"The West Health-Gallup Survey was conducted via the web from Nov. 11–18, 2024, with a nationally representative sample of 3,583 U.S. adults aged 18 and older. The margin of sampling error is ±2.1 percentage points at the 95% confidence level."
OK, so is this article about what three to four thousand people are experiencing? And speculatively applied to Americans as a whole? Or are they using another source to say what "millions" of people are experiencing? And that "billions" were paid?
The part about race was done better than usual. It just said what they borrowed.
At a group level, I'd still rather see borrowing broken down across these attributes: working vs non-working; health practices; education level; income-to-spending ratio; attempts to save money. I have a feeling those attributes will be better predictors.
Since some attributes are intrinsic problems (eg poverty) and some are consumer responsibilities, we could see if disparities still exist that justify policy change vs people's circumstances or actions being primary cause. How much is personal responsibility? How much requires small changes to existing system? What requires either supplementary measures or an overhaul?
Many people don't answer unknown numbers. Of those that do, many will hang up if it's a telemarketer or pollster. Some remaining percentage will do a poll. So, we've selected for those kind of people. Also, people with a phone.
Additionally, it being 3,000 people means you've proven something about 3,000 people in that category. So, all I'd report is that 3,000 people who are willing to do phone polls are experiencing these things. Who knows about Americans in general.
The idea of a probability based poll is that if you asked 3500 people about something that was a good random sample, you could extrapolate to the whole population with pretty good accuracy.
The traditional way to do that is call phone numbers at random, but you run into biases such if a man and a woman are at home probably the woman answers the phone -- so you have to weight the results based on demographic characteristics.
One problem is that people are less and less inclined to answer the phone when a stranger calls so it might be that the people who answer the phone are not representative of the general population, overall that hasn't seemed to be a big problem until Donald Trump entered the scene because whenever he's being asked about probability based polls seem to underestimate his level of support. Looking at the tracker here
The second tracker is simply reporting the results of probability-based polls whereas the first one is applying some weighting to (mainly) non-probability-based polls that use different methods but generally are starting from a sample which is less likely to be representative to begin with but doing more work with weighting to try to make up for it. It's not a huge difference (maybe that $74B is really $69B or $78B or something) but for an election in a particular state the difference between 49.5% and 50.5% is all the difference.
"The West Health-Gallup Survey was conducted via the web from Nov. 11–18, 2024, with a nationally representative sample of 3,583 U.S. adults aged 18 and older. The margin of sampling error is ±2.1 percentage points at the 95% confidence level."
OK, so is this article about what three to four thousand people are experiencing? And speculatively applied to Americans as a whole? Or are they using another source to say what "millions" of people are experiencing? And that "billions" were paid?
The part about race was done better than usual. It just said what they borrowed.
At a group level, I'd still rather see borrowing broken down across these attributes: working vs non-working; health practices; education level; income-to-spending ratio; attempts to save money. I have a feeling those attributes will be better predictors.
Since some attributes are intrinsic problems (eg poverty) and some are consumer responsibilities, we could see if disparities still exist that justify policy change vs people's circumstances or actions being primary cause. How much is personal responsibility? How much requires small changes to existing system? What requires either supplementary measures or an overhaul?
>OK, so is this article about what three to four thousand people are experiencing? And speculatively applied to Americans as a whole?
Are you implying Gallup did not conduct the study correctly? They are among the most credible sources you could hope to find.
Many people don't answer unknown numbers. Of those that do, many will hang up if it's a telemarketer or pollster. Some remaining percentage will do a poll. So, we've selected for those kind of people. Also, people with a phone.
Additionally, it being 3,000 people means you've proven something about 3,000 people in that category. So, all I'd report is that 3,000 people who are willing to do phone polls are experiencing these things. Who knows about Americans in general.
The idea of a probability based poll is that if you asked 3500 people about something that was a good random sample, you could extrapolate to the whole population with pretty good accuracy.
The traditional way to do that is call phone numbers at random, but you run into biases such if a man and a woman are at home probably the woman answers the phone -- so you have to weight the results based on demographic characteristics.
One problem is that people are less and less inclined to answer the phone when a stranger calls so it might be that the people who answer the phone are not representative of the general population, overall that hasn't seemed to be a big problem until Donald Trump entered the scene because whenever he's being asked about probability based polls seem to underestimate his level of support. Looking at the tracker here
https://www.natesilver.net/p/trump-approval-ratings-nate-sil...
Trump does a few points better than the tracker here
https://ropercenter.cornell.edu/presidential-approval [1]
The second tracker is simply reporting the results of probability-based polls whereas the first one is applying some weighting to (mainly) non-probability-based polls that use different methods but generally are starting from a sample which is less likely to be representative to begin with but doing more work with weighting to try to make up for it. It's not a huge difference (maybe that $74B is really $69B or $78B or something) but for an election in a particular state the difference between 49.5% and 50.5% is all the difference.
[1] Full disclosure: I helped code this one.