Unlikely Voter

Tweet

Scientific polling, based on the laws of probability and the compounding of likelihoods, is a mathematical activity. It’s all about the numbers. Without the numbers no poll has meaning. That’s why I highlight key facts like Margins of Error.

Your typical internal poll release is very low on numbers and instead is a one page memo. Those releases can be based on sound polling practices, but they are firstly designed to push an agenda. When I see this new Illinois poll, I am reminded of an internal poll release.

I’m not accusing the Chicago Tribune and WGN of fraud. I have no reason to suspect that. But when I have to wade through paragraph after paragraph of opinion and armchair analysis in order to get basic data about the poll, I’m suspicious.

When I write about polls here, I too add my own paragraphs of analysis, but I link to the polls directly, and right away, so that my readers can see the same raw data I’m reading and draw their own conclusions, or at least re-check my assertions.

The Tribune doesn’t give us the raw data when it goes on and on attempting to explain why Democrat Pat Quinn might have taken a lead on Republican Bill Brady in the Illinois governor’s race. Instead we are given the top line (Quinn 39, Brady 37, MoE 4), and cherrypicked crosstabs chosen to support the conclusions made by the author.

But without seeing the raw data, we have no way of knowing whether this poll has signs of being an outlier or of having an ineffective Likely Voter screen. We also can’t do our own checks for signs of other interpretations of the data. We can’t even check if the author made basic mistakes in reading the poll. Mistakes happen.

Last night on Twitter I mis-read a poll and ended up posting the wrong analysis, forcing me to retract when I saw my own mistake. We have no way of knowing if the Tribune’s analyst misread a table, so I have no way to build on this poll and think about it meaningfully.

I will instead ignore it. If I were making projections of Governor’s races, I would not weight it. I work with numbers, not words. I need data, not opinions. My rule about internal polls, and similar polls like this from the Tribune, is that without the data they are worthless. Otherwise why take the poll at all?