TonyWilliams wrote:
Here's the list:

LEAF --- CapBars- miles-M/kWh-Volts ---GOM

Red429 --- 10 --- 71.8 - 4.3 - ----------74

Blue494 ---- 8 --- 59.3 - 3.7 - ----------56

Blue534 --- 10 --- 75.7* - --- - 315.5----74 (ECO=84) (*Data edit 75.7 for typo)

White530 -- 10 --- 69.7 - 4.0 - ----------73

White272 -- 10 --- 66.1 - 4.4 - ----------68

Red500 ---- 9 ----73.3*- 4.4 - -342.5*---66 (*No turtle; 2 miles >VLB: Added 4 miles)

White626 --12 ----73.5 - 4.3 - -317.5----73 (CapBars were 10, reset 12, now 11)

Blue842 ---12 ----79.6 - 4.1 - --------- 76

Silver679-- 10 ----71.8 - 4.2 - -303.5--- 75 (18.2 miles after LBW)

Blue917--- 10 ----72.5 - 4.1 - -310.5 ---67

Black782-- 12 ----76.6 - 3.9 - -295.0 ---88ECO (Out4.0/In3.8; LBW 6.9, VLB 6.5)

Blue744 ---9 -----72.3*- 4.4 - -352.0*-- 63 (*No Turtle; 1 mile after VLB; added 5 miles)

Notably the 8 bar car also had the lowest M/kWH, any idea why that is? If we extrapolate to e.g. 4.2 M/kWh, which is the mean for the other cars, it would have gone 67.3 miles, which makes it less of an outlier in terms of range as it appears. This would also lower the overall correlation of observed range with e.g. capacity bars or gids.

Given the variation in (reported) efficiency, one should actually consider the quotient of actual range (lets assume that these values are comparable, i.e. every car was indeed driven to turtle) of range measured divided by efficiency.

The you would get the following list

[Car] [Apparent capacity (=range/efficiency)] [normalized capacity = (apparent cp- <apparent cp>)/std(apparent cp)

Red429 16.7 -0.43

Blue494 16.0 -0.94

Blue534 18.0 0.57

White530 17.4 0.12

White272 15.0 -1.70

Red500 16.7 -0.46

White626 17.0 -0.13

Blue842 19.4 1.63

Silver679 17.1 -0.13

Blue917 17.7 0.31

Black782 19.6 1.8

Blue744 16.4 -0.63

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

mean 17.3 std 1.3

So all tested cars (with this small sample size), fall within 2 standard deviations of the sample mean, so technically, no outliers there. If we now had results for supposedly healthy new batteries (e.g (e.g. for at least 12 (ideally 30 or so) brand new leafs) under the same conditions, we could actually tell which of the tested cars had

*significant * degradation. If we assume that 19.6 apparent capacity (Black782) is representative of the mean for a healthy battery, and we have the same variation as in our sample of 11 bad cars then we have

Red429 -2.59

Blue494 -3.18

Blue534 -1.42

White530 -1.95

White272 -4.07

Red500 -2.63

White626 -2.24

Blue842 -0.20

Silver679 -2.24

Blue917 -1.72

Blue744 -2.83

Which shows that 7 out of these 11 are below 2 std, i.e. are significantly degraded with respect to Black782.

Right now it actually appears that white272 is the worst case (despite a mere 2 bar loss).

Still, since we compute a quotient of two very noisy variables, the error on these values is actually going to be quite high. Also, the sample size is really small....What we really need are ~ 30 new cars tested on the same track.