I mentioned the election in passing; what a mistake. This person went into a vitriolic denunciation of Obama very like the Perigo piece.
I talked to an American tonight who expounded at length on Obama’s failings in a way that didn’t seem to me to bear much relation to observable reality.
I also have a very good friend whos political views I find ... disturbing.
Not sure about this one. There are electoral systems in the world where people actually physically line up behind the candidates, and they can see where everyone is. This seems crazy to us, but it does seem to work out OK.
That's pretty much how the Iowa Democratic caucuses work, and they have a significant influence on who gets to be the Democratic nominee as they're the first held. From wikipedia:
The process used by the Democrats is more complex than the Republican Party caucus process. Each precinct divides its delegate seats among the candidates in proportion to caucus goers' votes. Participants indicate their support for a particular candidate by standing in a designated area of the caucus site (forming a preference group). An area may also be designated for undecided participants. Then, for roughly 30 minutes, participants try to convince their neighbors to support their candidates. Each preference group might informally deputize a few members to recruit supporters from the other groups and, in particular, from among those undecided. Undecided participants might visit each preference group to ask its members about their candidate.
After 30 minutes, the electioneering is temporarily halted and the supporters for each candidate are counted. At this point, the caucus officials determine which candidates are viable. Depending on the number of county delegates to be elected, the viability threshold is 15% of attendees. For a candidate to receive any delegates from a particular precinct, he or she must have the support of at least the percentage of participants required by the viability threshold. Once viability is determined, participants have roughly another 30 minutes to realign: the supporters of inviable candidates may find a viable candidate to support, join together with supporters of another inviable candidate to secure a delegate for one of the two, or choose to abstain. This realignment is a crucial distinction of caucuses in that (unlike a primary) being a voter's second candidate of choice can help a candidate.
When the voting is closed, a final head count is conducted, and each precinct apportions delegates to the county convention. These numbers are reported to the state party, which counts the total number of delegates for each candidate and reports the results to the media. Most of the participants go home, leaving a few to finish the business of the caucus: each preference group elects its delegates, and then the groups reconvene to elect local party officers and discuss the platform.
The delegates chosen by the precinct then go to a later caucus, the county convention, to choose delegates to the district convention and state convention. Most of the delegates to the Democratic National Convention are selected at the district convention, with the remaining ones selected at the state convention. Delegates to each level of convention are initially bound to support their chosen candidate but can later switch in a process very similar to what goes on at the precinct level; however, as major shifts in delegate support are rare, the media declares the candidate with the most delegates on the precinct caucus night the winner, and relatively little attention is paid to the later caucuses.
Somewhat crazy, but it sounds like fun for a political junkie. STV, musical chairs, and arguing with the other lefties for the win!
But whatever method he used you can’t simply take the 92% and treat it like a dice roll. It is not necessarily true that in 8 elections where he predicts a 92% chance he will have a 50% chance of being wrong.
If being 92% certain of something doesn't mean there is an 8% chance of something else happening, what does it mean?
I assume it acknowledges that there is at least a chance of something else happening? So if you have enough elections, at some point, there will be a 50% chance of you not picking 100% of the races correctly. At what number of elections does that happen for elections, all of which have a 92% certainty?
I had understood, though I may be wrong, that the group of results from each possible sample of a population would be normally distributed.
And the polling are random (at least in theory).
The polling is random. But determining the outcome of the election based on polling, demographics, exit polls, and partial results isn't. Bad data going in will mess it up, but Nate Silver is using multiple polls - once you mix together a couple of dozen data sets, each of a couple of thousand people, you're starting to get some good polls.
And the polling are random (at least in theory)
But it isn't. When you poll people they don't answer randomly - that's the point. Each poll is influenced by the type of question asked, who is asking (male female) the economics of the area, the method used (cellphones/landlines/people on the street).
Statisticians know all this and hence don't use the same models as you would use for a dice roll.
One of the problems we have is that polls are reported as if they are simple dice rlls and that is one of the many flaws in the reporting of polls.
The best example is in New Zealand the margin of error that is quoted is complete bollocks.
That's why statisticians are impressed by the 538 group, the work they do to combine non-normal datasets and generate a number that experimentally matches reality is pretty amazing.
I'm not saying that they can't be wrong just that it isn't a simple stats case where if you repeat the dice roll they will be wrong sooner or later. Using their kind of analysis it is entirely plausible that they would (almost) never be wrong.
These are probably just stupid blow-hards, but a bit of historical awareness would come in handy for them.
I think so, on both counts.
