Skip to content

Latest commit

 

History

History
executable file
·
176 lines (130 loc) · 7.27 KB

todo.org

File metadata and controls

executable file
·
176 lines (130 loc) · 7.27 KB
Raw

Todo

[#A] Read the following papers

  • [X] cite:falmagne2019stochastic 
    @incollection{falmagne2019stochastic,
  title = {A stochastic model for the evolution of preferences},
  author = {Falmagne, Jean-Claude and Regenwetter, Michel and Grofman, Bernard},
  booktitle = {Choice, decision, and measurement: Essays in honor of R. Duncan Luce},
  pages = {111--129},
  year = {2019},
  publisher = {Routledge},
}
  • [X] cite:acemoglu10_opinion_dynam_learn_social_networ 
    @article{acemoglu10_opinion_dynam_learn_social_networ,
    author = {Daron Acemoglu and Asuman Ozdaglar}, title = {Opinion Dynamics and    Learning in Social Networks}, journal = {Dynamic Games and Applications},    volume = {1}, number = {1}, pages = {3-49}, year = {2010}, doi =    {10.1007/s13235-010-0004-1}, url =    {https://doi.org/10.1007/s13235-010-0004-1}, DATE_ADDED = {Thu Jun 17    16:16:44 2021},}
  • [X] cite:grofman04_downs_two_party_conver 
    @article{grofman04_downs_two_party_conver,
  author = {Bernard Grofman},
  title = {Downs and Two-Party Convergence},
  journal = {Annual Review of Political Science},
  volume = {7},
  number = {1},
  pages = {25-46},
  year = {2004},
  doi = {10.1146/annurev.polisci.7.012003.104711},
  url = {https://doi.org/10.1146/annurev.polisci.7.012003.104711},
  DATE_ADDED = {Sat Feb 6 12:27:55 2021},
}
  • My notes about axelrod model and quick look at flache
    • cite:flache2017
    • cite:maciel2018emergencia
  • I believe this paper is also useful cite:grofman1990efficient !
  • [X] see notes/ephnotes/degrootreview.pdf
  • [X] cite:holland1996hidden
    • Particularly, what he has to say about tagging;

[#A] spell out my model ideas

See last meeting notes

[#A] send email to grofman

Discuss the following points:
  • The connection between the tag idea with cite:holland1996hidden mechanism of tag
  • My idea about the path vs connection using some kind of decay factor
  • Ask him about literature that could point out how to use institutions as molds of consensus/divergence

Code the model iteration procedures ( ongoing);

  • Fix the pos stuff
  • Fix

write MWE of pos error

Write unit*tests for potential bugs;

[X] Optimize the code;

[X] Think about the consistency of the design

The last email got me thinking about the consistency of the design. The primaries as they stand lead to party position not mattering since the candidate selection is independent from the position of the party. This is a problem. On the other hand if I reinstate the radius the size of the radius will probably be the main factor… Or maybe not.

change voter party update

implement party position update!

Fix the initial condition to actually be ONLY an initial condition

Fix the incumbent behavior

-

fix the candidate choice sampler

  • Simply filter before sampling!

Add \(κ\) as field for agents

[#B] REREAD cite:&aldrich1983downsian

[#B] Read analysis papers

  • [X] cite:&siegel18_analy_comput_model
  • cite:&lavin2021simulation
  • cite:&borgonovo2022sensitivity

[#A] write down the model description

OK, let’s skip for a while writing out a system of equations All I really need is a clear statement in English with just enough math symbolism for me to be sure of exactly how the different pieces of the model work. Then I’ll worry about describing the comparative statics in a non-formal way.

KILL plot the within party shares and prop people like me

implement the new distributional assumption;

  • I’ll simply put a constant in one dimension and work on the other lol.
  • I’ll use cohen d to define overlapping distributions:
  • Thus if I have a distribution of (43,10) I gotta have the other as (56.5,10)
  • OVL=2Φ(−∣δ∣/2)

OKAY implement the new initial party id

implement visualization of party polarization

[#C] Create a representation gap incumbency advantage variant

Right now the model is not presenting any interesting dynamics. Winning doesn’t feedback into the system, nor the dissatisfaction of voters is taken into account. I’m thinking of coming up with possibly simpler model with death/birth dynamics

[#A] check if candidate sampling is correct in the hardwired example

[#A] Create time series for party loyalty colored by position, marker by party

[#A] implement turnout function

(6) make sure that the voting rule overweights those with stronger party ID; maybe you use a turnout function such that the stronger the party ID the more likely the voter is to vote in the primary. If it overweights those with stronger party ID and those with stronger party are disproportionately extremists then, over time, that should push the party nominations toward the extremes.

How I’ll implement that:

  • The subset of agents this function must work is party_supporters
  • the proportion of times I’ve voted for this party will be the strength of my party identification; 
    if rand() < strength_party_id
return(my_id)
end
  • Then I get for each party the supporters who will actually vote in this primary.

[#C] test initial vote

The higher the overlap between party distributions the more cross-party voting there should be. If that’s not true then there’s something wrong. Try running the model with different initial overlaps to see if the model does what it is supposed to do in the initial round before we have party loyalties kicking in.

[#A] implement more concentrated distribution

[#B] Run small pipeline comparing the three distributions

[#A] Pick candidates from turnout voters

[#A] Write pipelines to test for those 3 expected behaviors!

  • Extremists must evolve to have stronger party ids;
  • Candidates Should Become more Extreme;
  • Those who turnout should be more extreme WITH TIME!;

[#A] Weight party id in defining party position;

PROJ [#A] Redesign this

[#A] New design

Re 1. We need to fix the design issue that makes extremists only marginally more likely to be party loyalists. One way might be to manipulate the probabilities in a clearly non-linear way, so that unless you always have voted for a given party, your party loyalty remains low. Maybe also you need to build in both raw number of times of agreement and percentage of agreement. Also, I think we start somebody off with a party ID and always measure things in terms of deviation from that original ID. That should simplify the bookkeeping.

[#B] New Trackers / Data Collection

  • (1) the actual location of the candidates,
  • (2) the changes in the party ID of the voters as a function of their current and previous votes,
  • (3) vote choices vis- a-vis the candidates as a function of ideology and party ID,, so as to verify that the formulas for (a) change in party ID, and (b) the way in which the locations of the candidates is determined by current party ID and or immediate past voting behavior work the way they are support.