Former First Lady Michelle Obama is the only Democrat running ahead of former President Donald Trump in a hypothetical matchup, according to a new poll.
EDIT: I am wrong about the sample size. Yes, the sample is a little small, but not too far off. They’re registered voters rather than likely voters, which is not quite as good, but, again, no terrible.
The poll surveyed 892 registered voters and has a margin of error of 3.2%.
As FiveThirtyEight would say, that’s a bad use of polling. That’s a very small sample size, and there’s no indication that it’s representative in any meaningful way.
Even more important, Obama has said she has no interest in being the president; she’s not willing to run.
It is most certainly not a small sample size. It’s what allows for a margin of error of ±3.5%* at the 95% confidence level. Here’s a graph of the margin of error vs sample size for 95% confidence interval.
With an 11 point margin, there’s a clear separation of the upper limit bar for Trump and lower limit bar for Obama. For a single poll, assuming the rest of it was well designed and executed, this is an important spread. And the reasons are obvious if you look at the report. She’s able to get 10% more Democratic support and 20% more independent voter support.
Ipsos is a high quality polling company. They don’t make rookie mistakes like sample size. There may be other reasons beyond my reasoning that make this a bad use of polling, but sample size is not it.
* The source incorrectly reported the margin of error for the full survey, both registered and unregistered participants.
There are multiple ways. Statistical significance is largely used to determine whether a sample size is representative but it’s flawed on its own for some sample sizes as small effects can get exaggerated the larger the sample gets. Look up the methods for determining effect sizes and confidence intervals to determine the best route to go to see what minimum sample size is necessary to both have high confidence in the accuracy of the hypothesis and to ensure that the results have enough statistical power to detect the effect in question.
Anyone that doesn’t want to be President should automatically win. If you want it, you should be locked in a cold, dark room until the election is over. And maybe slapped a few times for good measure.
EDIT: I am wrong about the sample size. Yes, the sample is a little small, but not too far off. They’re registered voters rather than likely voters, which is not quite as good, but, again, no terrible.
As FiveThirtyEight would say, that’s a bad use of polling. That’s a very small sample size, and there’s no indication that it’s representative in any meaningful way.
Even more important, Obama has said she has no interest in being the president; she’s not willing to run.
It is most certainly not a small sample size. It’s what allows for a margin of error of ±3.5%* at the 95% confidence level. Here’s a graph of the margin of error vs sample size for 95% confidence interval.
With an 11 point margin, there’s a clear separation of the upper limit bar for Trump and lower limit bar for Obama. For a single poll, assuming the rest of it was well designed and executed, this is an important spread. And the reasons are obvious if you look at the report. She’s able to get 10% more Democratic support and 20% more independent voter support.
Ipsos is a high quality polling company. They don’t make rookie mistakes like sample size. There may be other reasons beyond my reasoning that make this a bad use of polling, but sample size is not it.
* The source incorrectly reported the margin of error for the full survey, both registered and unregistered participants.
You are correct, and I am not. I’ve edited my comment to reflect that.
A fancy guess is still a guess.
892 out of 160,000,000+ is a small sample size.
It isn’t.
Isn’t 1,000 usually the benchmark?
I depends on the size of the population you’re attempting to represent.
What’s the formula/ratio? Didn’t know there was one like this.
There are multiple ways. Statistical significance is largely used to determine whether a sample size is representative but it’s flawed on its own for some sample sizes as small effects can get exaggerated the larger the sample gets. Look up the methods for determining effect sizes and confidence intervals to determine the best route to go to see what minimum sample size is necessary to both have high confidence in the accuracy of the hypothesis and to ensure that the results have enough statistical power to detect the effect in question.
It looks like most Ipsos polls are a little over 1000, and most of them seem to use likely voters rather than registered voters.
I have edited my comment to reflect that I’m wrong.
Anyone that doesn’t want to be President should automatically win. If you want it, you should be locked in a cold, dark room until the election is over. And maybe slapped a few times for good measure.