Recently the VW 2.0L diesel 'cheat' was also evident in user reported mileage being significantly greater than the EPA composite numbers. This pattern is easily detected IF we could get from Fuelly, a list of all cars, not just diesels, to identify:
Vehicles with common engine
Statistically valid sample, more than 34 entries
User average MPG >> composite MPG from EPA
So I used the EPA site, fueleconomy.gov, and found the 2014 BMW 328d had 4 user entries showing 44.3 MPG versus 34 MPG composite EPA. I checked Fuelly and found 39 vehicles with 39.5 MPG, again significantly greater than the 34 MPG composite EPA. But the 2015, 7 vehicles, is less bad, 37.8 MPG.
What I propose is perhaps Fuelly could sort through the car database looking for the average mileage where there are at least 34 vehicles of the same year, make, and engine. Then compare the user MPG versus composite EPA MPG. Ratio the user MPG / composite EPA MPG and list the 10 models on each side:
User % >> 100% - means the EPA test might have been spoofed like VW
User % << 100% - means the EPA test might have been vendor spoofed like KIA and Ford
First, Fuelly doesn't have the official ratings on site, and it has an international user base. EPA is currently the closest to real world, but the system hasn't been completely clean up in terms of engine types. Looking for 2L diesels will also net the smaller 4 cylinder ones sold in Europe at this point.
Next, the EPA test does under estimate diesels. The EPA's own research has shown this. The cheater VWs go back to 2009MY. Before that, they still average higher, and these were years in which diesels didn't need to meet the T2B5 levels for NOx. Diesel Cruze users here exceed its ratings, and so does the BMW 320d. The Bimmer exceeds it by less, but the system has removed some, mostly higher, outliers.
EPA does better reflect better real world driving, but it does have some failings. One is that a manufacturer only needs to report the fuel economy of the 'most commonly equipped' model. The US doesn't get ratings of trims with sportier tires, larger wheels, and taller geared differentials that are factory installed. These will all get lower average than the most efficient one the manufacturer uses for the test.
We don't know the full story yet, this is more to do with N0x emissions rather than C02 which is easier to calculate and compare with economy. There may be no link at all with the cars getting worse economy and the emissions in question.
The NOx trap needs a regen cycle like the DPF. If this wasn't being done often enough, the cheater car will get better fuel economy. But it's a YMMV scenario. If a person's drive is mostly under conditions that don't much particles or NOx, then they won't need much regen cycles to begin with.
I don't see the benefit though in the cheating on SCR equipped cars. The only one apparent is in using less DEF. When they first appeared, there was negative views of having to add another fluid to your car, and quoted prices from Mercedes dealers didn't help. But the stuff is cheap and readily available, under a cent per mile added to the operating costs, and it doesn't involve the compromises of using a NOx trap; lower fuel economy and performance.
Honda and Mazda have tried to bring a diesel to the US, but Honda dropped the plan when they couldn't satisfactorily pass emissions without SCR. Mazda might still be trying, but it has been over 2 years since their announced intro date. Perhaps now, they'll go with SCR, but likely they'll just drop the plan.
What would be the ultimate goal of such reporting, if it were even possible?
I'm curious especially about vehicles that significantly over or under perform relative to a normal distribution. Vehicles on either side of what should be a Gaussian distribution are the ones that can give us a clue about their technology and/or what to test. Let me share an example using the current, 1.8L, Prius.
One of the first puzzles was the distribution of user mileage reports from the new ZVW30 (1.8L) prius versus the older model:
The dual peaks of the EPA user data suggested something was going on that I did not understand. The lower peak suggested it was not significantly better than the previous NHW20. Yet half of the EPA users were seeing significantly better performance. As we got more data points:
Why are there two groups?
It turns out the problem was the indicated mileage from the trip meter is higher than the pump measured values:
The double humped curve was just a question of how the mileage was recorded. Later, I found a systemic problem with the 'revolutions per mile' of the OEM tires versus over-sized replacement tires that not only give a true tripmeter MPG but also GPS/mile-marker distance.
Statistical analysis led to more insights that eventually led to this driving stunt:
Done over a three week period, July 2013, in town driving.
Selecting 4-lane, routes, and commute times to avoid road-rage.
Changing lanes so following traffic does not have to change.
Using flashers to let following traffic know I was going 25 mph.
Use cruise control to maintain 25 mph when moving.
Use "N" when descending a grade and engine is OFF.
Extend all trips to a minimum duration of one hour.
I understand the limitations of Fuelly but the biggest is not being able to get a comma delimited version of selected vehicle mileage records. With enough fields, we can looks for patterns that can provide insights and the ability to optimize performance.
ps. We get 52 MPG with our 2003 Prius and 52 MPG with our 2010 Prius. However, the 2010 does it +5 mph, in colder weather, and superior performance at speeds under 30 mph. I use the smaller, agile 2003 for work and around town and the 2010 for long trips and -40F or above +90F weather. Use the right tool for the right job.