You sure got me there, podnuh! Got no fucking idea what you’re talking about.
Oh, and while I’m at it, you might want to calm down and get your facts straight too.
And one way to achieve a representative sample is to use a large, randomized pool of respondents. That’s why, depending on sample size, we’re able to predict how likely the responses are to apply to a group at large. This is stats 101, it’s not complicated. By the way, the Heckman correction is quite relevant to the problem of non-random samples.
[
](http://nobelprize.org/nobel_prizes/economics/laureates/2000/public.html)
If you’re really curious and not just being an asshole, I could link to to a few online calculators that show you the margin of error depending on sample size and such, or something to point you to the conceptual framework that makes statistics work. But you are just being an asshole, right?
Ooooh, burn! You caught me! I included unnecessary capitals. Truly the Argument from Unnecessary Capitalization is solid.
It’s generally best not to demonstrate total ignorance of statistical methodology while committing some sort of Appeal to Authority (with you as the authority). Yes, a properly conducted statistical survey is indeed evidence of what a populace at large thinks/does/believes. A statistical analysis is a model of the populace at large. They are indeed a valid proof as to the nature or popularity of beliefs (I’ll get to your dodge in a moment, sit tight buddy) while at the same time one must accept that there is a degree of imprecision as they are a model and not a description of a populace at large. That’s why statistics work, if they didn’t have a high probability of being predictive, we’d simply ignore them as they wouldn’t be a proof as to anything. But as they do, we don’t ignore them. Again, this is basic. Only an idiot (or a disingenuous schmuck attacking a strawman) would take a statistical claim to be proof that a definite percent of a populace did/believed X, Y or Z. They are proof, however, when a valid sample is used with proper methodology and shows that it is highly likely that the population at large shares a rough isomorphism with the sample collected.
Now, your dodge. Of course people can misrepresent their beliefs, only an idiot would argue otherwise and only someone arguing dishonestly would use that as some sort of a gotcha. (And studies that secretly aim to get at someone’s 'true beliefs" can often lead to some seriously dodgy conclusions, like finding that people don’t like to hire people with non-standard names that ‘sound black’ and concluding that they’re therefore racist, but not checking as to whether or not they would refuse, in equal numbers, to hire people with non-standard names that didn’t ‘sound black’) All samples that rely on the respondent to answer truthfully of their own volition are vulnerable to error. That does not mean that if 57% percents of a well designed and conducted study, that is roughly representative of the populace at large, respond affirmative to the question “are members of race X inferior?” that we can’t logically declare that it is highly likely that roughly at least 57% of the populace at large shares such a sentiment.
That’s why, after all, statistics are based on probability and prediction and only a disingenuous jerk would object to them being used as proof for the probability of a distribution while pretending that anybody was using them as proof for a statement with no probabilistic caveats at all.
Is there anything nearly as scintillating as a brisk discussion of the arcana of statistical sampling? By the way, does anyone have any morphine?
Fresh poop, gleaming in the morning sun?
No one actually needs statistics in order to see that these former cheerleaders for the conquest of Iraq are a bunch of assholes.
If they want credibility, let em pay a trillion for that war they thought was the greatest thing since sliced bread. Pawning it off on my kids like they’ve done ought to be a hanging offense.
Sure, why not? It is 4/20… hey, waitaminit!
Me either. And, hell, you live there! I guess you didn’t notice any “Clean for Gene” campaign members beating the shit out of the police.
Let’s consider our respective contributions to this thread.
Me:
- Outlined specifically what I thought was flawed about this poll and the conclusions offered.
- Outlined specific changes to the poll that I thought would allow one to draw the relevant conclusions.
- Agreed that it would not surprise me if said poll produced results that indicated a higher level of racism among TPs, on average, than one would find in the general population.
You:
- Insulted me.
Yeah, if only I could live up to your standards…
Hasty Generalization is generalizing from a sample that is not representative of the population. But very small samples can still be representative of the population. Therefore, it is not necessarily a Hasty Generalization to generalize from a very small sample.
Consider the example I gave: A newly discovered species, of which only one specimen has been observed, where that specimen has six legs. Do you not consider it to be a virtually failsafe generalization to say that every member of that species has six legs?
