The 73% fall that is 15% once the cohorts are the same age

Dark navy briefing slide, masthead “Vehicle safety desk · fleet defect monitoring” on the left and “NHTSA Part 573 recall campaigns · by model year · file to 9 Sept 2026” on the right, headlined “Recalls on new model years have fallen by nearly three quarters.” Beneath the headline sits a bordered gold chip reading “Demonstration · drawn by us from published figures”. A standfirst says every figure is a straight count out of NHTSA’s own recalls flat file — the manufacturers’ Part 573 defect and noncompliance reports, one source, one field, one definition; that the axis is linear, zero-based, evenly ticked and unbroken; that every vehicle campaign in the file is counted and nothing is sampled, smoothed, indexed or rescaled, with every value printed above its own column; that NHTSA publishes the file and the bars are ours; and that the series peaks at 2,163 campaigns on model year 2019 and stands at 582 on model year 2026. The main panel, headed “Vehicle recall campaigns affecting each model year” and subtitled “Distinct Part 573 campaigns naming that model year · vehicles only · model years 2012 to 2026 · zero-based linear axis, evenly ticked, unbroken · every campaign in the file, nothing smoothed, dropped or rescaled”, is a column chart of fifteen gold columns on an axis running 0 to 2,500 in even steps of 500: 2012 at 1,809, 2013 at 1,913, 2014 at 1,969, 2015 at 2,102, 2016 at 2,085, 2017 at 2,039, 2018 at 2,059, 2019 at 2,163, 2020 at 2,081, 2021 at 2,121, 2022 at 2,012, 2023 at 1,715, 2024 at 1,421, 2025 at 1,090 and 2026 at 582. The shape is a long plateau across the middle of the chart and then a staircase down over the last four columns, the final one just over a quarter the height of the peak. A caption reads “Campaigns as filed · every vehicle campaign in the file · one scale throughout”. Five cards run down the right. “The fall, in one figure” prints 73% in large type over the words “fewer campaigns” and says model year 2019 is the peak of the series at 2,163 while model year 2026 stands at 582, that model year 2025 at 1,090 is already 49.6% below the peak, and that the three newest model years are the three smallest bars on the chart. “Read-out for the desk” says the series turns over after model year 2019 and falls away steeply, seven consecutive model years below the peak, ending 1,715, 1,421, 1,090 and 582 — and recommends treating the defect rate on new vehicles as falling, holding the campaign-management headcount flat into next year rather than growing it with the fleet, and moving the recall-readiness budget to the model years that still carry the volume, the 2015 to 2022 cohorts. “What one bar counts” explains that a bar is distinct NHTSA recall campaigns naming that model year, from manufacturers’ Part 573 defect and noncompliance reports, vehicle campaigns only so tires, equipment and child seats are out; that a campaign usually covers several model years, 2.76 on average over the campaigns that name one, and is counted once in each one it names, so a bar is campaigns rather than vehicles and the bars do not sum to the number of campaigns filed; and that a manufacturer has five working days to file once it decides a defect is safety-related, the date used here being the date NHTSA received the report. “Every value on the chart” tabulates all fifteen model years in two monospaced blocks with their campaign counts and the number of days each model year has been on file, from MY2012 at 1,809 campaigns and 5,365 days to MY2026 at 582 campaigns and 251 days. “Basis of measurement” credits the National Highway Traffic Safety Administration, Office of Defects Investigation, recalls flat file (FLAT_RCL_POST_2010), downloaded 12 September 2026, latest Part 573 report received date in it 9 September 2026, 13,608 vehicle campaigns; notes that model years before 2012 are left off because the file carries reports received from 1 January 2010, that model year 2027 has barely begun, and that the 117 campaigns with no model year are excluded; and says NHTSA publishes the file while the counts and the bars are ours. A method line runs along the foot: published records, as filed; one source, one field, one definition; zero-based linear axis, evenly ticked, unbroken; every vehicle campaign in the file; nothing smoothed, dropped or rescaled. A footnote credits the source and states that this slide is a demonstration drawn by us in the manner of an internal fleet safety briefing — the desk, the read-out and the recommendation are invented, and every figure behind them is real. — Showing the misleading chart

