I've put together a simulation of the remaining primaries, and Trump's final delegate totals. The state-at-large delegate allocations are based on polls, or estimated polls. District and unbound delegates are a problem, since there's virtually no polling at the district level and unbound delegates are a total crapshoot.
What numbers do you guys have for district and unbound delegates? I'll give mine:
>MD: worst case 15, best case 21, max 24
>PA: worst case 10, best case 35, max 54
>IN: worst case 15, best case 21, max 27
>WV: worst case 20, best case 30, max 31
>CA: worse case 80, best case 120, max 159
>Unbound/misc: Approx 10, max 50
I'll take your estimates into account in my model if you want, but it takes ~30 minutes to run the simulation enough times to get good results.
>>71759200
That's not all of the remaining primaries
An Indiana poll just came out. Trump leads by 6 points. This is huge.
http://www.realclearpolitics.com/epolls/2016/president/in/indiana_republican_presidential_primary-5786.html
>>71759502
>single poll
>only 6 points
>huge
>>71759763
>Only public poll since December
>Lead is greater than the margin of error
post the r source code for method critique faggot
>>71759840
>lead is barely above margin of error
>one poll
>trump has already lost in the past to Cruz despite leads like this
>>71759200
>I'll take your estimates into account in my model if you want, but it takes ~30 minutes to run the simulation enough times to get good results.
Get some better software.
>>71759200
>30 minutes
Fuck dude. Is it really less accurate if you knock it back an order of magnitude?
>>71759854
It's incredibly inefficient, and it uses CSV files with polling data.
http://pastebin.com/gh4Kbchj
>>71760496
It's still pretty accurate if I use 1/10 as many trials.
>>71759981
Cruz just came out saying gays should die, he'll lose the remaining elections. Cruz is unelectable.
You have endpoint bias because of your modeling of a discrete uniform RV as the rounding of a continuous uniform RV.
otherwise noice.
Thinking briefly, you should model this as non-markovian, adding a bias as the number of delegates increases. Call it momentum.
To the nonmathfags in this thread, first of all it sucks to be you hahaha, second of all what OP is saying is that trump has a 50/50 chance to get the required delegates.
>>71761676
I thought about adding a momentum factor, but I wasn't sure how to code it.
There's also the question of the relevancy of momentum, mathematical elimination of Cruz and Kasich, and other factors. Are the voters going to care that Trump is close to clinching it? As Trump has amassed delegates, the party hasn't coallesced around him, so the voters may not either.
The delegate density graph is bimodal, because Indiana is so close and worth 30 delegates. If Trump loses Indiana, he ends up on the left peak. If Trump wins Indiana, he ends up on the right peak.
Trump is leading Indiana, so if he ends up on the right peak and does reasonably well in Pennsylvania and California, he wins.
>>71759763
Remember, the difference between a 1% victory and a 1% loss over half the delegates.
>>71760695
Why not use r studio
>>71763985
I just started using r. I've used it for simple regressions, but I wanted to see if I could write real code in it.
Also,
>$995
>>71762252
Do you have a time series for the prior delegates?
You can run a HMM on it (conditioning maybe across something like geography or demo), and then apply that model to the future to get the momentum.
Should be a good exercise.
If you do it this way it will tell you if there is any momentum to begin with at all (with some assumptions).
I am in Indiana, and I was at the trump rally in indie this week. It was not particularly high energy. With the announcement of the new pro-tranny agenda? There are some trump supporters it is true but I am not optimistic. I have not seen a tranny here ever so his move was ill-timed.
>>71764563
Stochastic processes and time series stuff are all greek to me. Is that stuff inbuilt into r? I know that multiple regressions, anova, etc are built in.
>>71765151
oh, wait, you're actually retarded. my bad. i'll go back to shitposting now.
>>71761676
>To the nonmathfags in this thread
This is like basic University math, why so dapper?
>>71759763
>yuuge
Ftfy
>>71759200
There's literally a site devoted to this:
http://projects.fivethirtyeight.com/election-2016/delegate-targets/
>>71761323
Indiana is not a liberal state.
>>71764466
r studio is free dink, why buy a commercial license?
>>71764466
>>71766292
538 mapped a path for Trump to the nomination, and they're showing how closely he's following that path.
I'm trying to calculate the probability that Trump wins 1237. Related, but not the same.
>>71766870
>>71766831
I noticed that afterwards, sorry.
>>71767232
>sorry
That's not enough. I'm afraid you have to seduce my 3 inch beast.
>>71767232
dont be sorry homie grow a backbone ; )
one way to speed up r code is to write all the data into the r engine and then access it that way instead of reading the csv data repeatedly
loading/copying/handling data is one of r's weaknesses
also if you can vectorize your code then you will be able to speed up the process a lot. there is also a bridge that will let you write c++ functions and use them inside R which will speed up R 100x in some cases
>>71767676
read all csvs once, bind them into r data object and save the r object as a r binary file then you only need to read that binary file once into your memory and it should speed your code up some
also fucken checkem
>>71767676
I originally reread the CSV files every time I needed the data. Now I only read it once and put it in a list. That cut down the time by maybe 10-15%.
It would be nice to parallelize the trials, but r can't do that, as far as I know. The run time could be cut in half, easily.
Thanks for the tips.
>>71768077
r can run in parallel
snow, parallell are two that i know
>>71766402
The overton window has been moved too far for that to matter.
>>71759763
Hey fucktard new poll is out trump +8
>>71767676
sick dubs
