Complete data loading, add tests, bump version

Author: nilshg

Project.toml

@@ -1,7 +1,7 @@
 name = "SynthControl"
 uuid = "a73a6d50-3b99-11e9-2e1c-6f5007f3555d"
 authors = ["Nils <nilshg@hotmail.com>"]
-version = "0.2.0"
+version = "0.2.1"
 
 [deps]
 CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"

src/SynthControl.jl

@@ -6,7 +6,9 @@ import CSV
 using Reexport
 @reexport using TreatmentPanels
 
-export SynthControlModel, fit!, isfitted, load_brexit, load_germany, load_basque, load_smoking
+export SynthControlModel, fit!, isfitted
+export load_brexit, load_germany, load_basque, load_smoking
+export load_brexit_panel, load_germany_panel, load_basque_panel, load_smoking_panel
 
 include("synthcontrolmodels.jl")
 include("plotrecipes.jl")

src/load_data.jl

@@ -3,48 +3,127 @@
 """
     load_brexit()
 
-Returns a `DataFrame` with quarterly GDP data on 30 OECD countries, borrowed from the
-analysis of the effect of Brexit on UK GDP undertaken by the [Centre for European Reform](
+Returns a `DataFrame` with quarterly GDP data on 30 OECD countries, borrowed from the analysis of
+the effect of Brexit on UK GDP undertaken by the [Centre for European Reform](
 https://www.cer.eu/insights/cost-brexit-june-2019)
 
-The `TreatmentPanel` for Brexit should be specified as
-```
-TreatmentPanel("United Kingdom" => Date(2016, 7, 1), df;
-                       outcome = :realgdp, id_var = :country, t_var = :quarter)
-```
+With the Brexit referendum having taken place on 23 June 2016, the `TreatmentPanel` for Brexit
+should be specified as
+    ```
+    TreatmentPanel(df, "United Kingdom" => Date(2016, 7, 1);
+        outcome = :realgdp, id_var = :country, t_var = :quarter)
+    ```
 """
 function load_brexit()
     CSV.read(joinpath(dirname(@__FILE__),"..","data","brexit.csv"), DataFrame; stringtype = String)
 end
 
+"""
+    load_brexit_panel()
+
+Returns a `BalancedPanel` object based on the data set available through load_brexit(), with
+treatment assignment specified as `"United Kingdom" => Date(2016, 7, 1)`
+
+"""
 function load_brexit_panel()
     df_brexit = load_brexit()
     BalancedPanel(df_brexit, "United Kingdom" => Date(2016, 7, 1); 
         id_var = :country, t_var = :quarter, outcome_var = :realgdp)
 end
 
+
+"""
+    load_germany()
+
+    Returns a `DataFrame` with annual observations of per-capita GDP and covariates between 1960 and
+    2003. The data is used in Abadie, Diamond and Hainmueller (2015) to study the effect of German
+    unification on per-capita GDP. 
+
+    As German reunification occurred in 1990, the `TreatmentPanel` for use in synthetic control
+    estimation should be specified as:
+
+    ```
+    TreatmentPanel(df, "West Germany" => 1990,
+        id_var = "country", t_var = "year", outcome_var = "gdp")
+    ```
+"""
 function load_germany()
     CSV.read(joinpath(dirname(@__FILE__),"..","data","germany.csv"), DataFrame; stringtype = String)
 end
 
+"""
+    load_germany_panel()
+
+    Returns a `BalancedPanel` object based on the data available from `load_germany()`, with
+    treatment assignment specified as `"West Germany" => 1990`
+"""
 function load_germany_panel()
     df_germany = load_germany()
     BalancedPanel(df_germany, "West Germany" => 1990; 
         id_var = "country", t_var = "year", outcome_var = "gdp")
 end
 
