November 11, 2020

Judging soundness of elections from bellwether states / counties (a project for someone else)

This post is split into two sections. Morons and partisans will only read the first section, and either accept it or reject it based on which party they are wedded to. Prove you are not a moron, and not a partisan shill, by reading through the second section.

First, a glaring red flag of the soundness of the media-alleged results of the 2020 election (i.e., a Biden win), judging from the bellwether states of Ohio and Florida. Both handily went to Trump, just like last time, strongly suggesting he will be re-inaugurated in January 2021.

Each state on its own has historically had a high agreement rate with the national winner. Sometimes one will miss, sometimes the other will miss, but have they ever both missed in the same election? Only once, all the way back through 1848, when both states have been in the union. If you know about stolen elections, you can already guess which one it was -- 1960, when the outcome in Chicago was rigged by the corrupt urban Democrat machine, swinging Illinois and its EC votes to JFK, and deciding the national result against Nixon, who both Ohio and Florida voted for.

Back then, Chicago was still the "Second City" after New York in population, and Illinois carried 27 EC votes (out of 537 total). Today, that steal is equivalent to stealing Wisconsin, Michigan, and Maine's 2nd district, combined. Or Pennsylvania, Iowa, and ME-2, combined. That shows how outsized of an effect a single large city can have on the national result, if it is rigged.

When your only historical precedent for 2020 is 1960, the presumption is that 2020 has been rigged / stolen at the "counting ballots" stage, albeit more obviously this time because it's in 4 states (Wisconsin, Michigan, Pennsylvania, Georgia), and therefore leaving more fingerprints. The most flagrant fingerprint this time is when they halted their elections around 10pm on election night itself, and only resumed the next day, with a risible reversal of fortune for the leading candidate from before the halt.

Unlike 1960, though, the party trying to ram the steal through the next stage, where state legislatures choose their slate of electors, and not having that over-ruled by SCOTUS -- is the opposition party, not the dominant party. Dems were dominant during the New Deal era, but are the weak opposition party in the Reagan era. Today the presidential candidate targeted by the steal, the state legislatures in the stolen states, and the majority of the Supreme Court, are all from the dominant party of this era, the GOP.

Reminder to political LARP-ers: if you're going to try to steal the national election, you'd better make sure you're the dominant party of your era. Libs and leftoids still cosplay as the Democrats of the '60s-'70s heyday of Civil Rights and the New Left, when they could get away with whatever they wanted, like pushing a Republican president out of office. Unfortunately they find themselves today in the Reagan era, where the balance of power has shifted so far away from their party that their LARP-ing only sets them up for destruction by a reality check.

* * *


This bird's-eye-view of bellwethers reveals the importance of combining the separate signals into a single index. The "errors" that a bellwether makes -- when it goes against the national result -- ought to be randomly distributed across the years. And these errors are few in number because that's what makes them bellwethers -- they rarely miss.

So if a particular year shows a lot of the bellwethers making errors in unison, that calls the soundness of the national result into question. Specifically, some process above and beyond the lowest-level casting of ballots by voters, such as the ballot-counters rigging the outcome by adding illegitimate ballots for their guy, ignoring ballots for their rival, or whatever other methods.

Here's the obstacle for constructing the history of "bellwether hits and misses," though -- the list of bellwether states and counties is not the same over time. Some may be reliable over a long period, but others may show a heyday of bellwether status, before and after which they were not so reliable.

If you just want to pick a start and stop point, and see which states and counties had the highest hit rate, fair enough. That's just a description of the period, with no inferential applications in mind.

But to judge the soundness of a particular election, you have to put yourself in the perspective of that year -- therefore, all information from future years is excluded, and the year itself is excluded because it's under consideration. How far back you go from the year in question is up for debate, but it has to include the most recent election before, and trailing backward from there.

For example, let's go back to the stolen 1960 election. If we want to assign a "bellwether warning index" to 1960, we cannot select states and counties to make up the list of bellwethers based on their performance in any point after 1960, and we're excluding 1960 itself because that's what we're uncertain about. The list of bellwethers would be chosen based on their hit rate from some sufficiently long interval that went up through the most recent election of 1956.

