If you’re like me, you’ve been inundated with stories, conversations, and anxious questions about how this year’s elections will turn out. A lot of them, perhaps most, are poll driven. I have believed for months (and everything I’ve seen recently confirms my belief), that the results of this election cannot be known until next Tuesday — and that anyone who says they “know” exactly what will happen is deluding either you or themselves.
Yes, the Democrats will lose seats in the House, the Senate, and the Statehouses. That much the numbers make clear. But how many seats they will lose is not only unclear — it’s unknowable. Here are four key reasons why:
1. Voters haven’t decided who they are angriest at. They are angry — with good reason. But there is a key group of voters who tell pollsters that they prefer the approach the Democrats are taking on the issues, have more positive feelings about the Democratic Party, but prefer to vote for a generic Republican for Congress. Most pollsters are emphasizing the importance of the answer to that final question. When that answers conflicts with the party that a voter likes better, though, that voter has almost certainly not made up their mind firmly. They will be strongly influenced by whatever they read or hear between now and the moment they cast their vote.
2. Most pollsters can no longer reach a statistical sample of the entire population. In particular, the most common polls done for individual Congressional and Senate races, which don’t involve live callers but robo-calls, cannot reach the increasingly large percentage of the public who have cell phones but not land lines. Pollsters attempt to weight their samples to account for this bias. It’s an important one because “cell-phone only” voters tend to be more Democratic than voters with land lines. But the problem with any poll which has to be weighted is that the pollster must assume that previous patterns still hold — any new trends among the under-weighted part of the electorate cannot be detected.
3. Who is going to vote? Almost all of the polls this year have shown that registered voters are more heavily Democratic than whatever subset of voters a pollster includes as “likely.” But in reality there are actually three groups of voters in advance of an election: certain or almost certain to vote, very unlikely to vote, and potential but far from certain voters. Some of the people in the potential category may by 30 percent likely to vote, while other may be 60 percent likely to vote.
Note that I didn’t refer to any of those groups as likely voters. When the votes are actually counted, they will come from two of these groups: almost all of the first group, the certains, and a significant portion of the third group, the potentials. (Very few of the unlikelys end up at the polls.) A pollster thus has the challenge of figuring out how to define “likely” voters so that the sample surveyed reflects the same mix of certains and potentials as will show up on Election Day. Bluntly, this is guesswork. The kinds of screens that pollsters use to decide who to include as a “likely” voter actually don’t allow for building up such a voter universe. This is exacerbated when the segment of the electorate that most pollsters can’t call (cell-phone only) also includes a very large proportion of the potential but not certain voters. It happens that low income, minority, and young voters are more likely to be potential but not certain voters, cell-phone only, and Democratic-leaning.
Now I imagine pollsters will complain that I am underestimating their ability to perform mathematical tricks to compensate for these challenges. During each election cycle I suppose it’s true that at least one pollster lucks out and builds its sample of likely voters in a way that ends up accurately representing the actual voting electorate — or appears to. The problem is that it’s not always the same pollster. So this is a bit like astrology — if you consult enough horoscopes, one of them will be pretty accurate, but you can’t predict which one.
If pollsters could spend enough money on large enough samples, I suspect they would report their results with much more confidence — but they can’t, and they don’t. When huge numbers of races are “too close to call” or “toss-ups,” the statistical problems with polling simply swamp the predictive power of the polls.
4. Finally, campaign ads matter, get-out-the-vote efforts matter, and the news stories that break in the last few days before someone votes matter. It used to be that all of this came down to the last week, because most votes were cast on Election Day. There are still states like Ohio where this is true, and where if the news cycle were to shift in a decidedly Republican or Democratic direction for the next five days, it would make an enormous difference. But there are also states like Oregon where all or most of the votes are cast by mail during the month before the election. And, of course, in many other states somewhere between a quarter and a half of the votes are cast well in advance of Election Day, either by absentee ballot or through early voting. Pollsters really have no way (given their small sample sizes) to adjust for the fact that their likely voter sample includes a varying number of people who have ALREADY VOTED as well as the remainder who can still be influenced by events. And, needless to say, no one can predict how these Election Day voters will be influenced by events that have not yet happened. Your neighbor, your daughter, your colleague at work may have been ranked by the pollsters as not likely to vote — but if you take that person to the polls, their vote will count.
What this all means is that, at the end of the day, WE matter more than the polls. Because we are part of that mix of events that have not yet happened and which will shape both WHO votes and HOW the crucial segment of undecided voters breaks. Elections are still like old-fashioned battlefields. Typically, close elections are won by the side that keeps trying the hardest, and are lost by the side that loses heart and runs away.
It’s still in our hands.