These odds estimates are based on a Monte Carlo probabilities, where several variables are each part of the matrix.
If they ran 45,000 different scenarios for gravity, sun spot effect, solar wind effect, celestial bodies gravitational effects, the asteroid density, etc. they are saying that 1000 of them are coming up with Earth impacts in 2029.
The Monte Carlo simulation by JPL's Sentry System has the probability at 30% right now for this one asteroid based on 44 "potential impacts" between 2029 and 2092. I'm surprised the fear-mongering press didn't pick up on this number. MUCH more impressive than 1 in 45.
Consider that the same folks who dismiss 1/43 odds of an asteroid collision as just fine, will play the Lotto with a 1/17 MM chance, and consider those better odds.
Its a little more complicated than that, but roughly true. They have not been able to nail down some of the parameters that would allow them to get at some extremely precise orbital elements. They'll be hovering around 1:43 until they get some better sensors looking at it; right now, that represents the margin of error for the models based on a lack of important parameters e.g. the rotational period and axis relative to the sun. It isn't as though they can't produce exceedingly precise orbital predictions but that they do not trust some of the data they have so far. Some of the unofficial analyses on the same data that tries to adjust for dubious data or bad calibration and giving missing parametric data "statistically representative" values is putting impact probability at closer to 1:20. But we still need better data, which will require the asteroid to be visible in the northern hemisphere to a significant extent, something which will happen in a bit. Right now, the rock is hovering around the South Pole.