Wikipedia:Featured article review/Margin of error/archive1

The following is an archived discussion of a featured article review. Please do not modify it. Further comments should be made on the article's talk page or at Wikipedia talk:Featured article review). No further edits should be made to this page.

The article was removed 11:12, 3 March 2007.

---

Margin of error

Review commentary

Messages left at WikiProject Mathematics. Kaldari 06:19, 18 January 2007 (UTC) Messages left at Fadethree, Michael Hardy. Jeffpw 07:02, 18 January 2007 (UTC)

Only 1 reference and no inline citations. Use and abuse section seems excessively listy. Kaldari 06:11, 18 January 2007 (UTC)

• The arguments section needs to be prosified and needs more than the one inline citation currently present. Parts of this section, including those discussing polling in general, are better suited for other articles and could simply be deleted. Most of the rest of the article is pretty basic statistics and doesn't in my view require inline citations, though the article should list a couple of standard textbooks in the references section. Christopher Parham (talk) 05:04, 19 January 2007 (UTC)
• Comment: I have reorganized the structure for better flow, moving external links to Newsweek source as inline citations, added a very standard statistics book, and a bit of technical clean-up. If I find some sources/books, then I will add them, but I'm not an American so I don't know much about the US election section. I think this article was featured when footnote citations were not yet discovered, but the content is really good. — Indon (reply) — 10:04, 19 January 2007 (UTC)
• Comment Both "Calculations and caveats" and "Use and abuse" are argumentative. Opinions need to be sourced or left out. And even if we do source opinions that the press regularly misuse the margin of error, they must be included along with reliable sources on journalistic standards — manuals of style, mass comm textbooks, trade magazines, whatever — that explain how the margin of error is used and why. Currently all the article says is "I searched Google news and I don't like what I found." Melchoir 23:47, 23 January 2007 (UTC)
• Comment I've edited "Use and abuse" to "Use and misuse." I don't find "caveats" argumentative. I took out some of the opinion-ish tone that has arisen in certain sections. And I agree about the "I searched Google news" comment and edited that. I have added a reference, but I am comfortable with shorter reference lists when dealing with elementary statistics articles where the proofs are easily laid out on the page itself. I also trimmed the list and tightened some language to make it less list-y. This review process is useful. The article has exploded since I last visited. Fadethree 07:37, 31 January 2007 (UTC)
  • I'm sorry, but far more needs to be done. Yes, "Caveats" is argumentative, particularly the last paragraph. The one new citation is an opinion piece on a single incident, and it doesn't contain the quotation to which it is attached. This article currently contains two "frequently misused", an "often interpreted", a "sometimes used", a "continues to be inappropriately applied", a "many pollsters", a "most pollsters". These synthetic descriptions cannot be advanced without reliable secondary sources. "The margin of error grew out of a well-intentioned need to compare the accuracy of different polls." Does that come from a study on the early history of polling, or is it speculation? There are three places where "margin" is placed in quotation marks and it is emphasized that the margin of error isn't a margin. Leaving aside the question of whether that's even a meaningful statement, who cares? Seriously, who exactly thinks that's important, and why aren't they identified in the article? Who thinks that "statistical tie" and "statistical dead heat" are inappropriate? Melchoir 06:02, 1 February 2007 (UTC)
    • Gotcha. Those are helpful comments. I've done some work to rephrase and clarify things. I hope the article is now a little more convincing to you. Let me know what you think. Fadethree 06:34, 3 February 2007 (UTC)
• Comment Sorry for the late comment - I only just noticed this was being reviewed. I still have serious concerns about this article, as raised on the talk page back in May 2006. The article is not entirely accurate or comprehensive. For instance, margins of error depend strongly on the sample design used. At present this is only mentioned in one bullet point, and is ignored in most of the calculations shown. This means they are probably incorrect, since the Newsweek data was weighted. In any case, our underlying assumptions should at least be made clear throughout. The press release about the poll is not very clear, but reading between the lines, I think that they have allowed for the sample design, and our implication that Newsweek's 4% margin of error uses a 99% confidence level is incorrect. If I'm right, this would have flow-on effects throughout our discussion of the poll.

