Date: Sun, 02 May 2004 12:00:02 +0100 From: Johannes H Andersen <johsnospamfitter_n_o_s_p_a_m_.com> Subject: Re: Fuel consumption figures computer wrong
SAABurger wrote: > > You are right. The accuracy of the device is completely unstated. This point > stimulates me to make two related points > > 1. On Precision and Accuracy in general > In science and engineering there is a general principle that measurements > (and calculations therefrom) are presented to a Precision commensurate with > the precision of the data. For example, if you measure the heights of a > class of children to the nearest centimetre, then the resulting data should > be presented as 123cm, 129cm, 122cm etc. Presentation as 123.0cm, 129.0cm, > 122.0.0cm would be considered misleading as this would imply a measurement > precision of 1mm. In this light, I rather suspect that SAAB's display > precision of 0.1 units (imperial-mpg, US mpg or l/100km) is somewhat > inappropriate. When measurements of heights are made to nearest centimetre, you're pigeonholing the data into classes, hence you can't infer any probability of a height of 123.5 because such measurements are not defined. What you have is a discrete distribution on the classes. This is not the same as saying that the number of display digits is inappropriate. Clearly, the digital display takes out the subjective conversion into numbers, hence we can get a small edge by keeping the numbers intact, even when we know that the individual measurement error is larger. Reporting a single readout with all the digits may be meaningless, but meaningful if it is used in a statistical sample. I suspect that most of the error is bias. If that's the case then relative measurements may still be useful for showing the results of e.g. changes to a particular car rather than comparing two cars.
