Vehicle records
How Far the Official Fuel-Economy Rating Moves Inside One Nameplate
Across 3,621 groups of model-year 2015–2026 vehicle configurations matching on make, nameplate, model year, fuel type, drivetrain and transmission, 24.0% still differ by at least 5 rating points or a fifth in the official combined figure, and among MPGe-rated groups that share is 60.6%.
Published August 12, 2026 Eldhurst Data Insights Eldhurst LLC
Executive Summary
Shoppers compare cars by name. The government rates them by build. A single nameplate in one model year can carry many separately rated configurations, and the official combined rating belongs to the exact configuration, not to the name on the badge.
The obvious objection is that this gap is trivial — of course a hybrid differs from a gasoline version, and of course all-wheel drive costs a little economy. This paper tests that objection directly by holding the distinguishing attributes constant, one at a time, and measuring what survives.
The ladder is a comparability sequence, not a decomposition: each step regroups the catalogue under stricter matching rules, so both the groups and the qualifying denominator change and the steps do not partition the first figure into explained and unexplained parts. Grouping by nameplate alone, 48.4% of groups show a material spread. Under nameplate and fuel type, 35.8%. Adding drivetrain, 32.9%. Adding transmission, 24.0%. The share falls as matching tightens and does not fall to zero: under the six-field rule, nearly a quarter of groups matching on make, nameplate, model year, fuel type, drivetrain and transmission still differ materially in the published combined rating.
The split by rating unit is sharper than the headline. Among groups rated in MPG the material share is 19.6%. Among groups rated in MPGe — electricity, compressed natural gas, hydrogen and the gas-electric combinations — it is 60.6%. Groups rated in MPGe carry a material spread about three times as often as groups rated in MPG. This is only an observed difference between the two rating-unit strata. The governed fields do not establish powertrain classes or estimate a powertrain effect, and neither is claimed.
The most important limitation constrains how this may be said. The governed grouping key is the source's base_model, and 746 of the 870 material groups (85.7%) contain more than one distinct model string. A group is a nameplate family in the source's own terms, not a single marketed model.
This is a reference benchmark for shoppers and small-fleet buyers: a model-level fuel-economy figure is a weak summary of the separately rated configurations a nameplate carries, and it is weakest among the groups the publisher rates in MPGe.
Business Question
Question. Across officially rated vehicle configurations, how wide is the spread in the published combined fuel-economy rating inside a single nameplate among records matching on make, nameplate, model year, fuel type, drivetrain and transmission, and how often is that spread large enough to matter?
Audience. Drivers, vehicle owners, shoppers and small-fleet decision-makers.
Publication-value mode. Reference benchmark. The official tools return a rating for one vehicle at a time and compare vehicles a user picks; direct inspection of the government's find-and-compare interface confirms it publishes no summary of how far the rating moves across the configurations inside one name.
Specific reader utility. A reader learns whether a model-level MPG figure — the number quoted in advertising, reviews and casual comparison — can be relied on, and learns which rating-unit stratum it fails hardest for.
Positive takeaway. If a material spread persists among groups matching on all six observed fields — make, nameplate, model year, fuel type, drivetrain and transmission — a buyer must check the exact build before making a cost comparison, and a model-level number is not a safe input.
Credible null takeaway. If the spread collapsed once fuel type and drivetrain were held constant, model-level shorthand would be validated for most nameplates and the exceptions could be listed.
Why it matters now. The model-year 2026 ratings are published, and a single nameplate in one model year can carry many separately rated configurations under one name.
Data & Scope
Source. FuelEconomy.gov vehicle ratings, jointly published by the U.S. Department of Energy and the U.S. Environmental Protection Agency, as represented in the governed Eldhurst research data platform. The source's population noun is the vehicle configuration; a configuration is one separately rated build.
Universe. Listed configurations, national, model years 2015 through 2026. 15,455 configurations qualify, each carrying a distinct vehicle identifier.