The same fifteen model years on a cream background, kicked “The same fifteen model years · the same campaigns · one clock” and “Recall campaigns per model year, age-matched · file to 9 Sept 2026”, headlined “The 73% fall that is 15% once the cohorts are the same age.” A standfirst says nothing on the briefing slide is bent — its axis starts at zero, every campaign in the file is counted, and all fifteen numbers are right — and that what sets the picture is that a recall arrives years after the vehicle does, so each bar counts events gathered over a different length of time and the length is set by the date of the last report in the file: model year 2019 has been collecting recalls for 2,808 days and model year 2026 for 251, so watching both for 251 days turns the fall from 2,163 to 582 into a fall from 687 to 582, which is 15.3% rather than 73.1%, and at the age model year 2026 has now reached, model year 2019 had only 31.8% of the recalls it has today. The main panel, headed “How a model year’s recalls actually arrive” and subtitled “Cumulative campaigns against the cohort’s age · one line per model year · each line stops where the file does · the first chart read all fifteen of these lines at whatever height they had reached on 9 September 2026”, is a line chart with a zero-based vertical axis running 0 to 2,000-plus in steps of 500 and a horizontal axis of years since 1 January of the model year, running 0 to 15. Fifteen curves climb steeply in their first two years and then flatten without ever going flat; a key colours model year 2019 blue, model year 2026 orange-red and the other thirteen tan, and each line ends in a dot, ten of the fifteen carrying a model-year label. The short orange-red line for model year 2026 stops almost immediately, at 582; the blue line for 2019 runs to 2,163 at age seven and a half; the tan lines for 2012 and 2014 run out past age fourteen and thirteen, still rising, at 1,809 and 1,969. A dashed orange-red rule stands at 251 days, labelled “251 days — the age model year 2026 has reached”, and every curve on the chart is still in its first steep climb where that rule crosses it. Below, two smaller panels sit side by side. “The same fifteen, all at 251 days old”, subtitled “Campaigns filed by the cohort’s 251st day · zero-based, evenly ticked”, is a column chart on an axis running 0 to 800 in steps of 200: 368, 388, 461, 606, 667, 623, 608, then model year 2019 highlighted in blue at 687, then 583, 668, 680, 647, 628, 603, and model year 2026 highlighted in orange-red at 582 — a rise across the early years and then a plateau, with the last column barely below the middle of the range. “Or a fixed two-year window”, subtitled “Campaigns filed by 31 December of the year after the model year · complete for MY2012–MY2024 · the last two windows are still open, so those two are drawn hollow and read ‘so far’”, is a column chart on an axis running 0 to 1,600 in steps of 400: thirteen solid columns for the closed windows — 727, 861, 976, 1,169, 1,183, 1,130, 1,202, then 2019 highlighted in blue at 1,200, then 1,228, 1,295, 1,340, 1,270 and 1,221 — a series that rises by two thirds across the span the first chart drew as a collapse, followed by two hollow dashed outlines for model years 2025 and 2026, at 1,090 and 582 so far, whose windows do not close until 31 December 2026 and 31 December 2027. Eight cards run down the right in two columns. “The same pair, three ways” tabulates model year 2019 against model year 2026: as the first chart drew it, 2,163 against 582, a change of −73.1%; both at 251 days old, 687 against 582, −15.3%; both at one year old, 848 against no figure at all — and notes that the middle row watches both cohorts for the same length of time while the bottom row is the honest chart running out of data, since model year 2026 is 251 days old and cannot yet be asked what it looked like at one year. “What the clock is” says age runs from 1 January of the model year to the date NHTSA received the Part 573 report, that vehicles go on sale before that date and recalls arrive before it too — 139 to 315 campaigns per cohort — so the count at age zero is not zero and every cohort is treated alike, and that at the snapshot model year 2026 is 251 days old and model year 2012 is 5,365. “Nearly half arrive after year two” says model year 2015 had 1,169 campaigns by the end of 2016 and has 2,102 today, so 55.6% had arrived when the cohort was two calendar years old, 74.9% by four and 91.1% by seven, and that it is still not finished: in the twelve months to 9 September 2026 the 2015 cohort gained 31 campaigns and the 2012 cohort 12. “Every cohort is still gaining” tabulates new campaigns in the last twelve months — MY2012 +12, MY2019 +94, MY2024 +333, MY2025 +487, MY2026 +501 — and says that redrawing the first chart next September will shift its cliff a year to the right, because a trend does not know where the file ends. “The recalls did not stop being filed” says that counting the same campaigns by the year they were filed rather than by the model year they name gives a flow that is flat to rising — 583 vehicle campaigns filed in 2012, 880 in 2019, a peak of 988 in 2021, 950 in 2024 and 891 in 2025, with 562 in 2026 to 9 September — so nothing has gone quiet and the campaigns are landing on model years that have been on the road for years, which is where they always landed. “What to draw instead” says to pick an observation window every cohort has completed, say so in the caption the way you print a unit, and stop the chart where the window stops, since the honest chart is shorter and the missing bars are the finding rather than a gap; to draw the development curves beside it; to give an immature cohort its own style and label it incomplete, or project it from the mature curves and call the projection a projection; and never to take a percentage change between two cohorts of different age, because the difference in ages is doing the arithmetic. “Not an unfinished period” distinguishes the two by what the x-axis is: an unfinished period buckets events by when they happened, and only the last bin is short, whereas this one buckets them by which cohort they belong to, so every bin is a whole model year, what differs between bins is how long each has been watched afterwards, and the censoring reaches back a decade rather than one bin; nor is it survivorship, since no cohort is missing and each is showing a number still on its way up. “Basis of measurement” credits the NHTSA Office of Defects Investigation recalls flat file (FLAT_RCL_POST_2010), downloaded 12 September 2026, latest report received 9 September 2026, vehicle campaigns only, model years 2012 to 2026, each campaign counted once in each model year it names, 117 campaigns with no model year excluded, with the age clock, the age-matched counts, the two-year window and the curves ours. Five notes run along the foot: the same campaigns, the same file; one clock, stated; every cohort watched the same length of time; zero-based linear axes; the open windows drawn hollow and labelled. A closing footnote credits the source, states that both drawings are ours and that the desk, the read-out and the recommendation on the first are invented while every figure behind them is real, and adds that on a two-year window, across the thirteen model years whose window has closed, the same file runs the other way — 727 campaigns on MY2012 rising to 1,340 on MY2022 and 1,221 on MY2024, up 68.0% across the span the first chart drew as a collapse. — Showing the misleading chart

