On Saturday, July 22, 2017 at 8:56:47 PM UTC-4, dpb wrote:
On 07/22/2017 6:42 PM, Mad Roger wrote:
...

The main point for bringing up the otherwise unrelated issues of
speedometer accuracy and repeatability is that some people here seemed to
think if they ran a calculation more times, that the "accuracy" somehow
(magically?) gets better because (magically?) their reading is low as much
as it is high.
...

NO! That is not at all what any of those people said. You're simply
repeating the same contention which is ok as far as it goes in that a
point estimate is not the same as averaging or using other techniques to
increase the precision of the measurement.

They (and I in particular) only point out that averaging muddles out the
differences in intermediate filling levels.

+1

I'm not sure I;d even call it averaging, it's a different test
method. Two ways of doing it:

1 - Fill the tank once, drive until it's near empty, fill it again.
Your accuracy is greatly affected by your ability or inability
to fill it to exactly the same level. If you're off by a gallon
on a 15 gallon tank, it;s 7%.

2 - Fill it at the beginning, drive it a much longer distance,
through 10 tanks worth of gas where you have the pump reading
on all of those, then fill it the last time to as close to the
original fill as possible.

Method 2 reduces the inaccuracy due to not filling it to exactly
the same level by an order of magnitude. If you're off by a gallon
between the first and last fill, it's an error of ~0.7%.
You still have whatever the accuracy of the pumps are to deal with,
but I agree I'd trust that the pumps are going to be a lot
closer to the 0.7% accuracy than 7%. I'd think they are better
than 0.7%. And like you said, in between
the first and last fill, it doesn't matter if you fill it all the
way or only half full, etc.