posted
A friend of mine's running the re-election campaign of a congressman. He's thinking about allocating resources town by town, and wants to focus on areas which have tended to vote for the party of his candidate, but which might underperform come election time. He has pretty much every statistic under the sun, and wanted a workable formula to help inform his decision-making. He asked me to help him out, and so far I've come up with the following.

P (K1 + YK2 + GK3 + EK4) = X

K= The weight we assign to each factor. The more we care about one thing relative to another, the higher a K number we assign it.

P= The population of the town in question. Obviously the bigger the town is the more we care about it. I want to weigh this factor relative to the other factors (how much it matters how big the town is vs the other considerations) and relative to other towns (so that a bigger town beats a smaller town with identical other factors by the right amount).

Y= The percentage of people who voted for the other guy in our last primary. The more primary voters the other guy got, the more of our people we have energized against us. That means we have a lot of work to do someplace like that.

G= The percentage of people who voted for us in the town in the last general election. We want to run up the numbers in towns that have tended to support us in the past, and aren't too interested (for my purposes at least) in gaining ground where we're substantially behind. I'm not tasked with making strategy or I'd take some exception to that; but I just want to make the equation do what they want it to.

E= The percentage of eligable voters who didn't show up to the polls last time around. If 95% of people are out there and voting there's not too much for us to do in terms of turnout; if 50% of the population isn't voting we have some work to do.

X= The "score", or how interested we should be in putting resources into the town.

There were some other factors I thought about adding in (how we did in 2008 vs how we did in 2004 to illustrate what we could hope for in a good year, since 2008 numbers are going to be a good bit lower than we expect in 2010. Also maybe 2006 numbers because 2008 was a presidential election so turnout is going to tend to be much higher there). Looking back actually maybe I should make my E factor either depend or at least be influenced by the 2006 number.

Anyway, I'm definitely aware of the limitations of statistical analysis as an approach to these things. But he thinks it'd be helpful to have it laid out in mathematical fashion, and that he'll be able to assign what I'm calling K values in such a way that it's a useful tool. Any thoughts on how to make a better equation? Problems with mine that I might not have considered?

posted
K1 is meaningless, if it as it appears (just a simple arbitrary constant). You might as well set it to 1 without losing any generality at all.
Posts: 1768 | Registered: Oct 2000
| IP: Logged |

posted
K1 looks like a personal fudge factor of some sort.

The best I can suggest is that it might be useful to look at the party splits on other elected offices (past elections and current polling) and compare performance- if you have areas where other people of the same party consistently outperformed the candidate, you've got ground that should be somewhat easier to make up.

On the other hand, areas where your candidate significantly outperforms the rest of the party, you might need to defend some ground there.
Posts: 11997 | Registered: Oct 2005
| IP: Logged |

quote:focus on areas which have tended to vote for the party of his candidate, but which might underperform come election time

Shouldn't the Y term be negative?

if X='how much do we care about this town'

then if a lot of people voted for the other primary, you'd have to do a lot of work to get them to vote for your guy, so that town would have a lower score.
Posts: 359 | Registered: Nov 2001
| IP: Logged |

quote:K1 is meaningless, if it as it appears (just a simple arbitrary constant). You might as well set it to 1 without losing any generality at all

Nope. It's the value I'm assigning to the population relative to the other factors.

quote:Shouldn't the Y term be negative?

if X='how much do we care about this town'

then if a lot of people voted for the other primary, you'd have to do a lot of work to get them to vote for your guy, so that town would have a lower score.

I tend to think the opposite. I certainly didn't vote for McCain in the primary, but with a little work he could've gotten people like me to come out and vote for him in the general. Often we're just not excited and enthusiastic about someone we vote against in a primary, but if they talk to us a bit about how good they are (or how bad the other guy is) they can swing it.

A friend actually suggested to me that what I need to do is to normalize the non-K variables, putting them on a range from 0-1. That would mean that the K could just serve to add weight instead of also needing to translate between ways of expressing things. I don't have the mathematical chops to do this and was hoping someone else might. Suggestions as to how to tweak the equation in other ways (additional factors I might want to consider) are also more than welcome.

quote:Wait, we're coming up with a formula for distributing pork?

