Calling balls and strikes is difficult when you’re partially blind.
That’s the situation Senate prognosticators are in when it comes to this year’s races. In the early going, there just isn’t a lot of good polling data out there to understand the playing field.
Democrats need a net gain of two seats to pick up control of the Senate. CNN rates 11 Senate races as either competitive (i.e. leaning towards one party) or as a toss-up, including 3 Republican-held seats and 8 Democratic held seats. Most of these seats have very little non-partisan polling for them.
In the median race, there have been just two non-partisan polls of the expected or actual matchup occurring that have been released publicly since the beginning of the year. Now in some races there have been a ton of non-partisan polls such as Florida (12) or Missouri (7), but in other races there have been zero.
While a number of key Senate races haven’t been polled at all this cycle, every single competitive race had at least one poll taken in it by this point in the last midterm cycle in 2014.
We really have little idea, for instance, of the true chances that West Virginia Sen. Joe Manchin has at defending his seat. Any judgments we have of races with no public data relies on internal polling, the partisan leanings of the state or data (such as approval ratings) from mostly untested polling techniques. We already saw in the West Virginia Republican primary earlier this year how internal information isn’t always reliable.
More worrisome is the lack of high quality polling information from these Senate races. Only 2 (Florida and Tennessee) of the 11 races (18%) have gold standard polling. That is, pollsters who are non-partisan, use live interviews and call cell phones and are transparent about their data. Only one state (Florida) has had more than one gold standard poll taken in it. In 2014, 67% of competitive races at this point had been polled by gold standard pollsters.
While other polling data is certainly useful in understanding the general broad outlines of where these races stand, they are less precise on average than gold standard polling. In 2016, for example, gold standard Senate polls taken in the final three weeks of the campaign were about two points more accurate than non-gold standard polling.
Now, it could be the case that we don’t need this early data anyway. As I previously wrote about, we’ve been able to predict Senate races with increasing accuracy just by looking at the previous presidential vote. (This includes after controlling for what the early Senate polls say.)
However, there is no guarantee the 2016 presidential vote patterns will be as predictive of the 2018 Senate outcomes as the 2012 presidential vote patterns were as predictive of the 2016 outcomes.
During the last Democratic wave in a midterm (2006), presidential vote patterns were not at all predictive of the Senate outcomes once we took into account the early polling data. It could be the case that without Barack Obama as president, the relationship between presidential vote patterns and Senate voting recedes.
This year we’ve already seen Democrat Phil Bredesen polling ahead of Republican Marsha Blackburn in Tennessee, a state Republican President Donald Trump won by 26 points. Democrat Claire McCaskill seems to have a slight lead over probable Republican opponent Josh Hawley in Missouri, even though Trump won there by nearly 20 points.
It is entirely possible that Democrats are leading in the states they need to in order to gain control of the Senate. On the other hand, it’s entirely possible that Republicans are on the verge of expanding their majority because they are beating Democrats in the competitive states that are currently held by Democrats.
The truth is that we don’t know.