Grain. One listed vehicle configuration, grouped to (make, base_model, model_year, fuel_type, drive, transmission) for the primary comparison.
Inclusion rule. A configuration is measurable when its combined rating is present and positive and its base_model is non-empty.
Exclusions and their counts. Zero configurations in the window were dropped for a missing or non-positive combined rating, and zero for a missing base_model. Zero measurable configurations are missing fuel_type, drive or transmission, so the primary grouping forms with no loss.
Units, and why the grouping enforces them. The source defines the combined rating by primary fuel and reports electric and compressed-natural-gas vehicles in MPGe rather than MPG, and reports hydrogen fuel-cell ratings in MPGe. The window holds 15 exact fuel categories, 9 of them MPGe. Because fuel_type is inside the primary grouping key, every within-group difference is computed inside one exact fuel category. No value in this paper is ever subtracted or compared across MPG and MPGe.
Coverage gap. Listed configurations are not sales-weighted and do not describe what buyers actually purchase. A rating is a standardised laboratory test result, not an on-road measurement.
Source limitation carried forward. The revision flag mpg_revised is empty for every configuration in the window, while the publisher states that all older and some 2011–2016 estimates were revised. The empty field is therefore not treated as proof of a single rating regime; a sensitivity restricted to model years 2017–2026 is reported in Section 4.
Method
For each group, the spread is the difference between the highest and lowest combined rating among the configurations in that group, together with the interquartile range and the ratio of highest to lowest. A group qualifies when it holds at least two measurable configurations; the primary grouping yields 3,621 qualifying groups, well above the prespecified minimum of 200.
A group shows a material spread when its range is at least 5 rating points or its maximum-to-minimum ratio is at least 1.20. Both thresholds were fixed before any outcome was computed and neither was changed afterwards.
The comparability ladder is the method's core. The same statistic is computed four times, each time adding a grouping field that makes the configurations inside a group match on more of the six observed fields: nameplate only, then plus fuel type, then plus drivetrain, then plus transmission. The last is the primary comparison; the first is retained and reported separately as product-line breadth, and is never used for any conclusion about comparable builds.
Two further views are reported: the material share split by rating unit, and the distribution of range sizes within each unit, binned. Because MPG and MPGe are different units, these are reported separately and never pooled onto one scale.
Prespecified sensitivity. Restricting to model years 2017–2026 gives 3,055 qualifying groups and a material share of 24.9%, against 24.0% for the full window — a difference of under one percentage point.
There is no ranking rule and no decision rule. No make, model or vehicle is named or ranked anywhere in this paper.
Findings
The locked evidence artifact: the comparability ladder
| Grouping | Fields | Qualifying groups | Material groups | Material share |
|---|---|---|---|---|
| Nameplate only (product-line breadth) | 3 | 2,972 | 1,438 | 48.4% |
| Nameplate + fuel type | 4 | 3,381 | 1,210 | 35.8% |
| Nameplate + fuel type + drivetrain | 5 | 3,599 | 1,185 | 32.9% |
| Nameplate + fuel type + drivetrain + transmission | 6 | 3,621 | 870 | 24.0% |
Figure 1 renders this table.
The ladder answers the obvious objection on its own terms. Each step adds a matching requirement, and the material share falls each time — from 48.4% to 24.0%. It does not fall to zero, or near it. The steps are not an explained-share decomposition: adding a grouping field changes which configurations are compared and how many groups qualify, so the 24.4-point reduction cannot be read as any attribute's estimated contribution. What the sequence establishes is that the material share is still 24.0% when records are matched on all six observed fields, which is the most the governed record can match on.
The split by rating unit
| Rating unit | Qualifying groups | Material groups | Material share |
|---|---|---|---|
| MPG | 3,230 | 633 | 19.6% |
| MPGe | 391 | 237 | 60.6% |
This is the sharper result. MPG-rated groups carry a material spread about one time in five. MPGe-rated groups carry one about three times in five. These are observed rating-unit strata, not inferred vehicle or powertrain classes.
