And here we go... first Ford lawsuit for not meeting its EPA stickers

my bad, I meant to ask: Does the EPA certify the MFD MPG display?

 

Considering they dont measure based on actual gallons used, but emissions, wouldnt really matter.

 

Epa does not regulate any vehicle instrumentation.

Spedometers are regulated to within 5 mph by US DOT. I don't know of any regulation regarding mpg displays.

 

All true -- But.

Merkins sue at the drop at a hat, hoping for a payday. No lawsuit is too frivolous for at least 100 scum layers, if not more. Accountability is not part of their lexicon -- it is someone else's fault.

Perhaps Merkins and Corps deserve each other.

 

Yet it has no statistical power in regard to the C-max. You can tell from the curve shape.

 

That is not correct; the C-Max sample has plenty of statistical power. The data at fueleconomey.gov for the C-max as a mean of 40.08, a standard deviation of 3.991, which with an effect size of 7 and a sample size of 40. Thus is has has full power (>.99999) even for a two-sided test at p=.0001 -- the difference is statically very significant.

 

While this is quite true, we really need to look at validity of the statistics.

fueleconomy.gov and fuelly are internet polls. We should assume false records and self selection. The net result will be positive bias in the long run, but in the short run all the articles may cause the negative selectors to post.

Self selection cars - the c-max is marketed as more powerful than the prius v. This may bias buyers to those that will use the hp. Camry versus fusion should not have much of this bias though. This should result in negative bias for the c-max if we are considering driving the same way. It will not bias results however if we are trying to figure out what drivers actually get.

Measurement error. We do not know how people are reporting. MPG meters in the cars have positive bias built in. Calculation from fill ups on small numbers of tanks has a great deal of sampling error.

Seasonality and geography - hybrid fuel economy varies a great deal with the weather. There should be a negative bias based on the car only being sampled in the winter.

Given all these biases, it is difficult to tell if actual numbers will be higher or lower, but the biases do not even out. That's why objective tests like the epa are better, especially when suplemented by magazines that measure fuel economy at different speeds. I would expect a fairly large difference to the epa test compared to reports based on these factors.

 

Sagebrush, I think you are a good guy and enjoy most of your posts because they are informative and/or entertaining. But, as long as you are going to live in this country, please choose a less derisive term for the citizens of the United States than "Merkins".

 

I have every reason to think the C-max owner data will approximate a normal distribution, and currently the data does not come close.

GIGO

 

Agreed there are many potential sources of error.. but that does not impact a "test of power" which is well defined.
If one wants to argue the data has baises, I would agree the self-reporting bais and self-selection baises can be real effects. But with no way to judge the bias there is no estimates of its impact. We could just decide that measuring MPG is meaningless because of the baises, but that is not very satisfactory.

I'll add the EPA has attempted to adjust their standard test data (which generally yields higher actual MPG) to predict average usage. Maybe hybrids are more impacted than regular cars by seasonal, but seasonal variations are already built into the EPA's figures, but EPA does not report any std.dev (which they really should).


@ sagebrush.. I agree its not guassian, then again I see no reason to be able to presume it will be well modeld by guassian. Lots of other car MPG data is not (look at 2009 Toyota Prius MPG Reports | Fuelly or 2008 Toyota Prius MPG Reports | Fuelly.. both fail test for normality at [p=0.05.] Both are old enough with enough tanks per person there is little sampling or seasonal error left. There may be self-selection and reporting bais but we cannot even estimate that. There are plenty of good reasons to presume it will not be a Gaussian.
With multiple potential subpopulations (city, highway) and also (agressive/careful) its not hard to imagine a multi-modal distribution being very natural. This is not the difference between samples and the mean -- so CLT does not apply.

 

Certainly when it comes to try to do this at a lawsuit, there is nothing to stop the law firm from pumping bad figures into these internet polls.

Because of likely bad reported data, I would throw out outliers before computing a mean and median. As long as they are not being used for legal proceeding and likely biases are understood, I don't see a problem with using the numbers. If the locations and dates are known corrections should be applied for weather.