I'm also concerned that we never explicitly present a margin of error for Kerry's lead over Bush, which seems a natural approach for this example. The whole section on the_probability_of_leading seems a bit off topic at best. Maybe it could be moved to another article, although it seems to verge on original research. We are also not consistent in referring to the maximum margin of error, where that is meant. Some mention of software for complex survey designs is probably necessary. And I don't agree that references can be dispensed with here - it's not that elementary a topic, unless we simplify it to the point where the Newsweek poll would be beyond the scope of the article.

In my view, the article should not remain a featured article unless these sorts of issues are addressed. Avenue 14:59, 31 January 2007 (UTC)

The issue of complex sampling procedures is certainly important, but I think that going into it in the detail which you seem to favor would make the article even more weighty than it already is while not providing the reader much additional value for the interpretation of the margin of error. No matter what the sampling plan, the target for reporting is the radius of the 99% (or 95%) confidence interval for a proportion of .5. You keep saying that the margin of error is different, but it's the calculations that are different, the interpretation (as flawed and unhelpful as I think it is) remains the same. I don't see a problem leaving the calculations assuming random sampling in the article, and leaving you to make the valuable comments as you have already about complex sampling designs and how they would change the calculations.

Let me be clear that, as I understand it, we are not talking about confidence intervals around proportions here. We are talking about the margin of error as reported by polls in their footnotes or closing paragraphs, which is a very particular confidence interval of limited usefulness.

I agree that the tables presented for the probability of leading are excessive, and I'd be in favor of removing them, but I think the probability of leading is an incredibly important concept that belongs in an article about the margin of error for polling. After all, the probability of leading is exactly the inference that most political polls are trying to make. You even suggest as much where you raise the idea of a "margin of error for Kerry's lead over Bush." The probability of leading addresses this idea in a much more straightforward fashion than reporting the confidence interval for the difference in proportions.

Until you are sure that the Newsweek poll used a complex sampling design, I'm in favor of assuming that they did not. At best, they did use a simple calculation, and nothing needs to be changed. At worst, the calculation will be not be the one they used but still serve the purpose of illustration, and the interpretation will be unchanged.

I appreciate your comments. You raise some very important points. Fadethree 20:02, 31 January 2007 (UTC)

I think I might see where we are passing each other here. Your May 2, 2006 edit introduces the idea of a maximum margin of error. This leads me to believe that you are equating "margin of error" with "confidence interval," which we had discussions about in earlier versions of the article before it was featured. We ended up deciding that we were going to leave "confidence interval" to the "confidence interval" entry, and let margin of error refer to the specific use of the term in polling, where it is reported, for example, that the margin of error for such and such a poll is 4%. This is aimed to attract readers who read that and say to themselves, well, what does 4% mean? I think you would answer that question, well, that's the maximum margin of error, and then you'd go on to explain that margin of error is a confidence interval around a proportion, etc. If you accept that the article is about the maximum margin of error (which most polls just call, the margin of error), does that change your impressions about its accuracy?

I should note that I am not in favor of the term, maximum margin of error, and I have never heard it before. The reason I think it is misleading is because I could arbitrarily tweak the margin of error by changing the confidence limits, and even then the margin of error isn't a margin at all; it's not like the probability outside the margin is zero. I'd be in favor of rewording that note or deleting it altogether. Does that make sense?

Let me add, too, that I understand that different disciplines use different terms in different ways. Let's see if we can understand how we're using different words to say the same concepts or the same words to express different concepts. Fadethree 20:32, 31 January 2007 (UTC)

Thanks for clarifying your position - it explains a lot. I agree we are using "margin of error" to mean different things, and I do equate the "margin of error" for a particular result with the "half-width of the confidence interval". This is not just me; for instance, the American Statistical Association's pamphlet What is a Margin of Error? defines the margin of error by saying "To be precise, the laws of probability make it possible for us to calculate intervals of the form

estimate +/- margin of error.

Such intervals are sometimes called 95 percent confidence intervals and would be expected to contain the true value of the target quantity (in the absence of nonsampling errors) at least 95 percent of the time." I agree that the term "maximum margin of error" is not as commonly used as "margin of error", but it is not my invention either. It has been used in the mainstream press - e.g. in reporting on this USA TODAY Gallup poll - and by many other authors. A Google search for "maximum margin of error" gives plenty of examples.

