--- title: Viral Polls Are Not Representative Polls | daily.dev description: Viral polling posts on Bluesky tend to favor Democrats more than the average poll for that race, not represent typical polling. Analyzing 90 days of posts from... canonical: https://daily.dev/posts/viral-polls-are-not-representative-polls-mkyk38lff twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Viral Polls Are Not Representative Polls | daily.dev og:description: Viral polling posts on Bluesky tend to favor Democrats more than the average poll for that race, not represent typical polling. Analyzing 90 days of posts from... og:url: https://daily.dev/posts/viral-polls-are-not-representative-polls-mkyk38lff og:image: https://api.daily.dev/og/posts/mkyK38Lff.png og:image:alt: Viral Polls Are Not Representative Polls og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Viral Polls Are Not Representative Polls **[Hacker News](https://daily.dev/sources/hn)** · 21 min read · 0 upvotes · 0 comments ## Summary Viral polling posts on Bluesky tend to favor Democrats more than the average poll for that race, not represent typical polling. Analyzing 90 days of posts from @usapolling.bsky.social, the author finds reposts rise as a poll's margin diverges from the race average in a pro-Democrat direction, while the actual Dem-Rep margin itself barely matters. The pattern holds across most individual races, with the NC Senate race as an exception. A mixed-effects model shows TX-Senate posts reshare heavily when they favor Democrats, while the Pennsylvania governor's race gets little attention. Data was scraped via the atproto library and extracted using GPT-6 Luna through OpenRouter with a Pydantic-validated schema. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Does the number of reposts on a poll post correlate with how favorable the poll is to Democrats, independent of the actual margin? Yes, a mixed-effects model of posts from a Bluesky polling account found that reposts increase as a poll's margin diverges from the average margin for that race in a pro-Democrat direction, while the actual Dem-Rep margin itself was relatively unrelated to repost counts. The model used random intercepts and slopes by race, built on 90 days of scraped posts. _Developers digging into social data bias patterns like this can find similar analyses on daily.dev._ ### How can I use an LLM with Pydantic to reliably extract structured data like poll numbers from noisy social media text? Define a Pydantic schema requiring every extracted field to carry verbatim evidence excerpts quoted from the source text, then validate that each quote is a real substring of the post and that percentages match their quoted tokens exactly. One implementation routed requests through OpenRouter's Auto Router to GPT-6 Luna with zero-data-retention models and a 'High' cost tier, using the instructor library for schema-validated responses. _Anyone building reliable LLM extraction pipelines can track techniques like this via daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 99 comments across hackernews (as of 2026-10-27). **TL;DR:** The discussion largely accepts the post's core premise that viral/shared polls are biased by what gets reshared, but pivots into a long-running broader argument about whether any polling (including professional) can be trusted given non-response bias. **Sentiment:** 5% positive · 25% mixed · 70% skeptical **The case for** - The selection-on-reshares effect described isn't unique to polls — any content shared selectively skews toward surprising outliers. - Non-response and partisan non-response bias are at least partially addressed by established survey methodology (e.g. weighting, demographic adjustment) rather than being unsolvable. - Averaging across many polls is more reliable than trusting any single poll, viral or not. **The pushback** - Many argue the fundamental problem — people who decline to respond may skew systematically toward one side — can never be fully corrected for, since you can't measure the opinions of those who refuse to answer. - Corrections for non-response bias are inherently a lagging indicator that fails when underlying voter behavior shifts. - Some see the reliance on 'professional' pollster credibility as an appeal to authority rather than a real methodological guarantee. - One commenter questions whether the reshare effect is really about partisan favorability versus simply pollster name recognition/reach, confounding the study's claimed finding. **By community** - hackernews (heated): A tangent from the post's specific finding spirals into a long, contentious debate about whether polling in general can ever be trusted given non-response bias, with some defending professional methodology and others dismissing nearly all polling as unreliable. **Hottest debate:** Whether professional pollsters can meaningfully correct for non-response bias or whether the problem is fundamentally unsolvable. **Open questions** - Does the reshare-bias effect found in the post hold up once you control for which specific pollster produced the poll (since some have bigger followings or house effects)? **Highlights** > This filter isn't specific to polls. Anything that reaches people through reshares gets selected on how surprising or pleasing it is, so the visible sample skews toward the tails. Marketing has the same problem with screenshots of campaign results: the ones that travel are the outliers, and anyone calibrating on what they've seen shared ends up with expectations nobody can hit. Curious whether the effect survives controlling for pollster. Some pollsters have consistent house effects and some have much bigger followings, so part of the slope could be "polls from well-known outfits get reposted more" rather than "favorable margins get reposted more". > — [swipeharbor on hackernews](https://news.ycombinator.com/item?id=50260921) > This is totally wrong. Accounting for partisan non-response bias is a standard part of a pollsters job, not some insurmountable gotcha. Some are better than others at this. Polls aren't oracular, nor are they claiming to be. It's almost always better to look at an average of polls rather than a single one. There is a lot of public data on past elections that anyone can