# Survey Response Error: Variance v. Bias (Another Election Thread) **URL:** \ **Category:** Great Debates\ **Created:** [November 11, 2000, 2:10pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787 "2000-11-11T14:10:30Z")\ **Posts on this page:** 20\ **Page:** 6
**Author:** ![UncleBeer](https://avatars.discourse-cdn.com/v4/letter/u/977dab/32.png) [@UncleBeer](https://boards.straightdope.com/u/UncleBeer)\ **Post date:** [November 20, 2000, 3:34pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/101 "2000-11-20T15:34:43Z")
I dropped out of this debate before for a couple reasons. First, I felt that my points were being deliberately misinterpreted. That may or may not be the case and isn’t really germane to the discussion; it’s my problem to make myself understood properly. I was merely getting upset and that’s not conducive to debate. Second, I saw that someone was attempting to make an argument from a position of expertise that I lacked… This is an example of _argumentum ad verecundiam_ and is generally considered a fallacy; it’s not directly concerned with the debate at hand for this reason: I was not taking exception to anyone’s expertise or actual statistical analysis; in fact, I admitted my ignorance in this field. My points against RTFirefly’s conclusions were that his initial assumptions were invalid; I still hold that view. And finally, I was put off by RTF’s sweeping (and totally unwarranted) generalization that the Bush voters in Palm Beach are stupid. This type of accusation only shows the _bias_ of the analyst and should, in theory, discredit his investigation. I’m certain, he’d make this same claim if the situations were reversed. If we accept his initial assumptions and the following statistical analysis, I believe a circular argument has been thusly created: he has posted what he believes to be the source of the bias and then used those assumptions to create an analysis that supports his beliefs. It also seems to me that he’s been a bit disingenuous and has indirectly used the _argumentum ad verecundiam_, himself. I know in other debates here, he’s said certain posted statistics are flawed, or come from a biased source, and as such, he refuses to recognize them, possibly because he feels himself to be an expert in this field while the other participants are obviously not. That may be his prerogative, but it is also my prerogative to call certain presuppositions flawed. I have done so with his underlying assumptions here, and in my opinion, they were never satisfactorily justified. However, I’m arguing not from a position of expertise, but by disputing his initial assumptions from which he’s generated his analysis. Now, I’m rejoining this debate from a slightly different perspective. There is now a statistical analysis, which shows that it’s Bush who has been the victim of **fraud** in Palm Beach, not merely a **bias** Mr. Gore as previously asserted. A link is here. [Explicit statistical evidence of massive ballot tampering in Palm Beach, Fl](http://www.reagan.com/HotTopics.main/document-11.16.2000.0.html). This study also addresses and buttresses some of my prior claims of the invalidity of RTF’s initial assumptions. Consider the ballot problems in Palm Beach which all **add** Gore votes: (Also, please note the items below are merely representative points in the study; these points and others are detailed further and analyzed in the link provided. You’ll need to read it to address this properly.) [ul] [li]Palm Beach County has an error rate ten times larger than reported in any other county in the nation using paper punch ballots.[/li][li]One early excuse given to the national news media by Democratic “spokesmen” to explain the Palm Beach fraud was that people were 'exchanging" their double-punched ballots - and were given new ones…If so, over 26 people per minute “were confused” and voted twice for President.[/li][li]Only in Palm Beach (and in only the most heavily Democratic precincts) were 19,120 ballots rejected in 2000 for double punching… (This is a 4.4% error rate overall; in the rest of Florida there is less than 1/2 of ONE percent “double punch” error rate!)[/li][li]Only in Palm Beach did Gore gain 750 votes in a recount. In 50 out of 67 counties in FL, the actual change in the recount was 5-7 votes, and in 63 out of 67 counties, the total change was less than 30 votes either way. Further, in 63 out of 67 counties the “changes” were somewhat evenly divided between **all** the candidates, in rough proportion to the original number of votes. This is the statistically expected result, and represents a true and legal recount of the ballots without any change in the ballots themselves.[/li][li]In every precinct in Palm Beach where Gore got more votes than there are registered Democrats, Bush received less than 60% of the registered Republican votes. In NO precinct in Palm Beach County did Bush receive more than 80% of the number of registered Republicans.[/li][li]Only in Palm Beach did Bush receive less than 65% of the registered Republican voters. (Registered Republicans = 231,626 while Bush voters = 152,954.) On average, in every other county in FL Bush received more votes than there were registered Republicans.[/li][/ul] I contend this is a far more serious allegation, as well as a better designed study, and more thorough analysis, than RTFirefly’s. I’d like this addressed by you guys who were on the stampede to anoint Mr. Gore earlier. ---
