Astroprof's blog is full of teaching moments derived from whatever is going on in universe at the moment. As someone who teaches astronomy classes, he knows how to explain what's going on in an accurate but understandable way. A case in point is his current post on widely-discussed asteroid 2007 WD5 which will pass close to (but most likely not hit) Mars on January 30.
His explanation of what it means to say that there is a "4% chance" (or whatever) of the asteroid colliding with Mars is great, complete with very helpful diagrams. I won't repeat it here, but I would like to comment on what measurements can be used to determine the orbital paths of asteroids and other solar system objects. In the not-too-distant past, the positions of solar system objects were determined by measurements on photographic plates. If the object is close enough and moving fast enough relative to the Earth, a series of photos taken over days, weeks, or months will show a shift against the known background stars. Multiple 2D points can be mathematically fit to equations from which the orbital elements can be calculated (see P.S. below), but if the angular shifts are small, and if the object is not very bright, the relative error in the measurements can be pretty large, leading to the orbital position uncertainties that Astroprof describes.
Nowadays the "plates" are likely to be data from CCD arrays, and such optical data can sometimes be supplemented by radar data which can give not only angular position, but also velocity data derived from Doppler changes in the frequency of the return signal. But there are still errors in these data. Multiple measurements can be used to average out the errors and calculate more reliable elements. Measurements always have errors, but by understanding the errors, and by combining multiple measurements in the right way, scientists and engineers can get a good handle on the reliability of predictions made with the measured data.
P.S. Out of curiosity I searched for programs that can use astrometric observations and determine orbital elements from them. There are probably many others, and I haven't used any myself, but two free programs that look interesting are Find_Orb (which has a compiled Windows version for download) and Orb_Fit (which seems to be more Unix and do-it-yourself oriented, but perhaps more general). Another freeware one that looks great judging from screen shots and sample animations is ORSA, though it seems to be "some assembly required" to install needed data files. It seems to do a lot of cool stuff, not only orbit determination.