How large the spreads are
| Range bin | MPG groups | Share | MPGe groups | Share |
|---|---|---|---|---|
| 0 to under 1 | 651 | 20.2% | 44 | 11.3% |
| 1 to under 2 | 796 | 24.6% | 25 | 6.4% |
| 2 to under 3 | 594 | 18.4% | 26 | 6.6% |
| 3 to under 5 | 712 | 22.0% | 59 | 15.1% |
| 5 to under 10 | 432 | 13.4% | 110 | 28.1% |
| 10 to under 20 | 42 | 1.3% | 87 | 22.3% |
| 20 or more | 3 | 0.1% | 40 | 10.2% |
The two distributions have different shapes, not merely different means. Among MPG groups, 63.2% sit under 3 rating points of spread and only 1.4% reach 10 or more. Among MPGe groups, 32.5% reach 10 or more. The bins are counted separately by unit and must not be added together or plotted on a shared axis.
Interpretation, kept separate from the above
The direct findings are the shares and distributions in the tables. The interpretation is that a model-level fuel-economy figure is a lossy summary in about a quarter of the groups that match on all six governed fields, and a badly lossy one among the groups the publisher rates in MPGe. Both statements describe the publisher's listed configurations grouped by base_model, which is a nameplate family in the source's terms rather than a single marketed model.
Association Analysis
No inferential association test is run, and none is justified by the locked method.
The locked method specifies a within-group spread distribution, a fixed materiality rule, and a comparability ladder that matches records on successively more of the six observed fields. It authorises no significance test, correlation or model.
The reason is that these are all listed configurations, not a sample of them. A confidence interval or p-value would describe sampling variability that does not exist here, and would invite the reader to treat the catalogue as a draw from a market it does not represent — the catalogue is not sales-weighted.
The comparability ladder in Section 5 is the authorised test and it is reported in full, including the direction and size of every step. The unit split is a stratification of the same statistic, not an association between variables.
Association is not causation, and this paper claims neither. In particular, the ladder does not establish that transmission choice causes the remaining spread; it establishes only that holding transmission constant does not remove it.
Business Implications
The declared value mode is a reference benchmark, so the permitted implication is what a reader can compare and check.
A shopper or small-fleet buyer can:
- treat a model-level fuel-economy figure as a weak summary and look up the exact build before using a number in any cost comparison;
- expect the model-level figure to be least reliable for nameplate families the publisher rates in MPGe, where 60.6% of groups carry a material spread against 19.6% for those rated in MPG;
- use 24.0% as the observed share of groups matching on make, nameplate, model year, fuel type, drivetrain and transmission that still differ materially — not as a claim that the configurations inside such a group are identical in every respect, since the governed record carries only those six fields.
Anyone quoting fuel economy can:
- avoid attaching one rating to a nameplate without naming the configuration it belongs to;
- keep MPG and MPGe figures separate rather than presenting them on one scale.
What must not be inferred or automated. Nothing here supports a claim about any individual vehicle, about on-road or owner-specific fuel use, about reliability, or about what buyers actually purchase. No vehicle is recommended and none is criticised. The paper names no make, model or configuration, and its groups are nameplate families in the source's own terms rather than single marketed models.
Limitations
What the data exclude. Sales volumes, on-road fuel consumption, owner experience, price, and any vehicle not separately rated by the publisher. The ratings are standardised laboratory test results.
What the method cannot determine. Why a spread persists after transmission is held constant. Candidate explanations include tyre and wheel packages, weight and option content, and additional engine variants not separated by the four grouping attributes available — the governed table does not carry the fields needed to separate them, so the paper reports that the spread survives and does not explain it.
The grouping caveat, which is the sharpest limitation here. The grouping key is the source's base_model. Of the 870 material groups, 746 (85.7%) contain more than one distinct model string. A group is therefore a nameplate family as the source defines it, not a single marketed model, and no sentence in this paper may describe it as "one model" without that qualification.
