A dynamic multilevel Bayesian model to predict US presidential elections

栏目: IT技术 · 发布时间: 4年前

内容简介:Last update on Thursday June 11, 2020 at 07:37 AM EDTCode for a dynamic multilevel Bayesian model to predict US presidential elections. Written in R and Stan.Improving on Pierre Kremp’s

State and national presidential election forecasting model

Last update on Thursday June 11, 2020 at 07:37 AM EDT

Code for a dynamic multilevel Bayesian model to predict US presidential elections. Written in R and Stan.

Improving on Pierre Kremp’s implementation of Drew Linzer’s dynamic linear model for election forecasting (Linzer 2013) , we (1) add corrections for partisan non-response, survey mode and survey population; (2) use informative state-level priors that update throughout the election year; and (3) specify empirical state-level correlations from political and demographic variables.

You can see the model’s predictions for 2020 here and read how it works here .

File dictionary

In terms of useful files, you should pay attention to the 3 scripts for the 2008, 2012 and 2016 US presidential elections are located in the scripts/model directory. There are three R scripts that import data, run models and parse results:

final_model_2008.R
final_model_2012.R
final_model_2016.R

And there are 3 different Stan scripts that will run different versions of our polling aggregate and election forecasting model:

poll_model_2020.stan
poll_model_2020_no_partisan_correction.stan
poll_model_2020_no_mode_adjustment.stan

The model diagnostics displayed below are all results of the poll_model_2020.stan script.

Model performance

Here is a graphical summary of the model’s performance in 2008, 2012 and 2016.

2008

Map

A dynamic multilevel Bayesian model to predict US presidential elections

Final electoral-college histogram

A dynamic multilevel Bayesian model to predict US presidential elections

National and state polling averages and the electoral college “now-cast” over time

A dynamic multilevel Bayesian model to predict US presidential elections

States’ partisan leans over time

A dynamic multilevel Bayesian model to predict US presidential elections

Model results vs polls vs the prior

A dynamic multilevel Bayesian model to predict US presidential elections

Performance

outlet ev_wtd_brier unwtd_brier states_correct
economist (backtest) 0.0333707 0.0302863 49

A dynamic multilevel Bayesian model to predict US presidential elections

## [1] 0.02242826

Predictions for each state

state mean low high prob se
NC 0.503 0.454 0.555 0.527 0.031
MO 0.510 0.461 0.561 0.635 0.030
IN 0.486 0.434 0.537 0.330 0.031
MT 0.483 0.427 0.534 0.297 0.031
FL 0.519 0.472 0.569 0.734 0.030
GA 0.476 0.425 0.528 0.201 0.031
VA 0.527 0.479 0.574 0.818 0.028
OH 0.527 0.478 0.576 0.818 0.029
AR 0.472 0.423 0.524 0.190 0.031
AZ 0.470 0.418 0.523 0.172 0.032
WV 0.469 0.420 0.520 0.154 0.031
NV 0.534 0.483 0.587 0.859 0.031
MS 0.465 0.412 0.522 0.134 0.034
CO 0.535 0.485 0.583 0.878 0.028
ND 0.463 0.403 0.520 0.150 0.034
LA 0.461 0.412 0.514 0.102 0.031
TX 0.458 0.407 0.508 0.092 0.030
0.544 0.513 0.575 0.996 0.018
SC 0.455 0.408 0.508 0.066 0.032
SD 0.453 0.401 0.506 0.074 0.031
NH 0.552 0.503 0.602 0.956 0.030
PA 0.557 0.508 0.606 0.966 0.029
WI 0.559 0.507 0.606 0.974 0.028
KY 0.440 0.391 0.491 0.030 0.031
NM 0.560 0.502 0.614 0.951 0.032
MN 0.561 0.512 0.611 0.976 0.030
TN 0.438 0.388 0.491 0.030 0.032
IA 0.564 0.512 0.612 0.978 0.029
MI 0.565 0.516 0.612 0.981 0.028
AK 0.427 0.375 0.479 0.007 0.031
OR 0.575 0.524 0.624 0.993 0.030
KS 0.424 0.375 0.475 0.009 0.030
ME 0.587 0.537 0.635 0.997 0.029
WA 0.587 0.536 0.636 0.998 0.030
NE 0.412 0.363 0.462 0.001 0.029
AL 0.407 0.359 0.460 0.009 0.031
NJ 0.594 0.543 0.644 0.999 0.030
DE 0.619 0.569 0.666 0.999 0.028
CA 0.621 0.569 0.669 1.000 0.028
OK 0.379 0.330 0.432 0.001 0.031
CT 0.623 0.576 0.670 1.000 0.028
WY 0.375 0.325 0.426 0.000 0.031
MD 0.629 0.568 0.687 0.999 0.034
IL 0.633 0.586 0.679 1.000 0.027
MA 0.636 0.586 0.686 1.000 0.030
ID 0.352 0.301 0.404 0.000 0.031
NY 0.649 0.599 0.699 1.000 0.029
UT 0.344 0.297 0.393 0.000 0.029
VT 0.660 0.611 0.707 1.000 0.028
RI 0.669 0.622 0.716 1.000 0.028
HI 0.677 0.618 0.727 1.000 0.030
DC 0.908 0.880 0.934 1.000 0.015

