Another Monday Message Board. Post comments on any topic. As a reward to paying subscribers, I’m limiting comments to them. But if you want to comment for free, you can do so at my blog
1. Polling companies don’t make their money from election polls, which are essentially advertising for market research where they can charge customers for quality. Election polls are just good enough to print.
2. The cited error margins are those for true frictionless random samples, approximated by textbook scenarios of coloured marbles in large bags and the like. The real error bars are much larger.
4. Pollsters do their best to disguise their odd sample of respondents as fair representatives of the target population with heavy theatrical makeup, roughly like this.
- Step 1 is to buy a large stratified sample of say 10,000 voters from a data broker. The stratification categories used by the large CES survey are gender, age, region, education, race, “the interaction of education and race”, home ownership, and 2020 presidential vote. Other pollsters will have a slightly different list. Anyway the initial bag nicely matches the electorate in general, which is what they have paid for.
- Step 2 is to carry out the survey. The pollster gets lucky and gets 1,000 responses, so the raw error margin looks OK. Even more luckily, all 1,000 answer all the stratification questions - in reality, many hang up before it’s complete.
The problem now is that the crosstabs no longer match the initial large sample and the general population. The simplest case is gender, with three possible values (M, F, and won’t say). In the general population, the split may be 49%, 49%, 1%. In the respondents, it’s 45%, 53%, 2%.
-Step 3 is to correct in the target question (voting intention) by overweighting the Ms and underweighting the Fs and WSs. Rinse and repeat.
But each time they do this, the effective sub-sample size drops and the margin of error grows. Pollsters can’t correct for everything – and which categories they do pick is a matter of judgement, influenced by priors and herd effects.
5. To some extent, we can make the polls collectively more reliable by aggregation, increasing the effective sample size and narrowing the error margin. This only works however if the individual polls are independent. It seems unlikely that the essential reweighting choices meet this test.
6. “Monster” ha a striking example of a large category that no poll has tried to cover: personal experience of sexual assault. One of the presidential candidates has been accused of this by numerous women, and paid damages to one of them in a civil trial. You would expect an impact on voting – though whether it leads to higher or lower propensity to vote is uncertain.
Can this be fixed? The CES poll https://cooperativeelectionstudy.shinyapps.io/prez2024pre/ , with a much larger sample size (N= 78,000) and conducted by interviews not phone calls, suggests yes. But polling of this quality is much, much more expensive than is usual, and can’t be frequent. Down the road, we can imagine participation, perhaps in focus groups, as an onerous civic obligation like jury duty, properly rewarded.
*Footnote: “them” would be the usual pronoun but does not work in this sentence through ambiguity of the antecedent.
• EVs (BEV and PHEV) passed 50% of new car sales in July in China, earlier than expected. https://carnewschina.com/2024/08/07/chinese-new-energy-vehicle-car-sales-50-84-july-preliminary-figures-show/ So far this year, “sales of new energy vehicles have amounted to 4.991 million, which is a 34% year-on-year increase”. For comparison, cumulative PHEV sales in the US in the 13 years from 2010 to 2023 totalled 4.7 million. Chinese carmakers and suppliers must already be getting significant economies of scale in production, and Chinese households enjoying economies of networking in chargers.
That brings the world total of pre-commercial but fully sized EGS geothermal projects in progress to two, the other being Eavor’s smaller but more advanced scheme at Geretsried in Germany. The two companies use radically different technologies. Fervo stick with creating artificial hydrothermal reservoirs by fracturing deep rocks with high-pressure water and sand, which promises high yields - if you can durably get enough flow, the problem which many earlier ventures have failed to solve. Eavor bypass it, circulating water in sealed bored tubes heated by simple conduction from the surrounding rocks. This is a less risky and more conservative approach they have tested for years on a small scale. Both companies need to get their accurate drilling done cheaply enough to pay. These projects are risky, but reasonable long-shot bets with budgets in the hundreds of millions not billions. If either or both succeed, they will add useful flexibility to the portfolio. Geothermal is intrinsically almost perfectly reliable and despatchable, ideal backup for wind and solar, and geologically suitable locations are common. These features are worth paying a reasonable premium for.