**Author:** ![wring](https://avatars.discourse-cdn.com/v4/letter/w/838e76/32.png) [@wring](https://boards.straightdope.com/u/wring)\ **Post date:** [November 20, 2000, 4:31pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/102 "2000-11-20T16:31:06Z")
**UncleBeer** Please explain to me who the author is asserting is “Double punching” the ballots? Because if it’s the individual voters, then accusations of fraud are false. If it’s persons other than the actual voters, then please demonstrate for me how this would be done. The voter gets the unpunched ballot, and a sleeve. They enter the voting booth, slip the ballot in a PLASTIC sleeve which then only shows a single column of punchable areas. They vote by turning pages in the form that’s attached to the plastic sleeve, and identifies which hole to punch for which candidate. After completing the voting process, they place the ballot into a sleeve and the ballot, then is placed into a box. The precinct worker has no way of identifying how many holes, or indeed if any, and certainly not which ones, are on any individual ballot. The ballots then are fed into a machine which reads them, and in PBC, it rejected 18,000 as having two punches in the race marked President. When would the fraud have occured? before the ballot is handed to the voter? well, that doesn’t make sense since the poll worker wouldn’t have an idea of the party affiliation of the prospective voter, and may end up screwing up their own guys’ votes. After the ballot is voted and handed back? but it’s in a sleeve and again, the poll worker doesn’t know for whom the ballot is marked. Then it goes into the box. When the ballots are being fed into the machine for counting? (remember, counting went quickly) For this to happen, the nefarious poll worker **in view of other workers** would have to quickly find 18,000 of those vote cards with the “wrong guys” vote on it and quickly punch another hole into it. It’s a computer card. I would think it’d be very difficult to quickly eyeball the card and figure out for which person the card was marked (and remember, the holes were pretty close to each other), and do the deed. I agree that’s (the number of double punches) an anomoly. Interestingly enough, the republicans brought out information quickly that LAST election, there was a similar number of double punches. **Firefly** , if I read correctly was not claiming that there was fraud, but that the ballots’ design made it more likely that if an error by the voter would be made, that it was statisitically more likely to be made against a Gore vote than against a Bush vote. The article you cite accuses fraudulent acts. This is a serious accusation, and they attempt to support the accusation, not by an explanation of how it could be done, but by a statistical analysis of the election results. and fail to sugget not only that it \*\* could \*\* be done, but that it \*\* was\*\* done 18,000 times in the course of a single evening in front of witnesses. ---
**Author:** ![kabbes](https://avatars.discourse-cdn.com/v4/letter/k/e99b99/32.png) [@kabbes](https://boards.straightdope.com/u/kabbes)\ **Post date:** [November 20, 2000, 4:40pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/103 "2000-11-20T16:40:35Z")
> [@](#): > > \*Originally posted by UncleBeer \* > **My points against RTFirefly’s conclusions were that his initial assumptions were invalid; I still hold that view.** **and** > [@](#): > > \*\* > If we accept his initial assumptions and the following statistical analysis, I believe a circular argument has been thusly created: he has posted what he believes to be the source of the bias and then used those assumptions to create an analysis that supports his beliefs. \*\* Hmm. The trouble with this argument, **UncleBeer** is that a statistical analysis of pure data **requires no presuppositions on the part of the analyst**. The hypothesis being tested is : “The distribution of votes shows no evidence of bias” (we test the absence of bias, since this is the more fundamental position - the duty is on the analyst to prove that there is a bias rather than the other way around). To do this we take into account the expected number of votes for each candidate and a statistical distribution that the votes would correspond to. These come from analyses of other voting areas and previous ballots in this area. We then test whether the votes actually seen could have come about merely through random fluctuations from the expected positions. Depending on your initial parameters, the chance of this happening given no bias comes out to be somewhere between about 1/10000 and one in a trillion. This is not an acceptable result - we conclude that the ballot was biased. There have been **no** preconceptions that are not carefully displayed. These correspond to the distribution of expected results and are fairly well documented - voters’ habits really don’t change that much across time and geography. I myself am **not American** and what’s more (shock horror!) have little interest as to which right-wing rich boy takes the figurehead. I know little about the state of Florida other than that old people live there. But I can still do the analysis as a statistician. Statistically speaking **there is very good evidence for bias in that ballot**. I hope that this separates the emotions from the mathematics. To get annoyed and attack those who know what they are talking about is behaviour somewhat reminiscent of creationists in an evolutionary debate. Fortunately rather than do this you took a step back for a few days, thus showing your experience. The path here however **has** to be recognition that good statistics like good science isn’t an emotional rollercoaster displaying nothing but the biases of those who engage in it. I hope that you can agree with me on that. > [@](#): > > \*\* > It also seems to me that he’s been a bit disingenuous and has indirectly used the _argumentum ad verecundiam_, himself. I know in other debates here, he’s said certain posted statistics are flawed, or come from a biased source, and as such, he refuses to recognize them, possibly because he feels himself to be an expert in this field while the other participants are obviously not. That may be his prerogative, but it is also my prerogative to call certain presuppositions flawed. I have done so with his underlying assumptions here, and in my opinion, they were never satisfactorily justified. However, I’m arguing not from a position of expertise, but by disputing his initial assumptions from which he’s generated his analysis.