Sometimes our biggest mistakes teach the greatest lessons. Pundits, pollsters, and reporters in Kansas all learned an important lesson about our state’s electorate in 2014. August’s primary elections strongly suggested an electorate in an angry mood, ready to boot incumbent candidates.

An unknown gubernatorial candidate who spent pennies on the dollar took 37% of the vote. Subsequent polls were no kinder, showing both Governor Sam Brownback and U.S. Senator Pat Roberts in deep trouble. The narrative throughout the campaign was that Roberts had been in D.C. too long and Brownback had cut taxes too deeply. Kansas Republicans were supposed to have a bad Election night, even as things were looking good for the GOP nationwide.

Once the votes were counted, it was obvious we had all missed. Kansas Republicans did not have a good night, they had a remarkable one. Brownback and Roberts won, despite close races. Every other federal and statewide Republican in Kansas cruised to victory, and Republicans even picked up more seats in the state House. The pre-election narrative was destroyed, but why? Lay some blame before the pollsters. As data pundit Nate Silver indicated, polls across the country were skewed as much as eight points towards Democrats. Polls have a tougher time predicting close races, and having multiple close races may have driven uncertainty and electoral volatility.

While polls are good predictors of elections, they are by no means perfect. And new developments in campaigns have made polls less relevant and helpful in determining winners prior to votes being cast. The television age encouraged candidates and parties to mobilize segments of the population instead of the local mobilization that hallmarked the first century and a half of American politicking.

Since 2004, when Republicans amassed a national voter database to identify groups of voters they could activate, local mobilization has returned to prominence as a vitally important element of winning elections. Over the last ten years data-driven efforts at individual-level mobilization have been adopted by political electioneering organizations. Those organizations have refined and improved their voter databases since 2004, notably by the Obama presidential campaigns. Kansas Republicans got serious about data in 2010. Exit polls for 2010 and 2012 showed GOP turnout at near-record rates of 55%.

For the 2014 campaign, Kansas Democrats also got serious about their database use. Democratic efforts at individual-level voter mobilization may have blunted the GOP advantage, as the percentage of Republican turnout in 2014 dropped to 49% of all voters. Both Paul Davis and Greg Orman should have benefitted from the equalization, though. Sub-50% turnout should have helped them as well.

The key to understanding why election returns looked so different from the polls may lie in the electorate itself. Fort Hays State University’s Docking Institute of Public Affairs conducts an annual statewide poll, breaking down partisan identification into seven categories.

Since 2010, the Kansas electorate has become slightly more volatile, and that volatility likely explains the changing poll results and their variation from the results of the 2014 elections. Independents and those only casually leaning towards one of the two major parties have increased from two-fifths to half of the Kansas voting public. Independents have been shown, according to Hillygus and Shield’s The Persuadable Voter, to be more susceptible to campaign messages and volatile in their vote decisions. So as the campaigns ramped up, persuadable voters took in those messages and based on their evaluation at the moment could have supported one candidate one week and another the next.

Close races and skewed polls made predicting winners difficult, but late-race individual mobilization of an uncertain electorate, invisible to the watchful eyes of reporters and analysts, made for an election full of surprises.

Chapman Rackaway is a Professor of Political Science at Fort Hays State University.