On a minor point, it is not sensible to talk about the margin of error of a whole poll, even using your maximum margin of error definition, unless no analysis of subpopulations is undertaken. For example, the press release for the Newsweek poll quotes twelve different "margins of error", for subpopulations such as registered voters, men, women, Republicans, Democrats, and debate viewers. This seems to negate much of the apparent simplicity of your definition.

More generally, I believe that giving people a practical explanation of the drawbacks of the margin of error is important, much more so than deriving the formula for the maximum margin of error for a proportion under a simple random sample (for example). So we should expand our coverage of non-sampling errors. And at least where I live, the sample design is still ignored in poll reporting often enough that I think we should set a good example. This is why I am uncomfortable with presenting calculations that assume a simple random sample in the context of our example poll, which they don't apply to. (Our apparent confusion here over the confidence level in the Newsweek poll is an illustration of how this can lead people astray.) I also think we should be careful not to limit the focus of our article to polls. Sure, they're high profile, and this means they can provide an easily understood example of the concept, but there are many other sample surveys out there that are at least as important.

I agree the probability of leading is useful in understanding the Newsweek poll results. Can you give a reference to someone discussing this concept? This would help assuage my concerns about it being original research. -- Avenue 23:28, 31 January 2007 (UTC)

Sorry, I missed your comment about the Newsweek poll sample design. I can't how they could have avoided using a complex sample design (assuming they select households randomly first, then people, as is typical); either they interviewed interviewed all eligible people in the contacted households, in which case it is a cluster sample, or they interviewed a subsample of the eligible people (e.g. just one person per household), in which case there would be varying sample selection probabilities. Either way it's a complex design, and the standard formula for unweighted simple random samples wouldn't hold. In any case, they do say the sample is weighted, so that rules out the naive formula on its own. -- Avenue 00:19, 1 February 2007 (UTC)

Thanks again. I don't have too much time to respond to your points just just now, so let me just ask everyone's opinion about moving this article to FARC. Neither you nor I came saw this review until recently, and I think the day and a half of review has already proven helpful. I'd like to keep this in FAR for now and give us the chance to resolve these issues over the weekend. I know I'm being US-centric in saying this, but I'd really like to keep this article's profile high as the presidential race and its accompanying misinterpretations of polling data has already begun. Of course, it may be possible to get the same work done in FARC.

The key issues I think we're going back and forth about are scope and focus. I originally added content to this article because I was frustrated with misinterpretations of the single margin of error that newspaper articles report as a footnote in their coverage. I basically wanted to write about how it can be interpreted (not necessarily how it's calculated) and how it was a mostly unhelpful statistic that is prone to misinterpretations. The margin of error that you seem to be discussing is, I think, equally well addressed by the notion of a confidence interval for a proportion. In fact, it's synonymous.

One of the ways that we can address this is to run some kind of disambiguation page. For a time, there was, in fact, a note saying that this article discussed a margin of error for polling. Could we spin off another article or develop the confidence interval page further to address your concerns? If so, how should we relabel this article to keep its narrow scope clear while we restructure links to other articles? Fadethree 05:20, 1 February 2007 (UTC)

• Comment fails 1a, prose problems - some examples only from the final part of the article:

• ("The rest of this section"?) The rest of this section shows how the Newsweek percentage might be calculated.
• an assumption that if anyone does not choose Kerry they will choose Bush, and vice versa, i.e. they are perfectly negatively correlated

• perhaps something like, assumption that a voter must choose Kerry or Bush; the two possibilities are perfectly negatively correlated if, for the purpose of illustration, the possibility of a voter choosing another candidate is ignored.

• It's unnecessary to have convoluted English grammar making the math appear more difficult: This assumption may not be tenable given that a voter could be undecided or vote for Nader, but the results will still be illustrative.

• see above

• Redundant prose:

• It is evident that tThe confidence level has a significant impact on the probability of leading.
• Note that tThe 100% entries in the table are actually slightly less. Here is the same table for the 99% confidence level:
• This can be easily accomplished by taking a glass of seawater and then chemically analyzing the proportion of salt in that sample.