\*\* I’ve deliberately put this quote after my pompous diatribe in an attempt to highlight my point. There is a difference between a biased **source** or **study** , which introduces bias **at the data stage** and a biased **analysis**. In fact one way to check that the studies **are** biased is to analyse them in the above way. M’kay? \*\* > [@](#): > > And finally, I was put off by RTF’s sweeping (and totally unwarranted) generalization that the Bush voters in Palm Beach are stupid. This type of accusation only shows the _bias_ of the analyst and should, in theory, discredit his investigation. I’m certain, he’d make this same claim if the situations were reversed.\*\* As I said, political leanings of analyst shouldn’t affect a proper analysis one jot. However I don’t actually recall this accusation anyway? I’m afraid that I’m not replying to the rest of your post for now **Unclebeer** as I don’t have sufficient information to investigate it myself. Is it possible to link to the studies that you mention? regards, pan ---
**Author:** ![kabbes](https://avatars.discourse-cdn.com/v4/letter/k/e99b99/32.png) [@kabbes](https://boards.straightdope.com/u/kabbes)\ **Post date:** [November 20, 2000, 4:46pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/104 "2000-11-20T16:46:23Z")
Sorry - just noticed the link, **Unclebeer**. I’ll have a read of it. pan ---
**Author:** ![xenophon41](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/xenophon41/32/253_2.png) [@xenophon41](https://boards.straightdope.com/u/xenophon41)\ **Post date:** [November 20, 2000, 5:00pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/105 "2000-11-20T17:00:15Z")
Thanks for the comedy, **Uncle Beer**! Oh, I know some people here won’t get the joke, but I saw that this “analysis” had to be satire early in the report when I read: > [@](#): > > 19,120 double punched ballots -all in one race, with “errors” only occurring two Presidential races against Republican opposition - is 33,000 “too many” to be an accident. –by which logic I suppose one should have expected about _negative_ 13,880 double punched ballots. Sadly, some of the humor in this pastiche is lost due to the cleverly hidden statistical fallacies and convoluted language used. The reader really has to look hard at the math, statistics and logic to see where the author is leading one astray. On the other hand, the author brilliantly imitates the style of a radical right wing nut job! On the whole, I found the article to be a hoot. Well done. ---
**Author:** ![xenophon41](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/xenophon41/32/253_2.png) [@xenophon41](https://boards.straightdope.com/u/xenophon41)\ **Post date:** [November 20, 2000, 5:29pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/106 "2000-11-20T17:29:01Z")
Here’s another article from the same site: [Homosexual Recruitment of School Children](http://www.reagan.com/HotTopics.main/HotMike/document-11.15.2000.1.html). This one’s almost as funny, but in a really sick sort of way. Again, the humor’s kinda laid on a little thick here, almost as if… as if… Hey, this site’s startin’ to scare me a bit, **UB**. ☹ ---
**Author:** ![kabbes](https://avatars.discourse-cdn.com/v4/letter/k/e99b99/32.png) [@kabbes](https://boards.straightdope.com/u/kabbes)\ **Post date:** [November 20, 2000, 5:29pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/107 "2000-11-20T17:29:07Z")
Having read the report with growing amusement, I then returned to read **xenophon’s** comment. I couldn’t have said it better myself - bravo! Just the tonic at the end of a tedious day at work. I particularly like how the spoiled Democrat papers leads the author to suggest that this means anti-republican fraud, with some wonderfully distracting arguments. Nice. And for my next trick… pan ---
**Author:** ![UncleBeer](https://avatars.discourse-cdn.com/v4/letter/u/977dab/32.png) [@UncleBeer](https://boards.straightdope.com/u/UncleBeer)\ **Post date:** [November 20, 2000, 5:34pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/108 "2000-11-20T17:34:19Z")
Thanks for reasoned response, xenophon. Read on laughing boy. My first point in the next post addresses the supposed deficiency. And I see you are willing to discount this entirely on bias. How about we entirely discount the conclusion of this RTF on bias also. It’s well known that he’s a very committed liberal. I don’t see your point. ---
**Author:** ![UncleBeer](https://avatars.discourse-cdn.com/v4/letter/u/977dab/32.png) [@UncleBeer](https://boards.straightdope.com/u/UncleBeer)\ **Post date:** [November 20, 2000, 5:35pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/109 "2000-11-20T17:35:33Z")
All right, **wring**. The method of fraud first. What I understand the author to be alleging, from the article/study is that the “double-punched” ballots were supposedly exchanged for new ballots and then **both** ballots were actually counted for Gore; the “double-punched” ballots were not discarded as is proper. Additionally, as the study shows, the rate of “double-punched” ballots is significantly higher in Palm Beach. Now, put this together with, the anomalies in the rate of registered voter turnout and it becomes suspicious. **kabbes** : > [@](#): > > … a statistical analysis of pure data **requires no presuppositions on the part of the analyst**. I’m well aware of this; while statistics is certainly not my field of expertise, I do know that a pure data sample should be just that, pure. Just because this is the ideal, does not necessarily make it so. It’s is my contention that **RTFirefly** has selected a data set that he _knew_ would reflect the result he wished to obtain. (In my field, this is called “reverse engineering.”) I thought I’d made that clear already. He said up front that the ballots were difficult to interpret; this is not my opinion. As such, I see a biased data set. And then, he set out to build an analysis to prove that supposition. Apparently I haven’t made that clear until now. You go on to talk about emotions affecting reason and credit me for momentarily stepping back. You also say that “recognition that good statistics like good science isn’t an emotional rollercoaster displaying nothing but the biases of those who engage in it.” Certainly. I can wholeheartedly agree with that. I find it interesting, though, that you are just now commenting on this give the emotion displayed by RTF in the following quotes: _“At any rate, I don’t give a flying flip about the recount…”_ _“You missed the point entirely. Wanna try again?”