Bike Computer

Most bike Computers allow you to callibrate them to the millimeter, don't know if you need to be more accurate than that. Of course your computer is going to be no more accurate than your calibration, so make sure it is set for your wheel circumference and tire width, or the actual rolling distance.

Car odometers are notoriously inaccurate.

The width comes into play if you use a predetermined value rather than measuring your self. For instance, here are the predetermined circumference settings that Cateye gives for various widths of a standard 700 road bike wheel:

Firstly, don't trust your car readout, 10% accuracy from a car speed is unusually good, and many of them are calibrated to over report speed (and depending on the cmoplexity off the ECU, distance as well)

Second, where you place the magnet and sensor doesn't matter, provided the sensor ALWAYS "sees" the magnet as it passes.

The importance of the calibration is that the computer assumes that every time the sensor ir triggered the wheel has done 1 rotation.  The number you put in uis the distance the bike travels for 1 revolution of the wheel, so the circumference of the tyre basically.

I calibrated mine by riding a course I knew was 10 miles long exactly (in the UK the cycling federation has course instructions for road time trials which are well marked) and then did the maths to work out the correction from the setting initally in the computer to get to the correct setting.

If you're computer is accurate to 0.2miles in 20 I wouldn't bother fiddling - whether you do a 40 mile or a 39.6 mile training ride isn't going to make a huge difference to your training.