US Election 2016
Users browsing this topic:0 Registered, 0 Hidden and 0 Guests
Registered Users: None |
 |
| View previous topic :: View next topic |
| Author |
Message |
Wokko 
Come and take it.
Joined: 04 Oct 2005
|
| Post subject: |  |
|
I genuinely think if Joe Biden had run he'd be winning in an absolute landslide, despite him being creepy and a bit of a flake. A Biden/Paul contest would've been politics at its best. What we have is a circus, but if they're going to give us a circus then damn it, I'm going to enjoy the show.  |
|
|
|
 |
Wokko
Come and take it.
Joined: 04 Oct 2005
|
|
|
|
|
Wokko
Come and take it.
Joined: 04 Oct 2005
|
| Post subject: | |
|
Clinton had CNN debate questions leaked to her. Of course it's all just a vast right wing conspiracy.  |
|
|
|
|
Tannin 
Can't remember
Joined: 06 Aug 2006
Location: Huon Valley Tasmania
|
| Post subject: | |
|
| Wokko wrote: | | As shown in the leaked emails, the polls are pretty universally rigged |
Look, it's OK to be a statistical ignoramus. There is no law that says you have to understand anything about mathematics, survey sampling, or poll methodology. But it is, to say the very least, very unwise to open your mouth and tell the whole world that you haven't understood even simple basic terms like "oversampling".
For those with an interest in these matters, oversampling is a standard technique used to improve accuracy. It works like this:
Suppose you want to poll the members of Nick's Collingwood Bulletin Board about something and you can't afford to ask every single one individually. Let's say your question is "Which club do you hate more: Sydney or Carlton?"
First, you need to select a representative sample. You must choose the sample according to certain rules to ensure that it is indeed representative of the membership as a whole. In essence, you choose people randomly. (In practice, there is a whole field of study just in sampling design, but we can skip the technicalities - the important thing is that we end up with a truly representative sample of Nick's members. There are well-established guidelines and procedures for this and anyone with professional training knows them.)
Now the really neat thing about a properly selected sample is that you can use the numbers it generates to produce quite extraordinarily accurate pictures of the whole population - not just those you sample - and you can quantify your margin of error with remarkable precision. You need several years of study at tertiary level to properly appreciate the power of these mathematical methods, but to summarise: if you follow the rules properly, you get results so good that it seems like magic.
So what is the story with "oversampling"? Well, suppose you don't just want to know if the membership as a whole hate Sydney worse than Carlton, you also want to know if there is a difference between male and female members on this question.
You have polled 50 Nicksters on the main question (which, you have calculated using the standard methods, is a sufficient sample size to provide the accuracy you need). But of those 50, only 7 are female? Can we trust a finding from just 7 respondents? Probably not. Properly sampled data sets don't need to be nearly as large as most untrained people imagine, but 7 is unlikely to be enough. We probably need about twice as many. There are ways to calculate exactly how many we need for a given confidence level. Anyone with proper training can apply them. Standard stuff. I can't remember now whether you learn how to do this in first year or second year. Let's say that we do our sums and decide that we need at least 15 females for the accuracy we desire - but we only have 7. So what do we do?
We oversample. We select our 50 members at random, and then we keep on sampling extra Nicksters until we have 15 females alongside our 43 males. We ask our question. Then we adjust the weightings of the answers mathematically to compensate and thus keep the overall result correct. Now we can not only say how many Nicksters prefer Carlton to Sydney (none, I hope!), we can alse be specific as regards male and female Nicksters. If we want to, we can use the same oversampling technique to measure the opinion of Nicksters from NSW or Nicksters with red hair - whatever we like. It's a routine, normal part of polling, and something that anyone with even very basic training is familiar with.
What the Trump campaign is doing, of course, is quoting a standard technical term to people who are way too ignorant and stupid to know what it actually means, and telling them it means "fraud!"