_ and, _" OK, Unc, I yield to your higher wisdom. Humans aren’t fallible, and never make trivial, idiotic mistakes. Point well taken."_ These quotes all sound pretty emotional to me. And they’re from a guy you are trying to defend. Let’s not have any double standards here. If I’m supposed to be calm and rational, which I believe I was and still am, this applies to all parties, especially the party making the claim and study. More from **kabbes** in response to my allegation of **RTFirefly’s** demonstrated bias: > [@](#): > > As I said, political leanings of analyst shouldn’t affect a proper analysis one jot. However I don’t actually recall this accusation anyway? To answer this I’ll merely quote **RTFirefly’s** > [@](#): > > If we’d required every Bush voter in Palm Beach County to spell ‘potato’ correctly to have their vote counted (a task that any 7 year old should be able to handle), that would’ve probably knocked out a couple thousand Bush votes, and Gore would be the President-elect. Calling the opposition effectively stupid most certainly demonstrates a bias on the part of the analyst. As such, I suggest the entire analysis is flawed and should be disregarded as self-serving. I also earlier demonstrated the fallibility and subjectivity of memory. No one has taken issue with that. Therefore, I still submit this analysis of **RTFirefly’s** is further flawed. Fatally. ---
**Author:** ![UncleBeer](https://avatars.discourse-cdn.com/v4/letter/u/977dab/32.png) [@UncleBeer](https://boards.straightdope.com/u/UncleBeer)\ **Post date:** [November 20, 2000, 5:44pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/110 "2000-11-20T17:44:51Z")
And there’s more. Another absurd conclusion from RTFirefly on a different message board (italics mine): > [@](#): > > The people at the polls the other day were perfectly capable of lining up a candidate, an arrow, and a hole to punch, as long as they were on task for that task. However, the real task that brought them there was voting. _They weren’t really thinking about this other task that interposed itself, the one with the holes and arrows._ And it’s easy enough to goof up if you’re mind’s not completely on it: your eyes briefly move somewhere else, and they pick up the dividing line, rather than the arrow. So, from this I conclude, either bias on the part of the analyst, or he can _intuit_ the mindset of a voter from over a thousand miles away. If that’s the case, I suggest he talk to James Randi about that million bucks. ---
**Author:** ![kabbes](https://avatars.discourse-cdn.com/v4/letter/k/e99b99/32.png) [@kabbes](https://boards.straightdope.com/u/kabbes)\ **Post date:** [November 20, 2000, 6:00pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/111 "2000-11-20T18:00:47Z")
**Unclebeer** I’d like to take your last point about **RTF** first, if I may. My interpretation of that quote (in context) was as follows: [ul][li]There are idiots (who, for example, can’t spell “potato”) on **both sides** of the ballot.[/li][li]Idiots are likely to make mistakes[/li][li]The dodgy ballot paper means that some Gore idiots will screw up and vote for the wrong guy.[/li][li]However the Bush idiots are unlikely to end up voting for the wrong guy, 'cos their guy is top of the list[/li][li]Therefore Bush isn’t losing his idiot vote, whereas Gore is. And not being an idiot isn’t a prerequisite of voting rights, so this is unfair.[/ul][/li] I really honestly don’t think that he was accusing Bush supporters of being idiots with that statement. Your first point has more weight however. I’ll concede - **RTF** _did_ show bias and emotion in his subsequent arguments. In return though:[ul] [li]I didn’t. I’ve done the analysis from such raw data as I was able to get. I agree that there was bias.[/li][li]Others such as **jshore** agree too.[/li][li]It shouldn’t matter if the analyst is biased as long as the data is honest. In this case the data ought to be honest because we have the **actual votes** from this time, last time and other areas of Florida. Other analysts can then check the analysis and agree or disagree. The experiment is repeatable. I know that this is where you find a sticky wicket - you’re not so sure that the data is trustworthy. Fair enough. But what data **would** you find trustworthy? AFAIK these are the _actual results_. I’m gonna put my faith in 'em. Is that also hairy muff? [/ul][/li] Politically I’m well out of this one, so I can see the tempers getting riled on both sides. Understandable. Personally I think that you **all** have been remarkably restrained in this debate, given its importance and given its partisanship. So let’s not all blow up at eachother just yet, eh? pan ---
**Author:** ![RTFirefly](https://avatars.discourse-cdn.com/v4/letter/r/c77e96/32.png) [@RTFirefly](https://boards.straightdope.com/u/RTFirefly)\ **Post date:** [November 20, 2000, 6:05pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/112 "2000-11-20T18:05:27Z")
Unc, I must admit that I find precious little in the way of reasoned analysis in this fellow’s piece. Take his third paragraph, for instance: > [@](#): > > 19,120 double punched ballots -all in one race, with “errors” only occurring two Presidential races against Republican opposition - is 33,000 “too many” to be an accident. You see my problem