About This Site
The ProBirdFeeder Research Program
The guides on this site are built from other people's research. This section is where we produce our own: a structured dataset we compiled and published, analyses run on it, an open observation network, and test protocols anyone can run and submit results to.
The short answer
- Most bird feeding content, including much of ours, restates the same institutional sources. That produces accuracy without producing anything new.
- This section adds four things that did not exist before: a machine-readable trait matrix for all 78 profiled species, original analyses run on it, an open field-observation network, and eight replicable test protocols.
- Everything is reproducible. The dataset ships with the script that builds it from our published pages, so you can regenerate it and check us rather than take our word.
- Protocol pages describe tests to be run, not results already in hand. Where we have no data yet, the page says so rather than filling the gap with a confident number.
Why this section exists#
Our research standards page explains how a guide here gets sourced: a four tier hierarchy, peer-reviewed ornithology at the top, retailer blogs and AI output excluded entirely. That method is good at one thing, keeping the site accurate, and incapable of another, producing knowledge that was not already published somewhere else.
That gap is real and it is worth naming. If every page on a subject is a careful synthesis of the same handful of institutional sources, then the whole field is downstream of a small number of studies, and nobody notices the questions those studies never asked. Backyard feeding has a lot of those questions, because the interesting variables are local: your squirrels, your window, your weather, your seed.
The four arms#
Original findings
Results we derived ourselves from our own dataset, including a body-mass ceiling on tube feeders and the smallest set of feeders that reaches almost every feeder bird.
Open dataOpen datasets
A normalized trait matrix covering all 78 profiled species: mass, size, residency, food acceptance and feeder compatibility, as CSV and JSON under CC BY 4.0.
LiveThe Feeder Log
An open field-observation network. Readers log what fed at which feeder on which food, and the aggregate results are recomputed from those rows on every page load.
ProtocolsField protocols
Eight replicable test procedures with materials, controls, scoring scales, sample size guidance and datasheets, covering seed preference through window markers.
What the data already shows#
The trait matrix was built to be a reference table. Running the obvious analyses on it produced three results that cut against standard feeding advice, and all three are reproducible from the downloadable data.
Tube feeders have a body-mass ceiling near 44 g (about 1.6 oz). The heaviest species in our matrix recorded as a tube user is the Rose-breasted Grosbeak. 25 of the 73 feeder-using species sit above that line, and none of them are tube birds. Read the analysis.
Three feeders reach 93% of feeder birds, and a tube is not one of them. A platform, a suet cage and a nectar feeder together cover 68 of 73 species. Adding a tube feeder as the fourth brings in 4 more. Read the analysis.
Two foods cover 89% of feeder birds. Sunflower in a hulled form reaches 43 species on its own, and adding insect food (mealworms and suet's insect-eating audience) takes the pair to 65. Everything after that is single-digit gains. Read the analysis.
What we have not done#
Being straight about the limits is part of the point, so: the trait matrix is a compilation, not a census. It encodes what our own species profiles say, which in turn comes from the institutional sources in our source hierarchy. It is original as a structured, queryable, downloadable object, and as the substrate for the analyses above. It is not original observation.
The original observation is the Feeder Log, and it starts empty. Its results pages show sample sizes on every table and withhold any cell built on fewer than 5 observations. A research section that published thin numbers as findings would be worse than no research section at all.
The protocols are exactly that, protocols. They are detailed enough to run and they carry no results, because we have not run them at scale yet. When runs come in, the results will be posted with their n attached and their raw datasheets downloadable.
How to contribute#
Three ways, in ascending order of effort. Log a single observation in the Feeder Log, which takes about a minute. Run one of the field protocols and submit the datasheet. Or take the open data, find something in it we missed, and tell us through the contact page. Corrections to the dataset go through the same corrections policy as everything else here.