Data mining and gay rights

There is a big debate amongst the big gay rights organizations in Michigan. Should we push for a voter-led constitutional amendment to ban discrimination based on sexual orientation and gender identity? Some point to recent polls saying gay rights have a 68% approval rating. The support runs across Dems, GOP, and independents, all age groups, and all church goers.

An article by Jan Stevenson in Between the Lines looks at another way to determine if Michigan is ready. The technique is modeling. It has been used by several campaigns and has shown to be more accurate than polling. It uses data mining methods used by marketers. It collects a huge amount of data on voters, such as what groups they belong to, what magazines they buy, and education levels. It also takes in data from the Michigan Democratic Party Voter Activation Network. It adds in likely turnout, and crunches the numbers.

For this issue in 2016 it predicts a yes vote of only 42%. That means we have to change the minds of 365,000 voters by next November to win. And that means about 100 full time field organizers starting now. It means talking to 3.6 million voters, much of it in face-to-face conversation. All together it means 22 million contacts through direct visit, mail, phone, etc. The cost (from another article) might be as much as $22 million. Do we have those resources? Probably not.

So should we go with the poll data, which suggests a comfortable margin of victory and implies we go for it, or do we go with the modeling data, which suggests 2016 isn't the time?