with his arithmetic. > [@](#): > > ONLY in Palm Beach did Buchanan get less than HALF the of votes he received before in 1996. Buchanan’s “lost” votes in that county in 2000 were much greater than in any other district in Florida. (Buchanan received over 8,000 votes in 1996 Republican PRIMARY (where only registered Republicans can vote; but he received only 3,407 under the Reform Party from ALL voters in the 2000 Presidential election. That’s not only untrue, it’s wrong by MILES (to borrow his style). We’ve been around that one already in this thread. If he cites ‘facts’ that I know are wrong, pardon me if I’m skeptical about others he might cite. But let’s get to essentials. His thesis is this: > [@](#): > > There is a simple, single explanation for ALL of these 19,120 problems in ONE set of ballots in ONE district in ONE state. > > 15,000 Bush ballots, and 3,400 more Buchanan ballots, were double-punched to imitate a Gore vote, AFTER legitimate voters left their polling booth. In other words, he’s saying there were really 6800 Buchanan votes in Palm Beach County, of which 3400 got stolen by double-punching them to turn them into Buchanan-Gore ‘votes’. And 15,000 Bush-Gore double-punches, and the remainder being Gore-other double-punches. (His explicit numbers, not my interpretation.) (How they did this, he explains, is that they took 45,000 ballots in PBC, and punched 'em all for Gore. If they were already Gore votes, they stayed that way. If they were votes for candidate X, they became X-Gore double-punches.) There’s a clear explanation for why Buchanan originally got 6800 votes in PBC, he says: > [@](#): > > It is important to note that Buchanan’s running mate is the FIRST Afro-American to run for national office. Further, she is only the SECOND woman to run for national office. Given the demographics of the Palm Beach are, and its high concentration of minorities, it is perhaps surprising that Buchanan-Foster didn’t get more votes than they are credited with receiving. That one’s easily testable: we can check other areas with high African-American concentrations, be it in Broward or Bed-Stuy, and see what drawing power Foster had in those areas. I think you know the answer to this one. But the whole thing’s testable, of course: if the 19,120 double-punched ballots are examined, and there isn’t a preponderance of Bush-Gore double-punches, then his thesis is disproven, pure and simple. So far, all we’ve got is the 144-ballot sample, but at this point, it ain’t looking good for him: there were only 3 Bush-Gore double-punches, out of those 144. It’s possible that the precincts chosen for the initial manual recount were precincts other than where the alleged fraud was committed, but if so, the results from those precincts point toward Gore-Buchanan double-punching happening in droves without any help from anyone. Not to mention, the whole damned thing read like one of those crank pieces that every math department in the country recieves on a regular basis. Anyhow, we should find out more about the rest of the double-punched ballots at some point. When that happens, we’ll know for sure. ---
**Author:** ![xenophon41](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/xenophon41/32/253_2.png) [@xenophon41](https://boards.straightdope.com/u/xenophon41)\ **Post date:** [November 20, 2000, 6:31pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/113 "2000-11-20T18:31:52Z")
> [@](#): > > What I understand the author to be alleging, from the article/study is that the “double-punched” ballots were supposedly exchanged for new ballots and then **both** ballots were actually counted for Gore; the “double-punched” ballots were not discarded as is proper. Actually, **Uncle Beer** , he does _not_ make that allegation (probably didn’t think of it); he merely asserts that in order to try and cover up their fraud, some unnamed “Democratic spokesmen” made the claim early on that voters were exchanging double punched ballots for new ballots. The author then points out that the 19,120 double punched ballots were _not_ exchanged. An actual allegation made by the author is that, because there were 15,000 fewer votes in PBC for George W. Bush than there were registered Republicans, this means that 15,000 of the double punched ballots must surely have been Bush + Gore punches. He then offers this as “proof” of voter fraud! It’s not my purpose to instruct a moderator of the Straight Dope Message Boards in elementary logic, **UB**. Let me point out merely that the author supports his **conclusion** that Democratic operatives altered 15,000 Republican ballots with the **premise** that 15,000 ballots were intended for Bush and altered by Democratic operatives. This not only [begs the question](http://www.intrepidsoftware.com/fallacy/begging.htm), it also creates an inductive fallacy (fallacy of exclusion) in that the veracity of the premise is never called into question (do 15,000 of the double punched ballots actually contain a punch for Bush?). I would go on, Unc, but really it’s like shooting fish in a proverbial. Also, I still don’t see what you have against **RTFirefly** ’s analysis, but that’s for you and him to hash out. I’m not defending the OP (although I agree with it); I’m merely disputing the validity of the statistical “evidence” presented in the article you cited. ---