Doesn't mean we should be stocking up the shelter with canned food and ammo. :)
What's the point of the shelter if it's not stocked?
It is not necessarily true that in 8 elections where he predicts a 92% chance he will have a 50% chance of being wrong.
So no I stand by my first statment, it was a forgone conclusion.
Presuming you meant 9 rather than 8, I'm still not following you there. Yes, successive elections are not independent events, but if you've correctly worked out the probability each time, you've taken the dependence into account. Otherwise your claim of 92% likelihood is not accurate.
Perhaps I'm missing something subtle in here. What does "92% likely" actually mean, in this case? Because "foregone conclusion" to me means "100% likely".
ETA Snap Graeme.
I talked to an American tonight who expounded at length on Obama’s failings in a way that didn’t seem to me to bear much relation to observable reality.
It does make you wonder what the hell’s going on.
My theory (today's one) is that many middle Americans' financial futures and investment incomes have been so badly hit by the global economic meltdown, with no outlook for it improving, that they are scared; terrified of losing everything. So people who are usually reasonably rational are looking for someone to blame. Obama is President so he must be to blame; never mind that the mess was created by a couple of decades of financial and political chicanery on all sides.
Perhaps I’m missing something subtle in here. What does “92% likely” actually mean, in this case? Because “foregone conclusion” to me means “100% likely”.
My question too.
Using their kind of analysis it is entirely plausible that they would (almost) never be wrong.
If so, then they should not say that it's 92% likely, but whatever the likelihood actually is. Because "92% likely" means what it means. I'm not disputing they could be more accurate than that. Just that I don't see the point of saying a number that is not correct. If it's known to not be correct, why bother saying it?
ETA: I'd like to add that I'm genuinely interested in this question, Bart. You might very well be right, and understanding why would be useful to me.
There has to be something in the water,
or the air…
Something in the water might help!
I would just like to point out that when Corin Dann explained this story to One News viewers this week, he said that Nate Silver had forecast that "Obama will win by 92%".
This data journalism thing clearly has a way to go yet.
In other upcoming elections, can I predict that Xi Jinping will win Shangdong and thus become leader of China for the next ten years.
Now now, let's avoid the cliches
Thank you for that. I think.
And to David Hood
A (very) quick bit of digging into the results
Thank you very much for that too...PAS - explaining stuff in more detail since ages ago
Ok I'll try and explain what I think is going on again with the caveat that I am not a professional statistician and I don't know exactly what the 538 group is doing.
What I think 538 are doing is using a mathmatical model that takes data from polls and from economic data and from other data. Some of that data is hard, meaning average income is a hard number, the number of new cars sold in a county is a hard number, and some is soft meaning a poll result has probability of being wrong. The data is weighted, with some data being assumed to be more important to the outcome. And for numbers with variability (like the poll results) that variability will be carried through into the model.
They then run the model and it spits out a number eg 303 votes for Obama. Then they run the algorithm again instead using the poll result plus 5% and it spits our another number, or the same number.
They run that model thousands or millions of times with slightly different input numbers and slightly different weightings, sometime with more importance placed on economic data sometimes with more importance on one poll versus another. The reason they change the weightyings etc is because some of those weightings are guesses so they try the model with different guesses to see what happens.
Then they add up the number of times they see a given result, in this case 92% of the times they ran the model it said Obama won.
It is a LOT more complicated than that because they have spent ages verifying assumptions in their models.
But inherently they will have been conservative in the sense that data they are haven't tested as much will be weighted differently. Using an analysis like that if they get over 90% of the runs showing an event will occur then out in the real world it is a forgone conclusion. If I were a betting man I would put money on it.
I don't really have a problem with the MSM reporting it as Obama 92% to win because this is seriously complex maths and stats. But you simply can't treat it like a dice roll or coin toss and say they'll be wrong once every 8 or 9 elections. It just isn't that kind of stats.
What puzzled me talking to my guy last night was that he was ultra-focused on the ballooning fiscal deficit -- which is entirely understandable -- but I couldn't get him to engage with the fact that Romney's budget proposal, as repeatedly expressed, simply didn't add up.
The Econmist calculated that it would add $7 trillion to the deficit over 10 years, using the words "cloud cuckoo land" and "innumerate", pointing out that the only serious defence of it was the belief that once elected, Romney wouldn't really do what he said he was going to do.
(And in response The Economist was deluged with literally thousands of comments accusing the editors of being communists. I could not find a single comment that sought to rebut on the actual facts. It was mad -- and maddening.)
