About the internals of that new 54-55% FAKEPOLL (isn't it a crying shame that the only thing they have are FAKEPOLLS that they create?) Make sure to see last 3 paragraphs for English version that they left blank. DUH????
Sampling based on this: (Honestly...I'm not making this bullshit up....this is what they posted)
ˆ t i = P(Y t i = Clinton | Y t i ? {Clinton, Trump}) = logit-1 a t gender[genderi ] + a t age[agei ] + a t race×edu[(race × edu)i ] + a t edu[marriedi ]+ adivision[divisioni ] + a2012 × (2012 Obama vote)i + aorder × (survey order)t (1)
and
Given ˆy t i , we can compute an estimate of Clinton’s share of the two party vote among a group of interest G by weighting each ˆy t i by Ni the number of voters in that demographic cell derived from our estimated likely voter space: Yˆ t G = P i?G Niyˆ t i P i?G Ni .
plus this:
according to an auto-regressive AR(1) process: a t+1 demo[k] = µdemo[k] + fdemo[k](a t demo[k] - µdemo[k]) + t+1 demo[k]. (3) At each wave, the parameter estimates from the previous wave are shrunk towards the mean µdemo[k] by a factor fdemo[k], before mean-zero noise t+1 demo[k] is added.
Fine tuning with this:
We complete the Bayesian specification of the model with the following priors: a 1 demo[k] ~ student-t(µdemo, p sdemo 1 - f 2 demo ; ?) sdemo ~ N+(0, 0.1) fdemo ~ Beta(10, 1) µdemo ~ N(0, 1) t demo ~ student-t(0, sdemo; ?). Drawing fdemo from a beta distribution ensures that 0 < fdemo < 1, meaning that the AR(1) process determining the evolution of the parameters is stationary, with mean µdemo and variance s 2 demo 1-f 2 demo . The initial parameter values a 1 demo[k] are also drawn from the stationary distribution. Stationarity is important because we have no reason, a priori, to b
Figure 6: Correlation of Errors in our State-by-State Prediction and Huffington Post Pollster’s Estimates (left panel); Distribution of Errors in State-by-State Predictions in our Approach and Huffington Post Pollster’s E
(+a few more paragraphs of bullshit and nonsense and graphs from Venus and Pluto)
Then when it was time to post the results in English, they left it blank as show below:
13) How do you feel about the Democratic Party today? : 0 - Hate : 10 : 20 : 30 : 40 : 50 : 60 : 70 : 80 : 90 : 100 - Love
14) How do you feel about the Republican Party today? : 0 - Hate : 10 : 20 : 30 : 40 : 50 : 60 : 70 : 80 : 90 : 100 - Love 1
5) What is your political party affiliation? : Strong Democrat : Weak Democrat : Lean Democrat/ Independent : Independent : Lean Republican/Independent : Weak Republican : Strong Republican
Go try to read it yourself if you can and get a good laugh!
researchdmr.com
msn.com |
