Data and Original Analysis
Four reference tables and the methodology behind them, published openly so the reasoning can be checked rather than trusted.
The short answer
- The core dataset covers 34 feeder species, 13 foods and 13 feeder designs, and is published as JSON under a Creative Commons license.
- Cost analysis is done per pound eaten and per calorie delivered, not per pound bought, because hull weight is between 14% and 40% of a seed sold in the shell.
- Every derived number on the site is reproducible from the published dataset using the formulas on the methodology page.
- Where our confidence is low we say so rather than rounding it away.
Most bird feeding advice is qualitative, and for most questions that is fine. "Cardinals like sunflower" is true and useful. It stops being useful the moment the question is comparative: which of these two bags is better value, which food covers the most of my species list, which feeder design is worth the extra money.
Comparative questions need a table. So we built four, plus the methodology that makes them checkable.
What is in the dataset#
One JSON file at /datasets/bird-food-matrix.json holds everything:
- 34 species with scientific name, rough range, mass, ground-feeding tendency, flocking tendency, preferred feeder designs, and a 0 to 4 score against each of the 13 foods.
- 13 foods with kernel yield, energy density, fat and protein percentage, mess rating and a benchmark price.
- 13 feeder designs with the foods each handles, squirrel resistance, weather protection and typical capacity.
It is versioned, it is licensed CC BY 4.0, and it is the same file the bird food matcher and seed cost calculator read at runtime. What the tools use is what you can download.
Reading the grid: rows, columns and cells#
Three query patterns cover nearly everything anyone wants from the matrix.
Read a row when you start with a bird. Take the Carolina Wren. Its row scores suet 4, mealworms 4, sunflower hearts 3 and shelled peanuts 3, and nothing else above 2: in-shell sunflower 2, safflower 1, millet 1, cracked corn 0, nyjer 0. Its feeder list is suet cage, mealworm dish, platform and peanut mesh. Read together, those two fields say in one line what a species account takes three paragraphs to say. This is not a seed bird, and a wren will ignore your best seed tube in favor of a suet cage six feet away. Rows are what the species pages print.
Read a column when you start with a bag of seed. The millet column has six species at 4: Mourning Dove, Dark-eyed Junco, White-throated Sparrow, Song Sparrow, House Sparrow and Red-winged Blackbird. Twenty-two species rate it 1 or 0. That column is the entire argument about white proso millet at a glance. It is an excellent ground station food for native sparrows and doves, and it is simultaneously the most efficient way to feed house sparrows and blackbirds that exists. Whether that trade is worth making depends on your yard, not on the seed.
The safflower column makes the opposite case. Exactly one species rates it 4 (the Northern Cardinal) and seven rate it 3, which is modest reach, but it is rated 1 or 0 by House Sparrows, European Starlings, Common Grackles and Red-winged Blackbirds. Reach is not what that seed is for. Exclusion is, which is the point of safflower versus sunflower.
Cross the row against the feeder field, always. This is the query people skip and the one that explains most disappointment at a new feeder. Thirteen species in the matrix rate black oil sunflower at 3 or better and do not list a tube among the feeders they use, including the Northern Cardinal, the Blue Jay, the Mourning Dove and the Dark-eyed Junco. A food a bird loves, in a feeder it cannot stand on, produces nothing. The bird food matcher does this intersection for you, and the feeder comparison does it design by design.
One more pattern is worth knowing: read two columns against each other. Black oil sunflower is rated 3 or better by 25 species and sunflower hearts by 28. It is the same seed with and without a hull, and the three species that gain are the Song Sparrow, the Carolina Wren and the European Starling, none of which is built for cracking a shell. That difference is what the hearts versus in-shell comparison is arguing about.
The three findings worth knowing#
Hull is a quarter of a sunflower bag. Correcting for kernel yield moves black oil sunflower from $1.35 per pound to about $1.80 per pound eaten, and striped sunflower from $1.60 to about $2.67. The full working is in cost per calorie.
One food does most of the work. Black oil sunflower scores 3 or 4 for more species in the matrix than anything else, which is the quantitative version of advice everyone already gives. What the grid adds is the exceptions: the species where it scores 0 or 1, and what to do about them. See the preference chart.
Feeder categories differ more than products within a category. The gap between a platform feeder and a caged feeder is larger than the gap between a cheap and an expensive example of either. That is why our feeder comparison scores designs rather than models.