That everyone clapped enthusiastically is good evidence that “the view espoused by that person when he said that thing is frequently endorsed, and is considered to be part of the standard spectrum of beliefs within that group.” Indeed, that claim is not just a good generalization in the case you mentioned; it’s true. That Obama is the best candidate is “frequently endorsed” within the political group in question and “is considered to be part of the standard spectrum of beliefs within that group.”
ETA: Having now read your link, all I can say is that author has voiced a common misconception about hasty generalization. My cite is, basically, the example of a perfectly good generalization from a sample of one that I gave earlier. I’ll have to ask for a raincheck on a proper cite… no books around…
ETAA: Wikipedia has it better, but of course, that’s just Wikipedia…
When all you know is stats 101, nothing looks particularly complicated to you.
Political surveys are pretty much never random samples. The cost is too high and opinion changes too quickly to poll enough people to get there. That’s why polls use different sampling methodologies just so they can poll only, say, 1000 people. Simple random sampling will inevitable result in over- and under-sampling. Here are a few sampling methods for the attention span-impaired.
The conceptual framework that makes frequentist statistics work. This is, of course, subject to massive debate among statisticians. The validity of frequentist inferences is by no means beyond dispute. Know any Bayesians?
The margin of error only quantifies sampling bias. You might have learned in remedial stats that this is not the only form of bias, and is in fact one of the least insidious.
But you don’t even know what you don’t know. Because you are such a pig-headed, obstinate asshole, you can’t even bring yourself to admit it.
Fortunately for me, my employers do not agree with you. They continue to reward me handsomely for my econometric work. Imagine that. How much can you get for your wikipedia credentials?
It is actually surprisingly difficult to discuss these sorts of things with semi-educated people like you. You know just enough to be able to use the word “model” and “distribution” and sort of know what you are talking about, but unlike a person without your delusions of adequacy, you don’t recognize the limits of your understanding. I’d gladly have a technical discussion of this sort of thing. If you have a copy of Greene’s Econometric Analysis around, we can go to town.
Asking people and watching people is evidence, too. Keep the insights coming.
While you’re at it, professor, why don’t you tell us what a properly conducted statistical survey actually is. Let me give you a hint: there is a lot of dispute about this by, yanno, actual experts.
So now we have gone from “proof” to “highly likely that the population shares a rough isomorphism” in one breath. My work here is done.
Rewritten, “only someone arguing dishonestly would use the fact the inferences drawn from systematically unreliable data are highly questionable.”
Right after discussing selection bias, Damodar Gujarati writes in Basic Econometrics:
Gujarati’s emphasis. Much of the time, especially if you do surveys, you don’t have a great ex ante sense of data quality. You can design the survey and the sampling methodology as well as is humanly possible, but opinion changes fast and it is damned easy to get it wrong. In your ignorance you take for granted something that is actually quite difficult. It pays to be more cautious.
I am glad that no one’s life depends on your reasoning powers. Please tell me you aren’t in pharma research, biometrics, statistical crimefighting, or in any career that requires you to do more than mouth off about this sort of thing. I hope you don’t pretend to teach it, either.
Actually, that is exactly what it means. Selection bias is a gigantic and mostly unquantifiable problem. The actual literature is shot through with it. In case you are curious and are not just being an asshole, here is a fairly recent and quite good book on the subject. I quote the abstract below:
“Proof for the probability of a distribution?” Do you even read this drivel while you are writing it?
Everyone needs a hobby.
What were the issues? Too wide? Too narrow? pattern too busy?
Are you wearing regimental colours you’re not strictly entitled to?
ETA: 
Ohthankgod.
Yes, but the argument is circular. Nobody has shown that the folks that make the news are representative of the TP movement.
Consider finding a black swan, is that a virtually failsafe piece of evidence for all swans being black?
No, it’s proof that that specific group of people applauded it because they endorsed it.
And Edwards’ rallies? And Kucinich’s, and… That’s the point, if you reduce “frequently endorsed” to “endorsed by any faction in a group” then it’s essentially meaningless. What if racist TP member are a prevalent as diehard Kucinich supporters were during the primaries? Let’s say, for instance, one TP rally applauds some racist tripe about blacks and another boos someone off stage for using the word “nigger”. Does that show us that the TP movement is both a racist and an anti-racist movement?
