In every election cycle, some gain a reputation for accuracy and some gain a reputation for being garbage. So far in the current election cycle, some have been good enough, many from poor to bad, and a few with a “garbage” reputation haven’t too shabby. One exception to that latter point. Loras College.
Actual Iowa Caucus Results: Clinton 49.8%, Sanders 49.6%, and O’Malley 0.5%
In 2014 I read through the detailed polling report issued by Loras on the IA senate race. They had the science of polling and statistical analysis down pat. It really looked like a very high quality pollster. Except a monkey throwing darts at a board could have been more accurate if one were interested in projecting the results of that election. The last Loras poll of the 2016 IA caucus reported: Clinton: 59, Sanders: 30, and O’Malley: 7 with a MOE of 3.1 (the lowest among the pollsters and a function of the “objective” care they put into their polling). Needless to say, I ignored Loras polls in the past year.
“The gold standard” in IA is DM Register. Their last poll results were: Clinton: 45, Sanders: 42, and O’Malley: 3 with a MOE of 4.0. As the results for Clinton were outside one MOE and Sanders almost two MOE, that’s more like a lead standard.
Not to dump too much on the IA caucus pollsters, it’s more difficult to poll than primaries, but some with large MOEs (which make them look like worthless polls) did better. Quinnipiac had Clinton 46 and Sanders 49 (MOE 5.9), NBC/WSJ Clinton 48 and Sanders 45 (MOE 4.7), CBS/YouGov Clinton 46 and Sanders 47 (MOE 8.9), ARG Clinton 45 and Sanders 48 (MOE 5.0), ISU/WHO TV Clinton 47 and Sanders 45 (no MOE reported). Q did well in 2014 in a few general elections and as this might be it’s first IA caucus polling, it should be taken seriously.
The other pollsters had Sanders scraping 40% and some had Clinton above 50%. There is an explanation for why Q and others got closer to the mark for Sanders and were short on Clinton’s numbers and why the others were too short for Sanders and Loras was so dreadfully off. Sampling and LV screens. DM Register and Loras sample populations skewed more traditional and overlaying LV (prior caucus voter) exacerbated the sample skew. Q managed to get a better random sample but needed to adjust for an under-representation of “traditional” voters in its polling methodology. (Building in a second choice question would allow the statisticians to move the small change around.)
The caucus “base” voters (non-first time voters, near 105,000 in the past three competitive caucuses, and what the IA DEM machines deliver) is reliably near 80% for institutional/establishment candidates. Clinton received 78% of that vote. Being such an exciting candidate, an additional 5,000 first time voters showed up for her.
New Hampshire
Actual primary results: Sanders 60.4% and Clinton 38.0%.
The “base” NH DEM primary vote is twice that of the Iowa caucuses (NH population 1.3 million, Iowa population 3 million). Near 210,000 and first time voters are a smaller factor. A 2.5 MOE (random sampling error/fudge factor) is what pollsters of the NH primary should have worked to achieve in the last month before the primary. None did. So, I’m going to apply 2.5 points over/under to their results for both candidates to evaluate their performance.
One passed:
NBC/WSJ (2/3) Sanders 2.4 under, Clinton 0 over/under
Those that came closest to passing were:
CNN/WMUR (was told that it’s garbage) Sanders 0.6 over, Clinton 3.0 under
CNN/WMUR (2/6) Sanders 2.4 under, Clinton 3.0 under
UMass (tracking 2/6) Sanders 3.4 under, Clinton 2.0 over
CBS/YouGov (1/21) Sanders 3.4 under, Clinton 0 over/under
Most of the polls were under for Sanders and it was closer to 50/50 overs/unders for Clinton.
Looks as if NBC/WSJ and CBS/YouGov bested the competition in both IA and NH. (Might have to take a look at how soon they got there and their polling methodology.)
Only three pollsters even attempted to poll for the NV caucus. Gravis (poor reputation) got the actual results in its last poll but it was so far off the mark in December that CNN/ORC seems to have done a somewhat better job with its October and February polls.
SC GOP — just because the dataset comparing polls to results is very small so far — they all got it right that Trump would win. The pollster that got the shape of the results most correct was PPP (2/15). (Monmouth must have gotten lucky in NH. None of them got it right in IA or even came close to getting it right. A bit strange because it should be easier and not more difficult to poll for the GOP IA caucus.)
For the upcoming SC primary, all of the pollsters will either earn high marks or all of them will fail. For the most part, they have Bernie scraping along with about a third of the vote and Clinton as beloved there as Sanders was in NH. The only thing that I’d put any money on is that if Clinton matches Sanders’ performance in NH and Sanders matches Clinton’s in NH, he’ll be called LOSER, LOSER, LOSER! Far and wide which for some strange reason wasn’t what greeted Clinton after NH.
UPDATE (2/24/16) – Nevada
Not going to critique the NV GOP pollsters because there weren’t enough to bother with. What is curious (and haven’t seen commented on by others) are the entrance polls.
dKos (FP)
For what it’s worth, the entrance polls (which always change put the race at Trump 46, Rubio 25, Cruz 20, Carson 5, and Kasich 3. Can Kasich make it all the way to Ohio’s primary on March 15 after finishing behind Ben freaking Carson? [emp added]
The Nevada result was called at 9pm local time by the Associated Press. By 2.30am, when all precincts had reported, Trump had a remarkable 45.9% of the vote.
Marco Rubio and Ted Cruz, the two senators best placed to challenge Trump, battled it out for second place, with Rubio on 23.9% edging Cruz, who got 21.4%. [Carson 4.8% and Kasich 3.6%]
Were the GOP entrance polls in IA, NH, and SC as accurate as the ones in NV? (Maybe someone would like to take on the task to check that out.) The entrance polls that changed were for Sanders in IA and NV. Why would that be? Sort of like what I recall from the general election in FL 2000. Just something to ponder.