Tuesday, July 24, 2012

Introducing the National Pulse Index

In the run-up to the election, the Naked Dollar will keep you up to date on its two political models, the Electoral College Forecasting Model, and the National Pulse Index. Today we will talk about the latter.

I used to run a quant hedge fund. What we quants do is take lots of data and make sense out of it. So, how to use this approach to capture the ebb and flow of an election cycle at the national level? Is the national mood swinging Democrat or Republican?

The approach I came up with two years ago is a simple metric, and the purpose is to use data - not the media or the water cooler - to track short term swings in political momentum. As it happens, it's a quick way to determine if the media narrative has any merit.

Essentially, the Index aggregates polls from everywhere, from presidential down to the House level. The idea is to pull in a huge and continuous sample, exploiting the wisdom of crowds. Specifically, the Index compares each new poll with the previous poll from the same race. The other day, for instance, Rasmussen reported that Sherrod Brown, who's running for re-election for the Senate in Ohio, is up by 4 points. While this may seem like good news for Democrats, the last poll actually had him up by 16. I assign this a "minus 12" for the Democrats.

Each day, as new polls come out, they are compared to the previous polls from the same race. The numbers are added up to get a net score for the day. Then, I use a moving average of the last five days to smooth out the data. This gives us a very good picture as to which side has the momentum.

Allow me to make my case for why this works, and then we'll get to the numbers.

1. On any given day, this will capture the aggregate opinions of up to 20,000 people, whereas an individual poll (that might get one side or the other excited) can have as few as 300 people.

2. There really is a national "mood," and it really does matter which way it's moving. Opinions are infectious. It matters in Ohio what people are thinking in Alabama. They may not be perfectly in sync, but they do tend to move in the same direction at the same time. Thus, if Diane Feinstein just went from 20 points up to 30 points up, it doesn't mean anything for Feinstein - she's still going to win easily - but it does mean the mood in California is moving in a certain direction. This makes it likely the mood is moving in the same direction elsewhere.

3. Aren't some of these polls partisan and/or sloppy? Absolutely, but the Index takes them all, because it evens out. For instance, not long ago a Bloomberg poll came out showing Obama up by up by 13 points, which was a crazy outlier, and resulted in a +10 score for the Democrats (since the previous poll had shown him up by 3). A day later, though, another poll came out showing Obama to be down by two, which resulted in a -15 score. Bad polls tend to get balanced out, in other words. To the extent that they don't, they probably have useful information, which is why one doesn't want to make judgments about which polls to use.

4. What about the Generic Ballot polls, the ones where people are asked whether they intend to vote for a Republican or a Democrat? Don't those show the national pulse? Yes they do, but they don't come out every day, and they survey far fewer people (as few as 700).

So, the narrative over the last couple of months has been: good month for Romney/Republicans in June, good month for Obama/Democrats in July. But what do the numbers say?

This shows the swings. As you see, it's mostly an oscillator. Here, the positive swings mean Republican swings, and negative mean Democrat. Overall, the swings seem to favor Republicans.

Perhaps making this clearer is the following graph:

This time series started on April 1st. Each time a poll moves for a Republican, it's a positive, and vice verse. (Obama moving from +1 to +3 would mean the Index moves down 2 points, in other words.)

The trend is clear enough: momentum has been favoring Republican candidates, unrelentingly so. This is a tide that will be reflected in the presidential race, even if the those poll numbers don't reflect it yet. Interestingly, the media isn't picking up on any of this, probably willfully so. It also suggests that Obama's media buys are working against him.

The obvious caveat: we are a million years from election day, and momentum shifts in October will be more meaningful.

This is a lot of work, folks, so I hope somebody cares!


  1. Are you comparing each poll against the previous poll by the same company, or just each poll against the most recent released, regardless of company?

  2. @Zhukov - Polls are compared to most recent released for the same race. While this might seem to be problematic (different polling companies have different methodologies, after all), it's not. First, it is self-correcting over time. For instance, as presidential polls bounce back and forth between, say, Rasmussen and Gallup, biases are balanced as you go. Second, there are now so many polls released every day for so many different races (something that didn't happen even a few years ago) that the effects of differing methodologies balances out almost immediately.

    I have also been asked about the effects of "mega" biases, specifically that most polls are of registered voters and not likely voters. This does, in fact, produce a consistently Democratic bias, it doesn't matter to the Index, because the Index is only concerned with trends.

  3. Good stuff - hopefully you keep it up through to the election, will be very interesting to follow. How often do you plan to update?

    1. I will post an update weekly through September, after which I will post daily. I also have an electoral college model that I will be posting periodically as well.

  4. Thank you Mr. Johnston. Dana L.

  5. Looking forward to the weekly update (isn't that due today...)

  6. great and comprehensive work, Scott. appreciate all your efforts!