question - right now for understandable reasons, response rate for phone polling (and perhaps internet panel too) is extremely low. Does it make it a lot harder to get to a decent poll with say 3% margin of error?
Yeah, the number provided as the margin of error applies when your sample is truly random (and there are no other sources of error). Unfortunately, there are, and these can make the margin of error much higher.
They account for this by weighting their data - but it's an inexact science. As consumers of data, we have to accept that polls are imperfect and some could be more imperfect.
With all that being said, it's impressive how accurate they have been given all the challenges
question - right now for understandable reasons, response rate for phone polling (and perhaps internet panel too) is extremely low. Does it make it a lot harder to get to a decent poll with say 3% margin of error?
Yeah, the number provided as the margin of error applies when your sample is truly random (and there are no other sources of error). Unfortunately, there are, and these can make the margin of error much higher.
They account for this by weighting their data - but it's an inexact science. As consumers of data, we have to accept that polls are imperfect and some could be more imperfect.
With all that being said, it's impressive how accurate they have been given all the challenges
Do you think non-response bias is a legit thing? If it is, how can it be mitigated?
Oh yeah, it's enormous, and I think most data collected today substantially understates it.
The only solution is to stop getting so much nonresponse.
The preferred solution of modern pollsters is to say "meh" and try and weight their way out if it.
I think the long term solution will require a way to avoid so much nonresponse