Markos Moulitsas, publisher of Daily Kos, today charged that Research 2000, the polling firm that he has been using for a number of years, has been falsifying data for at least the last year and a half.
For a number of election cycles, some people interested in polling have thought that the results presented by Research 2000 were less believable than for most other polling companies. Their results, some thought, mirrored their clients’ wishes. Many polls that are paid for by an individual candidate or a political party have some sort of problem like this; as we know, people like to get good news, and rarely blame the provider of happy tidings even when they don’t work out in the end. Daily Kos is now leveling that charge at Research 2000.
A trio of statisticians had gone to Kos a few weeks ago to advise him that they had examined the Research 2000 data from a number of polls, and believed that the results could not be explained using probability analysis. They believed that the analyses presented by the polling firm were the result of data that had been invented by the firm. Some of their complaints readers might find boring, but one is pretty easy to understand. When two different groups are sampled, the results are independent of each other normally. That means that there is not a relationship between groups. But in the results examined by the trio, if one group had an even (0,2,4, 6, result, the other groups also did. The same held true for odd results. This should happen half the time; to happen all the time strained credulity.
Today, Kos published the statisticians’ findings and reported that he had terminated the relationship with Research 2000. He said that he had requested R2000 to present him its raw data, but they had delayed doing this for two weeks. This is not an unusual request that he made. As an aside, I still have the raw data for my dissertation in a box in my office, just in case someone wants to replicate the data, and that was obviously a long, long time ago.
Any future polling results used by me will be done without using R2000 results. Numbers have beauty concealed within them; to do what they are accused of doing, assuming the claim is true, is like Lazlo Toth attacking the Pieta with a hammer.