We cannot take the bellwethers of 2016 and apply them back to 1960, because some of the current ones were not bellwethers back then, and others from back then have not survived as bellwethers through today. It would be imposing our perspective on another time. Maybe the two will turn out to overlap, but to be as accurate as possible, each year requires its own proper list of bellwethers.

That still allows the analysis to proceed and give an interpretable result. Each year has its own list of bellwethers -- those that clear a sufficiently high threshold of hit rate (e.g., 90%), evaluated over a sufficiently long interval backwards including the previous election (e.g., 15 elections). The longer the interval, the lower the hit-rate threshold should be -- easy to hit 90% over 4 elections, impossible to hit 90% over 40 elections. I'd still say nothing lower than 80-85% even for periods over 50 or so years.

Then for the year in question, each bellwether result is compared to the national result. The total number of misses is added up for the year in question, and that is the index signal for that year. Weighting each component of the index by how reliable each bellwether is relative to the others, gets into over-fitting problems.

Then the index signals from each year can be plotted over time, and we could see spikes in the signal that flag certain elections as suspicious.

This is the point of a solid historical analysis -- we would not only detect a spike from 2020, which is obvious, we would see how unusual it is compared to the rest of our nation's history -- contra the cope that "shady stuff always happens". Moreover, we might detect other (rare) suspicious exceptions, like 1960, contra the other cope that no election has ever been purportedly stolen before.

The insane deviation of 2020 is convincing enough on its own that this is an attempted theft-in-progress, but an even more convincing story would show how rare such a signal is over history -- and how perhaps the only other large deviation is the known stolen election of 1960.

However, I will not be doing this analysis myself. Whoever wants to, can begin with the database sources, and the code for analyzing them, in this post from Adam Obeng. It is county-level, and he looked at 1952 through 2012, but you can substitute different years into his code for your own purposes. Still, that will only generate the list of bellwether counties for a single year. You have to add your own code to do that for each separate year, so you're not judging one period by the bellwethers of another. And you'll write your own code to see how well the bellwethers performed for a year in question, add the misses into an index for that year, and then plot those index signal values over time.

This cyber-professor is handing over such a project to his cyber-grad-students and cyber-postdocs because I'm not terribly passionate about it myself, it will be good practice for them, and they need correction from their current analyses of bellwethers, which use a single list of states or counties to compare elections across several decades, when each election needs its own list of bellwethers. But the basic sources and methods are already there in the post linked to above, so have at it.

I also want to impose some "doing the work" requirements on the "weaponized autism" crowd, who typically waste most of their time just scrolling a social media feed. So many man-hours utterly wasted on social media activity, striving to be talking heads (reacting avatars) and call-in guests (reply guys), just on a 21st-century media platform rather than radio and TV like the Boomers.

I also notice that "the autists" don't get much on a theoretical or conceptual level that ought to shape their interaction with data, even if the point is simple (like bellwethers not being the same states / counties across all years). Instead they consume "the data" as just another form of digital content to entertain themselves with for a take-cycle or two, and then totally forget about it, don't polish it, and never follow up on it.

Here's to hoping that they'll be more motivated to do the most persuasive possible analysis when their president's re-inauguration hangs in the balance.

6 comments:

  1. I read Obeng's post, and he didn't say anything about bellwhethers being evidence of fraud. In fact, he said counties with a hot-streak of 15 in a row still had roughly a coin-toss likelihood of getting the next election right. The conclusion was about how we can't conclude that much from them.

    ReplyDelete
  2. Correct, *I* said they indicate fraud, which they do. If his post did, I would simply quote the relevant parts and not claim it as an original insight.

    You also don't get the argument, as usual, despite it being simple and repeated many times -- it's not whether any particular bellwether misses, but whether a whole lot of them miss in the same year.

    He doesn't discuss combining their signals, since he's looking at the classifying angle, which considers them separately -- which ones have a high hit rate.

