It was a crazy Super Tuesday in Oklahoma...

And, as we all know, the pre-election polls didn't fair very well.  Not just ours, but all of them except for one, which may have foreseen a Sanders win but still underestimated his double-digit win. According to 538's aggregate of polling in the Oklahoma Republican primary, Trump had a 68% chance of winning.

Polling in primaries has always been harder because a larger portion of the electorate can and will change their minds about candidates with similar views, while in general elections it is less likely to occur.

This year was no different. But still, there is a lot to be learned about this year's presidential primary, even if all of the polls were wrong.

First, let's look at our results.

Our polling methodology for Super Tuesday consisted of three different data collections among three unique samples, unlike any other polling performed in Oklahoma before Super Tuesday.  ALL of them showed a similar outcome in both the Republican and Democrat ballots.

We started with data from our internal database of identified likely voters with landlines and surveyed using, first, an IVR technique on the first night, and then live interviewers for two consecutive days and nights. At the same time,  an online survey was conducted using sample data from SSI's online panel of voters. The third one was a cellphone sample purchased from SSI as well, and fielded with live interviewers the same two days and nights as the landlines. The table below shows the results of each of the four data collection methods.

IVR (landline) Online Live Interviewer (landline) LiveInterviewer (cellphone)

Trump 31.3%, 42.0%, 27.6%, 38.0%, 33.5%
Rubio 21.8%, 16.5%, 20.8%, 22.1%, 20.7%
Cruz 20.6%, 19.9%, 9.7%, 12.6%, 18.1%
Kasich 7.4%, 3.7%, 5.4%, 1.2%, 5.8%
Undecided 9.3%, 8.03%, 3.31%, 3.6%, 12.9%
Sanders 30.1%, 37.0%, 21.7%, 38.1%, 30.9%
Clinton 41.6%, 39.4%, 31.5%, 44.2%, 39.8%
Undecided 28.2%, 23.6%, 46.8%, 17.7%, 29.3%

So, what happened?

The polling wasn't really wrong, it was measuring, in essence, where the candidates were AT THAT TIME in the week before the election. But in the final weekend, both races changed dramatically, something that none of the polling was capable of capturing.

Below is a chart of the absentee and early voting results for both the Republican and Democratic primaries.  SoonerPoll was in the field starting the Tuesday of the week before the election through Thursday, the first day of early voting at the county election boards.

Had the election ended with early voting, our polling would have captured the result: Trump with a lead over his two closest rivals and Clinton besting Sanders.  But, if there is one thing we've learned about this entire presidential election season so far, this isn't your typical presidential primary election, there were other factors heavily influencing voters' final decisions.

Absentee, Early In Person, Combined

Trump 3,563 32.4% 6,380 34.0% 33.4%
Rubio 2,557 23.2% 4,880 26.0% 25.0
Cruz 2,896 26.3% 5,297 28.2% 27.5%
Kasich 585 5.3% 962 5.1% 5.2%
Totals 10,999 18,772 29,771
Clinton 3,828 53.7% 7,919 48.0% 49.7%
Sanders 2,751 38.6% 7,743 46.9% 44.4%
Totals 7,131 16,510 23,641

What's interesting to note is the movement just in these numbers alone.  Cruz moved from third in absentee votes to second behind Trump among those voting early in person, and Sanders closed the gap with Clinton to within a point.  This movement alone shows us the trajectory of the candidates going into the final weekend and how, after our fielding had concluded, both primaries were heading to completely different outcomes.

Exit polling in Oklahoma also tells a lot of this same story with 40 percent of Republican voters saying they made up their minds in the last week before the election, and 25 percent of Democrats making up their minds in the last week.  While the exit polling only offers "within the last week" as its closest time to the election, we would argue these voters were most likely making up their minds just within the last few days of the election.

What this does tell us is that support for the candidates was mostly soft support and voters were open to changing their minds even in the last few days.

So, what had the most impact on the voters' final decision in the last few days?  Money.

Of all of the candidates, Cruz -- who won the primary -- spent the most in television buys, $233,375, through a Super-PAC supporting his candidacy.  The pro-Rubio Super-PAC, Conservative Solutions, spent $166,177 to support the Florida Senator’s bid for the nomination.  Meanwhile, an anti-Trump Super-PAC, Club for Growth, spent the most money overall, $430,628 for commercials airing on Oklahoma television.

Direct spending from the campaigns showed that Cruz outspent Rubio in the week before the primary in Oklahoma. The Cruz campaign purchased $36,080 worth of advertising, while the Rubio campaign spent $15,815.

