Bad Fauci is killing America, he just doesn't understand per By Rep. Andy Biggs & Rep. Ken Buck

Fauci has admitted that the models he relies on are unreliable. The models, and their panic-inducing projections, have seemingly been revised down every couple of days. Fauci insists this because of his policy prescriptions, but time and data from the United States and other nations will reveal whether that is true.

These guys

thephan wrote:Bad Fauci is killing America, he just doesn't understand per By Rep. Andy Biggs & Rep. Ken Buck

Fauci has admitted that the models he relies on are unreliable. The models, and their panic-inducing projections, have seemingly been revised down every couple of days. Fauci insists this because of his policy prescriptions, but time and data from the United States and other nations will reveal whether that is true.

These guys

If they put the trillions in people's hands instead of buoying the markets maybe we'd all be less "inconvenienced".

Amen to that, brother.

thephan wrote:Bad Fauci is killing America, he just doesn't understand per By Rep. Andy Biggs & Rep. Ken Buck

Fauci has admitted that the models he relies on are unreliable. The models, and their panic-inducing projections, have seemingly been revised down every couple of days. Fauci insists this because of his policy prescriptions, but time and data from the United States and other nations will reveal whether that is true.

These guys

These are elected officials. Don’t make me do stuff!

Rasmussen has Trump at 43-56 approve/disapprove

The Savior wrote:Rasmussen has Trump at 43-56 approve/disapprove

Their numbers are always more favorable to Trump than other polls. I dunno why, JH could probably explain.

Swiggers wrote:

The Savior wrote:Rasmussen has Trump at 43-56 approve/disapprove

Their numbers are always more favorable to Trump than other polls. I dunno why, JH could probably explain.

I dont really know, but they are the only approval poll that uses a likely voter screen.

I can’t easily see the week over week change but to be -13 in a right meaning poll shows that the bump is over.

Rasmussen primarily reaches people by randomly dialing landlines between 5 PM and 9 PM on weeknights.

https://www.rasmussenreports.com/public_content/about_us/methodology

That explains a lot. People who A) have landlines and B) answer unknown numbers are definitely not a cross-section of the country.

Bucky wrote:That explains a lot. People who A) have landlines and B) answer unknown numbers are definitely not a cross-section of the country.

I would expect that population who overrepresent Trump supporters, as noted above. Those stats look doubly bad.

They do supplement their IVR stuff with online panelists. If done with care, you can do an OK job measuring public opinion that way. I don’t believe Rasmussen does their work with care. Their Twitter feed, btw, is right wing propaganda and bonkers. Didn’t used to be so bad when Scott Rasmussen was there, but he left the company bearing his name a few years ago and it’s really gone downhill.

What follows are ramblings on polling, which I’m not sure anyone cares to read, but I'm bored and the Super Bowl XLII rebroadcast doesn't start till 3.

Public polling and campaign polls are very different things. This is partially a function of cost, but mostly because they’re usually being done for different purposes. The most important data points in public polling are the key measures like the ballot or Trump’s job approval and whatnot. Maybe looking at some basic cross tabs like party or gender or age. In campaign polling, you care about that, but you’re also typically testing messages (pro for your side, anti for the other side) to see if you can move the ballot or your image or your opponent’s image. And you care about much more detailed cross tabs on key questions to develop target groups as well as generating specific targets for paid media or other forms of messaging through the message testing.

On a campaign poll, you set a bunch of quotas or targets on the front end. Let's say we're polling Pennsylvania. There are a bunch of things you know - number of registered voters in a county and ultimately a DMA, voter turnout in recent elections, registered voters by party, registered voters by age and gender. Then there are some things you have very good data sources to estimate, like race of registered voters and past voters. And then some things you have mediocre data sources for estimating, like education levels of past/likely voters. You use this information to set ranges of numbers of completes you want to get from men and women and voters 18-29 and voters 30-39 and white voters and black voters and you get the picture. If you don't do this, you will (typically) end up with too many old voters and too many female voters and too many white voters and too many college educated voters. This ensures you end up with a sample that hopefully closely resembles the universe you are trying to survey, and weighting on the back end to fully line up with your targets can be minimal.

These voters are called off the voter file, which has a trove of information about the people you are talking to. If one of you were in the PA sample that was ordered, you’d be asked for by name by the caller, and they would be able to append your age, gender, party registration, past vote history, as well as estimates of turnout models and likelihood to vote GOP/Dem and other more useless modeling to your survey responses. In my opinion a lot of the special, individual specific modeling within the file doesn’t add a ton beyond just knowing a person’s party registration, but the more basic stuff like past vote history is useful.

There’s still measurement error and you’re only talking to a small number of people and error grows as you drill down on subgroups, but if you set it up right you can learn a lot, and it tends to be pretty accurate on actually predicting the election results as well.

Polling like that is expensive as hell. Giving rough estimates, doing an n800 20 minute survey (fairly typical length for a campaign poll if you're testing messages, way long for most public polls), calling 70% cell, when you include the cost of phone sample, you're looking at about $30k in costs out of pocket before the pollster adds on anything for their fee. Calling cell phones is more expensive than landlines because you can’t auto-dial them (which is a big problem for Rasmussen, and they don’t call them).

