More notes from RootsCampDC...

 

DNC Data and Modeling: 2006 Lessons: Lessons from 2006 that will apply to the next cycle

(Viewed: times)

 

Everyone has an opinion about how the data should be collected and used, but they decided to survey local activists to find out how they used it in 2004 and how they wished things worked

 

"Fancy data" such as who owns a dog, a snowmobile, etc., often had more "excitement" from organizers than getting the _correct_ "simple" data like phone numbers and addresses, but in 2006 they were able to vastly improve these data

 

Partner with state parties as they had access to the longest voter histories, plus knowledge of the precincts

 

Republican Microtargeting myths: it's more about the PR of how good the Republican microtargeting is, to spread fear, than it is a successful strategy

 

Consumer info ("bourbon drinkers are Republicans, gin drinkers are Democrats", magazine subscriptions, car ownership) isn't sufficient to target -- you need to survey the consumers about political questions as well

 

List of consumers isn't microtargeting or modeling, it's just mailing to a list

 

Use local field staff to gather information about their precincts

 

Modeling in non-party-registration states: using census info, consumer info, historical IDs from state parties and from sec. of state, then you make a 10,000 person sample and do a poll

 

Microtargeting statistical techniques are generally used to determine GOTV targets, but depending on the questions on your survey, you could determine issue alignment as well

 

Put voters into tiers, to allow local campaign managers to determine for themselves how many voters they could afford to reach, while knowing which voters will be their best contacts

 

Requirement was that campaigns had to return the data to the DNC, to know how the scores were used, and how the campaigns adjusted to using them

 

Analysis of microtargeting data from the election is largely still to be done

 

You need to keep looking at the model throughout the election, because people's preferences change as the race progresses (in June, if one candidate is polled as ahead, and if that candidate is not ahead in September, the model needs to be updated accordingly, as more voters are now up for grabs and in need of being contacted)

 

Democrats had a 100 day program, versus the Republican 72 hour program, which means Repubs had no idea what was happening for the last 4 months


Page Information

  • 1 year ago [history]
  • View page source
  • You're not logged in
  • Tags: infotech data

Wiki Information

Recent PBwiki Blog Posts