Wednesday, July 22, 2009

Accuracy of the Model of the 2010 USN&WR Law School Rankings

I earlier offered a snapshot comparison of the scores generated by my model of the 2010 U.S. News & World Report law school rankings and the original. After Robert Morse, director of data research for USN&WR, asked me if I could quantify the fit between the two data sets, I realized that others might share his curiosity. Here, then, are the r-squared measures (more precisely, the squares of the Pearson product moment correlation coefficients) for each of the models I've done over the past few years:

2010 rankings: 0.999
2009 rankings: 0.999
2008 rankings: 0.999
2007 rankings: 0.997
2006 rankings: 0.995

What do those numbers mean? In brief, an r-squared closer to 1 (or –1) shows a closer fit between the two data sets. It might seem a bit absurd to report these results out to three decimals, but I wanted to make clear that the model has yet to obtain results absolutely identical to those reported by USN&WR. I daresay, though, that any r-squared above .99 shows a pretty strong correlation.

[Crossposted at Agoraphilia and MoneyLaw.]

1 comment:

Logan said...

"...r-squared closer to 1 (or –1"

r-squared cannot maintain a negative value. r can, but if it does once squared becomes positive.