As an aside, Nate Cohn and the NYT/Upshot’s polling is very much like typical campaign polling, which is why their project on House races last cycle was so fascinating and successful.

There are a bunch of other ways to collect public opinion data that is cheaper, and depending on what you're trying to measure, can be just as good or maybe even better. If you're looking to measure attitudes among adults rather than registered or likely voters, you won't want to call off the voter file, you don't need to match to turnout models, you can dial digits randomly rather than off lists, weight to census targets, and depending on response rates can be significantly cheaper. Many of the public polls that get the most attention use this method, which doesn’t make them right or wrong or better or worse than campaign polls, but it makes them different.

Online panels are growing and becoming increasingly useful for political work because some now have matching to the voter file so you can get the same info for building more sophisticated models. This is much cheaper than phone - can do that same n800 survey for about $10k matched to the voter file and maybe $6k if not. However, there are severe limitations on geographies with panels. You can do a survey in PA statewide, but you're not going to have enough panel to make a large n survey work if you're surveying New Hampshire. Or if you're trying to do a GOP primary poll in PA. And non-standard geographies that change, like congressional districts, are tough. There are also biases in the panels just like on different phone methods – panelists tend to be younger and more educated and more female than voters – so you need to be more careful in setting quotas here.

Two other interesting, newer methods that I think are going to become industry standards for inclusion are text to online or just full text/SMS surveys. You can start from voter list sample the same way as phone, so you have the info of the person you’re surveying, and direct them to take an online survey or conduct the survey if it’s short enough just through an exchange of text messages. Response rates are lower (about 1% of texts convert to completes online vs. 6-8% or so of phone calls), so there’s some worry that you’re getting people disproportionately interested in politics. But it’s a lot cheaper than phone because it is less labor intensive, and does a relatively better job at reaching younger voters and minorities and other tough to poll groups. Haven’t done a ton of this yet, but I think it’s going to be a bigger thing this cycle and moving forward.

Getting back to where this all started Interactive Voice Response, or IVR (what Rasmussen does primarily), is about the same price as online (and for Rasmussen if they do it in house with their own calling/dialing technology presumably even cheaper). It's worse than online in reaching a representative audience because landline usage in America is much more skewed demographically than online panel membership, and it’s tougher to fix through weighting on the back end because even if you boost younger voters in your IVR survey by weighting them up to the right share of the electorate, I don’t think it’s reasonable to assume younger voters who have landlines are representative of younger voters on the whole, which is the core assumption of weighting.

You can blend the methodologies of IVR and online panels to use the strengths of each (older, less educated on the one side, younger, more educated on the other), and come out with something halfway decent. But Rasmussen’s crosstabs are always screwy (Trump’s best numbers with minority voters invariably come from them). I don’t know who they use as their panel provider, but they might be going cheap there, they probably have too much of the sample coming from phone in the phone/online split because they can do that cheaper, and they probably don’t set very strict quotas up front because that adds to cost, which means more weighting, which means getting a few unrepresentative people can impact accuracy a lot. If you read their methodology page, you will notice it is vague as hell, which is not usually a good sign.

If you’re still reading, I’m sorry you’re as bored as I am, but a theory on why Trump might be doing worse on Rasmussen than normal is people are home and answering their phones more often than usual because they’re bored as shit, and maybe the demographic that ignores phone calls from 5-9 PM usually is one that leans Indy or Dem, is under 55, maybe more college educated, etc. So their typical special Trump fluffing weight sauce is being confounded by a bit higher response rate of those voters, hurting Trump. We’ll see how it moves over the next few weeks.

JH, just curious--any idea how incentives (even something really cheap like have respondents enter a drawing to win 5 $100 gift cards) help with SMS type response rates? I know a lot of market researchers do that, but as far as I know, public opinion types have not.

Do polling companies need optin for text if it's a political campaign?

TenuredVulture wrote:JH, just curious--any idea how incentives (even something really cheap like have respondents enter a drawing to win 5 $100 gift cards) help with SMS type response rates? I know a lot of market researchers do that, but as far as I know, public opinion types have not.

Haven't tried so not sure. I don't think we've ever offered incentives on political work for a survey.

Most of the stuff I work on is corporate, associations or non-profit clients, not direct for political candidates (though often politically adjacent in what's being studied). When we do incentives it's usually when we have a member list and the association is trying to get better response rates from their members. Same with universities and their students or faculty. We usually do small incentives for everyone (send Amazon codes) who completes rather than doing the bigger raffle incentives.

Werthless wrote:Do polling companies need optin for text if it's a political campaign?

Our main phone vendor doesn't do it because it's a legal gray area, and they don't think the current level of demand is worth the risk, but they're talking about offering it. A few of the other people we use do. I don't know exactly what the difference is, but think it has to do with how the texts are generated.

Pollsters seeing increased response rates

Other peoples' thoughts on the above and related matters

Fox News only post to be sucking Trump 96% of the time, so he’s pissed at them. Today he’s pissed at Wallace

Highlights or conclusion from the examination for those who can't access NYT?

Here’s a summation meme