analyze and see exactly how accurate polls have been in the past. If you're expecting polls to be a perfectly accurate prediction then you are misunderstanding their purpose. What they are great at is giving a probabilistic estimate of the outcome, within a certain range. It's likely that any key race you care about next week will end up within 4 points of the current polling average. It's extremely likely that it will end up within 8 points. That's useful information! > — [rurp on hackernews](https://news.ycombinator.com/item?id=50261188) > And there's no clear way to adjust for the most basic factor that people who vote for one candidate are less likely to say so. In order to account for that, you need you know how many people who are less likely to respond are voting for that candidate. And then you're back to square one. This is why professional polls have had a consistent bias over the last few years, as this hidden unobservable factor's effect has gotten stronger and stronger. > — [toomim on hackernews · 1 comments](https://news.ycombinator.com/item?id=50260891) > They do a variety of things based on professionalism, budget, ability to gauge that. My point is --> this problem is the most basic and fundamental problem that is the focus of every pollsters existence <-- Asking questions is extremely easy. Getting a 'truly random sample' is extremely hard. Trying to factor in those things is 100% of the work, that's what they do, it's their job. Some are better, some are lazier, some have better data, some are individuals, some are teams so it varies hugely. When the conditions are controlled they'll have a better idea than otherwise. Their job is not 'asking questions' it's 'figuring out how to get random samples'. > — [bluegatty on hackernews · 1 comments](https://news.ycombinator.com/item?id=50262225) > I think we should do our best to ruin polling data. Polls are the first step in a pipeline that effectively turns money into votes. The better that pipeline functions, the worse our democracy functions. Let the only accurate poll be the election itself. > — [\_\_MatrixMan\_\_ on hackernews · 5 comments](https://news.ycombinator.com/item?id=50261754) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=50260242) · 34 points · 99 comments ## Similar posts on daily.dev - [\[2510.25417\] Shifts in U.S. Social Media Use, 2020-2024: Decline, Fragmentation, and Enduring Polarization](https://daily.dev/posts/2510-25417-shifts-in-u-s-social-media-use-2020-2024-decline-fragmentation-and-enduring-polari-fyxhq0gyy) · Hacker News · 0 upvotes · 0 comments - [Forced to pick, Grok’s AI model chooses Republicans. OpenAI’s picks Democrats.](https://daily.dev/posts/forced-to-pick-grok-s-ai-model-chooses-republicans-openai-s-picks-democrats--80eg2wzzm) · TechCrunch · 0 upvotes · 0 comments - [AI disinformation is outpacing platform moderation before US midterms](https://daily.dev/posts/ai-disinformation-is-outpacing-platform-moderation-before-us-midterms-y2mf2kjnk) · The Next Web · 1 upvotes · 0 comments - [Beyond the Scroll: How Social Media Algorithms Shape Your Reality](https://daily.dev/posts/beyond-the-scroll-how-social-media-algorithms-shape-your-reality-bmzpwxszy) · Towards Data Science · 0 upvotes · 0 comments - [Leading AI models \(even Grok\) are all a bunch of leftist punks](https://daily.dev/posts/leading-ai-models-even-grok-are-all-a-bunch-of-leftist-punks-q6lzzlpua) · The Register · 1 upvotes · 0 comments --- Tags: [#data-analysis](https://daily.dev/tags/data-analysis), [#pydantic](https://daily.dev/tags/pydantic) [View this post on daily.dev](https://daily.dev/posts/viral-polls-are-not-representative-polls-mkyk38lff) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Viral Polls Are Not Representative Polls","url":"https://daily.dev/posts/viral-polls-are-not-representative-polls-mkyk38lff","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/viral-polls-are-not-representative-polls-mkyk38lff"},"datePublished":"2026-10-27T17:01:12.450Z","dateModified":"2026-10-27T20:19:51.933Z","description":"Viral polling posts on Bluesky tend to favor Democrats more than the average poll for that race, not represent typical polling. Analyzing 90 days of posts from...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/474385f67b502e87184c649cd37db3b7?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/474385f67b502e87184c649cd37db3b7?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hacker News","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/viral-polls-are-not-representative-polls-mkyk38lff","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"data-analysis,pydantic","timeRequired":"PT21M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"Viral Polls Are Not Representative Polls"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/viral-polls-are-not-representative-polls-mkyk38lff#faq","mainEntity":[{"@type":"Question","name":"Does the number of reposts on a poll post correlate with how favorable the poll is to Democrats, independent of the actual margin?","acceptedAnswer":{"@type":"Answer","text":"Yes, a mixed-effects model of posts from a Bluesky polling account found that reposts increase as a poll's margin diverges from the average margin for that race in a pro-Democrat direction, while the actual Dem-Rep margin itself was relatively unrelated to repost counts. The model used random intercepts and slopes by race, built on 90 days of scraped posts. Developers digging into social data bias patterns like this can find similar analyses on daily.dev."}},{"@type":"Question","name":"How can I use an LLM with Pydantic to reliably extract structured data like poll numbers from noisy social media text?","acceptedAnswer":{"@type":"Answer","text":"Define a Pydantic schema requiring every extracted field to carry verbatim evidence excerpts quoted from the source text, then validate that each quote is a real substring of the post and that percentages match their quoted tokens exactly. One implementation routed requests through OpenRouter's Auto Router to GPT-6 Luna with zero-data-retention models and a 'High' cost tier, using the instructor library for schema-validated responses. Anyone building reliable LLM extraction pipelines can track techniques like this via daily.dev."}}]} ```