Within the context of ALPA's polling? It's done by a very experienced pollster who ensures the sampling is done by demographic. Age, category, Regular, Reserve, commuter, etc. The larger the sample, the more accurate the data. Another control is the specificity of the question(s). The more binary the question (yes or no), the more accuracy.

The way is was explained to me, the pollster takes known demographic information, then develops a "call order" for which pilot to contact first, second, third, etc. within each group. Then the goal is the get as much feedback quickly before other events change the perspective of respondents. "How would you rate your contract compared to peer airlines?" changes the day UAL, American, or others get a deal, or the stock market crashes...so they try to get "same day" (my term) responses. So the timing is the final element of "controlled sample".

Based on the comments some reps have made in the past regarding polling, there are some who don't trust data gathered scientifically. I asked my former Capt rep in MSP what his problem was with the polling, and his reply was, "It's too neutral!"