    That's also why his remark is irrelevant about "bellwethers with a 60-year streak only getting the next election right half the time". The election he's predicting is not a single one for all of those cases -- e.g., if a bunch of counties had a 60-year streak through 1964, and they're all predicting '68.

    Some of their streaks went through '64, others through '56, or 2012, or whatever. Each of those cases is predicting its own distinct election, i.e. the one following its own distinct 60-year streak interval.

    This is silly -- no one uses a combination of bellwethers to predict different elections, but the exact same election, say 1960. And for that single election, some of the bellwethers will have had streaks going back a variety of lengths, and some may have had a miss or two along the way, at different miss-years.

    His example is wrong for that point anyway. Northampton PA did miss in '68 (D, when nation voted R), after a winning streak going back to 1908. But that was the only miss for awhile after. It correctly called '72 through '96, across both parties winning as well as a realignment in 1980. Only in 2000 did it lose bellwether status, turning mostly into a solid blue county -- although Trump did win it in 2016.

    And its winning streak goes back further than just 1908 -- correct in '04, though missing in 1900.

    So that county had only two misses during the entire 20th century, yet he uses it to illustrate the failure of bellwethers due to the '68 miss.

    ReplyDelete
  3. Here's the score for 2016 and 2020. For 2016 there are 16 bellwethers that had gotten 0 or 1 miss out of the past 15 (1956 to 2012). In 2016, 6 of those 16 missed, for a miss rate of 38%.

    For 2020, there are 10 bellwethers that had gotten 0 or 1 miss in the previous 15 (1960 to 2016). They're the 16 from above, minus the 6 that missed in '16, giving them 2 misses of 15, demoting them out. In 2020, with the results so far, and assuming Biden were the winner, 8 of those 10 missed (voted Trump), for a miss rate of 80%.

    Obviously a miss rate of 80% is a far stronger signal than 38%. But how to quantify that, using simple statistical tests?

    How about, is the miss rate from a given year compatible with it being a binomial process? As though each bellwether were a coin being flipped independently of the others, each with chance of "missing" at 50%. That's what the Obeng post would lead you to believe. He's making a somwehat different claim, using only those with *unbroken* 60-year streaks, covering a much broader period, but let's say your big-picture takeaway was bellwethers were no better at predicting a particular election than a coin flip.

    The 2016 results are not unusual if the process were binomial. Getting 6 or more misses out of 16 has a p-value of about 0.9. That's compatible with it being a bunch of coin-flips. (We're only going in the one-tailed direction of "missed this many or more" because bellwethers are chosen for having had high reliability, so only the direction of "lots of misses" counts as extreme and anomalous.)

    However, the 2020 results are rejected as being binomial. Getting 8 or more misses out of 10 has a p-value of about 0.05. There's nothing magical about any particular threshold of p-value (b/c you can easily p-hack your way cleanly below 0.05 by minutely tweaking the chance of a miss from 50% to 49.3%). The point is, 2020 not only leaps out as strange in itself, but certainly compared to the previous election.

    I won't be doing this for any other years to give the historical picture at the county level -- state level with Ohio + Florida is convincing enough. The rest of the project is for someone else to do, but the basic code is already there in Obeng's post and links.

    One aspect to investigate is using a shorter time-period, maybe 10 elections instead of 15, which would provide a longer list of bellwethers for each year, which had flawless records (or at most 1 miss). If you go back too far, and have too high of a hit rate cut-off, it could leave you with 10 bellwethers as in 2020, making statistical tests harder to run than if you had 20 or 30 bellwethers to get a signal from.

    40 years is roughly the length of a "party system" (Federalist, Jeffersonian, Jacksonian, Lincolnian, McKinleyan, New Deal, Reaganism). Sounds good -- for someone else to look into, rather than scrolling and shitposting on social media.

    ReplyDelete
  4. The take junkies are moving onto conspiracizing about the vote machine company, and drawing "webs of linkages" with all sorts of Reagan-era boogeymen.

