Internal Campaign Polling

Polls for Decision-Making

Internal campaign polls serve a fundamentally different purpose than public polls. Public polls ask 'who is winning?' Internal polls ask 'what should we do?' They test messages, identify voter segments, track the effectiveness of advertising, and measure the impact of campaign events.

A typical internal poll battery includes the standard vote question plus dozens of additional items: favorability of both candidates on specific traits (honest, strong, cares about people like me), reactions to tested messages and attack lines, issue priorities, and sometimes experimental questions that show respondents ads or news clips and measure the response. This data drives resource allocation: which states to invest in, which messages to use in ads, and which voter segments to target.

The Selective Release Problem

Campaigns sometimes selectively release internal polls to shape media narratives. If internal polls show the race is closer than public polls suggest, the campaign may release the data to encourage donor enthusiasm and attract media coverage. But campaigns never release internal polls that show them losing badly. This selection bias means that publicly released internal polls should always be treated with extreme skepticism.

The opposite problem is internal self-deception. If the campaign's own pollster consistently delivers good news, the temptation to believe it is strong. Romney's 2012 campaign famously 'unskewed' polls to convince themselves they were winning states they were losing. The best campaign pollsters deliver uncomfortable truths, and the best campaign managers listen to them.

A Worked Example: Reading a Message-Test Battery

An internal poll of 800 likely voters starts with the baseline vote question: 45% say they will vote for your candidate. The pollster then randomly splits the sample into four cells of 200 and exposes each to a different stimulus before re-asking the vote question:

How do you read this? With only ~200 respondents per cell, each cell's margin of error is roughly ±7 points. So the +2 lift from the attack message (C) is indistinguishable from zero — noise. The +6 from the healthcare message (B) is large enough to be plausibly real, and it edges out the tax message (A). The disciplined decision is to weight advertising toward the healthcare message and keep testing the rest, not to declare all three 'winners' because all three numbers went up.

This is exactly the kind of judgment that self-deception corrupts. If the pollster or manager wants message C to work, it is tempting to treat +2 as a victory. The 2012 'unskewing' episode is the cautionary tale: reading internal data through hope rather than through margins of error produced confident predictions of states the campaign then lost.