Nishitetsu — Japan
Three route buses were measured against a twelve-month rolling baseline — by some distance the most robust baseline in our records — and improved by +5.3%, +13.7%, +7.8%, a median of +7.8%. The measurement window was February to June 2020. Passenger numbers on Japanese city buses collapsed in exactly those months, and a lighter, less frequently stopping bus uses less fuel on its own. We cannot separate the two effects, so we do not claim this result.
A twelve-month baseline is the best measurement design on this site. It ran into the first months of the pandemic, and that is enough to make the number unusable.
| Operator | Nishitetsu (Nishi-Nippon Railroad), Nakagawa depot — one of Japan’s largest bus fleets |
|---|---|
| Vehicles | 3 route buses, QPG-LV234N3 |
| Baseline | Twelve-month rolling average, February 2019 to January 2020 |
| Application | 1 February 2020 |
| Measurement window | February to June 2020 |
| Measured quantity | Fuel efficiency in km/L |
| Result | +5.3%, +13.7%, +7.8% — median +7.8% |
| Why we do not claim it | The measurement window coincides with the collapse in bus passenger numbers in early 2020 |
Every measurement, not just the best one
| Vehicle | Engine no. | Baseline, 12-month average (km/L) | Feb–Jun 2020 (km/L) | Change |
|---|---|---|---|---|
| Route bus QPG-LV234N3 | #2788 | 2.806 | 2.955 | +5.3% |
| Route bus QPG-LV234N3 | #2789 | 2.690 | 3.059 | +13.7% |
| Route bus QPG-LV234N3 | #2790 | 2.902 | 3.128 | +7.8% |
| Median across the three buses | +7.8% |
Method before result
The baseline is the strongest element of any trial on this site: a rolling twelve-month average per bus, from February 2019 to January 2020. A full year of the operator’s own data absorbs season, weather, timetable changes and route variation in a way that no three-month baseline can.
The treatment was applied on 1 February 2020 and the buses were then measured from February to June 2020.
And that is where it falls apart. Those are the months in which passenger numbers on Japanese urban buses dropped sharply. A scheduled bus carrying substantially fewer passengers is lighter, stops less often at request stops and dwells for less time at each — all of which reduce fuel consumption per kilometre with no help from anything on the air filter. The size of that effect is not documented anywhere in the records we hold.
We are publishing the figures rather than suppressing them, with the confounder stated in the same breath. What we are not doing is adding them to any average or quoting the median as a result.
The limits of this trial
We publish these next to the result rather than in a footnote. If a limitation would change how you read the number, you should read it at the same time as the number.
Known limitations
- The measurement window coincides with the early-2020 collapse in bus ridership. Passenger load and stop frequency both affect fuel consumption on a route bus, and neither is recorded here. This alone is enough to make the figures uninterpretable.
- The baseline values differ between versions of our own summary table. An earlier version recorded 2.909, 2.889 and 2.913 km/L, the current one 2.806, 2.690 and 2.902. We use the current values and have flagged the discrepancy for resolution — until it is resolved, even the baseline is uncertain.
- Three buses out of a fleet of two to three thousand.
- There is no control group. Untreated buses from the same depot over the same months would have made this dataset genuinely valuable, because they would have absorbed the ridership collapse along with everything else.
- We hold a summarised record of this trial, not the operator’s primary data. The method, the periods and the per-vehicle figures are documented; the underlying fuel and distance logs are not in our possession, and we have asked for them.
Reading this evidence in context
This page is here as a methodological point rather than a result. The twelve-month rolling baseline is the design we would like every trial to use, and this one shows how completely an external event can still undo it.
It is also the clearest illustration of why we keep asking operators for a control group. Treated and untreated buses from the same depot over those same months would have made this the strongest dataset we own instead of an unusable one.
If Nishitetsu can supply passenger or ridership data for the period, the figures may become interpretable. We have asked.
Questions operators ask about this trial
Why show numbers you say cannot be used?
Could you not just correct for the drop in passengers?
What makes a twelve-month baseline better?
Provenance
- Operator
- Nishitetsu (Nishi-Nippon Railroad), Nakagawa depot, Japan
- Baseline
- Twelve-month rolling average, February 2019 to January 2020
- Application
- 1 February 2020
- Measurement window
- February to June 2020
- Source
- Japanese field-study summary — derived record
- Open items
- Baseline discrepancy between summary versions unresolved; ridership data for the measurement period requested; primary records requested
- Last checked
- 15 August 2026
The only baseline that matters is yours
Every trial on this site was run on someone else’s vehicles, duty cycle and baseline. A NanoEFX pilot measures yours — and we will help you design it with a control group so the result stands up.