    None of which makes the states / counties / cities in dispute jump out from the rest. None of which can be easily compared to previous elections to show how this one was in a galaxy of its own. None of which anyone in the vast middle of our admittedly polarized society would even begin to listen to -- let alone courts or elected officials.

    That initial burst of awareness and purpose on election night and the few days after has, once again, devolved back into Reaganite right-wing nutjob bullshit, by and for the consumers of polarized hyper-partisan media, including social media (talk radio shows and callers for the post-Boomer generations).

    I'll still be detailing how blatant the steal was, but I don't expect anything to come of it now.

    Especially with the GOP more and more abandoning Trump and his campaign. No Ty Cobb or Emmet Flood from the white-shoe Establishment law firms, or other heavy hitters from BIGLAW. All of his talking-heads-from-right-wing-media lawyers aren't going to accomplish anything.

    No punishment from the GOP, its elite sectors, etc., for the Georgia Republican executive branch throwing Trump under the bus with the non-audited recount.

    Chances that Trump is re-inaugurated, by overcoming the steal, now closer to 10-20%.

    Total implosion of the Republican party. I thought they'd at least go to bat for him since he already cucked to the Establishment throughout his entire first term -- why not get Paul Ryan / Jeb Bush policies for another term? They're too decrepit and moribund to even stand up to a flagrant theft of the highest office, when they have tri-fecta control of the relevant state (Georgia), or at least the state legislatures (Rust Belt).

    Now the Civil War will mainly be factions of the Democrats clashing with each other over who's going to take the helm after the moribund GOP just got stolen out of office, ending the Reagan era. Biden, Kamala, Pelosi, Bloomberg, Warren, Bernie -- it'll be the primary all over again, with no one on the other side to unify against, since Trump and the entire GOP will be dead.

    The Year of the Five Presidents -- Rome in 193 AD, and Murica in 2021.

    Will write more detail later. But you see.

    ReplyDelete
  5. GOP sectors (retail, manufacturing) turn against Trump's re-inauguration. They're threatening to de-fund the party, and even corporate relocations, if the GOP state legislatures don't certify the stolen elections in their states, beginning with Georgia:

    https://apnews.com/article/top-ceos-met-plan-response-to-trump-ae2790a11be9a73cd3216d17ce06e143

    Also today, Walmart CEO thanked "President-elect Biden," so that's another major GOP elite coalition member who is turning against Trump. Earlier, Walmart had only used the term as part of a larger group, the Business Roundtable. Now it's saying so through its own CEO.

    These GOP sectors are OK with the kabuki theater of the recounts, lawsuits by right-wing media talking heads (Sidney Powell, Rudy Giuliani, et al), etc., if that will mollify the mouth-breathing MAGA mob and Trump himself. But they are putting their foot down when it comes to who's actually getting inaugurated in January -- they demand it be Biden.

    Naturally they're joined by Democrat sectors like finance and media, but that doesn't matter because it's to be expected, and because they cannot threaten Georgia, as they do not control its patronage networks. Threatening California and Washington state, yes. Threatening the Deep South, no.

    When the big wigs of the GOP patronage networks are demanding that Trump not be re-inaugurated, that's the end of the story. Their puppets in the govt -- McConnell et al. -- are doing likewise.

    If the GOP wanted to stop the steal, the Pentagon would be threatening to re-locate Fort Benning unless the state GOP sent a Trump slate of electors. The agriculture cartel of the Plains states would threaten to withhold food shipments to Georgia, or at least the Atlanta metro area, unless they stopped the steal. Big Texas oil would jack up gas prices at the pump for the Atlanta metro area.

    None of these are happening. The only forces from GOP sectors of society are pushing in the direction of Biden, combined with every Democrat sector like finance and media pushing in the same direction. Analyzing the balance of forces doesn't even need a back of the napkin -- you can do it in your head. It's not Trump. It's Biden.

    ReplyDelete
  6. This didnt age well. My hope is diminished to 10% at this point

    ReplyDelete

You MUST enter a nickname with the "Name/URL" option if you're not signed in. We can't follow who is saying what if everyone is "Anonymous."