• “For more than two years, the floating offshore wind turbine TetraSpar at METCentre off the coast of Norway has been monitored by a bird camera. No bird collisions have been recorded.” Total bird detections were 21,138, 91% of them seagulls. Gulls are quite large, slow-flying and not very agile, though good at gliding on air currents. They are gregarious, and warn the flock of threats. They get on fine with humans, though the affection is not reciprocated, and are happy to exploit opportunities created by human activities like fishing and unguarded sandwiches. They have obviously learned how to stay safe around a big wind turbine blade. The 2,000 collision-free visits from other species of seabird suggest that this coexistence is normal. https://cleantechnica.com/2024/10/24/no-bird-collisions-at-offshore-wind-site-study-determines/
Why are election polls so unreliable?
Commenter “Monster” at Daily Kos, who obviously knows more than I do on this. has a good post on this. https://www.dailykos.com/story/2024/10/26/2279855/-Some-more-words-on-why-polls-are-wrong-with-a-deep-dive A few takeaways – my simplified interpretations, don’t blame him or her (*fn) for my mistakes. These are some of the failings of “good” polls, leaving out the deliberately skewed propaganda ones.
1. Polling companies don’t make their money from election polls, which are essentially advertising for market research where they can charge customers for quality. Election polls are just good enough to print.
2. The cited error margins are those for true frictionless random samples, approximated by textbook scenarios of coloured marbles in large bags and the like. The real error bars are much larger.
3. Response rates for many political polls are catastrophic. The highly rated Pew organisation reported 6% for phone surveys already in 2018. https://www.msnbc.com/opinion/msnbc-opinion/trump-harris-2024-election-polls-challenges-rcna176467 The normal response to a pollster`s call is to block or hang up. Those who reply are a freakish minority of political junkies.
4. Pollsters do their best to disguise their odd sample of respondents as fair representatives of the target population with heavy theatrical makeup, roughly like this.
- Step 1 is to buy a large stratified sample of say 10,000 voters from a data broker. The stratification categories used by the large CES survey are gender, age, region, education, race, “the interaction of education and race”, home ownership, and 2020 presidential vote. Other pollsters will have a slightly different list. Anyway the initial bag nicely matches the electorate in general, which is what they have paid for.
- Step 2 is to carry out the survey. The pollster gets lucky and gets 1,000 responses, so the raw error margin looks OK. Even more luckily, all 1,000 answer all the stratification questions - in reality, many hang up before it’s complete.
The problem now is that the crosstabs no longer match the initial large sample and the general population. The simplest case is gender, with three possible values (M, F, and won’t say). In the general population, the split may be 49%, 49%, 1%. In the respondents, it’s 45%, 53%, 2%.
-Step 3 is to correct in the target question (voting intention) by overweighting the Ms and underweighting the Fs and WSs. Rinse and repeat.
But each time they do this, the effective sub-sample size drops and the margin of error grows. Pollsters can’t correct for everything – and which categories they do pick is a matter of judgement, influenced by priors and herd effects.
5. To some extent, we can make the polls collectively more reliable by aggregation, increasing the effective sample size and narrowing the error margin. This only works however if the individual polls are independent. It seems unlikely that the essential reweighting choices meet this test.
6. “Monster” ha a striking example of a large category that no poll has tried to cover: personal experience of sexual assault. One of the presidential candidates has been accused of this by numerous women, and paid damages to one of them in a civil trial. You would expect an impact on voting – though whether it leads to higher or lower propensity to vote is uncertain.