Coverage and sensitivity weaknesses. The MPGe stratum rests on 391 qualifying groups against 3,230 for MPG, so the 60.6% figure is measured on a much smaller base and should be read as the property of a small and rapidly changing part of the catalogue. The mpg_revised field is empty throughout the window, so rating regimes cannot be separated using the publisher's own marker; the 2017–2026 sensitivity is the available substitute and it moves the headline by less than one percentage point.
Alternative explanation considered. The strongest is that the whole effect is an artifact of grouping together configurations that differ on fuel type, drivetrain or transmission. The ladder was built to test exactly that and it does not survive: material spread persists at 24.0% with fuel type, drivetrain and transmission all held constant, against a prespecified floor of 5% below which the finding would have been withdrawn.
The unanswered question that should guide follow-up. What accounts for the residual spread inside a group matching on all six observed fields? Answering it needs attributes this governed source does not carry.
References / Appendix
Authoritative sources
- U.S. Department of Energy and U.S. Environmental Protection Agency, Fuel Economy Web Services data description — defines the combined rating by primary fuel, states that electric and CNG values are MPGe, exposes the drive and transmission fields, and notes revisions to all older and some 2011–2016 estimates.
- U.S. Department of Energy and U.S. Environmental Protection Agency, 2026 Fuel Economy Guide — the current published per-vehicle answer, including its own statement that no test can predict fuel economy for all drivers and all conditions.
- U.S. Environmental Protection Agency, Basic Information on Fuel Economy Labeling — confirms the window label carries the rating for that exact build.
- U.S. Department of Energy and U.S. Environmental Protection Agency, Find and Compare Cars — the owner tool, verified by direct retrieval to offer per-vehicle values and user-selected side-by-side comparison only, with no summary of within-nameplate variation.
All external sources were verified on 12 August 2026.
Field definitions
| Term | Definition |
|---|---|
| Configuration | One separately rated vehicle build, identified by a distinct vehicle identifier. |
| Nameplate group | Configurations sharing make, base_model and model year. A nameplate family in the source's terms, not necessarily one marketed model. |
| Primary comparable group | Configurations sharing make, base_model, model year, fuel type, drivetrain and transmission. |
| Combined rating | The publisher's combined city and highway figure for that configuration, in MPG or MPGe according to primary fuel. |
| Material spread | A within-group range of at least 5 rating points or a maximum-to-minimum ratio of at least 1.20. |
| MPGe | Miles per gallon of gasoline equivalent, used by the publisher for electricity, compressed natural gas, hydrogen and combination fuels. |
Permitted and prohibited claims
This article states: the shares and distributions in Section 5, stated as properties of the publisher's listed configurations for model years 2015–2026.
Outside what this article claims: subtracting or comparing a value across MPG and MPGe; treating listed configurations as sales-weighted market data; inferring owner-specific or on-road fuel use; making a reliability claim about any vehicle; asserting that every configuration sharing a nameplate is cross-shopped; claiming the empty revision flag proves a single rating regime; and describing a base_model group as a single marketed model without qualification.
Figures
Scroll chart horizontally to read all labels.
Full chart description
Horizontal bar chart titled "Fuel-Economy Spread Persists After Six-Field Matching". Each bar is a grouping rule, matching records on progressively more of the six observed fields, from make, nameplate and model year alone at the top to all six fields at the bottom. Each bar shows the share of qualifying nameplate groups whose configurations differ by a material amount in the official combined rating. - Nameplate only (product-line breadth): 48.4% of 2,972 qualifying groups. - Nameplate and fuel type: 35.8% of 3,381 qualifying groups. - Nameplate, fuel type and drive: 32.9% of 3,599 qualifying groups. - Nameplate, fuel type, drive and transmission: 24.0% of 3,621 qualifying groups. Material spread means a within-group range of at least 5 rating points or a maximum-to-minimum ratio of at least 1.20. Ratings are standardised laboratory test values for listed configurations, not sales-weighted market data and not owner-specific on-road fuel use.