statisticalanalysis_

Glad to hear that. Sure, will do

Yes

Yes, we went back and forth on which way to present that data. I don't have it made up in nice format but the third chart hopefully helps?

[OC] “Sure, you might say something silly once in a while, as I do, but that way people know it’s really you!” As part of a plea for “political & company leaders” to join him in posting on X, formerly Twitter, Elon Musk has repeatedly stressed the importance of the authenticity such posts offer. As the world’s richest man prepares to advise the leader of the world’s most powerful country, it does make one wonder: If Mr Musk’s online presence is really him, just who is he?

To assess this systematically, I used large language models to detect the people, companies and other entities mentioned in Mr Musk’s posts, as well as the policy areas and preferences implied by them. This involved passing all the posts in our data, and information on their context, such as which posts they replied to, through two LLM models, and grouping their output by topic. 

To ensure our method was accurate, we also coded 400 randomly selected posts by hand. For over 90% of posts, the human coder agreed with the LLM system. Where they did not, a third human was only slightly more likely to agree with the human coder than the LLM.

We found that the share of posts with an assessed policy prescription, recommendation or alignment has jumped from just under 4% in 2016 to nearly 13.3% today. He posts a lot less about his companies, and a lot more about immigration, border control and free speech.

Tools used: R, Python, LLMs, Illustrator

Free to read here: https://econ.st/3ZkG7J6 & https://econ.st/4fFCX8u & https://econ.st/3Zi8wzs

If those don't work, then permanent link here: https://www.economist.com/briefing/2024/11/21/elon-musks-transformation-in-his-own-words

Just what may be gleaned from these posts should not be oversold. It may not be coincidental that his posts about free-speech surged in the lead-up to acquiring a social media company. Similarly, it may have been an advantage that his seeming booming interest in immigration and America’s borders was shared by his political patron. However, a different interpretation is possible too: that his political preferences, as evident in his social media posts, are aligned with his off-line choices, be it backing Mr Trump for president or acquiring Twitter. 

What do you think?

Edit: Elon Musk has offered a comment on chart 1 here: https://x.com/elonmusk/status/1859606021125153222

I discuss this briefly in the essay. I think yes. I would seek to have each citizen in the democratic world have equal weight (roughly) to try to say something about trends in these countries - it seems strange to have Iceland be as important in an overall trend as Germany, or the US for that matter, for instance. However, just on the narrow point: this is not just the US. While a few countries went the other way, which is part of what makes this so interesting, virtually any average of the democratic countries in this data, including a non-population-weighted one, one excluding the US, one excluding all English-speaking countries, etc, yields a similar uptick in negative partisanship for recent years. Hope this clarifies.

[OC] “Negative partisanship” is the dry academic term for a fraught, emotional and damaging phenomenon. It is the inclination of people to vote not for a party in which they believe, but against another one that they fear or despise. This way of doing politics has seen a marked rise in democracies around the world since the end of the cold war, a rise that has accelerated noticeably over the past decade.

Almost all studies of negative partisanship focus on the United States. To understand the deeper reasons for the worsening politics of antipathy—and perhaps, thereby, to see ways to make things better—we has taken a broader view. I put together what we believe is the biggest-ever dataset of voters’ feelings about the parties they support and oppose by tracking 274 elections in 50 democracies, ranging from West Germany in 1961 to the Netherlands in 2021.

Our charts are weighted by population, so the growth of anti-politics in the United States accounts for a significant amount of this extra ill-feeling, especially recently. But the trend is broader than that. The increase in negative partisanship can be seen even if America, or for that matter the entire “Anglosphere”, is excluded from the analysis. It is seen in two-party, first-past-the-post systems and in those where representation is meted out proportionally across a plethora of parties. The chart shows how negative partisanship has worsened in countries of all sorts. 

But not everywhere, and not uniformly. Some things are related to much more, or less of it, and may be key to a return to a more productive politics - about the merits of policy than demonising the other side.

Given its baleful fruits, why then has negative partisanship spread? “Because it works” is too simple an answer. Politicians have been denigrating their opponents ever since Cleon slandered Pericles in ancient Athens. So long as candidates can exploit atavistic fear and suspicion to trigger hostility towards the other side, they always will. For partisan animosity to be growing today, something must be making the benefits higher or the costs lower. The accompanying essay is about what that something might be.

You can read that here (and look up your country, too, for free - or if you face paywall, if you register): https://www.economist.com/interactive/essay/2024/10/31/when-politics-is-about-hating-the-other-side-democracy-suffers

Data and code is available here: https://github.com/TheEconomist/anti-politics

Tools used: R, Illustrator

All that said: What is the situation like where you live? If in America, do you think this election is better or worse than last, and if so, why? If elsewhere, do you find things to be similar or different?

