Timeline for Election modeling : How is this a valid approach? [duplicate]
Current License: CC BY-SA 3.0
16 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 8, 2017 at 11:11 | history | edited | kjetil b halvorsen♦ |
edited tags
|
|
| Dec 14, 2016 at 17:38 | comment | added | user1566 | Umm, you can't differentiate between models based on a single result, although, in this case, the higher percentage chance a model gave Trump, the better it was... for this one case. In some sense, there are only so many reasonable ways in which you can model election voting. | |
| Dec 14, 2016 at 17:31 | comment | added | irritable_phd_syndrome | @barrycarter. If you can't differentiate between models, then what good is it to make a model at all? | |
| Dec 14, 2016 at 16:37 | comment | added | Aksakal | @barrycarter, 70% was a day before the election, after the 3rd debate they were giving 90% chance to Clinton. the model is not invalid, it's useless | |
| Dec 14, 2016 at 16:18 | comment | added | user1566 | @Aksakal 538 was predicting a 70% chance for Hillary (some people may've misinterpreted this to mean 70% of people were voting for Hillary), not 90%. And, the fact that Trump won doesn't invalidate the model. A 30% chance isn't a 0% chance. Minority probabilities have to occur, otherwise it's not a model, it's a flat out prediction. The general answer here is that you look at sample errors in the polls and use those to compute total probability. I can provide more details if the linked paper above isn't enough. | |
| Dec 14, 2016 at 16:15 | history | closed |
Matthew Drury whuber♦ |
Duplicate of Probability of a single real-life future event: What does it mean when they say that "Hillary has a 75% chance of winning"? | |
| Dec 14, 2016 at 16:13 | review | Close votes | |||
| Dec 14, 2016 at 16:15 | |||||
| Dec 14, 2016 at 15:59 | comment | added | Matthew Drury | Additionally, this paper is good reading on single case probabilities: arxiv.org/abs/quant-ph/0408058 | |
| Dec 14, 2016 at 15:58 | comment | added | Matthew Drury | I believe the answers to the above question cover most of the points you're curious about. Assigning probabilities to one time events has been a fraught philosophical issue for a long time, but the utility of doing so has generally shown us that it's a good idea. | |
| Dec 14, 2016 at 15:03 | comment | added | Aksakal | 538 was predicting 90% chance (or something like that) of winning for Clinton. we have a sample of size ONE. there was a 10% chance that Trump wins, and he won. does this invalidate the model? no. it only shows that it's useless for practical purposes. | |
| Dec 14, 2016 at 14:52 | history | migrated | from politics.stackexchange.com (revisions) | ||
| Dec 14, 2016 at 14:51 | comment | added | Philipp | As suggested by a flag I will migrate this to cross validated. Feel free to send it back when it doesn't fulfill your quality criteria. | |
| Dec 14, 2016 at 14:30 | comment | added | indigochild | As much as I hate to say it, this is a great political science question - but a poor fit for politics.SE. Currently this is not within our scope. | |
| Dec 14, 2016 at 14:06 | answer | added | user5751924 | timeline score: 0 | |
| Dec 14, 2016 at 13:28 | comment | added | jwodder | This seems like a better question for Cross Validated. | |
| Dec 14, 2016 at 12:43 | history | asked | irritable_phd_syndrome | CC BY-SA 3.0 |