Bigger outside, smaller inside

The original claim

Cars grew on the outside and shrank on the inside.

Two testable halves. Exterior is easy, everybody publishes length, width, height, wheelbase. Interior is the problem. No long-running standardized European interior volume series was found. The US has one: the EPA has published passenger volume under a single method since model year 1984, defined in 40 CFR 600.315-08, measured per SAE J1100a with EPA modifications. So the exterior spine is European nameplates in European spec, and the interior control is the EPA database.

Pass one: two nameplates, every generation

Index both series to the earliest generation and watch whether they stay together. If the extra metal were buying cabin, they would.

They separate from the fourth generation and never reconverge. The Golf grew 32.7 percent in exterior envelope and 16.3 percent in passenger volume. The 3 Series grew 26.5 percent outside and 16.0 percent inside.

The strict claim failed immediately. Nothing shrank. But the share of the outer envelope that is actually cabin fell from 26.6 to 23.3 percent on the Golf, and from 23.4 to 21.4 percent on the 3 Series.

One number from that pass stands on its own. The Golf's boot in 2024 holds 381 litres. The Mk1's held 368. Over fifty years the car grew by a third and the boot grew by 3.5 percent, and for four consecutive generations it was smaller than the Mk1's.

Pass two: from 2010, widened sideways

Two nameplates over fourteen years is not a sample. So instead of a longer window I took a wider one: 25 nameplates sold in the US across the whole period, each with an EPA interior figure and an exterior spec at both ends.

Read it in three passes.

The left half is empty. 24 of 25 nameplates grew in footprint, the twenty-fifth by a rounding error. Nothing in this sample got smaller outside.

Below the horizontal axis sit nine cars. That is my claim stated literally, bigger outside and tighter inside. Real, but a minority, and six of those nine move by less than the rounding band explained below.

Below the diagonal sit nineteen. Those cars grew, and the growth did not become cabin. The gap between nine and nineteen is the finding.

Pass three: the start year stopped being a detail

I had asked for 2010 after seeing the long run. I then asked for 2005 after seeing 2010. Three start years, each chosen after seeing the result of the last one. That is specification search. Try enough windows and one will eventually agree with you, and the one that agrees tells you about the window, not about cars.

So the answer stopped being one number and became thirty. Every start year from 1989 to 2018, one rule, 30 nameplates, each measured against its own most recent observation.

The median interior change falls from plus 7.4 cubic feet at a 1989 start toward zero, reaches 0.0 from 2012 onward, and never goes below. The share of nameplates that shrank sits flat between 20 and 27 percent for every start year after 2002 and never exceeds 27 percent. That is a constant churn rate. Some nameplates lose volume at every generation change. It is not a trend that emerged.

For the record, 2005 is a worse window for my claim than 2010, not a better one. Of 27 nameplates measurable from model year 2005, seventeen gained interior volume, six lost it, four are inside the rounding band.

The correction

In pass two I reported this and I believed it: cars since 2010 got longer and lower, median height minus 18 millimetres, 16 of 25 with a lower roof. I called it the mechanism behind the feeling. A cabin that is longer and flatter measures the same and reads as smaller, because headroom is what people register as roomy.

Here is the chart I was looking at when I believed it.

It does not survive the wider window.

Measured from 2005, the same 25 nameplates gained a median of plus 10 millimetres of height, 14 up against 8 down. The mid-2000s were a low point for several designs. The Mk4 Golf, the B15 Sentra, the BL Legacy and the EF Sonata were all unusually low, so a 2010 start captures the tall phase that followed and reads its correction as a trend.

I made the same error I had just finished pointing at in the question.

Two nameplates do fall in both windows and are unaffected. The Corolla is down 51 millimetres since 2005 and the Camry 46. Those are real. The fleet-level roofline claim was not, and it is withdrawn.

The corrected claim

One row changes sign between columns. That is the row I withdrew.

What is left is narrower than what I started with and better supported. Cars got bigger, and the extra car did not become cabin. That holds at 1974, at 2005 and at 2010. It does not depend on where you start.

The clearest single case is the Corolla. The E170 recorded 98 cubic feet of passenger volume and was classed Midsize. The E210 that replaced it records 89 and is classed Compact. Same nameplate. Wheelbase identical at 106.3 inches, car fractionally shorter overall, roof 20 millimetres lower. Toyota's own launch material states the direction: hood lowered 1.4 inches, height reduced 0.8 inches, wheelbase maintained. Nine cubic feet of cabin traded for a silhouette, at no exterior saving.

What this is actually about

The cars are the worked example. The subject is the start year.

Every benchmark you are shown was measured from somewhere. Supplier quality trend, program timing baseline, headcount productivity, cost per unit, energy per unit. Somebody picked the first year. If they picked it after seeing the series, the number is manufactured, and it is manufactured without anybody lying.

Three questions that cost nothing to ask:

1. Which year does this start from, and who chose it?
2. Was it chosen before or after the result was known?
3. What does the same series look like from five years earlier and five years later?

