4 ms·

It's an increasing struggle to get a good sample for surveys, as the rate of people refusing has been going up continuously since surveys were invented. This ha

by pmyteh 7y ago

It's an increasing struggle to get a good sample for surveys, as the rate of people refusing has been going up continuously since surveys were invented. This has methodological implications for survey researchers that are quite serious. Some things aren't a problem, though. Survey companies definitely work outside 9-5, and the 2,000 population sample turns out to be more than enough: the sample size needed for a particular degree of statistical confidence is independent of the population size (except for very small populations) and remarkably few people are needed for a good sample. 2000 is actually pretty large for survey research, where ~1000 is pretty standard.

vmilner 7y ago
There's nothing like UK election polling for shaking confidence in getting representative samples from the UK population. https://en.wikipedia.org/wiki/Opinion_polling_for_the_next_United_Kingdom_general_election https://en.wikipedia.org/wiki/Opinion_polling_for_the_next_U... shows that two polls this week, each of ~2000 people, found the Conservatives to have either a 1% or a 12% lead over Labour.
- pmyteh 7y ago
  The British opinion polls are all over the place at the moment, yes. It's very likely that these are not sampling issues, though. British politics is currently in a period of realignment with the position of all parties in flux. Election polling relies heavily on turnout modeling: it's not enough to know what people's opinions are to predict elections, but also which of those people are likely to vote. And the different polling companies have very different models for that, and other 'artistic' parts of the modeling which aren't strictly related to their polling sample. It certainly doesn't imply problems for survey research as a whole (though, as I said, there are increasingly long-term challenges to the while edifice).

by pmyteh 7y ago

vmilner 7y ago

There's nothing like UK election polling for shaking confidence in getting representative samples from the UK population. https://en.wikipedia.org/wiki/Opinion_polling_for_the_next_United_Kingdom_general_election https://en.wikipedia.org/wiki/Opinion_polling_for_the_next_U... shows that two polls this week, each of ~2000 people, found the Conservatives to have either a 1% or a 12% lead over Labour.

pmyteh 7y ago
The British opinion polls are all over the place at the moment, yes. It's very likely that these are not sampling issues, though. British politics is currently in a period of realignment with the position of all parties in flux. Election polling relies heavily on turnout modeling: it's not enough to know what people's opinions are to predict elections, but also which of those people are likely to vote. And the different polling companies have very different models for that, and other 'artistic' parts of the modeling which aren't strictly related to their polling sample. It certainly doesn't imply problems for survey research as a whole (though, as I said, there are increasingly long-term challenges to the while edifice).