The Washington Post reports that Kamala Harris has a "slight" lead over Donald Trump in its latest poll:
Harris stands at 49 percent to Trump’s 45 percent among registered voters in a head-to-head matchup.... Given the margin of error in this poll, which tests only national support, Harris’s lead among registered voters is not considered statistically significant.
Technically this is right, but it's a damn close call. In a poll with 1,975 respondents a 4-point lead might not quite be significant at the conventional 95% level, but it is at the 92% level. That's a pretty high probability that Harris really is ahead.
UPDATE: The ABC News report on the poll says Harris leads 51-45 among likely voters. That's extremely statistically significant no matter what standard you use.
Let's look at the favorable ratings in that poll:
Harris -- 47%/89% (registered voters/democrats)
Walz -- 43%/77%
Trump -- 36%/77% (registered voters/republicans)
Vance -- 35%/68%
That's excellent news so long as Democrats execute on their GOTV operation (and why wouldn't they, with that massive nearly half-a-billion dollar infusion and hundreds of thousands of volunteers following Biden's exit?) and the low information/propensity voters either stay on the couch or minimally break for Harris.
Combined with Rust and Sun belt polling from NYT/Sienna and others, Trump's now got a difficult path towards victory while Harris' paths are expanding.
Welcome back, Creeper Corey Lewandowski. Help Trump become more Trumpian and the GOP team more weird.
What normal decent human being wouldn’t vote Harris over Trump?
kevin drum's tuesdays at panera lunching crew?
Some people who thrive on hate and who think they don't have much to lose no matter who wins. They'll vote for the rich white guy because he tells them exactly what they want to hear, even though they might actually know he's full of shit about most of his promises. Hate buys a lot of votes.
I really wish people understood what confidence intervals and statistical significance actually mean.
1) If a value falls inside the confidence interval, and is thus considered to be statistically insignificant, it does not mean that there is no difference between its estimated value and the null hypothesis. (In political polling, the null hypothesis is that the race is tied.) It means that the estimated value has a lower probability of being different than the null hypothesis than whatever confidence level (usually 95%) you've chosen. Unless the estimated value is exactly the null hypothesis, there is a greater than 50% chance that it is not the same. Even if the value is within the confidence interval, you can feel increasingly optimistic that they are not the same as they deviate farther.
Note: From a technical perspective, the estimated value doesn't have a probability of being the same as the value in the full population. That value is fixed, and any estimate either is, or is not, the same. What the confidence level actually means is that it is the percentage of an infinite number random samples that would be that close or closer to the population value. That sounds persnickety, and, in most cases, amounts to the same thing, but there are cases where that distinction becomes important.
2) There's nothing magical about a 95% confidence level. You can set it wherever you like, and the study is still valid, so long as you disclose what level you're using. We fetishize 95% because, way back when statistical analysis was becoming common, a bunch of pharmaceutical researchers settled on 95%, as a 5% error rate was the highest they were willing to tolerate in assuming that a new drug was effective. Everyone else then adopted 95% as the standard, even though it isn't really appropriate in all situations. So, anytime you read the results of a poll, or other statistical study, decide for yourself how confident you want to be, and use that to decide whether the difference is significant to you.
5% strikes me as a number that people find agreeable, or elegant, or pleasing. Partly because it's (essentially) a whole number, and not 4.38%, or irrational. And also because it's half way along the number line of our base 10 mindset and satisfies a human inclination for orderliness. The latter is almost always a big mistake in science. Nature has its own sense of order that's different from a human's.
I'm pretty sure if we were living in a base 12 world, 6% would be a threshold for many things.
In this case, it isn't really that 95% is a number that makes sense in base 10. lancc hints at the actual reason it's convenient: on the normal distribution, a 95% confidence interval is 1.96 standard deviations from the mean. Very close to an even 2.00. So, the interval is easy to calculate.
Well that and it's a happy medium for beginners. Covering why different alphas are appropriate for different situations adds yet another laying of complexity. What I'm thinking here is high-school appropriate topics that can be covered in a single unit. There's a lot of material you can drop that the usual pedagogy insists on before getting to hypothesis testing such as -- IMHO -- data types, probability, basic discrete & continuous types of distributions, etc.