If the answer flips, you have a property of the window. If it holds, you have a property of the thing.

I did not ask myself those questions until the third pass.

Method and limits

Interior: EPA passenger volume, model years 1984 to 2021 from the agency's vehicle database, plus manufacturer press specifications for 2022 to 2026 where the database ends. Base body style, sedans and hatchbacks, no SUVs, no wagons, no convertibles.

Exterior: manufacturer specifications, European spec for the European nameplates, US spec for the US cross-section so it matches the EPA figures. Every dimension cross-checked against a second source. Four disagreements over 20 millimetres were found and resolved in favour of the concurring majority, two of them the same failure mode: a model-level specification page returning wagon dimensions on a sedan row.

Rounding: the EPA passenger volume field is an integer. A recorded 95 means anything from 94.5 to 95.5. Differencing two such readings carries up to plus or minus 1 cubic foot, about 1 percent on a typical car. Applying that as a deadband to the 2010 window leaves 6 nameplates that genuinely shrank, 11 that genuinely grew, and 8 that are unresolvable noise. Every result on this page smaller than 1 percent is no signal, not evidence. Exterior and height figures are unaffected, they come from millimetre specifications.

What the measurement cannot see: EPA passenger volume is swept air. It does not see a centre console that grew, a belt line that rose, seat bolsters that thickened, or a windscreen base that moved forward. Each of those makes a cabin feel smaller at constant measured volume. This biases the result against my original claim, not for it.

Survivorship: the long-run nameplates all survived fifty years, which usually means they climbed a segment. The honest framing is "this nameplate grew", not "this class of car grew".

Cause: I am not claiming to know why. The regulatory drivers are documented, FMVSS 216a doubled the roof strength to weight requirement from 1.5 to 3.0 and made the test two sided. What does not exist in public literature is a measured series of pillar section or door thickness over model years. There is also a competing explanation: Whitefoot and Skerlos modelled footprint-indexed fuel economy standards and predicted they alone would push sales-weighted vehicle size up 2 to 32 percent. If a meaningful share of the growth is regulatory arbitrage on footprint rather than crash structure, then "cars got fat to be safe" is the wrong story even though the dimensions are right.

Exterior growth cross-check: three independent fleet studies put the EU width rate near 5 millimetres per year, JATO at 5.35 over 2003 to 2023, ICCT at 5.11 over 2001 to 2020, Transport and Environment at 5.00 over 2018 to mid 2023. The last of those has a known weakness, its two top-100 lists contain different models so sales mix shift is inside the number. The two single-nameplate rates computed here, 5.05 for the 3 Series and 3.58 for the Golf, have no mix problem and bracket the fleet figures.

Open gates

Two things here are not closed. They are on the page rather than in a drafts folder, because a gate that does not block is how the last error shipped.

Gate 1, unresolved. I state that no comparable European interior volume series was found. I have not confirmed that no EU or UNECE instrument requires such a declaration. Checked so far: EU type approval overview material and Certificate of Conformity summaries, none of which list interior volume among required dimensions. That is absence of evidence. Next step: read Annex I and the CoC template of Regulation (EU) 2018/858 directly, plus the scope of ISO 3832, which covers luggage compartment volume only. Until that is done this page says "no comparable European series was found" and not "Europe does not measure this".

Gate 2, constrained. The Golf Mk8 passenger volume figure of 91.9 cubic feet is a GTI. GTI sport seats are bulkier than base seats, so the Mk7 to Mk8 step may be a trim artefact rather than a design change. The Mk5 to Mk7 decline is like for like and is not affected. The figure is labelled GTI wherever it appears.

Pre-registered call

If a standardized European interior volume series is found, or introduced by regulation, the 30-start-year sweep is rerun on European market data and the result is published here whether or not it agrees with this page. No resolution date, this one depends on somebody else.

How this page was made

Partially AI generated, human supervised. The split, so you can weigh the page accordingly.

Mine: the hypothesis, including the bias I flagged in the first sentence. The three start years, which were my requests and which turned out to be the problem. The decision to widen instead of picking a fourth. Which findings ship and which are withdrawn. The final text.

The model's: pulling the EPA database and the manufacturer specifications, cross-checking every dimension against a second source, computing the sweep across 30 start years, rendering the charts, and drafting this page.

The charts are not AI-generated pictures. They are code rendering a dataset, deterministic, same input gives the same output every time. In IPTC's vocabulary that is `dataDrivenMedia`, not `trainedAlgorithmicMedia`. The distinction matters on a page whose whole argument is about where numbers come from.

What supervision actually caught, since a disclosure that only says "a human reviewed it" is worth nothing: the roofline finding in the corrections log was produced by the model, believed by me, and killed by widening the window. Two source pages returning wagon dimensions on a sedan row were caught by the second-source rule and not by reading the prose. The European interior volume claim is still open at Gate 1 because I would not let it be written as settled.

Every number on this page traces to a source below. If one does not, that is an error and it goes in the corrections log with a date.