The Polling Center: Lubbock Poll or Leave It? Leave It.

Well, as Ross Ramsey's post explains, we've heard from Michael Stevens, and I'm glad we did. Based on how he explained his method to Ramsey, his Republican primary poll should be taken with not a grain, but a shaker of salt.

This is not a sound sampling method, period.

The sampling method is neither random nor matched, and there is likely significant non-response bias, a classic problem in mail-only solicitation. (In English: That means there might be something about the people that don't respond to your poll — some particular characteristic — that throws off the results and makes the poll unreliable.) 

Given the problems in both sampling and response, there is no real margin of error.  You might do some calculations that might give you the reported +/- 1.89 MOE, but the methods don't meet the conditions to make such a calculation.

I haven't been able to reach Stevens (got voice mail), but based on what he told Ross, the only thing definitively in the weeds here is the reliability of this poll.