On trusting tables#
A table looks more certain than the evidence behind it. Our preference scores are a condensation of published sources onto a five-point scale, not a measurement, and the middle of that scale carries real uncertainty. Kernel yields are midpoints for lots that vary. Prices are benchmarks that will drift.
None of that makes the analysis useless: the conclusions are robust to the uncertainties, and we show the sensitivity where it matters. But it does mean the methodology is part of the data, not an appendix to it.
What is missing, and what we would add next#
The honest way to publish a dataset is to name what is not in it.
Volume. A 4 records that a bird chooses a food, not how much of it the bird eats. A chickadee at 11 g and a Mourning Dove at 120 g both rate black oil sunflower at 3 or better and consume wildly different amounts of it. Nothing in the file lets you compute demand, which is why every consumption figure on this site is a working estimate rather than something derived from it.
Season. Each species and food gets one number, and real preference moves through the year. The bird that lives on suet in January is carrying insects to a nest in June. See bird food nutrition for how far apart those two months are.
Region at any useful resolution. Range is a short text string such as "East, Midwest, South". It is not a map and not an abundance estimate, so it cannot tell you whether a species occurs in your county or how often you should expect it.
Competition. Scores describe a bird in front of a food, not a bird in a queue. Displacement by grackles, starlings and house sparrows changes what actually gets eaten, and none of it is modeled. A species that would rate a food highly alone may never reach it at a crowded station.
Species coverage. Thirty-four is a working set, weighted east. The FAQ below lists the main gaps.
Feeder build. The feeder records describe designs, not objects. Disassembly, port material, drainage, durability and price are all absent, which is why those have to be argued in prose. There is a related artifact worth knowing about: the caged and weight-activated designs score poorly on any species count computed from the file, because species records name the underlying design a bird uses rather than the squirrel-proofing built around it. Read those two as inheriting the reach of the feeder they enclose.
Prices. One benchmark per food, United States, large bag, dated. No regional variation, no bulk tiers, no small-bag penalty.
In priority order, what we would add next: a consumption dimension in grams per day per species, a seasonal modifier on each food, twenty more species weighted to the West and Southwest, and a build-quality axis for feeders. The first would change more conclusions than the other three put together.
Using and citing the data#
The file is published under a Creative Commons Attribution 4.0 license, which means you may use it in teaching, in software, in your own analysis and commercially, as long as attribution travels with it. Three requests, none of them license conditions:
- Cite the version, not just the site. The dataset carries a version string, currently 2026.2, and scores are revised when corrections are accepted. A citation without a version cannot be checked later.
- Carry the labels with the numbers. Preference and feeder scores are judgements condensed onto a five-point scale, kernel yields are working midpoints, and prices are benchmarks. If you republish them as measurements, you have made them claim more than we do.
- Send corrections rather than forking silently. A fix that reaches us reaches everyone using the file. The corrections process is public and dated, and contact goes to a person.
If you are building on top of the file, the derived formulas on the methodology page reproduce every published figure on this site exactly, so you can verify your implementation against our tables before you trust it against your own.
Frequently asked questions#
Can I reuse these tables?
Yes. The underlying dataset is published at /datasets/bird-food-matrix.json under a Creative Commons Attribution 4.0 license. Use it in your own work, teaching or software, and credit Bird Feeder Web with a link. If you spot an error, tell us so the fix reaches everyone using the file.
Why build your own dataset instead of citing existing ones?
Because the question people actually ask, which food should go in my feeder given the birds I have, cuts across the way existing sources are organized. Species accounts describe one bird at a time; food guides describe one seed at a time. Neither can be read as a grid. Building the grid required condensing both onto one scale, which is a judgement we document rather than hide.
Which birds are missing from the dataset?
Plenty, and the gaps lean western and southern. The 34 species are a working set of common feeder birds weighted toward the East and Midwest. Carolina, Mountain and Chestnut-backed Chickadees are absent; so are most western hummingbirds beyond Anna's, several western and far-northern sparrows, crossbills, quail and most of the desert Southwest list. Expanding coverage westward is on our list, behind adding a consumption dimension to the species we already have.
How do I look up the best food for one particular bird?
Read that species' row. Every species page prints its own row as a ranked food table, so you can see not just the top food but where each one sits and what the bird ignores. Then check the row's feeder list before you buy anything: preference and access are separate fields, and thirteen species in the matrix rate black oil sunflower highly while not using a tube feeder at all.
Sources and further reading
- Project FeederWatch Cornell Lab of Ornithology Species accounts and feeder-type guidance underlying the matrix
- FoodData Central US Department of Agriculture Composition values for seeds and kernels