First, there's meaningless precision, where way too many significant figures are used, either through ignorance or in an attempt to make the numbers seem more credible than they are ("surveys indicate people will drive 10.7% more miles this vacation than last"). A lot of this is due to basic statistical ignorance.

Then there are things which can be counted fairly precisely, but using this excess precision clouds the point of the argument, as in "paid subscriptions to Magazine X are down by 11.97% this year, compared to last year." Sure, that number of paid subscriptions can be counted accurately, but why not just say "about 12%" to make the point of the discussion?

And finally, there are far-off predictions that come with precision. I saw this one in the article "Charger Mats Curtail Clutter, Not Energy Use" in The Wall Street Journal, about wireless recharging devices. It said "an additional 280 gigawatts of new generating capacity will be needed in two decades just to power all of those new cellphone chargers, cable boxes and high-definition televisions, according to the International Energy Agency."

Really? I'd be a lot more comfortable is it said "between 100 and 500 gigawatts."

It is an allegedly factual–but actually absolutely unprovable–prediction which is based on entirely on extreme extrapolation from some very rough data points. As all engineers know, the combination of extrapolation and a shaky starting point is a recipe for missing the target by a very, very wide margin.♦