Be the first to star this exhibit

A fleet safety briefing we drew counts vehicle recall campaigns by the model year they name, straight out of NHTSA’s own recalls flat file: model years 2012 to 2026, every vehicle campaign in the file, a linear axis that is zero-based, evenly ticked and unbroken, and every value printed above its own column. The series peaks at 2,163 campaigns on model year 2019 and stands at 582 on model year 2026 — a fall of 73.1%, which the desk reads as a defect rate in retreat. Every number is right. But a recall arrives years after the vehicle does, so each column counts events gathered over a different length of time: model year 2019 has been collecting them for 2,808 days and model year 2026 for 251. Watch both cohorts for the same 251 days and the fall from 2,163 to 582 becomes a fall from 687 to 582 — 15.3%. On a fixed two-year window, across the thirteen model years whose window has closed, the same file runs the other way: 727 campaigns on model year 2012 rising to 1,221 on model year 2024.

National Highway Traffic Safety Administration, Office of Defects Investigation, recalls flat file (FLAT_RCL_POST_2010); latest Part 573 report received date in it, 9 September 2026. Vehicle campaigns only (13,608 of them), model years 2012–2026, each campaign counted once in each model year it names; the 117 vehicle campaigns with no model year are excluded. The age clock, the age-matched counts, the two-year window and the development curves are ours. Retrieved 12 September 2026, 2026Added September 12, 2026

01The claim

Vehicles are being recalled far less than they were. The figures are a straight count out of NHTSA’s own recalls flat file — the Part 573 defect and noncompliance reports manufacturers are obliged to file, one source, one field, one definition — and every vehicle campaign in the file is on the chart. The axis is linear, zero-based, evenly ticked and unbroken; nothing is sampled, smoothed, indexed or rescaled; every value is printed above its own column. Recall campaigns naming each model year peak at 2,163 on model year 2019 and fall away from there, seven consecutive model years below the peak, ending 1,715, 1,421, 1,090 and finally 582 on model year 2026. That is a fall of 73.1%, and the three newest model years are the three smallest columns on the chart. Read-out for the desk: treat the defect rate on new vehicles as falling. Hold the campaign-management headcount flat into next year rather than growing it with the fleet, and move the recall-readiness budget to the model years that still carry the volume — the 2015 to 2022 cohorts.