2012

Map

A dynamic multilevel Bayesian model to predict US presidential elections

Final electoral-college histogram

A dynamic multilevel Bayesian model to predict US presidential elections

National and state polling averages and the electoral college “now-cast” over time

A dynamic multilevel Bayesian model to predict US presidential elections

States’ partisan leans over time

A dynamic multilevel Bayesian model to predict US presidential elections

Model results vs polls vs the prior

A dynamic multilevel Bayesian model to predict US presidential elections

Performance

outlet ev_wtd_brier unwtd_brier states_correct
Linzer NA 0.003800 NA
Wang/Ferguson NA 0.007610 NA
Silver/538 NA 0.009110 NA
Jackman/Pollster NA 0.009710 NA
Desart/Holbrook NA 0.016050 NA
economist (backtest) 0.0327484 0.021624 50
Intrade NA 0.028120 NA
Enten/Margin of Error NA 0.050750 NA

A dynamic multilevel Bayesian model to predict US presidential elections

## [1] 0.02187645

Predictions for each state

state mean low high prob se
FL 0.497 0.450 0.542 0.468 0.027
VA 0.506 0.461 0.550 0.591 0.026
CO 0.510 0.465 0.557 0.645 0.028
OH 0.511 0.465 0.558 0.656 0.028
0.512 0.480 0.542 0.773 0.018
NC 0.488 0.442 0.534 0.343 0.027
NH 0.517 0.470 0.564 0.726 0.028
IA 0.518 0.473 0.561 0.749 0.026
NV 0.521 0.474 0.568 0.787 0.028
WI 0.524 0.478 0.569 0.805 0.027
PA 0.531 0.485 0.577 0.862 0.027
MO 0.463 0.416 0.510 0.095 0.028
MN 0.538 0.493 0.582 0.916 0.027
OR 0.539 0.492 0.585 0.919 0.027
AZ 0.459 0.412 0.504 0.072 0.027
MI 0.541 0.495 0.586 0.928 0.027
IN 0.458 0.413 0.504 0.062 0.027
NM 0.543 0.496 0.589 0.936 0.027
MT 0.455 0.409 0.502 0.056 0.028
GA 0.453 0.405 0.499 0.049 0.028
SC 0.441 0.386 0.496 0.043 0.033
NJ 0.559 0.510 0.606 0.975 0.028
SD 0.439 0.392 0.487 0.016 0.029
ME 0.561 0.515 0.606 0.982 0.027
WA 0.565 0.519 0.612 0.987 0.028
CT 0.571 0.523 0.619 0.993 0.029
ND 0.425 0.380 0.469 0.005 0.026
TN 0.425 0.378 0.472 0.004 0.028
NE 0.422 0.375 0.470 0.003 0.028
WV 0.418 0.368 0.469 0.002 0.031
MS 0.415 0.351 0.478 0.015 0.037
TX 0.415 0.373 0.461 0.002 0.027
CA 0.590 0.547 0.632 0.998 0.025
MA 0.590 0.541 0.637 0.996 0.028
LA 0.405 0.359 0.452 0.002 0.028
KY 0.405 0.354 0.455 0.000 0.030
KS 0.399 0.342 0.460 0.005 0.036
DE 0.604 0.549 0.659 0.997 0.033
IL 0.604 0.556 0.649 1.000 0.027
MD 0.607 0.559 0.653 1.000 0.027
AL 0.389 0.342 0.437 0.000 0.028
AR 0.378 0.333 0.425 0.000 0.028
RI 0.623 0.573 0.672 1.000 0.029
NY 0.623 0.576 0.668 1.000 0.027
AK 0.366 0.306 0.428 0.001 0.037
OK 0.339 0.290 0.393 0.000 0.032
HI 0.662 0.615 0.707 1.000 0.027
VT 0.674 0.622 0.721 1.000 0.028
ID 0.325 0.276 0.375 0.000 0.030
WY 0.317 0.250 0.391 0.000 0.044
UT 0.275 0.231 0.323 0.000 0.029
DC 0.899 0.848 0.938 1.000 0.023

2016

Map

A dynamic multilevel Bayesian model to predict US presidential elections

Final electoral-college histogram

A dynamic multilevel Bayesian model to predict US presidential elections

National and state polling averages and the electoral college “now-cast” over time