**Author:** ![RTFirefly](https://avatars.discourse-cdn.com/v4/letter/r/c77e96/32.png) [@RTFirefly](https://boards.straightdope.com/u/RTFirefly)\ **Post date:** [November 20, 2000, 7:11pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/114 "2000-11-20T19:11:45Z")
Unc, **kabbes** has it correctly: the intent of the ‘potato’ test wasn’t to insinuate that Bush voters were stupid. I wasn’t even claiming that some small percentage of Bush voters was stupid. There are more important things in life than being able to spell every last word correctly, and having some small set of even the simplest words that we misspell doesn’t show we’re dumb. I assume that all but a small percentage of Bush voters would get it right. But, simply out of human fallibility, the relative unimportance of the task in everyday life, and sheer surprise that that was one of the requirements for voting, a small percentage of Bush voters would bungle spelling ‘potato’ under those circumstances. (Ditto for Gore voters; fallibility knows no party.) The goal of the example was to make only one small point: that if you put a requirement on one group of voters, but not another, in order to have their vote counted, it will reduce the number of voters from that group who actually get to vote. It doesn’t matter how low the bar is set: if one group finds a bar, and the other group doesn’t, it will take away votes from that one side. (They’ll be boozing away in the bar, rather than voting. ;)) That fit in to a larger argument, of course, but I’m not gonna restate it here. And **Unc** , I apologize for getting a bit heated at one point in this discussion. But neither that, nor my partisan views (nor yours, for that matter) have anything to do with the accuracy, or lack thereof, of either of our arguments. The arguments are sound, or they’re not, regardless of where we’re coming from. ---
**Author:** ![RTFirefly](https://avatars.discourse-cdn.com/v4/letter/r/c77e96/32.png) [@RTFirefly](https://boards.straightdope.com/u/RTFirefly)\ **Post date:** [November 20, 2000, 10:33pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/115 "2000-11-20T22:33:19Z")
> [@](#): > > If we accept his initial assumptions and the following statistical analysis, I believe a circular argument has been thusly created: he has posted what he believes to be the source of the bias and then used those assumptions to create an analysis that supports his beliefs. After reading this, Unc, I came to the conclusion that I may not have put forth very clearly the _structure_ of my argument, and if so, the fault is mine. I indeed posted what I believed to be the source of the bias, but then I created a theory based on that belief. This theory was falsifiable. One never quite proves a theory, but one hopes it predicts enough otherwise unlikely events, and that no falsifying results turn up. Anyhow, let’s review the bidding: 1. We became aware of this whole mess the day after the election, when we found out that a bunch of people in Palm Beach County had come to the conclusion, the previous day, that though they’d intended to vote for Gore, they’d inadvertently voted for Buchanan. 2. Photos of the Palm Beach County ‘butterfly ballot’ were posted on Internet news sites. They tended to show the arrows (indicating which hole to punch) shifted upwards relative to the holes they were supposed to be pointing to. Neither (0) nor (00) proves anything at all by itself, or even together. They’re enough evidence to tell us what theories might be put forward, what arguments might be made. But by themselves, they’re just a pointer. T) The **theory** I’m trying to marshal evidence for is that the upshifted paper portions of many of the ballots misled people into making certain types of wrong choices of what holes to punch when attempting to vote for Gore, while reducing any confusion that Bush voters might have otherwise encountered. A visual inspection of the ballot suggests that if one intended to vote for Gore, and somehow goofed up, the most likely ways of doing so would be to either punch the hole indicating a Buchanan vote, or to punch both the Gore and Buchanan holes. Therefore, the _statistical evidence_ that one would look for in order to prove my theory would be (1) an implausibly high Buchanan vote, and (2) the Gore-Buchanan double-punch gettng an implausibly high proportion of the double-punches. By ‘implausibly high’ I mean ‘extremely unlikely, statistically speaking, to have happened by chance.’ In the case of (1), we need to add, ‘given other indicators of potential Buchanan support in Palm Beach County.’ In both cases, that evidence was found. In the case of (1), we were able to show Buchanan’s Palm Beach County vote totals to be extremely unlikely, statistically speaking, in three different ways, thanks mainly to the Pittsburgh researchers who did the actual work for two of those ways. But as you allude to, we knew of the Buchanan vote total from the start; it’s certainly possible that we could have designed the theory with the supporting evidence in mind. I personally don’t feel that’s much of a problem; it wasn’t as if we were choosing between this bias and some other kind of bias. It was either this bias, or no bias at all. And the only real factual evidence we had - the Buchanan vote - strongly supported bias. Still, there’s something much more emotionally reassuring if your theory turns out to be supported by events, previously unknown, that your theory predicts. Which was why I mentioned the double-punched ballots in my OP, and several times since. Along the way, I spelled out clear standards for what statistical evidence from the double-punches that I’d consider strong enough to ‘prove’ my theory. And then **nebuli** tripped across Thursday’s N.Y. Times story giving the numbers of double-punches in the three(?)-precinct sample that was hand-counted to determine the need for the countywide hand recount. They contain Gore-Buchanan double-punches in numbers extremely unlikely to have happened by chance. QED. That’s the outline of my argument, Unc. I don’t think there’s a lick of circularity there, although you’re welcome to try to pick it apart on other grounds. Again, sorry for my frayed temper last week. ---