• This can be accomplished by chemically analyzing the proportion of salt in a sample of seawater.

• Why tentatively? This can tentatively be called the Probability of Leading.
• "very approximately equal"? The margin of error applicable directly to the "lead" is very approximately equal
• "note that" occurs throughout the article - it's redundant, not necessary, and we shouldn't tell our readers what to note.

• These are samples only. Again, we find the math is made to appear inaccessible to the average reader not because of the math, but because of the English. Serious copyedit attention is needed by a fresh set of eyes - I suggest the Math Project attempt to get some good copyeditors involved.
• Referencing has something goofed up - books listed as Further reading are included in footnotes - if they are used as References, they should be listed as such. Why is the Bush-Kerry MSN article given in External links - is it used as a Reference? If not, what is its relevance?

Move to FARC. SandyGeorgia (Talk) 04:38, 2 February 2007 (UTC)

I've tried to address these issues, mostly by trimming things out. Let me know what you think. Fadethree 21:10, 2 February 2007 (UTC)

I changed the Further reading to agree with WP:LAYOUT and added an ISBN on the reference book. I read one section as an example - I found missing wikilinking (sample size - important to the topic), and made prose changes to further reduce redundancy. That's one section only; then I went back to glance at the lead, since I noticed that the first section introduces confidence intervals without defining them. I found that the WP:LEAD gets into too much detail about confidence intervals, which should be handled and defined in the first section - the lead should be a stand-alone summary of the entire article, not the place for defining concepts. I'll keep going after you've done more work. Is there a reference chapter for the entire section "More advanced calculations behind the margin of error"? The tipoff to the possibility of original research is the sentence, "This is perhaps optimistic, but if care is taken it can be at least approximated in reality". Section headings violate WP:MSH; I'll add that these kinds of basic issues (e.g.; following WP:MOS) reflect my ongoing frustration with the Math WikiProject. I do wish the Project would make a concerted effort to make sure their Featured Articles conform with basic guidelines like WP:MOS, and get a good copyeditor on board to review all of their FAs, as many of them coming through FAR suffer from prose problems that seem to be getting through FAC. Perhaps reviewers are intimidated by the math, which is made harder to understand because of prose redundancies which make the math look more convoluted than necessary (sorry, that was a general rant, not aimed at you or this article in particular :-). SandyGeorgia (Talk) 23:53, 2 February 2007 (UTC)

I've made edits to the lead, though I'm happy with your suggestion to move some of its content to a later section. One of the reasons that confidence interval comes up so early is to distinguish the margin of error as a special case of the confidence interval. For those who are used to thinking of them as synonymous, that's something that they'd be confused about quickly. I've taken out the advanced calculations; they were correct but may belong to another article. I really appreciate your comments. There is no excuse for bad prose, but I think a small part of it arises from honest differences in disciplinary norms. As you know, math articles occasionally use passive voice and first-person viewpoints. Oh, and they're often badly written. It's a little bit of assuming that the readers are as fluent in jargon as we are, and it's a little bit of laziness and deliberate obfuscation. I'm do my best to keep things clear. Fadethree 07:20, 3 February 2007 (UTC)

• Comment I've done a fair amount of editing. Avenue, I've tried to delineate the boundaries and state the assumptions of the article as clearly as possible up front. Let me know how you like this framing. Sandy, I've knocked around almost all the prose that you found cluttered. Please let me know if anything else is clear. I'd appreciate anyone's help with the references; I'm not as familiar as I'd like to be about reference conventions. Fadethree 21:48, 2 February 2007 (UTC)
  • If you stick something in for references and let me know, I can help clean them up to correct formatting style - just get something in there. SandyGeorgia (Talk) 23:53, 2 February 2007 (UTC)
    • Thanks for your help, Sandy. I'll see if I can get more references in where appropriate. Fadethree 06:34, 3 February 2007 (UTC)
  • Thanks, Fadethree. I think getting the boundaries and assumptions sorted out up front will help us progress the substantive issues. I agree that the article shouldn't go into too much technical depth, although we may have different ideas about what bits should qualify. I liked the way you clearly distinguished between the two meanings of margin of error, but I felt the way they were described was misleading. I've tried to improve it, and tidy up the rest of the lead section (especially the paragraph on confidence levels, which seemed to ramble). It would also be good to shorten the caption for the graph, and preferably make it a bit narrower, but apart from that, the lead is looking much better.
  • I think the rest of the article still has a way to go to meet FAC 1a and 1c though. A few examples:

• The fact that the (maximum) margin of error is the radius of a specific confidence interval is not really that important, but it seems to be stressed throughout.
• It's confusing to assume a 99% confidence level in the example when most other polls[1] use 95%, and this one probably does too.
• Bayesian assumptions and interpretations are only hinted at, not described or linked to, and I don't believe they apply directly to the calculations shown in our article anyway. These formulae give confidence intervals, not credible intervals.
• The sentence "When the sample is not random, the margin of error must be estimated through more advanced calculations." sounds like we might be suggesting the sorts of things argued against by the AAPOR. A random sample is not the same as a simple random sample.

• And that's just from a quick look through the first quarter of the article. -- Avenue 14:08, 3 February 2007 (UTC)

Thanks, Avenue. I like your edits. Some replies:

• The reason I stressed that the margin of error is synonymous with the radius of a confidence interval is because anyone who knows what a confidence interval is but not a margin of error will come closer to understanding it. I think that showing that two things that may be presumed different are in fact the same is a useful endeavor. I'm not strongly against deemphasizing this point, but I thought I'd make a case for its inclusion.
• I'll take a look at redoing the figure for a 95% confidence interval. That seems to be the biggest timesuck to switching.
• The Bayesian issue is actually quite subtle, and, as you probably know, there's a long history of Bayesian vs. frequentist interpretation of intervals. The fact is that we have taken a Bayesian interpretation of confidence intervals in this article, and that is what I wanted to acknowledge in case there were any hard core frequentists lying around. The frequentist interpretation of confidence intervals is not that the probability of the "true" value being somewhere in a range is 95% (the true value doesn't have a distribution; it's known) but that the confidence intervals of multiple different samples will contain the true value 95% of the time. This interpretation is awkward, which is why I chose to fall back on Bayesian assumptions. Since so many people take a Bayesian interpretation of confidence intervals, and I don't think this is the place to go into it in too much depth, maybe we can just delete the qualification or refer people to the CI article.
• Can we change "random" to "simple random sample" to address your last point?
• Thanks again. I'll try and find that old figure. Fadethree 17:00, 3 February 2007 (UTC)

• I've changed the baseline example and figure from 99% to 95%. Fadethree 20:14, 3 February 2007 (UTC)

FARC commentary

Suggested FA criteria concerns are citations (1c), and list sections (2). Marskell 05:32, 4 February 2007 (UTC)

• Comment: Plenty of work done here; just keeping it on schedule. Marskell 05:32, 4 February 2007 (UTC)
• Comment nice work so far, Fadethree - from a quick glance, the prose is much tighter and cleaner. I haven't had time for a complete read, but can someone please review the section headings per WP:MSH - they seem very repetitive. Also, a page number (or range) is needed on note 6, Sudman and Bradburn. SandyGeorgia (Talk) 01:22, 5 February 2007 (UTC)
  • I tried to eliminate the redundancy in section headings while eliminating the "listiness" previously noted. I added page numbers for the Sudman and Bradburn reference. Editing references to conform to style is always appreciated. Let me know what you think. Fadethree 23:07, 6 February 2007 (UTC)
• Comment. I think the listiness has been fixed and the referencing has definitely been improved. Kaldari 06:52, 7 February 2007 (UTC)
• Comment Sorry, I've been away and I do try to be more helpful in these things… I agree that the article has improved but I still have some accuracy/POV issues. Maybe I'll just take them up at the talk page when I get the time. Melchoir 22:28, 8 February 2007 (UTC)

• Melchoir, are you still planning to work on this? I was waiting for your edits before re-reading. SandyGeorgia (Talk) 15:13, 14 February 2007 (UTC)
  • Well, let's try this. I'll boldly take out the statements that I don't think can be defended. We'll see whether it causes controversy. Melchoir 07:33, 15 February 2007 (UTC)
    • Thanks for your edits, Melchoir. I'll reword a couple of the removed sentences so that they can be defended, add them back, and provide a defense. But not till this weekend, if that's okay. Fadethree 08:31, 15 February 2007 (UTC)

Ah, I didn't notice that you removed large parts of the "incorrect interpretations" section. To clarify, do you think that everything that you removed is indefensible, or just that there are statements that are not defended?

