of cases and a mechanism.

Data and where it came from

I read five kinds of material. I did not run any code for this post, and I did not use the Lab.

A national atlas series. Finland ran three breeding bird atlases: 1974 to 1979, 1986 to 1989 and 2006 to 2010. Each covered about 3,800 grid cells of 10 by 10 km. The analysis covered 114 southern species and 34 northern species [1]. An atlas is a planned survey, but its coverage still changed between rounds.

A winter count. Audubon's Christmas Bird Count (CBC) asks volunteers to follow routes in a 15-mile circle and count every bird they see or hear in one day [2]. Audubon's analysis of 40 years of these counts reports an average northward shift of about 35 miles per species [3]. Note the two verbs in the CBC rules: "see or hear". A count of birds reported on a CBC is a count of birds seen or heard by whoever turned up that day. I list the effort question for it below.

A review of the bias itself. Sanczuk, Lenoir and Staude (Nature Climate Change, 2026) report a geometric bias in the sampling used by range shift studies. Studies favour sampling along latitude, which makes a latitudinal shift more likely to be observed, as warming predicts [4].

Guidance on eBird. eBird holds about 150 million checklists from 1.1 million eBirders in 253 countries, as of its 2 billion observation milestone [5]. The eBird best practices guide says participants "sample near their homes, in easily accessible areas such as roadsides", and that checklists differ in duration, distance and observer number [6].

A trend comparison. Walker and Taylor used eBird for 22 migratory species in southern Ontario, 1970 to 2015, and compared the trends with the Breeding Bird Survey. They used species per checklist as a stand-in for effort [7].

I could not open the Wisconsin atlas paper on poleward shifts of breeding birds. I also could not open the full text of the Sanczuk paper. I cite the Sanczuk paper only for what its abstract said.

Method

I did not fit a model. I read how each study handled effort and recorded what happened to the shift.

For the mechanism, I use one hand calculation. It is not data. Suppose a bird is present in a block, and each complete checklist detects it with probability $p = 0.02$. The chance that a block with $n$ checklists records the bird at least once is:

$$P(\text{recorded}) = 1 - (1 - p)^n$$

With $n = 10$, this is $1 - 0.98^{10} \approx 0.18$. With $n = 100$, it is $1 - 0.98^{100} \approx 0.87$. The bird did not change. The block went from 18% to 87% on coverage alone, a factor of about 4.7. This is why I ask for effort next to every count. It is also why a shift of an atlas edge needs the unsurveyed cells counted.

Result

Finland, southern species. Their range moved poleward by 1.1 to 1.3 km per year in all three pairwise atlas comparisons. The authors reduced coverage effects by using a weighted mean latitude that takes the changing set of surveyed cells into account [1]. The paper gave no confidence intervals in the results I read, so I cannot show an error band. This shift looks the sturdiest of the cases.

Finland, northern species. The shift was 0.67 to 0.81 km per year. The authors say that cells in the far north "remained relatively frequently unsurveyed" in the first atlas, so the expansion of northern species "may thus, partly, be due to a better census". They add that the evidence is "much less clear than for southern species" [1]. A better census where the first atlas was thin is the pattern my thesis predicts. The paper does not say how many kilometres of the shift are effort.

Wintering birds in the CBC. The 35 miles is an average over many species and 40 years [3]. I could not read how that analysis handled party-hours and circle changes. A common CBC approach divides counts by party-hours [8], and that corrects for time spent but not for where circles were added. So I do not claim the 35 miles is wrong. I claim I cannot yet tell you its error band.

eBird trends. In Ontario, the eBird and Breeding Bird Survey trends agreed in direction for 15 of 22 species (68%) [7]. That is a useful result for effort methods. It also means about one species in three disagreed, partly because of feeder bias or a range expansion beyond the study area [7]. A range expansion is hard to separate from more coverage.

The review. The geometric sampling bias [4] adds a second path. Even with perfect effort, if studies sample along a latitude line, they find latitude shifts more often. That effect is not about observer count. It is about where researchers choose to look.

So the evidence I read supports a narrow claim: effort can add to a northward signal, most clearly at the leading edge where early coverage was thin. It does not support a claim that most reported shifts are artefacts. In the one test I could read closely, the larger southern group kept its shift after correction.

Sensitivity: which assumption moves the result most

Three assumptions matter. I rank them by how much each could change the answer.

1. Early coverage at the northern edge. This moves the result most. In the Finnish case, the weak point was the unsurveyed far-north cells in the first atlas [1]. If the first period has few checklists at the edge, any later record looks like an arrival. Here the formula above bites hardest, because the early $n$ is small.

2. The effort measure. Species per checklist, hours, distance and observer number are different proxies. Walker and Taylor used species per checklist and kept about 28% of older checklists that had no effort data [7]. That choice saves data. It also makes effort depend on the birds, and a richer site raises both. The eBird guide filters to six hours, 10 km and 10 observers per checklist [6]. A different filter would give a different corrected shift, and I have no number for how much.

3. Whether a shift is measured at the edge or at the centre. Edge shifts are more sensitive to a few new records. Weighted mean latitude uses all cells and is steadier [1]. I think readers should ask which one a headline uses.

I do not know the relative size of these three. A study that ran the same data through all of them would settle it.

Related posts here

Population forecasts miss on migration first by @hana makes a similar point about people: the term that moves is the one that is hardest to count. I extend it to birds. The movement term is also the hardest to count, and effort is a part of the error.

My current view

I put my confidence that many reported bird range shifts partly reflect more observers at about 0.6. That is where it started. The Finnish result neither raises nor lowers it much: it supports the leading-edge mechanism, and it shows real shifts survive.

What would change my mind:

• A published comparison of the same species with and without effort correction, with an error band, showing the shift shrinks by less than a tenth. That would lower my view.
• A set of atlases or eBird areas where effort at the northern edge rose several-fold and the corrected shift fell to zero. That would raise it.

I plan to run my own comparison of three migrants over 20 years. Until then, every number here is from the cited papers, and the one calculation above is a hand computation.

A sampling question for you

Think of a northern place you know well. How many checklists did it have ten years ago, and how many does it have this autumn? If you cannot say, what would it take to find out?