Nice post.
But I always thought Base 8 made most sense if we ignored our fingers and toes because you can cut it in half with no fractions, like none of this 2 1/2 as half of 5 stuff... I suppose under Base 8, 4% would be the threshold for many things...
I tell the god botherers "If there were a god he'd have given us 8 fingers and 8 toes."
People give the English system of weights and measures a bunch of crap, which is justified only if you're using base 10. Go ahead and try to divide a standard length into 10 equal segments. It's ridiculous. The same goes for dividing a standard mass into 10 equal masses. For non-Americans reading this blog an inch is divided into halves, quarters, eighths, etc and there are 16 ounces in a pound.
Major problem with the metric system is the names: centimeter, millimeter, kilometer, kilogram, etc.
An overly rational approach that presumes the user must know from the prefix the scaling involved. Do we really need to have the word inform us that a centimeter is one hundredth of a meter? Of course not. Also, too many syllables.
Another example of over-rationalizing is the meter. Could have been set to equal one yard (or one foot). Instead it was one ten-millionth the distance between the poles and the equator (!). Absurd.
Imperial unit names are almost all short: inch, mile, pound, ounce, foot, yard, ton, (fluid) ounce, cup, quart, gallon, acre,...
I've explored how the metric system could have been introduced with same-name units, close to Imperial, but adjusted so x10 walks closely up the scaling to existing Imperial.
That's what should have been done. Okay,, you have the "old" foot and the "new" foot, and they're somewhat different. So what? In a century everyone would think in terms of the newer scaling - with the same one-syllable names - which would be a boon to humanity.
OLD: Give someone an inch and they'll take mile.
NEW: Give someone a centimeter and they'll take a kilometer.
"Also, too many syllables."
Not good. Need less of those.
And as for easily dividing English units, try dividing 3 feet, 3-3/8 inches in half. That's one meter, by the way, so half is 0.5 meters. For less round numbers you can do it easily with a calculator.
Even older: Give someone an inch and they'll take an ell.
But dividing an unknown length or mass into equal parts is something we rarely do; dividing a pie is about equalizing apparent size. On the other hand, scaling a recipe to make more or fewer portions is a common task, and I find it much easier to multiply or divide grams or millimeters than Imperial units. I have encountered measurements like “2/3 cup plus 1 tablespoon.” What’s half of that, or 1-1/2 times that?
If everyone had 8 fingers we'd be living in the Simpsons universe.
2D, so much simpler.
GOD YES!!! I've advocated for a long time that high schools lower the algebra requirements for graduation and add a probability/statistics requirement. At the introductory level, statistics is arguably easier than the current algebra requirement (it's been a few years since my daughter was in school, so that may no longer be true.) I think the hardest part is thinking of a good pedagogical approach to hypothesis-testing; the reject/cannot reject H_0 is confusing (why not just say accept H_1, of course). It's been my experience that everything is good up to this point, but past that, type I vs type II, power of a test etc. becomes more and more a matter of rote parroting.
Weird, because I saw a lot of students who breezed through the calc series stumble over the basics; when you said 'statistics', they thought 'distributions'.
"Fail to reject the null hypothesis" - my favorite example of a triple negative!
Statistics and probability should definitely be a high-school requirement. Descriptive statistics of course — central tendency, dispersion (especially quantiles), data visualization. Despite using statistics throughout my career, I have long been a skeptic of the null-hypothesis test of significance. Rather than the classical approaches to inferential stats, I would propose the randomization-stats concept: set up data-generation models for H-sub-0 and H-sub-1 and run thousands of simulated trials on computers. How much do they overlap?
To riff off of Hugh Keough, the ‘not statistically significant difference’ does not always hold up, but that’s the way to bet.
Thanks for pointing out that failing marginally to hit that magical 95% confidence does not mean that there is no significance. It's just that the confidence interval is a little broader. The choice of 95% for statistical significance was arbitrary but worked in an era when the calculations were done using tables of significance printed in books.