I know this is a platitude, but a statistic is just a number until it is interpreted. There are defensible and indefensible interpretations of a statistic. I think it is worthwhile to delineate the domain of reasonable interpretation. I have and hopefully can further reduce the argumentative tone of the article, but if your objection is that the domain of interpretations is not interesting, I'm not sure there's much I can do but respectfully disagree. In fact, I think the common misinterpretations of this statistic are exactly what make this an interesting topic and an interesting article

I think that everything that you removed should be improved, and I will work to do that. But I hope you agree that a description can be communicated both by saying what something is and by saying what something is not. Fadethree 09:06, 15 February 2007 (UTC)

Well, to be scientific about it, all I can really say is that the statements were currently undefended. But I can offer my reasoning line by line:

• Kerry and Bush are "statistically tied" or are in a "statistical dead heat". The fact of two percentages being within a margin of error of each other is elsewhere offered as the definition, or the criterion for usage, of these phrases. Logically it is absurd to then claim they are also incorrect.
• The difference between Kerry and Bush percentages is only notable if it is greater than 3%. This is not an objective statement that can be correct or incorrect. Strictly interpreted, I wouldn't agree with it. But it's an abbreviation of a correct statement: the difference between the percentages implies that the lead is reflected by the poll, with an insignificant possibility of error that scales with the confidence level used to calclate the margin or error, if it is greater than a statistic that is of the same order of magnitude as the reported margin of error.
• Any change in the percentages is trivial unless the change is greater than 3%. Again, it depends on what you consider trivial. Again, if you expand out the statement, it becomes correct: the likelihood that the change in percentages is of a different order of magnitude than the reported change, or in the wrong direction, is insignificant (and scaling with the confidence level used to calclate the margin or error) if the change is greater than a statistic that is of the same order of magnitude as the margin of error.
• Because Nader got 2% and the margin of error is 3%, he could potentially have 0%. This is surely a fallacy but a trivial one. I considered keeping it, but then we'd be left with a list of one uninteresting item. And there are nearby, correct statements: Nader could potentially have a percentage that is so much lower than 3% that for it would be more accurate to think of it as 0%. It could even be that Nader has no supporters besides the ones who responded to the poll, however unlikely that would be.
• The margin of error is the same for every percentage, i.e. 47% ± 3%, 45% ± 3%, 2% ± 3%. Since the margin of error is defined as a poll-level statistic, it is the same for every percentage. I guess from the numbers that it was meant that the radius of the confidence interval is the same for every percentage. That would be a true fallacy, but it is less interesting than the associated true statements that the confidence interval scales linearly with the margin of error; it can be calculated knowing only the margin of error; and the constant of proportionality is sufficiently close to 1 for a wide range of percentages and purposes. The incorrectness of the fallacy could be brought to light by emphasizing the true statements instead.

Of course, besides the factual issues, there were also the usual problems that the claims were original research, it had not been verified that the statements were sufficiently common to justify comment in the first place, and the implication that they represent a systematic problem was POV. Melchoir 16:17, 15 February 2007 (UTC)

• I have to say remove, at least for the current version. Melchoir's deletions have greatly improved the article, as has all the work put in since this review started, but there's still a lot more that needs fixing. I've reworked some of the easier bits, but there's plenty more that is currently just plain wrong.

One of the more fundamental issues is the reliance on Bayesian interpretations of frequentist calculations. I don't see myself as a hardcore frequentist, but I suspect this can't be justified for surveys in general. I think we should either go the frequentist route throughout, or warn people that we're fudging things by interpreting frequentist calculations in a Bayesian fashion, and that there may be some traps hidden there for the unwary. The Bayesian interpretation is much more natural, which makes the article easier to write and understand, but we shouldn't gloss over the distinction completely, or the potential dangers of ignoring it.