Only very ignorant and stupid people would fall for this moronic trick, of course. But then, where is the problem in that? You'd have to be pretty damn ignorant and stupid to believe that you could build a wall between the USA and Mexico and have the Mexicans pay for it, after all. If you can get away with one stupid lie, says Trump, why not a two of them? Or a hundred and two?
_________________
�Let's eat Grandma.� Commas save lives! |
|
|
|
|
Wokko
Come and take it.
Joined: 04 Oct 2005
|
| Post subject: | |
|
| Have you read the emails or just gone off half cocked? Tell me how oversampling Dems +30 in Arizona improves accuracy. |
|
|
|
|
David 
to wish impossible things
Joined: 27 Jul 2003
Location: the edge of the deep green sea
|
|
|
|
|
watt price tully 
Joined: 15 May 2007
|
| Post subject: | |
|
| Tannin wrote: | | Wokko wrote: | | As shown in the leaked emails, the polls are pretty universally rigged |
Look, it's OK to be a statistical ignoramus. There is no law that says you have to understand anything about mathematics, survey sampling, or poll methodology. But it is, to say the very least, very unwise to open your mouth and tell the whole world that you haven't understood even simple basic terms like "oversampling".
....... |
There's a lot of information about oversampling out there & you're spot on about the methodology. That is one of the reasons why there is such variation in the polls that are averaged out by "real clear politics" averages.
The scores & scores of polls are not all equal. I gave an example of margin of error earlier & oversampling is another feature that is not used in some/many of the polls. When oversampling may take into account the views of say hispanics and others in the polling to get their view of matters that does not makes the polls invalid, indeed the opposite is true. In the same way, if they want to get a view of people living in bunkers then they need to oversample people in bunkers when doing a general poll.
Mind you it reminds me of the joke about a pollster asking a question to three people; one from China, one from the US and one from Israel.
The interviewer asked:
Excuse me, can I get your opinion on the food shortage"
The person from China asked "what's an opinion?"
The person from the US asked "what's a shortage?"
The Israeli asked "what's excuse me?"
Then again maybe it wasn't related to oversampling.
_________________
“I even went as far as becoming a Southern Baptist until I realised they didn’t keep ‘em under long enough” Kinky Friedman |
|
|
|
|
David
to wish impossible things
Joined: 27 Jul 2003
Location: the edge of the deep green sea
|
|
|
|
|
Tannin
Can't remember
Joined: 06 Aug 2006
Location: Huon Valley Tasmania
|
| Post subject: | |
|
| Wokko wrote: | | Tell me how oversampling Dems +30 in Arizona improves accuracy. |
In exactly the same way that oversampling female Nicksters improves accuracy. The 7 females in our original sample of 50 is too small a number to draw valid statistical inferences from. We can say that (e.g) "80% of Nicksters prefer Sydney to Carlton" within a certain confidence interval, but we can't say "but only 57% of female Nicksters feel the same way" because the sample size is too small and chance plays far too large a part.
The part played by chance in sample selection can be calculated very accurately, provided only that the selection is carried out according to certain well-established rules. It is trivially simple to calculate the necessary sample size to provide any desired accuracy. Larger samples are more accurate (all else being equal) but also (of course) more expensive, and as the sample size increases, the improvement in accuracy with each extra individual decreases rapidly. A sample of 2000, for example, is only just barely superior to a sample of 1000, where a sample of 20 is vastly better than a sample of 10.
Calculating the minimum necessary sample size for a given accuracy under given conditions is a fundamental part of statistical training. If I remember correctly, you normally learn how to do that around the middle of the first year of a statistics degree, though of course you revisit the subject from time to time later on as you advance to more complex and difficult methods.
Let's return to our example. How can we improve our ability to measure the views of female Nicksters re Carlton v Sydney? Well, we could double the sample and ask 100 Nicksters instead of 50. But, of course, that also doubles the expense and makes the whole thing take longer. So we don't want to do that - especially as it makes little difference to the accuracy of our overall answer (because of the effect of diminishing returns with increasing sample sizes) , only to our answer re female Nicksters.