**Author:** ![jshore](https://avatars.discourse-cdn.com/v4/letter/j/b5e925/32.png) [@jshore](https://boards.straightdope.com/u/jshore)\ **Post date:** [November 21, 2000, 1:17am UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/116 "2000-11-21T01:17:21Z")
That Cook imbecile overplayed his hand. If he had stuck only to numbers that were not so easy for me to verify (like the number of registered voters of each party in the county), I wouldn’t have been able to say that his numbers are fraudulent. But, then he had to go and talk about the 1996 Buchanan vote…Whoops, I have the numbers there, straight from the Florida Department of Elections Website! Let’s analyze the following quote from that article: > [@](#): > > ONLY in Palm Beach did Buchanan get less than HALF the of votes he received before in 1996. Buchanan’s “lost” votes in that county in 2000 were much greater than in any other district in Florida. (Buchanan received over 8,000 votes in 1996 Republican PRIMARY (where only registered Republicans can vote; but he received only 3,407 under the Reform Party from ALL voters in the 2000 Presidential election. (Pat Buchanan has relatives who lives in Palm Beach County, and this local support greatly increase the number of local voters who choose Buchanan, compared to every other region of the country. The only statistics that are factually correct in this whole paragraph are the 3407 and \>8000 vote numbers. All the rest is bullshit and lies. In particular, let’s take a look at the Buchanan votes in a few large Florida counties (not, by the way, selectively chosen to illustrate my point). In each case the first number is Buchanan’s votes in the 1996 primary and the second number is his votes in the 2000 election: Alachua: 2152 262 Brevard: 7927 570 Broward: 10966 789 Dade: 6349 561 Hillsborough: 7267 845 Orange: 7203 446 Palm Beach: 8788 3407 Pinellas: 12116 1010 You’ll notice a few things about these numbers: In contrast to the claim quoted above, Buchanan received less than half as many votes in 2000 compared to 1996 in ALL of the counties. In fact, Palm Beach is an outlier because he received 39% there of what he got in '96 and no more than 12% in these other counties! And, in terms of “lost votes”, last time I checked, say, 12116-1010 was greater than 8788-3407. Finally, as for Palm Beach being a hot-bed of support, Buchanan received 15.3% of the vote in PBC and 18.1% in Florida as a whole in '96…It is only in this 2000 election that PBC turned out to be a hot-bed for Buchanan! In conclusion, **Uncle Beer** , you really ought to be more careful where you go dredging for information. That article is really an insult to our intelligence. Perhaps next time it might occur to you that a site called “[www.reagan.com](http://www.reagan.com)” might not be so great at getting its facts straight…just as was true for its namesake! ---
**Author:** ![jshore](https://avatars.discourse-cdn.com/v4/letter/j/b5e925/32.png) [@jshore](https://boards.straightdope.com/u/jshore)\ **Post date:** [November 21, 2000, 2:41am UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/117 "2000-11-21T02:41:24Z")
Well, I’ve been looking a little more at the claims in that article (downloading more data from the Dept of Elections) and while none are quite as far out of whack as the Buchanan votes, they are still seriously flawed. Let’s look some more! > [@](#): > > To compare similar counties in Florida, look at the 10 counties in Florida where there are more than 100,000 democratic voters for president. In every one of these counties - EXCEPT Palm Beach County - Gore received FEWER votes than Nelson, the democratic Senate candidate. Hmmm…two of the ten biggest counties are Broward and Dade. In Broward, Gore outpolled Nelson by over 9000 votes. In Dade, he outpolled Nelson by almost 24000 votes. Guess there must have been fraud in these counties too! 😉 [It is true that Nelson tended to outpoll Gore slightly in most of the counties in Florida, but the differences were always pretty small and, as I understand it, Dade and Broward are the two counties which are quite demographically similar to PBC.] > [@](#): > > ONLY in Palm Beach did Bush receive LESS than 65% of the registered Republican voters. (Registered Republicans = 231,626 while Bush voters = 152,954.) On average, in every other county in FL Bush received MORE votes than there were registered Republicans. PBC is amongst those that have the lowest ratio of Bush voters to registered Republicans, although Broward’s number as a percentage is practically identical (Registered Republicans = 266,829; Bush voters = 177,279). In general, the relation between party registration and which Presidential candidate is voted for is all over the map! In some of the other counties, there are even more dramatic swings the other way: For example, in Baker County, there are 10,261 Dems and 1,684 Reps, but the vote was 2392 for Gore and 5610 for Bush. Even in large counties like Duval, you have strange things like this: 211,762 Dems and 152,008 Reps but 107,680 for Gore and 152,082 for Bush. [I guess it is only evidence of fraud when it goes the other way though! ;)] Why might things work out this way? One possibility is that people often never change their party affiliation even if they change their voting patterns. But, whatever the cause, anyone who would try to base statistical analysis on data that is all over the map like this is clearly majorly fucked in the head! I think it is sad when people can spread such lies and BS to millions of people over the internet. The moral of the story: **Check your facts!** ---
**Author:** ![nebuli](https://avatars.discourse-cdn.com/v4/letter/n/0ea827/32.png) [@nebuli](https://boards.straightdope.com/u/nebuli)\ **Post date:** [November 21, 2000, 3:20am UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/118 "2000-11-21T03:20:04Z")
This is only a **possible** inaccuracy, but given some of the other statements in that report, skepticism may not be inappropriate: The author said that in 1996 15,000 ballots in PBC were disqualified for doublepunching. Last week I heard a PBC official say on the radio that the 15,000 figure was a combined total of doublepunched and not-punched ballots, with the greater portion being the not-punched. Given the now widely recognized problems caused by machines being unable to read the infamous hanging, swinging or triad chads it seems to me very likely that many of the 15,000 would have been in one of those categories. Since the election in '96 was not close no one investigated to see what those disqualified ballots really were not punched or merely incompletely puched. Sorry I cannot remember the official’s name or the station. **If** what he said was accurate, those supposed 15,000 doublepunched ballots in '96 are a mirage. ---