Can this be fixed? The CES poll https://cooperativeelectionstudy.shinyapps.io/prez2024pre/ , with a much larger sample size (N= 78,000) and conducted by interviews not phone calls, suggests yes. But polling of this quality is much, much more expensive than is usual, and can’t be frequent. Down the road, we can imagine participation, perhaps in focus groups, as an onerous civic obligation like jury duty, properly rewarded.
*Footnote: “them” would be the usual pronoun but does not work in this sentence through ambiguity of the antecedent.
More placebo antidotes to election gloom
• The IEA revised upwards its near-term forecast for global new renewables: 5,500 GW by 2030. This will bring total renewables to 2.7x 2022 levels, close to the PR target 0f 3x cynically or wishfully adopted by COP-28 less than a year ago. 80% of this will be solar. https://www.iea.org/news/massive-global-growth-of-renewables-to-2030-is-set-to-match-entire-power-capacity-of-major-economies-today-moving-world-closer-to-tripling-goal
• China officially installed 160 GW of solar in the first three quarters of 2024, 20.9 GW in September alone. https://cleantechnica.com/2024/10/26/china-adds-160-gigawatts-in-first-3-quarters-of-2024/
• EVs (BEV and PHEV) passed 50% of new car sales in July in China, earlier than expected. https://carnewschina.com/2024/08/07/chinese-new-energy-vehicle-car-sales-50-84-july-preliminary-figures-show/ So far this year, “sales of new energy vehicles have amounted to 4.991 million, which is a 34% year-on-year increase”. For comparison, cumulative PHEV sales in the US in the 13 years from 2010 to 2023 totalled 4.7 million. Chinese carmakers and suppliers must already be getting significant economies of scale in production, and Chinese households enjoying economies of networking in chargers.
• “The [US] Bureau of Land Management (BLM) has approved the Fervo Cape Geothermal Power Project in Beaver County, Utah, which will use innovative technology to generate up to 2 gigawatts (GW) of baseload power….” https://www.blm.gov/press-release/biden-harris-administration-takes-major-steps-accelerate-clean-energy-geothermal
The 2 GW is hype of course. Fervo’s actual project https://fervoenergy.com/fervo-energy-breaks-ground-on-the-worlds-largest-next-gen-geothermal-project/ is for a still significant 400 MW, with 24 planned wells and 320 MW of PPAs signed with Southern California Edison.
That brings the world total of pre-commercial but fully sized EGS geothermal projects in progress to two, the other being Eavor’s smaller but more advanced scheme at Geretsried in Germany. The two companies use radically different technologies. Fervo stick with creating artificial hydrothermal reservoirs by fracturing deep rocks with high-pressure water and sand, which promises high yields - if you can durably get enough flow, the problem which many earlier ventures have failed to solve. Eavor bypass it, circulating water in sealed bored tubes heated by simple conduction from the surrounding rocks. This is a less risky and more conservative approach they have tested for years on a small scale. Both companies need to get their accurate drilling done cheaply enough to pay. These projects are risky, but reasonable long-shot bets with budgets in the hundreds of millions not billions. If either or both succeed, they will add useful flexibility to the portfolio. Geothermal is intrinsically almost perfectly reliable and despatchable, ideal backup for wind and solar, and geologically suitable locations are common. These features are worth paying a reasonable premium for.
• “For more than two years, the floating offshore wind turbine TetraSpar at METCentre off the coast of Norway has been monitored by a bird camera. No bird collisions have been recorded.” Total bird detections were 21,138, 91% of them seagulls. Gulls are quite large, slow-flying and not very agile, though good at gliding on air currents. They are gregarious, and warn the flock of threats. They get on fine with humans, though the affection is not reciprocated, and are happy to exploit opportunities created by human activities like fishing and unguarded sandwiches. They have obviously learned how to stay safe around a big wind turbine blade. The 2,000 collision-free visits from other species of seabird suggest that this coexistence is normal. https://cleantechnica.com/2024/10/24/no-bird-collisions-at-offshore-wind-site-study-determines/