If getting people to go door-to-door and make phone calls is pork, then you got it. Obviously talking about campaign resources and not governmental stuff here.

quote:Originally posted by Paladine: If getting people to go door-to-door and make phone calls is pork, then you got it. Obviously talking about campaign resources and not governmental stuff here.

Ah, my bad. Although the cynic in me can't help pointing out that your criteria for deciding where to campaign the hardest during the election would also be ideal for working out where to send the most pork before the next election
Posts: 2570 | Registered: Jul 2004
| IP: Logged |

quote:K1 is meaningless, if it as it appears (just a simple arbitrary constant). You might as well set it to 1 without losing any generality at all

Nope. It's the value I'm assigning to the population relative to the other factors.

All K1 effectively does is scale the value of X, which is entirely arbitrary in the first place. The function of scaling the population relative to the other 3 terms can be absorbed into the other 3 constants. You have 4 parameters (P, Y, G, E) so you need 3 wieghting factors to determine the relative importance of them. You'd need 4 weights if the overall scale of X was absolute in some way (if you want to interpret X as an absolute amount of money), but as I understand it, X is just being used as a value asigned to each area to identify which ones score high (relative to others) as targets for focussing campaigning.

Look at it this way: if K2 = K3 = K4 = 0, then X = P x K1. It is fairly obvious that in that case K1 does nothing aside from turn the weight from population into an arbitrary number directly proportional to population. If all you care about are the relative scores, you've added nothing useful. (K1 remains redundant no matter what values you use for the other Ks, it is just more obvious in that case).

BTW have you considered factoring in variability of turnout from one election to another in some way? That might indicate areas where there are more people on the vote/can't be bothered borderline where campaigning could have a larger effect on boosting turnout in your favour.
Posts: 1768 | Registered: Oct 2000
| IP: Logged |

posted
Am I correct in saying, the final outcome of this should be a ranking that roughly indicates where he would garner the most voters (in terms of heads of the population) for e.g. a fixed number of man-hours spent campaigning?

In other words, if one town has a score of 50 and another town has a score of 25, it indicates that he will gain more voters from the first town if his people spend 100 hours campaigning for him than if they spend those 100 hours in the second town—without necessarily indicating how much more?
Posts: 2147 | Registered: Nov 2004
| IP: Logged |

quote:Am I correct in saying, the final outcome of this should be a ranking that roughly indicates where he would garner the most voters (in terms of heads of the population) for e.g. a fixed number of man-hours spent campaigning?

In other words, if one town has a score of 50 and another town has a score of 25, it indicates that he will gain more voters from the first town if his people spend 100 hours campaigning for him than if they spend those 100 hours in the second town—without necessarily indicating how much more?

You're right about what the goal is. If we could model the equation in such a way that it told us more about how many more votes could be gained, obviously that'd be better. I do think that it has at least *some* implications for that question though, since a bunch of towns are being measured against each other. It might not have much to say about "how much more" in absolute terms, but the nature of the scale is such that the size of the difference should tell us something, no?

quote: The function of scaling the population relative to the other 3 terms can be absorbed into the other 3 constants.

Yeah, what I'm thinking now is something along these lines. I've can derive normalized values for the non-K variables by finding the mean and standard deviation of the data set I'm working with. Then the K doesn't have to translate between the scales on which each is expressed, and only has to determine what weight I'm assigning each factor. I think I can solve the population problem just by multiplying P through to everything.

quote:BTW have you considered factoring in variability of turnout from one election to another in some way? That might indicate areas where there are more people on the vote/can't be bothered borderline where campaigning could have a larger effect on boosting turnout in your favour.

I definitely have, but at present I'm just trying to get the general skeleton of the basic equation down. Once I do that I can refine it with things like this (which I do think are very worthwhile). What I'm doing right now is extrapolating from the 2008 general election and the more recent primary contest. I have data from other years also, but there are so many things that can cause changes from election to election, and demographics change significantly if you go too far back.
Posts: 3235 | Registered: Jul 2004
| IP: Logged |