Data
====

1. earnings.data.R
  - N         : number of observations
  - n_eth     : number of ethnic categories
  - n_age     : number of age categories
  - height    : height
  - age       : age category
  - earn      : earnings
  - eth       : ethnicity category
  - x         : heights
  - x_centered: centered heights
  - y         : adjusted y value

2. pilots.data.R
  - N          : number of observations
  - n_groups   : number of groups
  - n_scenarios: number of scenarios
  - group_id   : group id
  - scenario_id: scenario id
  - y          : score

3. radon.data.R
  - J     : number of counties
  - N     : number of observations
  - county: county number
  - radon : radon measurement
  - u     : county-level uranium measure
  - x     : house-level first-floor indicator
  - y     : log of the home radon level

Models
======

1. One predictor
  radon_complete_pool.stan: lm(y ~ x)
  y_x.stan                : lm(y ~ x)

2. Multilevel model with varying slope and intercept
  earnings_vary_si.stan: lmer(log(earn) ~ 1 + (1 + height | eth))
  radon_inter_vary.stan: lmer(y ~ u + u:x + (1 + x | county))
  radon_vary_si.stan   : lmer(y ~ 1 + (1 + x | county))

3. Multilevel model with several group level predictors
  earnings_latin_square.stan: lmer(y ~ 1 + (1 + x | eth) + (1 + x | age) 
   				       + (1 + x | eth:age))
  pilots.stan               : lmer(y ~ 1 + (1 | group) + (1 | scenario))

4. Above models with Matt trick
  earnings_latin_square_chr.stan: lmer(y ~ 1 + (1 + x | eth) + (1 + x | age) 
				           + (1 + x | eth:age))
  earnings_vary_si_chr.stan     : lmer(log(earn) ~ 1 + (1 + height | eth))
  pilots_chr.stan               : lmer(y ~1 + (1 | group) + (1 | scenario))
  radon_inter_vary_chr.stan     : lmer(y ~ u + u:x + (1 + x | county))
  radon_vary_si_chr.stan        : lmer(y ~ 1 + (1 + x | county))