**Author:** ![RTFirefly](https://avatars.discourse-cdn.com/v4/letter/r/c77e96/32.png) [@RTFirefly](https://boards.straightdope.com/u/RTFirefly)\ **Post date:** [December 1, 2000, 3:05pm UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/119 "2000-12-01T15:05:19Z")
A while back, **UncleBeer** was making a point of my bias in these matters. And, of course, I’m partisan as hell - I’ve made my (extremely negative) views on George Dubya Bush well known. Nothing’s wrong with being partisan, in and of itself, IMHO. The problem comes when one suspends critical judgment on the arguments made, and evidence put forth by, one’s own side. We can’t examine everything that comes our way, but at the very least, anyone making arguments for or against a position on some partisan issue should have a ‘Too Good To Be True’ filter. If something pops up in the news that seems to favor your side of the debate a bit more than seems likely, and you want to use it to back up your arguments, zing the bad guys, or whatever, then you need to check it out and make sure it’s legit, before relying on it. Some examples: in another thread, a poster here said that of the 19,120 double-punched ballots, 16,500 were Gore-Buchanan double-punches. ‘Great news if it’s true,’ I replied. ‘Do you have a cite?’ (Substantiation never materialized, of course.) This was exactly the sort of smoking gun I wanted in this debate, and would still love to find. But it just smelled a little too good to be true, and I refrained from relying on that ‘fact’. I did the same thing with the early reports that Bush had signed a law in Texas that gave hand-counts preeminence, and accepted dimpled chads as evidence of voter intent. That one turned out to be true, but it was still so exactly what the doctor ordered (for my side of _that_ debate) that, again, I didn’t want to trust in that without a cite. Sorry if I seem to be patting myself on the back here, but the instances I can most quickly recall of exercising appropriate skepticism in this manner are my own, because I was there. The [study cited by the Washington Post](http://www.washingtonpost.com/wp-dyn/articles/A8025-2000Nov30.html) regrading the ‘butterfly ballot’ is an example of this. I think the study may have some parts that are statistically robust, such as its measure of the relative confusion levels of different styles of ballots. But the part that I really wish was statistically strong, isn’t. It’s a bummer, but that’s life. The study, by a Canadian psychologist, takes the _Canadian_ ballot choices, and puts them into the Palm Beach butterfly ballot configuration to test how well different groups handle the ballot; their major-party candidates took the places of Bush and Gore. The article says: > [@](#): > > In one version of the study, 20 percent of participants who tried to vote for the second candidate in the left column - the slot that featured Vice President Gore’s name on the Palm Beach County ballot - voted mistakenly for the first candidate in the right column, where Patrick J. Buchanan’s name appeared in the Florida county. You look at that and say: _twenty percent!!_ Omigod, that shows that the Canadians had just as much trouble with this as we say the Floridians did!" Then you read down. ‘Twenty percent’ sounds like it has some real numbers behind it, doesn’t it? Nope. It’s got numbers, but they’re a weak reed: ‘twenty percent’ represented three errors out of fifteen tries. Three errors doesn’t mean shit, and we all know that. This ‘twenty percent’ is worthless, vacuous, and meaningless. Even the _Washington Post_ reporter who wrote the story, and his editor, should have known _that_. If we had an award for making it look like a study says more than it does (we could call it a ‘Limbaugh’ :)), the Post would get one for this story. Percentages derived from small samples should never be stated as percentages; the actual numbers should be given instead. We use percentages, in the _language_, to distill complex ratios down to a familiar scale: if there had been 415 errors in 2083 tries, _that_ would be 20%, in the manner we use it. But three out of 15 needs no ‘simplification’ to percentages; anyone can grasp 3 out of 15. Rather, using a percentage instead of the numbers simply conceals the weakness of the numbers. It’s bad journalism - and when that sort of stuff is used in debate, it’s bad advocacy. OK, you get the idea. End of rant. ---
**Author:** ![RTFirefly](https://avatars.discourse-cdn.com/v4/letter/r/c77e96/32.png) [@RTFirefly](https://boards.straightdope.com/u/RTFirefly)\ **Post date:** [December 5, 2000, 1:49am UTC](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787/120 "2000-12-05T01:49:04Z")
…I’m putting it in here. The [Miami Herald](http://www.herald.com/thispage.htm?content/archive/news/elect2000/decision/104268.htm) had some statisticians do a precinct-by-precinct analysis of the nonvotes in FL, and came up with the conclusion that if all the people who had problems voting, had been able to vote properly, Gore would have carried the state by about 23,000 votes. This doesn’t surprise me, and I find their assumptions properly conservative. At this point, it won’t make a lick of difference, of course: it’s over, and if Bush won’t be a man and concede, then Gore should. [Previous page](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787.md?page=5) [Next page](https://boards.straightdope.com/t/survey-response-error-variance-v-bias-another-election-thread/40787.md?page=7)