Thread: What is the realistic accuracy & precision of typical consumer MPG calculations (tripmeter miles/pump gallons)
View Single Post
  #116   Report Post  
Posted to alt.home.repair,rec.autos.tech
dpb dpb is offline
external usenet poster
  
Posts: 12,595
Default What is the realistic accuracy & precision of typical consumerMPG calculations (tripmeter miles/pump gallons)
On 07/22/2017 11:39 PM, wrote:
On Sun, 23 Jul 2017 02:40:10 -0000 (UTC), Mad Roger
wrote:

On Sat, 22 Jul 2017 19:44:26 -0500,
dpb wrote:

I got curious myself on what the numbers revealed and looked at the NIST
numbers again.

I computed an empirical cdf and compared it to normal...statistics from
the 20,036 observations are below:

,,,

min: -50
max: 146
mean: -0.0788
std: 3.7681
median: 0
mode: 0
....

For a typical 20-gallon fill, how many gallons off can reality be, plus or
minus from the indicated reading on the pumpmeter?
The poorest pump checked in that data would be +/- 44.6 oz per 20
gallon tank - the average about +/- 12 ounces.
....

Actually, the extrema limits in the sample are much worse: -50 and +146
in the 5 gal test collection. I simply computed the table for typical
exceedance limits; didn't include the endmost points there. The 0.999
point, for example, is the int(20036*0.999)-th observation or 20015 so
there were 21 more observations above that value. The last two,
however, were really, really outliers that skew things quite a lot. The
last five observations were

NIST(end-5:end,
ans =
28 1
29 1
30 4
56 1
127 3
146 1


It's quite an oddity that there were 3 observations at 127; illustrating
again that "random data aren't" or more correctly that one can always
find patterns visually even in random data.

Of course, if the pump is putting out 5.6 gal/5 gal reading, you'll only
have "pumped" 20*5/5.6 -- ~17.8 gal when the tank is full already...

Hmmm...I had noticed the outlier on the positive end and looked at it;
wonder what the LH tail looks like now...

NIST(1:5,
ans =
-50 1
-45 1
-36 1
-35 4
-32 1


Pretty similar pattern, just not as extreme. There are two outliers
separated quite a bit from the bulk of the rest of the observations, but
they're only 10 cu in differential to nearest whereas there's 70+ on the
other end.

Interesting that from a customer viewpoint you're just about as likely
to get more than you're paying for as under and on the extreme ends by a
lot more than by what you get shorted.

I didn't try to find what Canadian limits are -- I presume they must be
somewhat more stringent in order to match, more or less, the size of the
measurement interval? You know?

--
Reply With QuoteReply With Quote