+"""
+    load_basque()
+
+    Returns a `DataFrame` with annual observations of per-capita GDP and covariates between 1955 and
+    1997. The data is used in Abadie and Gardeazabal (2003) to study the effect of ETA terrorism  
+          on per-capita GDP in the Basque country. 
+
+    To undertake analysis simliar to that in Abadie and Gardeazabal (2003), the `TreatmentPanel` for
+    use in synthetic control estimation should be specified as:
+
+    ```
+    TreatmentPanel(df, "Basque Country (Pais Vasco)" => 1970,
+        id_var = "regionname", t_var = "year", outcome_var = "gdpcap")
+    ```
+"""
 function load_basque()
     CSV.read(joinpath(dirname(@__FILE__),"..","data","basque.csv"), DataFrame; 
     stringtype = String, missingstring = "NA")
 end
 
+"""
+    load_basque_panel()
+
+    Returns a `BalancedPanel` object based on the data available from `load_basque()`, with
+    treatment assignment specified as `"Basque Country (Pais Vasco)" => 1970`
+"""
 function load_basque_panel()
     df_basque = load_basque()
     BalancedPanel(df_basque, "Basque Country (Pais Vasco)" => 1970;
         id_var = "regionname", t_var = "year", outcome_var = "gdpcap")
 end
 
+"""
+    load_smoking()
+
+    Returns a `DataFrame` with annual observations of cigarette sales and covariates for 39 US
+    states between 1970 and 2000. The data is used in Abadie, Diamond and Hainmueller (2010) to
+    study the effect of Proposition 99 on cigarette sales in California
+
+    To undertake analysis simliar to that in Abadie, Diamond and Hainmueller (2010), the
+    `TreatmentPanel` for use in synthetic control estimation should be specified as:
+
+    ```
+    TreatmentPanel(df, 3 => 1989,
+        id_var = "state", t_var = "year", outcome_var = "cigsale")
+    ```
+"""
 function load_smoking()
-    CSV.read(joinpath(dirname(@__FILE__),"..","data","smoking.csv"), DataFrame; 
+    df_smoking = CSV.read(joinpath(dirname(@__FILE__),"..","data","smoking.csv"), DataFrame; 
     stringtype = String, missingstring = "NA")
+    select!(df_smoking, Not(:Column1))
 end
+
+"""
+    load_smoking_panel()
+
+    Returns a `BalancedPanel` object based on the data available from `load_smoking()`, with
+    treatment assignment specified as `3 => 1989`
+"""
+function load_smoking_panel()
+    df_smoking = load_smoking()
+    BalancedPanel(df_smoking, 3 => 1989;
+        id_var = :state, t_var = :year, outcome_var = :cigsale)
+end

src/synthcontrolmodels.jl

@@ -86,7 +86,10 @@ function fit!(s::SynthControlModel; placebo_test = false)
 
     # Objective function
     # Hat tip to Mathieu Tanneau who suggested switching to a quadratic formulation
-    obj(w) = dot(y₁₀ .- vec(w'*yⱼ₀), y₁₀ .- vec(w'*yⱼ₀)) + 1e6*(1.0 - sum(w))^2
+    # !# Check why yⱼ₀ transpose is needed - is this consistent with mathematical formulation of
+    # model? Appears so, the notation for y₁₀ based on the outcome matrix and x₀ based on a vector 
+    # of covariates are incompatible 
+    obj(w) = (y₁₀ .- yⱼ₀'w)'*(y₁₀ .- yⱼ₀'w) + 1e6(1.0 -sum(w))^2
 
     # Initial condition and bounds
     initial = [1/J for _ in 1:J]

test/runtests.jl

@@ -27,3 +27,26 @@ using SynthControl, TreatmentPanels, DataFrames, Dates
   @test isfitted(s)
 
 end
+
+
+@testset "Data sets" begin
+
+  df = load_germany()
+  @test only(df[df.country .== "USA" .&& df.year .== 1961, :gdp]) == 2929
+
+  bp = load_germany_panel()
+  @test typeof(bp) <: BalancedPanel
+
+  df = load_basque()
+  @test only(df[df.regionno .== 1 .&& df.year .== 1957, :gdpcap]) == 2.60361314875437
+
+  bp = load_basque_panel()
+  @test typeof(bp) <: BalancedPanel
+
+  df = load_brexit()
+  @test only(df[df.country .== "Australia" .&& df.quarter .== Date(2009, 1, 1), :realgdp]) == 1.04
+
+  bp = load_brexit_panel()
+  @test typeof(bp) <: BalancedPanel
+
+end