This letter:

http://www.stat.columbia.edu/~gelman/research/published/kana...

effectively shows that the original paper ignored endogeneity, mis-interpreted coefficient estimates and data-mined.