More concretely, I'm still noticing new errors in the text every time I read through it, so there's still a lot of tidying up to do. For instance, by saying finite population corrections would be irrelevant for a survey in a school, while presenting a poll with 1000 respondents, we imply that schools have at least 10,000 students. We should also provide better citations for the theory; no offence to the folks at Research Solutions, but I think even they'd agree that they are not an authority on this subject.

It would be nice to take the example a bit further before we introduce any formulae, because some readers will not get past them. I'd also like to avoid the assumption of perfect negative correlation in the section on comparing percentages; it might not be too bad in this example, but it comes across as sloppy and doesn't belong in an encyclopedia article. It would be good to eliminate the uncertainty around the Newsweek margin of error calculations as well. -- Avenue 13:17, 15 February 2007 (UTC)

• Okay, I don't think there's anything further I can do. I will say that I am a bit baffled by your philosophy, Avenue, which seems to be that every assumption that statisticians casually make in practice should be defended with down to the axioms. Statisticians are not obsessive mathematicians, we make assumptions based on convenience and based on the fact that the difference doesn't matter until the third decimal point. Your continuing assertion that things are "incorrect" is of course true but trite, my mantra is the same as that of many statisticians, "all models are wrong, some models are useful." If you are going to continue nitpicking all statistics articles to a standard of mathematical certainty, they will not only be complicated but boring and, worse, inconsequential for all practical purposes. Still, I'm happy with the results of the review, though I'm sad to see the article leave FA. Fadethree 19:57, 15 February 2007 (UTC)

• I'm sorry you feel that way. I should have said that I agree with you that some coverage of common misconceptions and misinterpretations would be useful, even though I'm happy that what was there before was deleted. I do feel that a certain degree of "nitpicking" is appropriate for featured articles, but I wouldn't characterise my comments and edits as insisting on anything like mathematical rigour. Regarding the dual frequentist/Bayesian interpretation, while I agree that this sort of joint approach is becoming common in statistics generally, I don't believe it's as straightforward in this context as you seem to think. For instance, a recent paper by two prominent Bayesian statisticians highlighted finite population sampling as an area where there is "a fundamental philosophical and practical conflict" (p. 35) between the Bayesian and frequentist approaches, which hasn't yet been surmounted. Glossing over this completely in this article seems like a bad idea to me. -- Avenue 01:18, 16 February 2007 (UTC)

• Comment — there is still a very solid chance this article will keep featured status. It only needs additional reinforcement from references, and maybe a small copy-edit to weed out any lingering major issues. Implementing Avenue's suggestions will help, as well. — Deckiller 12:20, 18 February 2007 (UTC)

What's up here folks? I hate dropping the ball so close to the goal line. Marskell 07:44, 21 February 2007 (UTC)

Personally, I think it's passable; the issues can be resolved outside of FAR(C). — Deckiller 13:24, 21 February 2007 (UTC)

I personally don't believe it's FA quality yet, although I think we're more than halfway there. Lately I've focussed on tidying up the prose, but there are still a number of substantive issues to address. -- Avenue 16:45, 21 February 2007 (UTC)

Some attention to flow might help; the definition is halfway into the page. Once sections are ordered more logically, a review of wikilinking may be in order. SandyGeorgia (Talk) 16:02, 24 February 2007 (UTC)

I agree some reorganisation would be useful. There are also a couple of topics mentioned in the lead section that haven't yet been expanded on in the text (non-sampling errors, effect of sample design). Along with the Bayesian/frequentist issue, these are the main things I think keep this article from reaching featured quality at present. There's also some more tidying up to do, and it would be nice to add more graphics and references, but that's not as crucial. -- Avenue 15:09, 1 March 2007 (UTC)

• Remove. After a good start at improvements at the beginning of the review, work on this article seems to have stalled. The review is running overtime, and issues have not been addressed. SandyGeorgia (Talk) 17:05, 1 March 2007 (UTC)

The above discussion is preserved as an archive. Please do not modify it. Subsequent comments should be made on the article's talk page or in Wikipedia talk:Featured article review. No further edits should be made to this page.