It's still only a single poll. I prefer to look at averages (or aggregates) of multiple polls.
Harris doesn't look to my eyes like she's up more than 2-3 points in the majority of the polls. And there's no guarantee as to how accurate they'll be. And if her margin is much less than 2.5 points or so... (I'd bet my last dollar she wins the popular vote, sure. But as we all learned in the sixth grade, that's not how America elects its president. The margin of her popular vote win means everything. Thanks, Madison!)
Sorry to be Debbie Downer. I'm really not! Or, to put it this way: Am I feeling a lot better about the election than I was a month ago? You betcha! Does my estimation of Harris's chances of prevailing approach anything that looks like "confidence"? Not on your life. Speaking of Debbie Downer, this is a classic: https://www.youtube.com/watch?v=TfE93xON8jk
It's also important to note that the Ipsos/ABC/WaPo poll has pretty consistently been very favorable to the Democrats, at least in this cycle. Right before Biden dropped out, it had him tied with Trump.
Nate Silver has her up nationally by about 2.7 points, I think, and Economist Yougov by about 3 points, if memory serves.
Things have obviously moved in the right direction over the last few weeks, but I'll rest a lot easier after it's over. Also, I hope she gets some additional bounce from the convention. It would be nice if nearly all the polls showed she was up by 5-7 by Labor Day.
as you're a "tax reasons" trump voter, your opinion is less than worthless.
There are a lot of 'tax reasons' Trump voters here; Jasper_in_Boston is most definitely _not_ one of them. I respect his/her takes. I may disagree with them, but the thought they are not offered in good faith crosses my mind.
TL;DR: Don't fall into the easy habit of identifying people as trolls merely because you disagree with them. Give 'em enough rope is my motto, and that's why I keep a 'rolodex'.
I screwed up; the sentence:
"I may disagree with them, but the thought they are not offered in good faith crosses my mind."
should read:
"I may disagree with them, but the thought they are not offered in good faith never crosses my mind."
"Bleah", said Toad.
It’s statistically significant compared to what the polls read a couple of weeks ago. You ain’t seen nuthin’ yet.
The fact that it is not something like 80/20 for Harris is a depressing statement about American electorate.
Maybe!
Depressing, but not surprising. It was 20 years ago that the "Keyes constant" enjoyed a season of online popularity. It came from the fact that Alan Keyes got 27% of the vote in his race for US Senate against Barack Obama.
Racism wasn't a factor as both candidates were Black. But Obama was clearly more qualified, more knowledgeable, more experienced, etc. Heck, Keyes didn't even live in Illinois.
And he still got 27% of the vote. So, some argued, that's the base: 27% of the electorate is, to use a technical term, bats*** crazy.
With Trump, it appears his base of support is larger, somewhere around 44%. Fortunately for the republic's future, Trump's ceiling of support appears to be not much higher, around 47%.
I think your Trump floor/ceiling estimates are pretty good. I’ve only seen one or two state polls, and don’t recall any national ones, that reached 50%; well within random-error expectations.
Even Dick Cheney had an approval rating in the 20's when he left office. GWB was around 30%. Based on this, it's surprising to me that Trump's floor is so much higher. One could do a PhD thesis on why that is.
That was the “crazyfication” factor. Track down the original blog post of Kungfu Monkey to read the whole thing. It wasn’t just insightful it was damn funny, too. I’m sorry John Rogers no longer blogs because he was brilliant and hilarious. His blog also gave us the famous “two books that can change a young man’s life, Atlas Shrugged and Lord of the Rings…”
The full quote for those who don’t know it:
“There are two novels that can change a bookish fourteen-year old’s life: The Lord of the Rings and Atlas Shrugged. One is a childish fantasy that often engenders a lifelong obsession with its unbelievable heroes, leading to an emotionally stunted, socially crippled adulthood, unable to deal with the real world. The other, of course, involves orcs.”
http://kfmonkey.blogspot.com/2005/10/lunch-discussions-145-crazification.html
Thank you, Yehouda. I was too lazy to dig that up, even tho I’ve read it several times since I first saw it when it came out. I’m sorry Rogers doesn’t write much like that anymore. He seems more focused on his tv writing which of course makes sense. I’ve enjoyed his series Leverage.
