Contrary to conventional intellectual wisdom there are not many good ones, but this packs some real force:
Starting from probabilistic simulations of likely presidential election outcomes that are similar to the output from election forecasting models, we calculate the likelihood of disputable, narrow outcomes under the Electoral College. The probability that the Electoral College is decided by 20,000 ballots or fewer in a single, pivotal state is greater than 1-in-10. Although it is possible in principle for either system to generate more risk of a disputable election outcome, in practice the Electoral College today is about 40 times as likely as a National Popular Vote to generate scenarios in which a small number of ballots in a pivotal voting unit determines the Presidency.
And note this, which explains a good deal of the debate and rationalizations — on both sides:
This disputed-election risk is asymmetric across political parties. It is about twice as likely that a Democrat’s (rather than Republican’s) Electoral College victory in a close election could be overturned by a judicial decision affecting less than 1,000, 5,000, or 10,000 ballots in a single, pivotal state.
That is from a new NBER working paper by Michael Geruso and Dean Spears.