In recent posts I’ve outlined how we plan to use our Reuters/Ipsos survey (http://goo.gl/GbM7hX) data to forecast turnout for the 2016 presidential election (http://goo.gl/47vYBG) and shown where we stand – with our way too early estimate – as of May 2015 (http://goo.gl/jUWYh7). In this post, I’d like to bring in some other proof points, expand our analysis a bit and update our turnout projections.
New Proof Points for the Ipsos Turnout Model
The earlier analysis looked at our data vs. state turnout levels in the 2012 and 2014 elections. It showed a substantial correlation between our data and actual turnout and it lets us construct a model to predict turnout. Going further, can we bring in additional external data points to improve this modeled relationship between our data and turnout? Yes we can. The Census Bureau (http://goo.gl/JvMu7j) collects information on voting in its massive Current Population Surveys and provides detailed demographic data on the voting population. The most current data available from the Census is for the 2012 election, so I’ll focus the rest of the analysis there. Below is a chart showing turnout levels (of the total 18+ population) by various demographic groups.
Expanding the 2016 Turnout Forecast Analysis
As I’ve shown in earlier posts (http://goo.gl/hNdhcu), the profile of those who show up to vote has a huge impact on which candidate is going to win or lose the election. Equally interesting, understanding who is and is not voting starts to suggest why a candidate wins or loses by diagnosing the groups where they have strong or weak support from. Looking at the gender, age, racial and geographical makeup of the potential electorate can give us an early sense of who will be voting next fall.
However, before looking at 2016, lets first examine our past data collection and see if our data is in the right ballpark. Below I’ve charted what the October-November 2012 Reuters/Ipsos Poll data projected turnout would be vs. the actual turnout levels from the Census.
So our projections were not perfect, but they are not bad either. We over-estimated the overall turnout by about 2 ½ percentage points. Most of that overage appears to come from three main sources: young people (18-34 year olds), men and Hispanics overstating their intent to vote.
Our Hispanic estimate was off by the largest amount but is also probably the easiest to explain. We do not typically interview in Spanish meaning the Spanish-language portion of the Hispanic population is not represented in our survey sample. Census does not have voting by preferred language but they do have voting by citizenship status – which we can use as an imperfect proxy for English-language use. Here Census shows 48% of Hispanic citizens voted in 2012 – much closer to our estimate of 44.9. We’ll need to work on better estimating Hispanic turnout, but this is a good start.
2016 Turnout Projection Update – June 2015
Again with the caveat that this analysis is very early and most of these data points are almost certain to change before the 2016 election, here is our current projection of turnout next fall along with last month’s projection and the actual 2012 turnout.
Our turnout projection has moved up about 2 ½ percentage points since last month indicating the volatility of these early estimates. With more candidates officially entering the race, it’s not surprising that people are currently more engaged.
Now for the interesting bits; what does our model predict for 2016 by demographic characteristics? I’ve charted that below against the 2012 projections. I’m using the 2012 projection instead of the actual so we can have an “apples to apples” comparison. Just keep in mind that there is some variance in these estimates compared to actual turnout.
Our model indicates that about 18 months before the election, enthusiasm for voting is less pronounced than in 2012 across all demographic groups. This gap is especially pronounced among Gen Xers (35-54 year olds) and African Americans, but is also apparent among women and Southerners. I’ll leave discussion of why these gaps currently exist to others but I want to note that we do not yet know the baseline trend for this set of data. It is entirely possible that these groups do not engage until closer to the election and we would have seen the same data points 18 months before the 2012 election.
We’ll continue tracking this over future months to see how the forecast changes.