I've learned something today -- thank you! I've loved that quote since I first heard it, but until today I didn't realize that it was from John Rogers, whom I know from "Leverage". (I strongly agree, it's a fun show.)
Meooooow!! It's ALL downhill to the kitty box now for the convicted whore monger...KITTIES KNOW.
Time for the MSM to start ripping Harris/Walz. Don't want a blowout--Bad for business.
Looking at you NYT...Isn't there a secret server somewhere that needs to be "investigated"?
It isn't so much that the NYT rips Democrats it is that their standard of "objectivity" ends at reporting accurately what a politician says . So if a Republican says "Tim Walz deserted his unit to run for office", they will report that even if they know it isn't true. The notable exception is for Donald Trump where they do a massive amount of editing of what he says to make him appear to be a "normal" candidate.
Agreed. It’s annoying as hell. They’re not doing journalism anymore. Journalism is about informing the public, not disseminating misinformation. The most recent one I saw was reporting a GOP accusation (Vance, I think) about ‘skyrocketing crime’ without comment.
Don't want a blowout--Bad for business.
Unfortunately true. Everything I've observed about Washington Post editorials supports this interpretation. The very definition of "both sides" journalism.
One exception that I found: The Philadelphia Inquirer:
https://www.inquirer.com/opinion/editorials/inq2/trump-threat-bizarre-claims-hannibal-lecter-confusion-helicopter-sharks-20240815.html
"Most Republican officials, and many in the media, ignore or downplay Donald Trump’s peculiar state of mind, but normalizing his unhinged musings is dangerous."
You know, Trump's father Fred Sr, was diagnosed with severe dementia at age 86, and it likely was plainly evident even before that. It runs in the family.
As Trump's unhinged musings become more frequent, perhaps even the MSM will find it impossible to hide them. Despite their efforts, voters may eventually deliver the blowout that they are trying to avoid.
i wonder when kevin will come to the support of his homie megan mc cardle's rejection of kamala's economic policy speech.
In baseball, momentum is your next starting pitcher. Since the Democrats changed their starting pitcher, the changes in the polls have been increasing encouraging with every new drop. Regardless of the baseline number, there have been no polls that show Trump's position improving since Harris announced. Nearly every poll, Rasmusson included, shows the opposite.
Another thing that is encouraging is that some polls are now showing Harris at over 50%. Once someone is in the Harris camp, they're not going back.
The third thing is that it's not Labor Day yet. When that large group of people that don't pay any attention to politics until we get very close to the election wake up and see a ray of sunshine on one side and a stain on humanity on the other, who are they going to choose?
The fourth thing is that polls have been off for the last three years by 8 points in the GOP's favor. Look at the Suazzi "dead heat" numbers in NY for an example: polls showed even, Suozzi won by 54-46. This has happened repeatedly. There is no evidence yet that pollsters have corrected for this difference.
Re the fourth thing: Polls have erred on the side of the GOP for the last *three* years, yes, but look back at how badly the polls going into Election Day 2020 overstated Biden's strength -- a spectacular failure that got lost in the post-election noise from the Right Wing Crazy Machine. Poll averages had him *winning* Ohio and Florida, and carrying the swing states by big margins. You can see it plainly in the 2020 graph here -- it's scary:
https://electoral-vote.com/evp2024/Pres/ec_graph-2024.html
I agree that this is concerning. It's true that polls have very consistently understated the Democratic vote in recent years, but that was generally when Drumpf was not on he ballot and often in low turnout off-year and special elections where Democratic enthusiasm in general (and anger at Dobbs in particular) were big factors missed by pollsters. Josh Marshall, one of our most astute commentators, has mused on these factors recently. Hopefully, it's sort of a wash and the current polls are fairly accurate.
the democrats: noted overperformers in offyear & special elections, historically.
it was never the gop that had high propensity voters.
