How accurate are the Estimated Dynamic NTRP ratings?

A question I get periodically is regarding the accuracy of my Estimated Dynamic NTRP ratings.  They are just an estimate, but feedback I get from folks that look at them is that they are quite accurate.  When chatting with an opponent at a recent mixed match, I was told it was spot on in predicting early start bump ups.

But a recent comment to my post on the availability of Estimated Dynamic NTRP reports pointed out that some players were bumped up to 4.5 that had estimated ratings below 4.0 and thus shouldn't have been.  So I thought I'd take a look at that exact list to see what I could learn.

There were 19 women in Northwest Washington that were on the 2012 early start list that were bumped to 4.5.  Of these 19, my current ratings show 12 having an estimated rating at or above the 4.0 threshold, and 16 of the 19 at or above 3.95.  Given my ratings are an estimate and 0.05 is probably an acceptable margin for error, predicting 84% of the bump ups isn't too bad.

Of the 3 that were not predicted, one had a very good record but played with some good partners which can hold ones rating down, another had a good record and won some playing up at 4.5, but also lost some she shouldn't have at 4.0, and the last hovered right around 3.9 but had a couple bad One Doubles results that weren't used for early start ratings which could explain the difference.

Also, one of the players started as a self rated 3.5.  My ratings both predicted their DQ at 3.5 and the bump up to 4.5.

I'll try to take a look at others to see how the early starts were predicted, but 80+% isn't too bad.