[bwilson4web]

(Originally posted in Prius_Technical_Stuff-rjw)

Sampling bias is always a problem and was recognized as a problem with the old GreenHybrid.com database. For example, the CleanMPG crowd spun off from GreenHybrid starting in late January 2006. Until then, they advocated 'driving style' to the point of violating the stated goals of "hybrid" in GreenHybrid. So looking at the original, GreenHybrid mileage database, we found a number of outliers.

GreenHybrid's owner, Jason, took pains to trim the outliers. I confirmed his good judgement by an independent analysis that revealed the effects of these outliers in the database. So if I were trying to come up with a credible 'average' MPG, the upper and lower 5%, a total of 10% of the data, should be trimmed or handled as an exception. Certainly we can see that pattern in the graph:

(One latent bias is the 2010 Prius has two summers but only one fall, winter, and spring in the data. This may be part of the bulge on the high side of the distribution.-rjw)

Now one thing true about the 2008 and 2010 model years was the absence of Tsunami production effects but there was an economic depression/recession that started in September 2008. There were some peripheral events that would have increased the EPA site participation: new model introduction; the implosion of GreenHybrid.com when the database was replaced and new policies; and advocacy of the EPA web site in PriusChat as a good place to save mileage records.

One test would be to compare the population counts against the annual sales:

• 2008 - 158,600 (Wiki)
• 2010 - 103,300 (Wiki Jan-Sept) + 69,000 (est. 2009) ~172,300

Sample sets and rates:

• 2008 - 125 ~= 0.000788
• 2010 - 150 ~= 0.000871

Now MyHybridCar.com has 47, ZVW30 entries which might have included some 2011 model year Prius. The 2010 model entries here probably reduced the EPA web site count of 2010 Prius (I post mileage to both.)

Now we know the ZVW30 has an indicated mileage error that is ~6% more optimistic than pump and odometer calculations. (There is no requirement for any given metric method at the EPA web site.-rjw) But I've not studied any similar bias in the NHW20 and the NHW11 is . . . well curious.

I've been conducting a long term, mileage test of oversized tires, so my indicated MPG is 6% depressed (as is the odometer and MPH indicator.) But I've calibrated my metrics and save true values in MyHybridCar.com. As for the other NHW11 owners, well I've never tried to study the population. Most NHW11s are already well past the OEM tires.

If I had the time and interest, I would increase the sample sets by including Fuelly.com. I could also include MyHybridCar.com, Ecomodder.com, the new GreenHybrid.com, and CleanMPG.com. But good, fast, cheap, I'm not sure the additional precision would significantly change the results.

So sometimes you just have to hope that the sampling 'noise' from the source (www.fueleconomy.gov) is sufficiently random that we can draw reasonable inferences from the data.

Bob Wilson