02The trick

Every count on that slide is right, the axis really is zero-based, no campaign in the file is missing — and the chart still cannot support the sentence in its headline, because the fifteen columns are not fifteen comparable measurements. A recall does not arrive with the vehicle. A manufacturer files a Part 573 report when it decides a defect is safety-related, and that decision can come seven years after the car was built: on this file a model year keeps gaining campaigns for well over a decade. So each column counts everything that has arrived so far, and “so far” is a different length of time in every column. Model year 2019 has been collecting recalls since 1 January 2019 — 2,808 days, counted to the last report in the file. Model year 2026 has been collecting them for 251. The x-axis is doing two jobs at once: it is a category axis, and it is a clock running backwards. Give both cohorts the same clock and the finding evaporates. At 251 days old, model year 2019 had 687 campaigns and model year 2026 has 582 — a gap of 15.3%, not 73.1%. Model year 2025, which the slide draws half a peak below 2019, had 603 at the same age. Run the age match across all fifteen and the series is a rise through the early years and then a plateau: 368, 388, 461, 606, 667, 623, 608, 687, 583, 668, 680, 647, 628, 603, 582. The cliff is gone, because the cliff was never in the data — it was the right-hand edge of the file. The same thing said the other way round: model year 2019’s column on the first chart is 2,163, and at the age model year 2026 has now reached it stood at 687. That is 31.8% of what it would eventually become. Nothing about the 2026 column tells you it is 31.8% of the way through its own history, and nothing about a bar ever can. Two checks confirm the diagnosis. The first is the oldest cohort: model year 2012, fourteen years on, picked up another 12 campaigns in the twelve months to 9 September 2026, so not a single column on the chart is final. The second is the flow. Count the same campaigns by the year they were filed rather than by the model year they name and nothing has gone quiet at all: 583 vehicle campaigns filed in 2012, 880 in 2019, a peak of 988 in 2021, 950 in 2024 and 891 in 2025. That series is flat to rising over the years the first chart draws as a collapse; the campaigns are simply landing on model years that have been on the road for a while, which is where they always landed. And the failure runs in the direction that flatters, every time. Late-arriving events are usually bad events — defects, claims, defaults, complaints, revisions — so an immature cohort always looks like an improvement, and the newest cohort, the one the meeting is about, looks like the best of all. (NHTSA publishes the file; both drawings are ours, and the desk, the read-out and the recommendation on the first are invented.)

03The fix

Give every cohort the same clock, and say what the clock is. The redraw keeps all fifteen model years and every campaign behind them, and adds three things the briefing had no room for. First, the development curves: cumulative campaigns against the cohort’s own age, one line per model year, each stopping where the file does. That picture is the whole mechanism in one frame — every curve climbs steeply for two years and the oldest is still climbing at fourteen, and the dashed rule at 251 days crosses all fifteen of them while they are still in that first steep climb. It also shows what the first chart was really doing, which was reading fifteen different lines at fifteen different points and drawing the heights side by side. Second, the same fifteen model years age-matched at 251 days, which is all the age the youngest has: 687 for model year 2019 against 582 for model year 2026, and a plateau rather than a cliff across the whole span. Third, a fixed two-year window — campaigns filed by 31 December of the year after the model year — which is the convention a desk would actually use, and which runs the other way: across the thirteen model years whose window has closed, 727 campaigns on model year 2012 rising to 1,340 on model year 2022 and 1,221 on model year 2024, up 68.0% over exactly the years the first chart drew as a collapse. Model years 2025 and 2026 are still inside their windows, which do not close until the end of this year and the end of next, so they are drawn as hollow dashed outlines at 1,090 and 582 so far — the treatment this page recommends for a cohort that has to appear before it is finished, and the reason the solid part of the series stops at 2024. In general: pick a window every cohort has completed, print it in the caption the way you print a unit, and let the solid part of the chart end where the window ends — a series that stops a year or two short of the file is more useful than one that runs to the edge and is wrong at the end. Draw the development curves beside the bars, since they cost one panel and settle the question. Where an immature cohort must be shown, give it its own style and label it incomplete, or project it from the mature curves and call the projection a projection; actuaries have done exactly this for a century, under the name development factors. Print the as-of date, because every number on a chart like this is “as at” something. And never take a percentage change between two cohorts of different age: that arithmetic is being done by the calendar, not by the world.