Open Field Protocol

How Long a New Feeder Takes to Be Found

The site's answer is a few days to three weeks, which is honest but useless for planning. This protocol replaces that range with a distribution: how many days to the first confirmed feeding visit, how many to regular use, and which yard conditions move the number.

The question this answers#

The published guidance on this site is that a new feeder in a yard with birds nearby gets its first visitor in roughly three to ten days, the roster fills through weeks two and three, and things settle around week four. Loosely: a few days to three weeks.

That range is a summary of experience, not a measured distribution. Nobody can tell you the median, the spread, how long the slow tail runs, or how much of the variation is explained by distance to cover rather than season. This protocol collects the raw intervals in a standard form so the range can be replaced with numbers that carry their own uncertainty.

Why the existing evidence is thin#

Three things break the usual accounts. The first is definition drift: one person counts a chickadee that hovered and left, another waits for a bird that sits and eats. Those are different events, often days apart.

The second is detection. A chickadee visit lasts under ten seconds, one seed grabbed and gone, so a watcher who glances at the window occasionally can miss days of visits and report a late date.

The third is survivorship. People who see nothing for a month change something and stop counting, so slow runs never get written down, and averaging only the runs that worked shortens every published estimate.

What you need#

• One new feeder, filled with fresh seed of a single type, and a tape measure.
• A recording setup: a phone on a small tripod shooting time lapse or motion clips, a trail camera, or a doorbell camera with the feeder in frame. Live watching works but costs more time and biases more.
• A piece of tape on the seed tube to mark the seed level on day 0.
• A clean patch of ground or a tray under the feeder so hulls are visible.
• The datasheet, filled in daily rather than reconstructed at the end.

How to set it up#

Fix time zero at the minute the filled feeder is hung. Everything is measured in days from that point.

Then grade the events in advance, so you are not deciding in the moment:

Record the site fields before day 1: distance to cover in feet and meters, feeder height, distance to the nearest window in feet and meters, feeder type, seed type, habitat class, and whether another feeder is active within 100 feet (30 m) and how long it has been there.

Record one more field honestly: whether chickadees, titmice or nuthatches have been seen in the neighborhood in the past year. The scout species question is one of the main things this dataset can test, and it needs a yes or no from before the run starts.

The procedure#

1. Hang the feeder and start the camera. Same position for the whole 28 days.
2. Cover the two peak periods. Record or watch 60 minutes from first light and 60 minutes before dusk, when most feeder traffic happens. A camera running all day is better and removes the sampling question.
3. Watch from back in the room if you watch live: at least 6 feet (1.8 m) from the glass, no movement, no backlit silhouette. Log whether a person was visible at the window in each session.
4. Score every event by grade, with clock time and species if you can identify it. First grade 0, first grade 1 and first grade 2 each get their own date field.
5. Check the indirect evidence daily: hulls below the feeder, droppings on the perch, seed level against the day 0 tape. It proves use without dating it, so it never sets the grade 2 date on its own, but it does tell you your schedule is missing visits.
6. Change nothing for 28 days. No moving, no seed swaps, no second feeder. Top up with the same seed when it runs low and log the top-up.
7. If the seed spoils, stop. Wet, clumped or sour seed is a health risk. Replace it, log the date, and restart at a new day 0 or mark the run void.
8. Close the run at day 28, whatever happened, and record whether grade 3 was reached.

What you record#

One row per run, built from a daily log. The outcomes are the three first-event dates, the derived days_to_first_feeding_visit, the regular-use date and the censored flag. The covariates are the site fields above plus season, region, observation method, and whether you scattered a little seed on the lid where it is visible from above.

Record what the first feeding bird was, even if the identification is only "small gray songbird, probably a chickadee". Say that rather than upgrading a guess.

Controls, and the ways this goes wrong#

The observer effect. The strongest artifact here is a person at the window. Motion behind glass reads as a threat, and the effect is largest in the first week, exactly when you watch hardest. Use the camera.

Intermittent watching. Ten random checks a day will find a dove and miss a chickadee entirely. Run a camera, or hold the fixed dawn and dusk sessions.

Fiddling. Most people intervene between day ten and day eighteen, which resets discovery and turns one clean run into two broken ones. Write the day 28 date on a calendar and leave the setup alone.

Unlogged confounds. A neighbor taking a feeder down mid-run, a hawk arriving, a cat starting to patrol, a heavy acorn year: all belong in the notes field on the day they happen.

Telling a real null from a dead feeder

A genuine null is a working feeder that has not been found yet. A dead feeder will never work, and the grade pattern separates them: no events of any grade means nothing has discovered it, while plenty of grade 0 and 1 events without a grade 2 means birds are there and are not eating.

• Stale seed. Seed opened last season smells rancid to a bird long before it does to you, and it is the commonest cause of an untouched feeder. Sunflower should look glossy and smell clean.
• Bad siting. More than about 20 feet (6 m) from cover, or a balcony three floors up with no tree nearby, slows discovery badly.
• Cat or hawk pressure. Approaches without landings, feather piles, or a cat regularly under the feeder. Birds return once the pressure stops, but the run is compromised.
• Mechanical failure. Ports blocked by fines or damp seed, a spinning perch, a leaking lid. Check the feeder by hand at day 14.

If you find one of these, log the cause and void the run rather than submitting it as a null. Nulls only inform when the setup was sound.

How many runs before the number means anything#

A single run gives one draw from a wide distribution, so it tells you about your yard and nothing more. A stable median needs on the order of 30 runs within a comparable stratum, and testing a covariate such as distance to cover needs several times that, because the strata multiply fast: four seasons, three habitat classes, feeder present or absent nearby.

Censored runs are why this needs volume. The runs where 28 days passed with nothing carry real information about the tail and are handled properly by survival analysis. Discarding them is the bias that produced the loose published range.

You can contribute several runs, at different times of year or different sites, as long as each uses a position with no feeder in the previous 90 days and each is logged separately.

Submitting your results#

Fill in the datasheet for this protocol, one row per run, and send it through the Feeder Log. Give your location no more precisely than state or province.

Submit runs that ended at day 28 with nothing, runs where the first bird arrived within an hour, and voided runs with the reason recorded. All three shape the distribution. What cannot be used is a date recalled weeks later, so if the daily log lapsed, mark which dates are estimates.