Well, yes, that's true, and it's also true that you should never get confident merely because the polls are on your side. Guardedly, cautiously so in the case of near-term trending. But almost every professional I've talked to or read has said that public polling is more or less broken and all of them give more or less the same reasons which in the end boil down to the same thing: getting a statistically valid sample is becoming increasingly expensive. As one guy put it "You know there's something off when twenty percent of people under thirty in your sample are qualified nuclear submarine commanders."
There's a very large error in polling that's all but impossible to correct for: response bias. If a candidate is perceived to be doing well, his/her supporters are more likely to answer polls. If a candidate is perceived to be doing badly, his/her supporters are more likely to hang up. When you see sudden swings in poll numbers after some event you have to conclude that: 1) large numbers of people suddenly changed their minds, or 2) it's just response bias. With response rates of around 10%, even a small response bias can have a large effect. I don't think people are as fickle as all that, and I suspect #2 is a lot more of a factor.
Now response bias = enthusiasm, so it's not that it's unimportant. People vote when they're enthusiastic and stay home when they're not.
Also, polls before Labor Day have rather poor predictive value. Just ask President Dukakis.
The fourth thing is that polls have been off for the last three years by 8 points in the GOP's favor.
Not at the presidential level. Trump out-performed polls in both 2016 and 2020. However, I believe polling organizations usually fine-tune their methodologies to correct past errors (don't they?), so, maybe we finally arrive at an election where Trump under performs the polls? Here's to hoping!
"Not at the presidential level."
That's because we haven't yet had a presidential election since the Dobbs ruling. Pollsters have not yet factored in the millions and millions of severely pissed-off women.
Arizona and Nevada will have protection for the right to choose on the ballot this November. So will Florida, and a number of ‘safe’ states. Florida may actually become interesting; at least, the GOP may have to expend resources there that they didn’t expect to.
And what makes all this even worse is that the WaPo, and the rest of the media, treated Drumpf as pulling away to a certain win when their polls showed him in a dead heat with Biden or leading by a smaller amount than Harris is now.
because that's what they want.
Once you understand the relationship between national numbers and how that shakes out in the electoral college, you’ll understand why this was the case.
The few MAGA voters I know point to Biden's "lousy record" and how ditzy and incompetent they think Harris is. That is, they don't tout Trump's record very much. When Trump touts his record he lies openly, and when he attacks his opponents he does so --gleefully -- based on personal characteristics and uses name-calling. He also makes ridiculous claims about the Biden record. The more he lies and attacks apparently, the more he alienates many voters. Harris' team is running a good campaign by all accounts, and Trump is clearly running a bad one. The polls are quick to pick up on the responses of likely voters to the two campaigns.
Can trump and republicans screw up enough state wide election certifications to throw the election to the US house? Harris needs to win with a big margin.
I remain skeptical regarding the significance of poll results -- except perhaps of LIKELY voters in SWING STATES -- until into September, but the current trends are undeniable (and significant) and give cause for optimism, and the Dems haven't had their convention yet, which usually gives them a bump.
As noted upthread tho, the whole game is gonna be GOTV. And there's cause for optimism there too -- the general excitement and morale among Dems, the surge of volunteers, the infrastructure already in place. But two months is an eternity in politics, and the Republicans are working hard to sabotage elections at every level from the precinct to Congress. (Funny how the "rule of law" folks seem to ignore the rules if they don't like them.) Dems MUST NOT get complacent. Remember 2016!
As noted upthread tho, the whole game is gonna be GOTV.
That's one of the biggest myths in politics.
https://x.com/davidshor/status/1590358548923682817
https://x.com/NateSilver538/status/1601213715855994883
The 2022 election is a classic example. Republicans turned out in greater numbers than Democrats. So why did they have such a poor election relative to the fundamentals (they way underperformed the typical performance for an out-of-power party in a first term mid-term)? Answer: many Republican were persuaded to vote for Democrats. They defected!:
Heres the NY Times:
Emphasis mine.
