The problem with polling is very few people "sampled" respond and the key is a heuristic likely to

[karynnj]

Vote model.

Pollsters try very hard to get to results that match what actually happens. None want to be off be a large amount. All use stratified samples where they hope to divide the population in homogenous groups, as far as who they will vote for. This lowers the margin of error for the stratified sample because the biggest variance is at 50/50.

Also, if the people within a strata are predominately for one or the other it makes an uncomfortable assumption less problematic. That assumption is that the people who do respond are similar to those who do not. However, note that the margin of error would not reflect any bias due to that assumption being wrong.

The other problem is everything ultimately depends on the likely to vote assumptions. Until election day, it won't be completely clear who is not turning out to vote, though we get some hint from early voting.

[yankee87]

1. Polling discrepancies

No one under the age of 70 answers their phone if number not recognized. Also, no one had land lined. Another discrepancy is like me, I don’t live in the state of my area code on my cell. I constantly get sales calls from my old state. Pollsters need a new paradigm on how to poll.

[agingdem]

2. people have given up their landlines in favor of cellphones

Last edited Thu Nov 3, 2022, 01:36 PM - Edit history (1)

and rarely do they answer any call that's not in their "contacts"...pollsters can't get to them

[Hugh_Lebowski]

3. Stated in the most general way possible, the problem with polling is that nearly all statistical

analysis relies on the concept of 'random sampling' in order to be 'valid'.

And it's actually a lot harder to conduct random sampling ... than it sounds like.

Add the additional complexity that polls have, which is that you can never know the persons future behavior. It's not like you're sampling 'what is the average eight of a 2nd grader on January 1 of the school year'.

That is a number that is 100% measurable at that specific moment in time. It's a simple statistic, it is what it is.

When you have a sample that is essentially just a hard number, and you truly gather a random sample, there's really not a lot of wiggle-room, the 'laws' of statistics are mathematical laws. You can calculate that average height with a GREAT degree of accuracy, esp. if your sample is large.

The assessment of polls, however, is a MUCH trickier proposition, even though they're based on the same mathematical laws ... randomness is hard to achieve in the first place, AND you're not measuring a hard number like 'height'. You have to account for the fact that some people say they're going to do something in the future, which they may or may actually do.

[Journeyman]

4. I refuse on principal to participate in polls of any nature . . .

They don't provide insight to a group dynamic; they're used most often to promote a pre-determined narrative.