It just seemed to me in the end that my guy simply hated Obama.
I don't really have a problem with the MSM reporting it as Obama 92% to win because this is seriously complex maths and stats.
Not sure I agree with that. Reporting it as 92% likely has a meaning that anyone with higher than year 10 maths can probably understand. Which is most of the population. So you're misreporting it to them, and patronizing everyone else. If there is actually a word for what the 92% number means, it should be used. IMHO.
While it is pretty thouroughly explained, I might as well offer my take as well.
538 is based on a kind of modelling generally called a Monty Carlo method, where you run the same test (election) lots of times with different random numbers (voter support) to see how likely each potential outcome is. In 538's case this is running mock elections many, many times see how changing the support in each state affects the final outcome.
The expert knowledge comes in knowing what range of numbers to use as inputs, in this sense how far the polls and other input data should be randomly shifted to reflect potential election day results. Basically, assessing how much the polls are potential wrong by. I suspect this is where the republican data modelling broke down, in the expert judgement about how far wrong (and in what direction) the polls are.
There is also some model building expert judgement in the way in which all the inputs are not independent- For example, if California had gone to Romney, then it is much more likely most other states would also have gone to Romney than it is that California would be an isolated incidence.
So by assessing how often an Obama result came up in simulations, and the strength of the input data, Nate Silver derived his headline figure of 92%. This figure is technically an oversimplification, as no indication of the confidence of the parts of the input was made (Nate Silver's model is a private one, unlike some of the University based ones which were predicting an Obama win as more likely but also providing the estimates of the strength of the conclusion and the data assumptions ).
I think at one point Nate Silver did say he was taking the more conservative options- essentially the Republican analysts were so certain of victory he reduced Obama's chance of winning on the assumption that they might know something he didn't. This is one reason that until close to election day Silver was rating Obama lower than some of the other such sites.
With a headline statistic like 92% you can say that given enough elections under the conditions set in the model, Romney would win 8% of the time (in that sense like the coin toss analogy). This is what the Electoral Vote Distribution graph on 538 shows- the distribution of potential results from the model.
The percentage headline figure ultimately derives from an estimate of how different the polls are from final turnout. If you assumed the polls were a totally accurate representation of the results of voting day, then Obama had a 100% chance of winning when giving him 51% support. If you assume there is some uncertainty (and it is in the professional polling analyst's judgement here) in some of that uncertain time Romney will win, and some Obama (and this is where the repeated analysis of possible outcomes comes in). Because of the uncertainty, If I run 9 elections with these assumptions (92% chance of victory) in the normal course of events I would expect my assumption to wrong somewhere between 0 and 9 times, but typically 1 time, more likely between 0 to 2 times.
The Econmist calculated that it would add $7 trillion to the deficit over 10 years, using the words “cloud cuckoo land” and “innumerate”, pointing out that the only serious defence of it was the belief that once elected, Romney wouldn’t really do what he said he was going to do.
Yup. And if you’re even half-way serious about being a fiscal conservative, David Stockman calling bullshit on Paul Ryan’s much-touted Budget plan should have raised more red flags than Mao’s birthday in Beijing. Anyone who thinks Stockman is now, or ever has been, some kind of born-again Keynesian ‘Communist’ is either a disingenuous GOP hack or functionally illiterate.
Here is an interesting article about how back in 1952 the press were saying the election was too close to call while the computer (The Univac) kept spitting out results that it was going to be a landslide.
Actually running these is slightly beyond my abilities, but I like the concepts.
They are fairly easy to do in excel (anyone keen could google monty carlo excel) but the real trick is correctly judging the probabilities you assign to potential outcomes, for which you generally need in depth knowledge of the area.
Please don't mention Baysian Statistics. It will start a firefight around here.......
92% Bet the house on it.
75% Bet the cat
51% Bet a dollar.
The best description of probablility / uncertainty I ever saw it this:
Dr C H Meyer NBS(1935ish)
on his recently calculated Heat Capacity of Ammonia
“We think our reported value is good to 1 part in 10,000: we are willing to bet our own money at even odds that it is correct to 2 parts in 10,000. Furthermore, if by chance our value is shown to be in error by more than 1 part in 1,000, we are prepared to eat the apparatus and drink the ammonia.”
Mitt Romney sings his concession speech :)
Furthermore, if by chance our value is shown to be in error by more than 1 part in 1,000, we are prepared to eat the apparatus and drink the ammonia.”
That is a scientist with deep committment to his research!
Laughed out loudly -[scared the neighbour's dog-