Reading the Polls: how to home-brew political hot-takes
A few weeks ago, the polls were suggesting that Trump was suddenly even. Wow, how this has changed. Should you tune into your cable news outlet to see some talking heads say words at each other? Naw. You should make your own news. It really isn’t too difficult to cook up evidence based “hot-takes” that are more interesting and more accurate than those you see on television.
How do you read the polls? What matters and what is “inside baseball?”
The polls you want are those that map the algorithm of the election. Results in this country tend to be mapped to individual locations and groups, not to the country as a whole. If you are younger than 30 he 2000 election may seem like the distant past to you. Summary: Gore won the popular vote nationally and lost because Bush won the electoral college. Hayes and Harrison also won despite losing the popular vote and John Quincy Adams was installed by the House of Representatives. There is the relatively simple question of who won the most votes, but then the larger structural question of which votes are actually being counted. So, should you care about top line national polls? Sure, they are an index of excitement, but not necessarily an accurate map of where the election is going.
Some polls are better than others. A number of academic and popular sites track this sort of information. Don’t get too worked up over an individual poll. Sam Wang and Nate Silver aren’t magical horses, they are aggregating and weighting a number of polls. Some of the important factors here include the quality of the pool and the recency of the poll. A decent poll from today is worth more than a great poll from a week ago. A bad poll from today on the other hand might not be worth more than the older great poll. Aggregated models can provide a smooth approximation of the current poll position. Look for a poll of polls.
The best models will combine polls of polls and map those to the election process itself. Bayesian statistics are important here as they evaluate the relative change in probability over time for an event. This isn’t your seventh grade coin-flip probability model. An election isn’t a magical coin that could go any way on any day, it is a real thing where the prior probability effects the current and future probability. A good model won’t change much from day to day unless a major event happens, like a party convention. If you like to be excited by the horse-race these models aren’t for you.
Which poll numbers matter longer run?
Look at the deeper questions asked by the polls. The ANES (an National Science Foundation funded poll) surveys with dozens of questions. These can tell you about the underlying affective conditions that make a candidate stronger or weaker. If you read the most recent polls, the surge for Clinton is expected (the DNC just happened), the inside baseball, the talk that the insiders, the people who really know, care about is the perception of the candidates personality. Personality evaluation is a strong model, which can predict the outcome of the election as this evaluation synthesizes a number of positions and perspectives. For the Trump campaign, the news that Clinton is at +57 on the question of presidential temperament, and Trump is at -63 is devastating. Negative evaluation of character can over-ride Party ID (the strongest of the classic predictors of voting), especially for Republican voters. These evaluations tend to be powerful and durable.
Take the next step. Which strategy are the campaigns taking? Are the candidates building themselves up or are they tearing each other down? Are they focusing on policy positions or personal attacks? Are there secondary ballot issues or local factors that over-ride synoptic campaign strategy? 2000 was a positive election compared to 2016. North Carolina has a major showdown brewing about HB2 which will drive voter turnout and potentially sway the election in ways totally unrelated to national politics. In 2004, the presence of Gay Marriage amendments on state ballots primed the polls for Bush, although it would be wrong to say it was the only factor deciding the election. Every election has a theory, a short explanation that describes who is winning and why. You can easily evaluate the quality of these theories by squaring them with the polls and the model of the election you are studying. It would be fair to say that both campaigns are attempting to drive the negatives of their rival up, on the positive side Clinton focuses on institutional competence and Trump on outsider status. Despite the negativity of this cycle, every campaign position has both positives and negatives, take account of them comprehensively. Beyond the existence of polarities, candidates can effect their positions and their rivals in both ways. Trump might do something that makes Clinton look bad, he also might make himself look bad. Clinton could make herself look good, she could also make Trump look good.
Don’t get too paranoid, but still question your premises. Last week there were plenty of hot-takes that supposed that the DNC hack was actually good for Trump because all publicity is good publicity. Turns out that for someone with extensive public brand awareness, bad publicity is bad publicity. This is where your home-brewed hot-takes can be especially good. Instead of getting caught up in the deep strategy of which master-minds are “leading” the fifth-dimensional chess game of the CNN feed, focus on basic, simple relationships that make linear sense. Avoid opposite day themes in your theories of the election. No, losers don’t magically win.
Formulating the hot-take:
What is the theory of the election?
What is the hot-take worthy event? Does it actually matter? It is perfectly legitimate to say that a non-story won’t “move the needle.” For example, negative stories about Clinton and coal won’t win her West Virginia, but was she going to win West Virginia anyway?
Does it fit with an even slightly rigorous reading of aggregated polls? If a poll from a company or site you have never heard of has a strange result, ignore it.
Can you formulate the explanation effect a deeper issue of candidate personality or position?
Example: Wow, Trump’s bluster has turned on him, his approach to driving up Clinton’s negatives is making him appear to lack the temperament to be President. His attacks on Clinton may actually hurting himself more than they hurt his rival at this point. Seems like it might be time for a new strategy.
Example: the electoral college is looking pretty clear at this point, Trump’s positive motion in Ohio and North Carolina seems to have stopped and Georgia may be in play. No one really should be that worried about courting Green party loyalists at this point, if anything they should be encouraged to turn out and vote.
There you go — free hot-takes that you could formulate yourself. Home-brewing is just for beer anymore.
