I don’t care which party you belong to or if you belong to no party, I think it’s pretty clear our political modeling is very broken. The recent Georgia runoffs have illustrated that yet again. Many who have typically championed the use of polling in elections are finally beginning to realize that there are problems, even admitting that election modeling is broken. Yet, they are still charging ahead in the belief that big data all by itself is the way of the future.
This is silly in a couple of different ways. One is simply that on one hand, they are saying that their techniques are fundamentally broken and on the other that they know exactly how to fix it using the exact same kind of information they currently don’t know how to use. Another is that they seem to be relying on the idea that they just need more data, as though merely having more data will solve problems. As stated, that is a rather silly notion. If all you have is bad data, it doesn’t matter how much you have, it’s still bad and will lead you to bad conclusions. It also helps if you know what to do with it. Otherwise, piles of good data don’t help you anymore than piles of bad data.
How so? Often polling sizes are extremely small. A sample of one or two thousand is taken to represent a population of millions. This is totally ridiculous. There is no way that kind of sample can be truly representative. For one, it is very tempting to do your survey in a limited area. And while no area is truly monolithic in its voting patterns, it’s well known that cities tend to vote one way and everywhere else another. If you over sample a particular geographic region then you are going to overrepresent one point of view while underrepresenting the other. Even if you try to sample over a large area, it’s still hard to consider such a poll accurate unless the sample size is much more significant. There is also the fact that different political inclinations will affect whether or not a person will even talk to the pollsters. Conservatives in general are much less likely to even talk to most pollsters, which creates a massive deviation in the polls.
What about knowing what to do with certain data? One of the things that polling agencies have been doing is looking at local shopping habits to predict how a given area will vote. The idea would seem to be that if lots of people are in the line at the local Starbucks, then they will be voting for the Democrat and all the people at Cabela’s will be voting Republican. Certainly that fits into stereotypes. Yet, if you are trying to be accurate in your polling then you need more than stereotypes. After all, republicans also drink coffee (even from Starbucks) and Democrats have even been known to go hunting and fishing on occasion. That person ordering the tofu wrap with the long hair may be walking out to a pickup with a couple cattle panels in the back. The neighbor who just loaded half a dozen guns into the back of a minivan for a day at the shooting range may be more at home sitting in the corner of a coffee shop eating a scone while reading Voltaire. Individuals are much more complicated than what one or two shopping preferences could ever entail. You simply can’t make accurate predictions based on these kinds of models. To really develop an accurate election model, you need to be able to get down to the individual level. What motivates a person to vote a certain way, or at all? If a person has voted for one party the last fifteen years but suddenly switches, the why is incredibly important.
How does a pollster develop this kind of connection and collect this kind of data? Through TARTLE of course. A pollster can sign up with us and connect with people willing to share their political preferences and go into why they think one policy might be better than another. Such information could be particularly useful to pollsters or campaign managers, both of whom are interested in determining what messages resonate with which voters. The kind of connection that TARTLE offers could also be a great help to those interested in forming a different party, one looking to deal with the concerns of most people, rather than trying to get people to care about the things the party is most concerned with.
What’s your data worth?