It seems the manufacturers and the lawyers are putting way to much faith in these tests. YMMV!

On advertising, I'll let you decide what is more distorting - that the c-max gets 47 mpg highway, or the Tundra can tow a space shuttle. These things actually happened EPA test, and publicity stunt, but lets face it, you aren't going to be able to tow an almost 150 ton object at normal speeds in a tundra. It will definitely void the waranty. Drive a c-max at 55 mph at 70 degrees and you'll beat 47 mpg. Yes just saw the tundra commercial again.

If we compare the EPA to say the JC08 it does a much better job. It does definitely have errors with higher speeds, that may be a factor of politics. Lower speeds plus a correction factor helps vehicles with relatively higher drag, and this may be the result of Auto PACs wanting to sell SUVs, then successful lobbying of the regulator. This systemic error appears to be what helps the highway mileage figures of the c-max.

I don't think we want to or should expect to get the EPA to estimate mileage ranges, but perhaps they can publish the results of all 5 tests on their website, and look at the city and high speed highway tests to more accurately simulate driving in the 2010s.

 

Great discussion guys!

drinnovation,
Have you calculated the SD for the Fuelly.com 2013 C-max Hybrid? I'm very curious.

I think that most Fuelly.com users are fairly honest. Some can't follow simple directions in entering data and some fudge their numbers.

Here's an example of "fuzzy" numbers:

Silver C-MAX (Ford C-Max) | Fuelly

 

I had not, but here you go:
Mean 38.966, Stdev 4.656,
with n=56, it is very statistically significant

Not sure you you think that users is "fuzzy". . looks like they owned the car for a while then drove it to family for xmax.

 

Look here.
The site used to have distribution curves, but you can at least see the sample size, median, average, and SD for the Prius and Civic when sample sizes are over 1000.

Fuelly is certainly bimodal: owners who post to show off, and owners who post to complain. The data screams non-random sampling.

 

They don't answer how good a fit it is to a Normal. I downloaded (scraped rally) the 1000+ samples of the prius data from that site and then applied a Kolmogorov–Smirnov test for normality. It gives a KS goodness of fit of 53.57, and thus can reject that data as being from a Normal distribution with p< .01

 

Where are the Mythbusters when you need them?

 

Another link
Fuel economy labeling of motor vehicles revisions to improve calculation of ... - Google Books

And another, all Prius at Fuelly
Toyota Prius MPG Reports | Fuelly

 

Looking at those stats, I guess I did pretty good in my 2007 getting 53 lifetime in it. Same with the Camry, got 38 in that one too. Got to give credit to Toyota, they underestimate the EPA of what the cars true potential is so the majority of drivers can actually meet the minimum, unlike Ford........ 47 MPG my a$$

 

That the EPA used a normal distribution for its correction does not show much.. they needed to use something to "correct" their bias. More significantly note they did not actually directly use it.. they use the "coefficient of variation" (which is usually used when the data is not actually normally). They used the relation of the CV to the differences between measured data and the " normal distribution they fit to the data", and then used that to adapt their estimates. If they truly believe the data was normal, they would have used a simpler correction based directly on the normal fitting.

And I just applied a KS test on the each of the 2012, 2011, 2010, 2009 fuelly prius data as well as all of those combined. Each is gain is rejected as being normal at p=.01 (i.e. its statistically very significantly not normal).


Turning it around, Why do you believe it should be normal? Nothing in a MPG measurements, let's one invoke the Central-limit theorem..

Some things to consider:


If you think usage (fuel/mile) would be normal, then MPG would not be?
Which could be normal? fuelused/mile (a linear measure) or MPG (an inverted)?

Since we know from statistics on travel, daily milage is highly non-Gaussian, it is far closer to a generalized Pareto distribution. (That is, mixed across vehicle types). Short trips produce lower MPG for any particular vehichle, so there is natural bias in underlying the MPG distribution -- making it likely asymmetric.

 

0+1=2