Closing in on Referendum C! Building a better model.

The first part of building any scientific model (aside from coming up with a hypothesis) is collecting data. My last few days have been filled with the unfortunate experience of collecting new data for the model on Referendum C.

Some of our data raises interesting questions, such as, can we use even year election results to predict an odd year election, or do we need to focus on odd year election results only. Quite frankly we aren't sure yet. If we just go off of the odd year results, the future looks bleak for Ref. C. For example, in 2003 Amendment 33 (the racetrack gambling amendment) polled in a statistical tie in the months of July and August, then suddenly in a late October Ciruli poll it collapsed down to 19%. None of the 2003 ballot issues polled over 35% by October, surprising since Referendum A (the water bonds referendum) polled in the mid 60s in July. Additionally, we saw undecided voters break on average of 4-1 against the ballot measures. That means that for Referendum C to pass it will need to poll about 48% or 49%. Anything lower than that will be the kiss of death.

Can we use 2004 data? Maybe, but until we run the numbers we don't know. Even year elections have higher turnouts than odd year elections, this turnout differential makes a huge difference (h/t to the guys over at soapblox, their election data is great and they turned me on to the drop off that happens between the top of the ticket and the referenda items). Until I can quantify that differential and use it properly, I probably can't use even year data, its just too dissimilar. On example of this is the difference between FasTracks, which passed last year with basically 60% support and Guide the Ride, which was basically the same thing, but it lost with only 40% of the vote in 1997.

Campaign funding also is a big issue. Campaign finances are a statistically significant variable in legislative races (not the most significant however, voter registration numbers rule that roost). Finding the campaign finance numbers for prior year elections here in Colorado, quite frankly, sucks. The Secretary of State's website doesn't allow you to search by campaign issue, so I have to jump through hoops to find any information. Even once I find the name of the issue committee, I have to add up the campaign expenditures because they don't have a running total. I assume that we will end up finding out that the pro side must outspend the con side by a factor of greater than 10-1, that seems to be the conventional wisdom. However, maybe we'll prove the conventional wisdom wrong, I just don't know yet.

Until we work out what data to use, and find the funding numbers, it wouldn't be prudent to update our model. Just from my gut, it looks like C has an uphill battle. I can't find an example of a off year ballot measure winning without polling in the 50s.