Why Doesn't the Party with the Most Registered Voters Win Every Time
We took a look at "Likely Voter" versus "Registered Voter" modeling - and what data & technology those smart polling people use behind-the-scenes - so you didn’t have to!
With the November 3rd election quickly approaching - and millions of Americans in a state of collective exhaustion with political news - many of us have been bombarded with stats and analysis about the latest Presidential Polling numbers. What makes following the ups-and-downs so wild is partly due to the enormous variance in the results (some show Joe Biden beating Donald Trump by as much as 16 points, and one shows him losing by 1 point).
But why do polls differ so much from each other? Aside from surveying different voters - and often using creatively worded ways of asking the same questions - why don’t all the polls come to roughly the same conclusion? Why don’t they just count up all the registered voters from each party and call their predictions a day. Databases of party registrations are widely available - some free, some for purchase - and you can even FOIA any of the ~3,000 counties in the US for the latest party & voting data (assuming you have the time or the appetite to do it).
It’s actually a super interesting data integration puzzle.
So we downloaded one of these databases (so you didn’t have to!) to see if we could unpack it ourselves. The databases show every person in the US who’s voted, their party affiliation, and a host of other information about them and their home district. One thing was immediately clear: Some of the big battleground states (ex Pennsylvania, Florida, and even Texas) have FAR more registered Democrats than Republicans. So why don’t those states always go blue?
Party Registration Is Not Destiny
It seems like an obvious point, but people just don’t turn out to the polls at the same rates. Turnout rates can vary wildly by the demographics of the voter; whether it’s a presidential, mid-term, or local election; and of course by who’s actually on the ballot. Some voter polling relies heavily on party registration data (called “Registered Voter” modeling or “RV” for short) while others develop more sophisticated models to help predict turnout likelihood (called “Likely Voter” modeling or “LV” for short). Sophisticated LV modeling obviously has a big advantage over RV modeling. As is often noted in politics: Demographics is Not Destiny. Texas is a great example.
Texas Monthly ran a great piece recently describing Latino voting trends in Texas. As expected, even despite a large independent streak, the majority of Latinos are registered Democrats. But the degree to which they turn out and vote for Democrats varies along a couple dimensions.
Empowerment Matters: “Feeling empowered and having a general sense of belonging” was a big motivation to vote, but when “government seemed distant...many struggled to directly connect its policies and actions to their lives” and they subsequently failed to turn out.
Neighbors Matter: In addition, “nonvoters tend to be surrounded by other nonvoters. Whereas voters often discuss politics with other voters, believe they have a right to be heard, are better able to directly relate government policy to their lives, and believe they can influence political outcomes.”
Mobility Matters: Perceptions of whether they had opportunities to move up also affected likelihood to vote. “This may partly explain why the border regions and San Antonio—a city whose Mexican Americans have a long history of racial segregation and inequality—have lower Latino turnout than, say, Dallas or Houston, where there is more economic mobility.”
When we looked at the actual party registration data in Texas, we were equally surprised. The geographic split is extremely well defined, with Democratic clusters tightly aggregated around the border and in major cities. If 100% of registered Democrats & Republicans in Texas voted in every US Senate and Presidential election, Texas would ALWAYS go blue.
But it doesn’t.
The Real Clear Politics average for Texas is currently 2.3 points in favor of Trump (about 600,000 people if we’re talking RV here) and the FiveThirtyEight forecast gives Trump a 66% chance of winning. Trump won Texas in 2016 by 9 points. Romney won it in 2012 by almost 16. Texas does not have a history of being Democrat friendly.
It’s Not Just Where You Live, but How Old You Are
According to Pew Research Center, geography is an ultra strong predictor of vote choice “with urban voters breaking Democratic by about a three-to-one margin (73% to 25%)...Republicans had about a two-to-one advantage over the Democrats with rural voters” in both 2016 and 2018. In recent years, big cities tend to go blue and rural areas tend to go red. Said another way “the most Democratic areas tend to be heavily developed, while the most Republican areas are a more varied mix: not only suburbs, but farms and forests, as well as lands dominated by rock, sand or clay.” This is a New York Times quote for the record, not ours.
But age matters even more. According to US News & World Report “In the United States, the oldest citizens are the most likely to cast their ballots, which gives them political clout beyond their numbers alone.”
In 2018, voter turnout by age was distributed pretty clearly:
18 to 24 - 30%
25 to 34 - 37%
35 to 44 - 44%
45 to 64 - 55%
65+ - 64%
Urban areas have enormous concentrations of registered Democrats, but older voters are actually much more likely to turn out at the polls.
If you were going to build a “Likely Voter” model, you’d definitely want to account for someone’s age. You’d also want to account for whether they live in a big city, whether they went to college, whether they’re white, and whether they felt they had access to economic and political opportunity.
And you’d want to take into account that even though Pennsylvania has 700,000 more registered Democrats than Republicans, it’s two bluest most populous counties - Philadelphia and Allegheny (home of Pittsburgh) - had some of the lowest voter turnout rates in 2018. Whereas some of the reddest population centers like Snyder and Perry County had very high turnout rates.
Florida - oftentimes seen as a microcosm of broader US elections given its demographic makeup and the influence the state can have - has one of the oldest populations in the country. And older people consistently vote. Sumter County - home to The Villages a retirement community with a population of over 50,000 retirees and a median age of 67 - had a 79% turnout rate in 2018. In Miami Dade County - with a population of 2.7 million and a median age of 40 - turnout was 57%.
So to make good political predictions, and in fact to make any good marketing decisions where demographics matter, having access to data is not the only thing that matters. Pollsters and marketers alike need the technology to join disparate data sets together, and the skills to make sense of it all.
This is what we focus on at Latticework Insights. Reconnecting the dots that matter. If you’d like more information about this report please contact us at info@latticeworkinsights.com.
** All Data courtesy of L2 Political and the US Census American Consumer Survey, unless otherwise specified