It has become clear that the social media data that I used in my previous forecast became antiquated quite quickly, due to the relatively large margin between the South Carolina estimate and actual result.
Thanks to a good friend’s tip, I began to utilize data from Google Trends to develop new estimates for Super Tuesday. Looking in retrospect, the correlation between the Google data and results in the first four states is striking once one has massaged the data set to produce usable measures (I use relative search frequency averaged over the couple days prior to the election). Demographics are taken into account as well. The new estimates are as follows:
This produces an estimated delegate allocation as follows (super-delegates are not included in this table):
Once we have the data from Super Tuesday, we will all be able to hone the accuracy of our models even further.