loading
Agriculture and food / a field record with its pair restored
Each Pair Being Duplicates
The Morrow Plots' 1927 table prints two untreated quarters beside the west MLrP and east MLbP treatment labels. Restore the pair the 2024 release averages away, then test the treated position difference without turning an unrandomized field into a causal experiment.
The published field is not a randomized trial. Its north half is untreated, its south half is treated, and each half is split east and west. The question here is narrower and more physical: what changes when the two untreated quarters are allowed to be two recorded places again?
In the 1904 to 1926 print, the untreated east led west by +5.01 bu/A while the designated treated east led west by +4.47 bu/A; subtracting that same position gap leaves -0.54 bu/A across 60 grain plot-years.
01 / The operable base
Put your hand on one year
Pick a printed year and plot. The four cards are the four entries in Bulletin 300 Table 8. The first two are untreated duplicates. The lower pair are the historically designated treated sides, labelled MLrP west and MLbP east in the table.
Loading the local record
The page reads only frozen local files.untreated east minus west...
designated treated east minus west...
position-adjusted gap...
The frozen record is loading.
19041926
The slider stays inside the quarter-level Table 8 window.
east gap = NE - NW
treated gap = SE - SW
corrected gap = treated gap - east gap
treated gap = SE - SW
corrected gap = treated gap - east gap
The 2024 release stores an averaged rate for the older quarters. The toggle repeats that rate so the lost duplication can be seen, not silently reconstructed. In that view there is no independent east-west untreated witness, so the position-adjusted result is unavailable.
02 / The dismissal, answered by a further result
An average is not the same as a pair
Averaging is useful for reporting a rate. It is also capable of erasing the very position difference that tells you whether the treated comparison has a fair geometric witness. Every selector below is a live parameter of the engine. Change one, then watch the record move.
Recomputing from the frozen rows.
The analysis will name its current rule.
...
The anchor is loading.
The anchor verdict is loading.
By plot
By crop
Every included grain row, after the correction
positive adjusted differencenegative adjusted difference
raw designated treated lead = mean(SE - SW)
position-adjusted lead = mean((SE - SW) - (NE - NW))
The MLrP and MLbP labels identify the historical treated sides, not one material over every year. Aref and Wander report rock phosphate and bonemeal through 1919, triple superphosphate from 1920, and Bulletin 300 says further phosphate application was discontinued in 1925. The later release comparison below is the NPK and manure layout. These are treated-side record comparisons, not a causal estimate.
position-adjusted lead = mean((SE - SW) - (NE - NW))
The MLrP and MLbP labels identify the historical treated sides, not one material over every year. Aref and Wander report rock phosphate and bonemeal through 1919, triple superphosphate from 1920, and Bulletin 300 says further phosphate application was discontinued in 1925. The later release comparison below is the NPK and manure layout. These are treated-side record comparisons, not a causal estimate.
03 / Further record, same instrument family
The later record keeps the duplicate visible
From 1955 through 1967 the DataBank release no longer has the older four-quarter shape. It has a west untreated baseline at NA, an east untreated pair at NC and ND, a west treated baseline at SA, and an east treated pair at SC and SD. This is the release's later NPK and manure layout, not the old rock phosphate and bone meal pair. The comparison below uses the available values. A pair recorded once is one observation, not two invented copies.
loading
1955 to 1967, grain rows only
The release pair counts will appear after the local CSV loads.
east gap = mean(available NC, ND) - NA
corrected treated lead = (mean(available SC, SD) - SA) - east gap
corrected treated lead = (mean(available SC, SD) - SA) - east gap
04 / The source does not speak with one mouth
When the witnesses disagree
The 2024 file is a valuable machine-readable release, but it is not a transparent replacement for the printed record. These 11 retained witness rows include the older publication conflicts and every grain disagreement found when the full Table 2 transcription is checked against Table 8. The page shows them beside the computation so a mismatch cannot hide inside a clean average.
The 2024 descriptor says: "Separate yield rates for the untreated subplots have been reported piecemeal in annual Agricultural Experiment Station bulletins..." It says those figures may be incorporated into a future dataset version. This page makes that older record operable, while keeping the descriptor's Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 (CC BY-NC-ND 4.0) boundary intact.
Prior art sets the boundary. Aref and Wander (1997) already reported phase-level comparisons of the untreated NE and NW plots and of the SW and SE phosphorus sources. They also describe the lack of randomization and replication. This page is a year-by-year reconstruction and extension of the visible record, not a first comparison.
We searched the full text of Aref and Wander (1997), Bulletin 300, and the Illinois DataBank release on 2026-09-14 and did not find a year-by-year reconstruction that restores every printed untreated pair beside both treated quarters and carries the comparison into the release's separately recorded subplots.
Read the paper at Aref and Wander's 1997 article. Its copyright remains with its publisher and authors. No paper bytes are redistributed here.
05 / A true instrument has a refusal position
Ask what the record cannot answer
Choose a tempting question. The engine refuses years before the quarter split, years for which this page has no quarter-level print, the post-1967 layout, and hay rows whose witnesses disagree. Before 1904 the printed record gives one annual yield per original plot, not north and south averages. A refusal is part of the result.
Choose a question, then ask. The response will be a refusal or an available record row.
06 / The check
The numbers have to survive being wrong
The browser check is loading the frozen record before it can recompute the displayed figures.
Transcription instrument. Table 8 was read once from the page images and once from the Internet Archive word boxes. The image settled every digit retained here. The print order NW, SW, NE, SE is kept in the files, with named keys added only so the engine cannot confuse layout order with compass order.
The Table 2 completeness audit will appear after the local record loads.
Bound choices. Bulletin 125 says treatment reached plot 5S for the 1904 crop and plots 3S and 4S for the 1905 crop. The default therefore excludes 1904 plots 3 and 4. The page also exposes the release coding, the Table 2 witness, the half-tenth rule, the crop pool, the later annual extension and the position correction.
Scope. Nothing here says bone meal caused a gain. The field is unrandomized, and phosphorus is entangled with the other south-half treatment. The result is a record comparison with its spatial correction visible.
The page will name the local files after they load. No upstream service is needed at runtime.
The build verifier repeats the anchor, the rendered-figure scan, the live-choice checks, the planted-fault test, the refusal assertions and the provenance hashes. The site publishes that verifier through its own check apparatus.