Science and Journalism Abandoned in Favour of Public Relations and Fantasy (or Buzzwords)
Inaccurate political surveys and AI [sic]
Recent: Dr. Andy Farnell's Article on Societal Disorganised Attachment and the Role of Social Control Media
A READER has sent us "the following article on political surveys. Though it is math-oriented, no formulas appear. It does a great job explaining why polls have become inaccurate. You may want to provide a link on your site."
To quote the article: "The main problem with polling recently is that the small number of people who respond to polls - they have a 1-2% response rate these days - don't accurately represent the overall voting population."
"So if you're a pollster, you try to weight your sample based on demographic data to make it more representative. For instance, the polling errors in 2016 were ascribed to not weighting for education level: More-educated voters were more likely to respond to polls, those voters favored Clinton, and pollsters hadn't corrected for it. If you know that voters without college degrees are, say, 20% less likely to respond to your poll than those with college degrees, you could weight voters without degrees more heavily to correct for that."
"But weighting can only go so far before it introduces problems of its own."
Fake science has given us all sorts of fake security and pure junk like Microsoft Windows. It also meant that voting machines (running Windows) could not be trusted and this, in turn, caused people not to trust elections. So it put democracy itself at risk/peril (people losing trust in election outcomes).
"The article is math-oriented but does not have any formulas," the reader told us. "It is a great eye-opener which goes down to the basics to get to the point. It should be of interest to people who have worked with statistics. Many people start their programming careers with programs that calculate basic statistics."
So not only voting itself is under attack; even the concept of objective surveys (or polls) has been replaced by PR stunts and lousy maths. There are now "left-wing" polls and "right-wing" polls.
"Whether you link it or not is up to you," the reader said, "and your decision should be made based on the merit of the article alone. If you can say good things about the article please do so; that would be nice."
"One reason I recommend the above article to you and readers of your site is that the problem discussed in the article happens in AI too. I notice that you often discuss the problems of artificial intelligence."
"See, the article is about inaccurate polls caused by a shortage of accurate survey data. No amount of fancy math will fill the gap. The pollsters use weights to make adjustments, but it is not working well."
"It is the same with AI. The system may be creating output based on insufficient input, in which case, the "intelligence" is bound to be inaccurate (save a miracle). I notice that AI systems don't tell users how much data is behind given output. They don't give confidence intervals, error margins, etc. which are well established figures in traditional statistics."
"Why trust AI when it can't tell you how much it trusts itself?" █