Trump, on the other hand, didn't purchase any television time in Oklahoma.

Meanwhile, Sanders nearly doubled the amount Clinton spent in the state, $689,590 to $378,229, clearly demonstrating Oklahoma was a state Sanders targeted. In contrast, Clinton spent $585,510 in Texas where she won compared to Sanders who only spent a mere $31,880.

Oklahoma, as the strategy has now been played out, was to be a proving ground for a particular kind of Sanders voter -- the white, working-class Democrat who felt abandoned by the national party.  In the coming weeks, the same strategy would help him win Kansas and Nebraska, and almost Missouri and Illinois.

In politics, money has always made a difference.  Not every candidate who spends the most money wins, but candidates who don't spend anything are almost destined to lose.  This is particularly the case for Trump, who led, not just in our polls but all Oklahoma pre-election polls, but didn't spend a dime on local Oklahoma television.

Is it any wonder then that the winners of the Republican and Democratic primaries, Cruz and Sanders, spent the most time in Oklahoma the final week, had the best ground games, and spent the most money in television advertising in the final weekend?

What no one knows yet in the Democratic primary, is the effect of Independents being able to vote, for the first time ever in Oklahoma history, with Democrats.  It has been assumed to be a sizable portion of the turnout, perhaps due in large part to the fact that fewer votes overall (-19.6%) were cast in the Democratic primary.  If so, this means an awful lot of registered Democrats did not vote and the numbers could have been a lot worse had it not been for Independents.  In neighboring Texas, Democrat turnout was down more than half (-50.3%) and next door in Arkansas, the Democrat turnout was down 31.7%.  Neither state allows Independents to vote in the Democratic primary.  In the coming weeks, more election board data will be made available that will tell us for sure.

Turnout is where art meets the science of polling, and determining turnout, particularly Democratic turnout, is becoming more and more difficult in Oklahoma.  We had anticipated higher turnout for Republicans and lower turnout for Democrats, but the profile of the lower turnout among Democrats is not matching up with the model.

In Oklahoma, Democrat registration, as a percentage of those registered, is down and Republican registration is up, mostly from a migration of disenfranchised conservative, middle-aged Democrats with good voting patterns.  Over the last 20 years, these once-Democrats were joining with Republicans to give the GOP large majorities in both state house and senate and a sweep of statewide elected officials in 2010, at a time when Democrats still out-numbered Republicans in registration.

The remaining Democrats are younger voters with less prior voting behavior and difficult to determine in turnout models, liberal and moderate Democrats, and the older Democrats, who may still lean conservative but are set in their ways and less likely to change parties.

We can see this in the change in registrations from the state election board where just more than half of all migrations is Democrat to Republican.  The table below shows the changes that occurred in just the week before the presidential primary registration deadline on January 23, 2016.

Registration Changes 1/15/16 -1/23/16 (Was a…Now a…)

Republican Democrat 1,081 13.0%
Democrat Republican 4,247 51.2%
Independent Republican 1,233 14.9%
Independent Democrat 754 9.1%
Democrat Independent 440 5.3%
Republican Independent 546 6.6%

Total Registration Changes: 8,301

If we were to model lower turnout among Democrats in Oklahoma, the overall age of the electorate becomes older in the model, but this was not the case on Super Tuesday.  In fact, older, well-established voting behavior Democrats were sitting at home, and younger voters were making up a larger percentage of the Democrat electorate but are not necessarily turning out in larger numbers.

Polling this presidential election isn't just causing problems here in Oklahoma.  In Michigan where Nate Silver called it “among the greatest polling errors in primary history,” Sanders beat Clinton by 1.5 percent with the polling average showing Clinton by 21 points.  Some of the analysis is similar to Oklahoma:  not polling close enough to the election and underestimating Independent turnout, so the problems are not just isolated in one particular state.

In conclusion, the primary cause of why our polling missed is NOT because of our methodology. If this were the case, one or more of the three very-different samples we used would have shown varied results.  Better turnout modeling will help and further research in this area is needed, but the chaotic changes in registration in both parties, and voting behavior among Oklahoma Democrats is still in flux.

Timing of fielding in relation to election day played the most critical role, but polling just a couple of days out from a primary election and reporting the night before election day seems impractical and pointless. Nonetheless, all pre-election pollsters are feeling a similar gut check these days and fine-tuning the craft always remains a work in progress.

In order to provide complete disclosure and transparency of our work on this Republican and Democratic presidential primary, the entire SPSS dataset and supporting documentation, including all weighting tables, will and can be made available to all interested parties upon request.