Monday, September 13, 2010

Last Primaries on Deck...Another Tea Party Shocker in the Brewing? Our Tracking System for 2010

Exciting Primaries to the End
Tomorrow marks the end of the primary season in the 2010 and Wednesday the unofficial start of the general election campaign (although obviously, the parties have been jostling for position for months if not years already), with primaries in Delaware, Maryland, Massachusetts, New Hampshire, New York, Rhode Island and Wisconsin.

Of these races, by far the most interesting is in Delaware, where Tea Party activists are once again trying to upstage an established Republican moderate. The GOP party establishment has to be somewhat worried, as moderate Rep. Mike Castle (R) was as close to a cinch to win the general election as a Republican can get in blue Delaware, holding double digit leads against the Democrats. Christine O'Donnell and the Tea Party had different ideas, however and we could have a huge upset brewing against the long-established and well respected Castle, with O'Donnell showing a narrow 3 point lead in the only poll available in the race (a survey conducted over the weekend by Public Policy Polling, a Democratic Polling firm whose numbers I normally do not use in my averages, but who, as the only poll available here, warrants some attention) and all the enthusiasm on their side. O'Donnell, if she wins, will be a tough sell in Delaware and is currently polling behind in the general.

In New Hampshire, establishment Republican Attorney General Kelly Ayotte is still the front-runner in a 4-way race, leading by 3-7 points in recent polling, but Tea Partier Ovide Lamontagne, currently running second in polls, has an outside chance at the upset. An upset by Lamontagne would put the race, currently tilting Republican, at least back to a toss-up and probably to a slight Democratic advantage, although a Tea Partier certainly would have a better chance in purplish New Hampshire than in Delaware.

My Guide to the General
Starting approximately this weekend, once all the primaries are settled, I'll be publishing the spreads based on the aggregation of polling. In the interest of full transparency, here are my basic rules for how I will calculate the averages:

(1) Only reputable, non-partisan polls will be used. Partisan-affiliated firms, such as Public Policy Polling and Strategic Vision will be excluded. Reputable polls include such firms as Gallup, Rasmussen, Survey USA, Research 2000, etc.

(2) The polls used for averaging will be as follows:
a. If there have been at least 3 qualifying polls in the past week, the lesser of all the polls from the current week or the 5 most recent polls will be used. More than 5 polls will be used only if polls are tied for being most recent.

b. If there are less than 3 qualifying polls in the past week, the 3 most recent polls in total will be used, including polls up to 3 weeks old.

c. If there are not 3 qualifying polls in the past 3 weeks, all polls from the past 3 weeks will be used, along with older polls. The polls older than 3 weeks will have an adjustment factor applied based on either the movement of other polls during that time period (if data are available) or the movement of generic polling data nationally (if race specific data are not available.)

(3) Averaging Methodology
Three methodologies will be employed for the qualifying polls used.
a. Simple mean -- the unweighted average of all polls
b. Simple median -- the median reading of all polls used (the middle value, or in the case of an even number of qualifying polls, a simple average of the two middle values)
c. Sample-Weighted Average -- the average of all polls, weighted based on sample size. For instance, a poll of 1,000 voters would receive twice the weight as a poll of 500 voters.

The simple mean of methods a, b and c will constitute the "average of averages" that I will use for race determination.

(4) Rating Denotations:
Note: I will be eliminating the "Toss-Up" category, as has been my habit, now that we have moved to the statistical projections.
0 -5 points = "Slight Lean"
5 -10 points = "Lean"
10 - 20 points = "Likely"
20+ Points = "Safe"

The categories don't mean as much, now that we are moving into mathematical models...the difference between a 4.9% lead and a 5.1% lead is arbitrary.

My methodology sounds very complex, but I've done a lot of modeling that substantiates the three averages. The sample-weighted average is the most scientific method, but using the median and simple mean help to counterbalance the disproportionate impact that one outlier poll can have on the race. In the 2008 Presidential election, this method clearly outperformed either of the averaging methods used by itself.

I welcome your feedback on the methodology or any suggestions for improvement.

First projections, approximately this weekend.

Get ready for game time, boys and girls.

If you like this site, tell your friends.

No comments: