++Addition2++In the 15 states that have gone since March 15, Trump has outperformed his RCP average in 14 of them, doing 5.4 points better than predicted. Indiana, the last competitive election of the nominating campaign, has been added to the table.
++Addition++The table in the body of the post has been updated to include the results from April 26 and May 3. Trump outperformed his RCP average in all five states, bringing the number of states he has outperformed polling predictions in up to 26 compared to 10 states where he fared worse than expected. Trump now averages 2.6 points better in actual results than in aggregate poll estimates.
Nate Silver continues to weep.
The stubborn belief that Trump has generally underperformed polling expectations came about after Iowa, a state where he in fact did do worse than the RCP average (which only includes polls taken less than two weeks out from the actual vote) predicted he would. He also fared more poorly than expected in Oklahoma and Kansas*. Excepting the cuck corridor, though, doing better than expected has been the rule rather than the exception.
Of the 30 states for which polling was conducted close to the preference vote, Trump did better than the RCP average expected he would in 20 of them and more poorly than expected in the remaining 10. Through New York, his average real performance shakes out to 1.6% more of the vote than polls predicted he would receive [Update: Through April 26, his average reals performance comes to 2.4% more of the total vote share than polls have predicted he would receive].
This tendency has been accentuated over the last month and a half. Of the nine states that have held their preference votes since March 15, Trump has exceeded RCP averages in eight of them, the only exception being North Carolina, where his polling average was 41.3% and his actual share of the vote was 40.2%. Even in Wisconsin he did marginally better than polls suggested he would. The story there was of Kasich's collapse, an implosion which redounded to Cruz's benefit.
The data are presented in the following table in descending order of actual performance vis a vis predicted performance:
Nate Silver's 538 continues to perpetually calibrate as his site misses the mark again and again (538's expert panel predicted Trump would get 71 delegates in New York; he picked up 90 of the state's 95). Silver uses what he calls a "polls-plus forecast" that tries to take endorsements and the inverse of national polling into account (see here for more details if you're so inclined, but the formula is garbage so I'd suggest you save your time).
On the Republican side, endorsements have been toxic. Initially, ¡Jabe! enjoyed a huge endorsement advantage. After he was gutted, Rubio claimed the endorsement crown. Then he got sliced up and Cruz, who is in hail mary mode, now has the dubious distinction.
Almost without fail--I can't recall seeing a single state where it's been otherwise, but I'm not going to dig through the archives to make absolutely certain--this "polls-plus" forecast shows Trump doing worse than the polling average alone does (see Indiana to get an idea of the usual gap; polls give Trump a 77% chance but polls-plus only a 45% chance).
This is in spite of the fact that Trump mostly outperforms polling averages! So the "polls forecast" slightly underestimates Trump's performances and then the "polls-plus forecast" underestimates them even more, yet Silver keeps putting more emphasis on the latter.
* I made several hundred dollars betting on the outcome in my home state. It's a closed caucus and it went for Huckabee in '08 and Santorum in '12. Kansas always goes for the theocrat.