What is this blog?
This is a record of the work of two politically-engaged students at Dartmouth College. In December 2013 I began a project with a very simple goal in mind: to predict the results of the Senate midterm elections in November. After all, if Nate Silver can do it, why can't I?
Turns out it's actually a bit more complicated. I started out by myself by creating a model for poll aggregation to calculate a weighted average of the polls in a single election. These weights were based on things like the number of days since the poll was conducted, the sample size of the poll, and the partisan lean of the poll. I'd come up with results like "Democratic incumbent Kay Hagan is leading probable Republican challenger Thom Tillis in North Carolina, 43 - 40." But I wouldn't be able to interpret that all that well. Was that a significant lead, considering that we'd only had three polls in and there was still almost a year to go before the election? If I were a bookie, what kind of odds would I put on Tillis?
That's where Brian Li comes in. I know nothing about programming. Well, that's not true, strictly speaking; I know how to count parentheses and I can create a Hello world program in Python. (It's print "Hello, world!") But that's it. I had Brian create a program to run simulations and determine probabilities of victory from the averages I had created. That's the basis of what we have here: calculated probabilities of victory, turned into predictions of "toss-up" or "lean/likely/safe Republican or Democrat".
I don't mean that it's a simple one-to-one conversion of "if it's above 90% it's safe, and if it's between 80% and 90% it's likely". We recognize that these numbers show a snapshot of a continually changing political environment. So we use these numbers as a strong pointer in the right direction--but we also take into account the national mood, the state's natural partisan leans, and of course, the candidates themselves. We know that all of these things are important, which is more than most pundits can say. Here's the summary of the examples, politely summarized by WonkBlog's Ezra Klein. What's the difference between us and them? We go by the numbers. And not in a vague, eyeball-the-polls way that most of those who predicted a Romney victory did; we have hard formulas, statistics, and rules about how we come up with our results.
So I guess if you ask: what is this blog? The answer is that it's a reaction to an apparent refusal on the part of the media to put at least part of their trust in empirical observations and science.
More to come. Here are our current projections:
I don't mean that it's a simple one-to-one conversion of "if it's above 90% it's safe, and if it's between 80% and 90% it's likely". We recognize that these numbers show a snapshot of a continually changing political environment. So we use these numbers as a strong pointer in the right direction--but we also take into account the national mood, the state's natural partisan leans, and of course, the candidates themselves. We know that all of these things are important, which is more than most pundits can say. Here's the summary of the examples, politely summarized by WonkBlog's Ezra Klein. What's the difference between us and them? We go by the numbers. And not in a vague, eyeball-the-polls way that most of those who predicted a Romney victory did; we have hard formulas, statistics, and rules about how we come up with our results.
So I guess if you ask: what is this blog? The answer is that it's a reaction to an apparent refusal on the part of the media to put at least part of their trust in empirical observations and science.
More to come. Here are our current projections:
 |
| March 30, 2014. |
No comments:
Post a Comment