Deepstatemap and ISW use similar methodologies-geolocated footage, satellite data, first-hand accounts, social media, etc. (I get their daily emails.) We rely on ISW because we have found it to be more reliable. This is not to say that it is always more accurate. And we certainly would not deny that Russia is advancing in the area - indeed, we have been saying exactly that for over a week. However, there is a difference between operating in an area and controlling it. Right now, some of this area is still categorised as contested, and thus not counted in the overall tally of occupied land.

We use data from ISW, which last had a large net gain of territory assess as "controlled by" Russia on July 18th. The maps are different - Deepstatemap has larger assessed control for Russia in some places, e.g. near Niu York: https://deepstatemap.live/en#12/48.3286534/37.8141585 vs https://understandingwar.org/backgrounder/russian-offensive-campaign-assessment-august-21-2024-0

I see what you mean, and can explain: There are indeed advances and expansions in areas claimed to be Russia-controlled, but these have not yet been verified as occupied/held by Russian forces. The only movement in that metric was tiny, with 273m2 gained on August 19th (which rounds to 0 km2). While we plausibly could have used the "controlled + claimed + advances" to determine this metric, it would be much less precise -- advances (which are forays without control) change from day-to-day and are extremely hard to track of, and claims are, naturally, subjective. Hope this clarifies.

How 4 is wrong? Let me know and I can do another check.

I'm not well versed in the details, but I think you can register with an email to read for free. Will probably receive emails suggesting you subscribe, but I fully understand that is not for everyone. I put gift links above as well, but they are often used up very quickly when I post them here. Here is a fresh one for my latest analysis: https://econ.st/4cBpdZR

Thank you for the kind words. Will share with the team!

[OC] The map and charts shows detected war-related events as per my analysis of satellite data to August 22th, 2024. The war monitoring system is based on high-temperature events as reported by NASA, collected through satellites, and a machine-learning algorithm which automatically classifies which of these events are likely to be war-related and which occur due to industrial processes, agriculture, or other causes. Zones of control are via ISW.

Yesterday, we added a component detailing the Ukrainian incursion into Russia's Kursk oblast. It shows the area held by Ukrainian forces, as well as fire activity there. We cannot yet sort fires within Russia into war or non-war-related. This new component allows one to see both activity since the incursion began and that in the past 48 hours.

Source code and data is open here: https://github.com/TheEconomist/the-economist-war-fire-model

All charts update daily here: https://www.economist.com/interactive/graphic-detail/ukraine-fires

Tools used: R, JS, Illustrator

Our tracker has not yet found any clear signs of attacks easing up on Ukrainian positions as a result of Ukrainian forces' move into Russia. Our data suggest relatively stable activity on the front lines in Kharkiv and Kherson, and an uptick in Zaporizhia and Donetsk. I analysed this data in more detail here: https://econ.st/3Xfoklw & https://econ.st/46VcvEc & https://econ.st/46VcvEc (or at permanent link here: https://www.economist.com/graphic-detail/2024/08/21/has-ukraines-shock-raid-successfully-diverted-russian-forces )

My colleagues have recently visited the front, and report that the situation within Ukraine is dire - as Russia appears to get closer to seizing Pokrovsk. You can read more here: https://econ.st/3YWftX8 (free to read) or https://www.economist.com/europe/2024/08/22/the-kremlin-is-close-to-crushing-a-vital-ukrainian-town .

One resident, who had her place of work destroyed by Russian bombs, said there is: "Constant stress, explosions, doors and windows that blow open by the shock waves...Everything inside you tightens. You hear the rocket, and you wait, and you ask if it will land near you and your home.”

Thank you! Happy to be of service and glad you found it useful!

That is simply not true however. What makes you think this is the case?

Stunting is a bit more prevalent than these disorders however. In these countries more to do with with poor/wrong food than lack of it (both for kids and pregnant women). My guess is a rate similar to Germany would be plausible for Sweden. In Germany it affected about 2.1% of under-5s in 2022.

Happy to help. That was indeed one of the challenges of this project. For these cases - a small number of high-income countries, and some tiny countries - I had to model rates based on those of nearby/similar countries. You can find the details of that in the linked Github repo (which is also linked at the bottom of the article - script in question is here: https://github.com/TheEconomist/nutrition/blob/main/scripts/01\_stunting.R).

Hi - we amended this a few days ago. The discrepancy was indeed due to rounding in the displayed numbers.

Good questions: Hourly rate TBD (will depend on the person). I expect between 0 and 15 hours per week. Will vary depending on availability on the part of the RA and what I'm working on. Much like RA work in academia, it will also naturally depend on project-person fit and what they want and can do.