Yeah, this is why I tend to doubt your claims that you’re a bigshot (earning big bucks, so I should trust your fallacy of argument from authority :rolleyes:). If you honestly don’t understand why a valid statistical study is proof for the probability of the distribution it finds? If you really don’t get why it’s proof of a rough isomorphism that can be used to predict future events and evaluate current events? Well, I’d just be a bit curious to see what commercial application you could put your ‘skills’ to.
“So Maegy old buddy, you finish that study?”
“Sure did, thanks for the money.”
“… um, so what did you find?”
“Oh, nothing, statistics doesn’t offer us any proof about the probability of the distributions we find applying to a larger data set. There are really no conclusions we can draw from the data we collected, even on a probabilistic basis. The work you paid me for proves nothing and doesn’t offer any basis to make probabilistic determinations about where to invest time or resources”
“Wait, we’re paying you for this?”
“Yeah, isn’t it awesome.”
“But statistics have long shown that they have a predictive power when the studies are designed properly. The ability to predict and describe reality shows that there is an isomorphism between our data set and the larger set of items which is was taken to represent and, I mean, that’s why we’re paying you. We’d like to have some studies done with a probative value that can help…”
"You’re such an asshole! You don’t know anything, you pig headed fuck, you can’t admit it! You don’t even know enough to know what you don’t know, I hate you!
“…”
“So, can I have another check now for more research?”
Yeah, keep on keeping on. I don’t doubt that you’ve taken a stats class or two in your day. Have fun.
Not giving people special favors = good
Treating everyone equally = bad
Got it.
Large TP rallies are not likely to be representative of the TP movement?
If it’s the first swan anyone’s ever seen, then it’s pretty good evidence. Not as good as the number-of-legs case, since we know that color can vary within a species more often than leg number.
What faction of the TP movement was at the rally?
If you’re at a Kucinich rally, it’s clear that the genral population in question is “Kucinich supporters”, not “Democrats”
But I’ve never seen any indication that the rally in question was for this or that particular kind of TP movement member. The opposite: It was a gathering for the TP movement itself, just as the Republican National Convention is a gathering for the Republican movement itself.
Conversely, your misuse of words and concepts causes me to doubt that your IQ is in the double digits.
The idea that I am committing some sort of argument from authority is also quite remarkable. Your own pseudo-arguments are backed up by nothing but your repeated obstinacy. This is is your very own kind of fallacy, the argumentum ad carborundum.
Find me a reference that suggests that a study is “proof for the probability of a distribution”. Any reference. You have so grossly distorted useful concepts like “probability” and “distribution” beyond almost any recognition. I think I kinda get what you are saying, but towards you, I am somewhat less than charitable.
What I really don’t get is why people like you do not recognize when they don’t know anything. It’s godddamned obvious that one can draw inferences from analysis of data, and it even works really well under certain circumstances. Statistical mechanics, QC for industrial processes, medical research, etc. It works really well when the data generation process is understood and a lot of data is collected.
You need to be a lot more guarded in your inferences when you do fucking political surveys. At best they can “suggest”. The “isomorphism” is very, very rough. They are useful when they are right, but very often, they just aren’t. Do yourself a favor and read the Berinsky book.
I’m still curious about your thoughts on research design. How do you determine that a survey is well-designed? What are your criteria? Can you determine that a survey is well-designed from reading the newspaper? Do you read empirical work in political science journals? Which ones? Can you point to a survey you think is well-designed and another that isn’t?
I’m definitely going to keep on keeping on. I’ve been promoted several times, and now I don’t even have to come into the office anymore. People bring up results I derived years ago, stuff I barely even remember doing. So I am going to go ahead and keep doing what I apparently do quite well. I analyze (and draw inferences from) billions of global credit transactions.
You, on the other hand, should keep your mouth shut and perhaps even read a book or two.
What’s oversampling? (Wikipedia only has an entry for a concept in signal processing…)
I agree completely with this.
When your sample has too many of a certain population. Like if I wanted a random sample of all Americans and got 80% Black people. They’d be oversampled.