Summary

• Loss et al. (2014) “estimate that between 365 and 988 million birds (median = 599 million) are killed annually by building collisions in the U.S. [United States]”. Bird-safe glass is “specially designed to make glass a visible obstacle to birds”, and therefore decrease the number of birds killed by building collisions. In this post, I illustrate that (replacing standard with) bird-safe glass may change the welfare of arthropods much more than that of birds.
• I calculate wild insectivorous birds eat 23.4 k arthropods per bird-year, which implies the birds live for 22.4 bird-minutes for each arthropod they eat.
• I conclude bird-safe glass can easily increase or decrease welfare. I believe it may impact arthropods way more than birds, and I have very little idea about whether it increases or decreases the welfare of arthropods.
• I recommend research on i) the welfare of soil animals and microorganisms, and ii) comparisons of (expected hedonistic) welfare across species. I think progress on ii) is difficult, but necessary to find interventions which robustly increase welfare. I also see lots of room for progress on ii) to change funding decisions even neglecting soil animals and microorganisms. For welfare range proportional to “individual number of neurons”^“exponent”, and “exponent” from 0 to 2, which covers the best guesses that I consider reasonable, the welfare range of shrimps is 10^-12 to 1 times that of humans.
• I am sceptical that targeting non-soil animals is a great way to build capacity to increase the welfare of soil animals later. I believe the most cost-effective ways of building capacity to help any given group of animals will generally be optimised with such animals in mind. I would also expect much more investigation of the extent to which interventions targeting non-soil animals are building capacity to increase the welfare of soil animals if this was key to whether they are increasing or decreasing animal welfare.

Context

Loss et al. (2014) “estimate that between 365 and 988 million birds (median = 599 million) are killed annually by building collisions in the U.S. [United States]”. Bird-safe glass is “specially designed to make glass a visible obstacle to birds”, and therefore decrease the number of birds killed by building collisions. Mal Graham has argued there is significant uncertainty about whether bird-safe glass increases or decreases the welfare of birds and other animals. In this post, I illustrate that bird-safe glass may change the welfare of arthropods much more than that of birds.

Arthropods eaten by birds

Here are my calculations.

I assume the target birds have a mean mass of 26.4 g. I get this from the mean between the means for white-throated sparrows, dark-eyed juncos, ovenbirds, and song sparrows, which are the most commonly killed according to Loss et al. (2014). “Of 92,869 records used for analysis, the species most commonly reported as building kills (collectively representing 35% of all records) were White-throated Sparrow (Zonotrichia albicollis), Dark-eyed Junco (Junco hyemalis), Ovenbird (Seiurus aurocapilla), and Song Sparrow (Melospiza melodia)”. I set the mean mass for each species to the mean between the lower and upper bounds from All About Birds.

I suppose wild insectivorous birds eat 421 Mt of arthropods per year. I get this from a weighted mean of the lower and upper bounds from Nyffeler et al. (2018). I put a weight of 2/3 on 396 Mt, and 1/3 on 472 Mt. They say the “true value” is “most likely at the lower end of this range”. Nyffeler et al. (2018) says “the standing biomass of the global community of insectivorous birds might be on the order of ≈ 3 million tons”. So I infer wild insectivorous birds eat 140 kg of arthropods per year per kg of birds. Combining this with my mean mass for the target birds, I determine that wild insectivorous birds eat 3.70 kg of arthropods per bird-year.

I estimate the target birds eat 6.33 k arthropods per kg of arthropods. I get this for an individual mass of arthropods of 0.158 g, which is the mean of a lognormal distribution with 8th and 99.99th percentile of 0.481 mg and 1.65 g. Gemini 3.1 Pro on 31 March 2026 estimated 8 % of the arthropods eaten by the 4 bird species I mentioned above are lighter than 0.481 mg, and 0.01 % are heavier than 1.65 g. Dark-eyed juncos eat aphids, and these have a mean mass of “0.4809 mg” according to Maia et al. (2020). Song sparrows eat grasshoppers, and I found a fact sheet mentioning a mean mass of female three-banded grasshoppers of “1,654 mg”.

I calculate from the above that wild insectivorous birds eat 23.4 k arthropods per bird-year, which implies the birds live for 22.4 bird-minutes for each arthropod they eat.

Discussion

Bird-safe glass may impact arthropods much more than birds

My intuition based on the numbers above alone is that bird-safe glass may change the welfare of arthropods much more or less than that of birds. I explore this further below, but ultimately arrive at the same conclusion. For simplicity, I ignore changes in the population of arthropods, and in the welfare of birds due to these dying in some other way due to avoiding collisions with buildings.