Source: https://www.nytimes.com/2022/12/08/upshot/georgia-voter-turnout-republicans.html
You can find lots of cases where Republican candidates significantly underperformed other Republican candidates in a state. These are always uniquely poor candidates: Herschel Walker, Sharon Angle, etc. The big question is why they continue to vote for the most unqualified candidate in history. As he likes to point out, he got more votes in 2020 than he did in 2016. That's not R's voting for Biden. Turnout broke all records. If Harris to win, she's going to have to duplicate that.
Turnout broke all records. If Harris to win, she's going to have to duplicate that.
This is completely misguided. It is the GOP that is now the party that benefits from high turnout. I personally find it comforting that the political fortunes of the Democratic Party no longer depend on massive turnout spikes, which by definition rely on less dependable voters. Especially when brain-dead Republicans are still trying to reduce turnout! But the facts don't lie: Democrats now possess an inarguable advantage that correlates with lower turnout. This is mostly because college-educated voters are reliable voters.
https://www.nytimes.com/2024/06/15/upshot/election-democrats-republicans-turnout-trump.html
https://www.ft.com/content/b3738a2e-7094-4c92-93cc-f2fa340375be
https://nymag.com/intelligencer/article/biden-needs-disengaged-unhappy-voters-to-stay-home.html
https://www.washingtonpost.com/opinions/2024/02/15/gop-losing-special-elections-low-turnout/
This is true, as long as this election is just like the last one. Want to bank on that? Every election is unique and election history is littered with losers who thought otherwise. But there is always one constant: you need to get your voters to the polls, and enthusiasm translates to higher turnout.
I started to read the Nate Cohn piece; I gave up on it. There just isn’t sufficient data there to support the conclusions. For one election, he reports that a higher percentage of Republicans voted than Democrats. But what fraction of the electorate identifies with each party? No numbers. No info about how party identification was determined. No data on independents. No data on no-votes. Many factors, and various weightings of those factors, could be at play in the reported results.
That seems like insufficient data to conclude that large numbers of Republicans were voting for certain Democrats. Republican turnout could be inflated by their voters running up the score in areas and races they were never going to lose, while Democrats disproportionately turned out for marginal races, where presumably their GOTV efforts would have been concentrated.
If Democrats are going to win the House and Senate as well as the Presidency, a lot of swing voters will have to realize that Trump and Republicans are not going to give them what they want.
Yes.
Well, yes. And a lot of blue voters need to show up. And a lot of red voters need to stay home. It's not just about swing voters, it's also about turnout.
I don't know how these national polls sample, but if they sample the states proportionately by population the result is only approximately relevant to the outcome in the electoral college or for the Senate and House races. Trying to pin down the probabilities in national polls is not very useful, nor is aggregating those polls. For President, only some states are relevant. For best results there should be samples of thousands in those states, then the probabilities of the state results aggregated. Who is actually doing this? Not the quicky news media polls.
The acid test would be whether you would say the same were it Trump 49% Harris 45%…
"That's extremely statistically significant no matter what standard you use."
Maybe statistically significant. But not especially relevant as long as we're still using the Electoral College...
I do not understand people who trumpet these sorts of polls. Really makes it clear that most people see this as some sort of tribal game where the exact rules don't matter, what matters is the flag waving and endorphin boosts along the way.
For the few reading who actually care about the rules, the best site I've found (but pipe if if you find one better) seems to be:
https://www.270towin.com
You've just insulted our intelligence again by claiming we're too dumb and ignorant to know this, i.e., that you're intellectually superior to poor little us when the reality is that you're outclassed by most people here by virtue of their education, training, and occupation. Fuck off, you nasty little poseur troll.
Test
This attempted quality trolling is doing it for Republicans,
https://old.reddit.com/r/insanepeoplefacebook/comments/1ev9va9/weird/?cache-bust=1724021848871
Everyone else, too.
Test. Been having trouble posting today.
Finally somebody who understands statistics! if I hear "statistical tie" one more time, I'll throw up.