From 7bc747d1b4f34b43b7c2ad8e7643b3e64f0dcc21 Mon Sep 17 00:00:00 2001
From: Kinto <kinto.behr@gmail.com>
Date: Wed, 30 Aug 2023 10:44:19 +1000
Subject: [PATCH] Cleanup

---
 README.md        |  2 +-
 README.qmd       |  2 +-
 models/mrp2.stan | 60 ------------------------------------------------
 3 files changed, 2 insertions(+), 62 deletions(-)
 delete mode 100644 models/mrp2.stan

diff --git a/README.md b/README.md
index f40afbb..958a76a 100644
--- a/README.md
+++ b/README.md
@@ -152,7 +152,7 @@ therefore, predict an ALP loss in 2019.
 
 Given that this is exactly the opposite of what one might guess looking
 at raw polls alone, and that 2019 was a notorious miss in terms of
-election predictions – with many analysts predicting an ALP win, we can
+election predictions, with many analysts predicting an ALP win, we can
 feel very satisfied with this result.
 
 # Diagnostics.
diff --git a/README.qmd b/README.qmd
index 11ea965..403b02d 100644
--- a/README.qmd
+++ b/README.qmd
@@ -148,7 +148,7 @@ ps_sim |>
 
 Thus we see that of our 10,000 simulations, the ALP manages to achieve a majority (i.e. win >75 seats) in under 40% of them. We would, therefore, predict an ALP loss in 2019.
 
-Given that this is exactly the opposite of what one might guess looking at raw polls alone, and that 2019 was a notorious miss in terms of election predictions -- with many analysts predicting an ALP win, we can feel very satisfied with this result. 
+Given that this is exactly the opposite of what one might guess looking at raw polls alone, and that 2019 was a notorious miss in terms of election predictions, with many analysts predicting an ALP win, we can feel very satisfied with this result. 
 
 
 # Diagnostics.
diff --git a/models/mrp2.stan b/models/mrp2.stan
deleted file mode 100644
index ce77b1f..0000000
--- a/models/mrp2.stan
+++ /dev/null
@@ -1,60 +0,0 @@
-//
-// MRP component of model.stan. 
-//
-// Note that in the main model we use the national polls effectively to produce
-// an estimate of national swing. We then take the previous TPP result for
-// each division modified by the national swing as a rough prior for the expected
-// TPP result. 
-//
-// In this model, we simply take the prior TPP result for each division as
-// our rough prior.
-
-data {
-  int<lower=0> n_records;
-  int<lower=0> n_covariates;
-  int<lower=0> n_divisions; 
-  array[n_records] int<lower=0, upper=n_divisions> record_division; # map obs to division
-  array[n_divisions] real<lower=0, upper=100> tpp_div_prev;
-  
-  array[n_records] int<lower=1, upper=2> sex_record;
-  array[n_records] int<lower=1, upper=4> age_record;
-  array[n_records] int<lower=1, upper=4> educ_record;
-  array[n_records] int<lower=0> tpp_record;
-}
-
-parameters {
-  real epsilon;
-  array[4] real b_age;
-  array[4] real b_educ;
-  array[n_divisions] real<lower=0, upper=100> tpp_div_curr;
-}
-
-transformed parameters {
-  array[2] real b_sex = {epsilon, -epsilon};
-  array[n_records] real eta;
-  
-  // Offsets
-  array[4] real o_age = {-2, -1, 1, 2};
-  array[4] real o_educ = {-2, -1, 1, 2};
-  
-  for (n in 1:n_records) {
-    int d = record_division[n];
-    int a = age_record[n];
-    int s = sex_record[n];
-    int u = educ_record[n];
-    
-    eta[n] = inv_logit(tpp_div_curr[d]/100) + b_age[a] + o_age[a] + b_sex[s] + b_educ[u] + o_educ[u];
-  }
-}
-
-model {
-  for (n in 1:n_records) {
-    int d = record_division[n];
-    tpp_div_curr[d] ~ normal(tpp_div_prev[d], 5);
-    tpp_record[n] ~ bernoulli_logit(eta[n]);
-  }
-  
-  epsilon ~ normal(0, 1); 
-  b_age ~ normal(0, 2); 
-  b_educ ~ normal(0, 2); 
-}