A dynamic multilevel Bayesian model to predict US presidential elections

States’ partisan leans over time

A dynamic multilevel Bayesian model to predict US presidential elections

Model results vs polls vs the prior

A dynamic multilevel Bayesian model to predict US presidential elections

Performance

outlet ev_wtd_brier unwtd_brier states_correct
economist (backtest) 0.0864787 0.0602528 47
538 polls-plus 0.0928000 0.0664000 46
538 polls-only 0.0936000 0.0672000 46
princeton 0.1169000 0.0744000 47
nyt upshot 0.1208000 0.0801000 46
kremp/slate 0.1210000 0.0766000 46
pollsavvy 0.1219000 0.0794000 46
predictwise markets 0.1272000 0.0767000 46
predictwise overall 0.1276000 0.0783000 46
desart and holbrook 0.1279000 0.0825000 44
daily kos 0.1439000 0.0864000 46
huffpost 0.1505000 0.0892000 46

A dynamic multilevel Bayesian model to predict US presidential elections

## [1] 0.03073332

Predictions for each state

state mean low high prob se
FL 0.502 0.457 0.547 0.509 0.027
NC 0.496 0.448 0.542 0.455 0.027
OH 0.492 0.447 0.537 0.386 0.027
IA 0.491 0.443 0.541 0.384 0.030
NV 0.514 0.465 0.562 0.676 0.029
0.517 0.487 0.547 0.846 0.018
PA 0.519 0.474 0.566 0.747 0.028
CO 0.520 0.475 0.566 0.763 0.028
NH 0.521 0.475 0.567 0.764 0.028
AZ 0.475 0.425 0.525 0.204 0.029
MI 0.527 0.482 0.572 0.844 0.027
WI 0.528 0.485 0.574 0.846 0.027
VA 0.528 0.483 0.572 0.857 0.026
GA 0.471 0.426 0.518 0.148 0.028
MN 0.534 0.486 0.580 0.884 0.027
NM 0.540 0.495 0.587 0.918 0.028
SC 0.456 0.410 0.501 0.052 0.027
ME 0.551 0.506 0.597 0.968 0.027
MO 0.446 0.401 0.492 0.025 0.028
OR 0.555 0.509 0.600 0.976 0.027
TX 0.442 0.397 0.488 0.018 0.027
MS 0.441 0.394 0.488 0.021 0.028
WA 0.573 0.528 0.619 0.996 0.027
CT 0.573 0.526 0.616 0.995 0.026
AK 0.426 0.380 0.470 0.003 0.026
IN 0.426 0.379 0.472 0.002 0.028
DE 0.576 0.532 0.621 0.997 0.027
NJ 0.580 0.532 0.626 0.997 0.028
MT 0.413 0.368 0.460 0.001 0.028
LA 0.412 0.368 0.456 0.002 0.027
KS 0.412 0.362 0.460 0.001 0.029
IL 0.589 0.542 0.635 0.999 0.028
TN 0.407 0.363 0.453 0.000 0.027
RI 0.594 0.545 0.643 0.999 0.029
SD 0.405 0.359 0.452 0.000 0.028
UT 0.395 0.325 0.452 0.000 0.034
NE 0.393 0.347 0.440 0.000 0.028
NY 0.610 0.566 0.655 1.000 0.027
AR 0.390 0.345 0.437 0.000 0.028
AL 0.389 0.344 0.434 0.000 0.027
ND 0.387 0.339 0.435 0.000 0.029
KY 0.378 0.334 0.422 0.000 0.026
CA 0.624 0.581 0.666 1.000 0.025
MA 0.631 0.585 0.676 1.000 0.027
MD 0.638 0.592 0.682 1.000 0.026
ID 0.359 0.317 0.407 0.000 0.028
WV 0.358 0.316 0.402 0.000 0.026
HI 0.653 0.605 0.700 1.000 0.028
OK 0.346 0.302 0.391 0.000 0.027
VT 0.676 0.629 0.719 1.000 0.026
WY 0.290 0.251 0.335 0.000 0.026
DC 0.906 0.870 0.936 1.000 0.018

Cumulative charts

Calibration plot

A dynamic multilevel Bayesian model to predict US presidential elections

Licence

This software is published by The Economist under the MIT licence . The data generated by The Economist are available under the Creative Commons Attribution 4.0 International License .

The licences include only the data and the software authored by The Economist , and do not cover any Economist content or third-party data or content made available using the software. More information about licensing, syndication and the copyright of Economist content can be found here .


以上所述就是小编给大家介绍的《A dynamic multilevel Bayesian model to predict US presidential elections》,希望对大家有所帮助,如果大家有任何疑问请给我留言,小编会及时回复大家的。在此也非常感谢大家对 码农网 的支持!

查看所有标签

猜你喜欢:

本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们

Flow-Based Programming (2nd Edition)

Flow-Based Programming (2nd Edition)

CreateSpace / 2010-5-14 / $69.95

Written by a pioneer in the field, this is a thorough guide to the cost- and time-saving advantages of Flow-Based Programming. It explains the theoretical underpinnings and application of this program......一起来看看 《Flow-Based Programming (2nd Edition)》 这本书的介绍吧!

JSON 在线解析
JSON 在线解析

在线 JSON 格式化工具

HTML 编码/解码
HTML 编码/解码

HTML 编码/解码

RGB HSV 转换
RGB HSV 转换

RGB HSV 互转工具