CROWDPAC DATA MODEL
The Crowdpac data model combines three sources of publicly available information about candidates:
Money - which individuals or organizations have contributed to the candidates' campaigns, and which campaigns the candidates themselves have contributed to, as reported to federal and state regulatory authorities. This gives us a good indication of their overall political position.
Speech - what the candidates say: the bills they sponsor or co-sponsor (if they are currently in office or have been elected before); the words or phrases they use most, as reported in legislative text and floor records, and candidate statements made on official websites, Facebook profiles and via official tweets. This gives us a good indication of their political priorities.
Votes - the candidates' voting record (if they are currently in office or have been in office before). This helps increase the accuracy of our predictions - from around 92% to 94% - and to estimate candidates' position on specific issues.
It works like this: to calculate overall scores for candidates - both incumbents and new candidates - we rely on campaign finance records. Donors to political campaigns tend to support candidates who share their policy preferences and/or personal interests, and screen out those who do not. This generates large amounts of information on where candidates stand. In analyzing the patterns of who gives to whom, our data model is able to make inferences about the issue positions of both candidates and donors. Additional information on candidates' personal contributions made to other campaigns are incorporated to improve the model’s predictions. As a result, it represents a new way of forecasting how a candidate would likely vote and legislate if elected to office.
To calculate scores on specific issues, for incumbent candidates, we use Congressional voting records; for non-incumbent candidates with no voting record, we compare their donors with the donors of incumbents.
We calculate scores for as many candidates as possible. It depends on the candidate, but often a single donation that candidate made to another candidate is enough for the model to calculate an overall score. In most cases, a candidate must receive 50 donations before we can calculate issue scores. In all cases, the contributions need to have been reported to federal or state regulatory authorities. The more data we have on a candidate, the more confident we can be about our scores. On each candidate's profile page, we provide a summary of the availability of data for that candidate.
You can find a more detailed academic explanation of our data model, including a paper prepared by co-founder Adam Bonica, Assistant Professor of Political Science at Stanford University here: http://data.stanford.edu/dime/
= new reply since forum marked as read