I guess I am roughly indifferent between 1 s of excruciating pain, like “severe burning in large areas of the body, dismemberment, or extreme torture”, and losing 24 h of fully healthy life. I speculate birds cause 1 s of excruciating pain to each arthropod they eat. So I estimate the decrease in the welfare of each arthropod is equal to that from them losing 24 h of fully healthy life (= 24*60/22.4). This is a loss of 64.3 fully-healthy-arthropod-years per bird-year (= 24*60/22.4).

One can clarify the comparison above using the tentative (expected) welfare ranges in Bob Fischer’s book about comparing welfare across species. That of chickens is 5.48 (= 0.40/0.073) times that of black soldier flies (BSFs). Welfare range is defined there as the difference between the maximum and minimum welfare per unit time among “realistic biological possibilities”. I do not expect the welfare range of the target birds to dramatically differ from 5.48 times that of random arthropods they eat under the methodology of Bob’s book. I would be surprised if the welfare range of the target birds was meaningfully different from that of chickens, or if that of random arthropods they eat was dramatically different from that of BSFs. In addition, I think it is reasonable to assume that individual welfare per fully-healthy-animal-year is proportional to the welfare range. So I suppose 1 fully-healthy-bird-year has 5.48 times as much welfare as 1 fully-healthy-arthropod-year. In this case, the decrease in the welfare of the arthropods eaten by the birds corresponds to a loss of 11.7 fully-healthy-bird-years per bird-year (= 64.3/5.48).

I speculate the lives of the target birds are 50 % as good as fully healthy lives. So I conclude from the above that the effects on arthropods are 23.4 (= 11.7/0.5) times as large as those on the birds.

My practical conclusion is that bird-safe glass may change the welfare of arthropods much more or less than that of birds. The final comparison above suggests the effects on arthropods are much larger than those on birds. Nonetheless, there is large uncertainty in the change in the living time of birds and arthropods, change in the welfare of birds due to these dying in some other way due to avoiding collisions with buildings, change in the welfare of arthropods due to these being eaten by birds instead of dying in some other way, and welfare comparisons across species.

Bird-safe glass can easily increase or decrease welfare

I conclude bird-safe glass can easily increase or decrease welfare. I believe it may impact arthropods way more than birds, and I have very little idea about whether it increases or decreases the welfare of arthropods. I do not know which species of arthropods are the most important to determine the change in the welfare of arthropods given the large uncertainty in welfare comparisons across species. I can see the most important arthropods being the smallest, largest, ones with intermediate mass, or any combination of these. To make matters worse, I have almost no clue about whether any species of wild arthropods has positive or negative lives in a given biome. I am also very uncertain about which arthropods become more or less abundant as a result of changes in the population of birds.

What now?

I recommend research on i) the welfare of soil animals and microorganisms, and ii) comparisons of welfare across species. I think progress on ii) is difficult, but necessary to find interventions which robustly increase welfare. For instance, ones that focus on the greatest sources of suffering across all species. I also see lots of room for progress on ii) to change funding decisions even neglecting soil animals and microorganisms. In Bob’s book, the tentative welfare range of shrimps is 8.0 % of that of humans. However, for welfare range proportional to “individual number of neurons”^“exponent”, and “exponent” from 0 to 2, which covers the best guesses that I consider reasonable, the welfare range of shrimps is 10^-12 (= (10^-6)^2) to 1 times that of humans, as shrimps have 10^-6 times as many neurons as humans.

I would prioritise the above research over the “ecologically inert” interventions which have been proposed so far. I suspect interventions decreasing the pre-slaughter pain of farmed invertebrates are the closest to robustly increasing welfare (in expectation). However, I still do not know whether electrically stunning farmed shrimps increases or decreases welfare due to potentially dominant effects on soil animals and microorganisms. Furthermore, I would say such interventions may increase welfare only negligibly due to their target invertebrates having a super narrow welfare range, as it would be the case if shrimps had a welfare range equal to 10^-12 times that of humans.

Targeting non-soil animals is a great way to build capacity to increase the welfare of soil animals later?

I am sceptical. I believe the most cost-effective ways of building capacity to help any given group of animals will generally be optimised with such animals in mind. I would also expect much more investigation of the extent to which interventions targeting non-soil animals are building capacity to increase the welfare of soil animals if this was key to whether they are increasing or decreasing animal welfare.