So instead, we oversample. We add extra female Nicksters to our survey to get the number we need for a valid result. In my example, I said that was 15 altogether. (The actual number for a sensible result would probably be more than that. I haven't calculated it - although of course this is a step you always take if you are doing it in real life - but you will find that the number is often quite a bit smaller than the layman would expect. But in real life you don''t guess it, you do the maths.
From this point there are two ways you could proceed. Either:
(a) Calculate the overall answer using your original 50 respondents and ignoring the added 8 females.
(b) Include all 58 respondents but weight the females at 7/15ths of a "vote" each so as not to skew the results.
Please yourself which method you use. Either is valid, though the second one would generally be preferred. This is basic stuff any first-year student could do - and most certainly would do in an end-of-year exam. (Or fail the unit, of course. Some people do.)
In short, oversampling is an established technique. It is not just legitimate, for certain purposes it is essential.
The ability of the Trump campaign to pretend that it is something weird or dishonest is quite bizarre. I mean, it's one thing to produce a lie, it's another thing entirely to produce and stick to a lie which is so obviously and easily proved false.
_________________
�Let's eat Grandma.� Commas save lives! |
|
|
|
|
Wokko
Come and take it.
Joined: 04 Oct 2005
|
| Post subject: | |
|
| Yes, I understand statistics. Have you read the wikileaks emails pertaining to Democrats using media polls to skew results? They are not using legitimate oversampled they are oversampling in categories that do not relate to the areas they're polling. |
|
|
|
|
Tannin
Can't remember
Joined: 06 Aug 2006
Location: Huon Valley Tasmania
|
| Post subject: | |
|
News to me.
_________________
�Let's eat Grandma.� Commas save lives! |
|
|
|
|
think positive
Side By Side
Joined: 30 Jun 2005
Location: somewhere
|
|
|
|
|
Morrigu
Joined: 11 Aug 2001
|
| Post subject: | |
|
5 days 2 hours 13 minutes 15 secs
Till it ends!!!
_________________
“The greatness of a nation and its moral progress can be judged by the way its animals are treated.” |
|
|
|
|
watt price tully
Joined: 15 May 2007
|
| Post subject: | |
|
| Morrigu wrote: | 5 days 2 hours 13 minutes 15 secs
Till it ends!!! |
AEST or US time M?
_________________
“I even went as far as becoming a Southern Baptist until I realised they didn’t keep ‘em under long enough” Kinky Friedman |
|
|
|
|
David
to wish impossible things
Joined: 27 Jul 2003
Location: the edge of the deep green sea
|
| Post subject: | |
|
| Wokko wrote: | | Yes, I understand statistics. Have you read the wikileaks emails pertaining to Democrats using media polls to skew results? They are not using legitimate oversampled they are oversampling in categories that do not relate to the areas they're polling. |
Can you provide links to these? The only emails I've seen have had the associated claims thoroughly debunked in the Atlantic and elsewhere.
I have a feeling something different is going on with these Arizona results, though. There, it seems, simply more Democrats were polled. But isn't this entirely plausible? Are you saying that Republican votes should have been oversampled in such a scenario, or that they shouldn't have been?
The claim here is that the pollster deliberately produced a fraudulent result but then, for some reason, couldn't be bothered to at least fudge the data in their totally publicly accessible press release to make the results seem more legit. That's clearly ridiculous on any number of levels; and if Democratic-leaning pollsters are doing that, why wouldn't Republican-leaning pollsters do the same?
I may not know much about statistics, but I can tell a red herring when I see one.
_________________
"Every time we witness an injustice and do not act, we train our character to be passive in its presence." – Julian Assange |
|
|
|
|
|
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You cannot download files in this forum
|
|
The Rules
FAQ
Usergroups
Calendar
Search
Log in
Register
