Mileage Database Glitches
 

GH Folks,

I have had my HCH II for over a year. One of my regular checks in that time is to compare the data on the Prius II vs the HCH II. Lately (last two months), the difference in overall mpg (average) has gone from about .6 mpg to 1.5 mpg in favor of the Prius. This does not make sense. I looked in both databases, and saw that there are a number of results that really affect the overall MPG's in both cars. This is especially true of the lowest MPG cars. Some are obvious data entry errors, and many are one tank from a driver who drives fast and drives in a colder part of the US. Another problem is that some entries show 3.000 total miles - BUT no tanks.

I wonder if some entry limitations would make the entire databases for all cars more realistic. For example, requiring that a minimum of three tanks be entered, and not accepting total miles without the tanks which correspond to them. It crosses my mind that "bashers" of a particular car could intentionally enter false data on fictitious cars to make them appear to be very poor at gas mileage.

Just my two cents worth. I try to keep all my calculation as exact as I can. I enter the gas and miles to the decimal point that I have - I don't round off tenths etc. My gallons of gas go to the thousandth, and miles to the tenth. I DO round the subsequent calculations up or down to the closest integer.

I wonder if other hybrid owners have seen this?

Re: Mileage Database Glitches
 

Since the upper and lower quadrilles are dropped, the average comes really from the middle, 50%. The outliers would have to do something to make that big of an effect.


One of the things I like most about GreenHybrid is the lack "bashers" arguments between advocates of different model cars. Although not impossible, it would be difficult to find someone trying to screw-up the database by entering false records.

I've looked at some of the Prius I outliers and tried to help them do better. But I've yet to find any change to the calculation method that would really make a difference. For example, I used the Prius I database and cut out all entries with less than 5 tanks. The difference in the averages of the MPG were 45.8 to 46.1, less than a 1% change. I carried it out to a minimum of 19 tanks and the changes were very modest. However, I did notice a step function around 15 tanks suggesting that above the number, the low MPG folks tended to 'opt-out.'

Perhaps you might propose some rules and we could model it and see what the real data shows?

Bob Wilson

Re: Mileage Database Glitches
 

I posted this at the Prius section about a month ago - when someone else got bent out of shape about the same thing. Even if several "bashers" or "false hypermilers" posted to one model, the median would not move more than 0.1mpg. I like medians - they resist outliers...

What might be cool is if the dbase could show mpg trends over time by model. A median chart with monthly buckets would be nice!

-----------------------------------------------------------------------------------------------------

The database does pull the median (50% tile) of the owners averages. I ran a stats test in Minitab using the dbase. The results do show that there is a statistically significant difference in the median mpgs posted. Prius is higher by +1.40 mpg with a confidence interval of +0.90 to +2.00 mpg. Is this a practical significance - probably not. This dbase is a subset of hybrid owners who have found this site and have posted their mileage data.

Both vehicles seem to suck a little (gas), though the Prius appears to suck less. :D

------------------

Mann-Whitney Test and CI: MPG Prius, MPG HCH

N Median
MPG Prius 1155 47.500
MPG HCH 492 46.050


Point estimate for ETA1-ETA2 is 1.400
95.0 Percent CI for ETA1-ETA2 is (0.900,2.000)
W = 998602.0
Test of ETA1 = ETA2 vs ETA1 not = ETA2 is significant at 0.0000
The test is significant at 0.0000 (adjusted for ties)

----------------

Also interesting is the fact that there are 19 Priuses at 47.5 today (5/15/07 - of the 1155), so the median may be pretty insensitive to changes at high/low extremes - even new ones. There are about 6 HCHIIs at their median.

Re: Mileage Database Glitches
 

Bob,

Your analysis isn't exactly accurate. Not only are the top and bottom quarters dropped, but every car besides the median car(s) is dropped.

Re: Mileage Database Glitches
 

Ahh, using a granularity of 0.1 miles per gallon?

Bob Wilson

Re: Mileage Database Glitches
 

I'm not sure what you mean.

The median figure (rounded to the tenths) is selected. And the main figure given on the holistic data pages is rounded to a whole number.