Merge pull request #187 from ayushpatnaikgit/jnkremove

Remove jackknife and use bootstrap in ratio estimation.

src/Survey.jl

@@ -17,7 +17,6 @@ include("SurveyDesign.jl")
 include("bootstrap.jl")
 include("mean.jl")
 include("quantile.jl")
-include("jackknife.jl")
 include("total.jl")
 include("load_data.jl")
 include("hist.jl")

src/ratio.jl

@@ -6,30 +6,19 @@ Estimate the ratio of the columns specified in numerator and denominator
 ```jldoctest
 julia> apiclus1 = load_data("apiclus1");
 
-julia> clus_one_stage = SurveyDesign(apiclus1; clusters = :dnum, weights = :pw);
+julia> clus_one_stage = SurveyDesign(apiclus1; clusters = :dnum, weights = :pw) |> bootweights;
 
-julia> ratio(:api00, :enroll, clus_one_stage)
+ratio(:api00, :enroll, clus_one_stage)
 1×2 DataFrame
- Row │ Statistic  SE       
-     │ Float64    Float64  
-─────┼─────────────────────
-   1 │   1.17182  0.151242
+ Row │ ratio    SE       
+     │ Float64  Float64  
+─────┼───────────────────
+   1 │ 1.17182  0.130834
 ```
 """
-function ratio(variable_num:: Symbol, variable_den:: Symbol, design::SurveyDesign)
-    statistic = wsum(design.data[!,variable_num],design.data[!,design.weights])/wsum(design.data[!,variable_den],design.data[!,design.weights])
-    nh = length(unique(design.data[!,design.cluster]))
-    newv = []
-    gdf = groupby(design.data, design.cluster)
-    replicates = [filter(n -> n != i, 1:nh) for i in 1:nh] 
-    for i in replicates
-        df = DataFrame(gdf[i])
-        push!(newv, wsum(df[!,variable_num],df[!,design.weights])/wsum(df[!,variable_den],df[!,design.weights]))
-    end
-    c = 0
-    for i in 1:nh
-        c = c+(newv[i]-statistic)^2
-    end
-    var = c*(nh-1)/nh
-    return DataFrame(Statistic = statistic, SE = sqrt(var))
+function ratio(variable_num::Symbol, variable_den::Symbol, design::ReplicateDesign)
+    X = wsum(design.data[!, variable_num], design.data[!, design.weights]) / wsum(design.data[!, variable_den], design.data[!, design.weights])
+    Xt = [(wsum(design.data[!, variable_num], weights(design.data[! , "replicate_"*string(i)]))) / (wsum(design.data[!, variable_den], weights(design.data[! , "replicate_"*string(i)]))) for i in 1:design.replicates]
+    variance = sum((Xt .- X).^2) / design.replicates
+    DataFrame(ratio = X, SE = sqrt(variance))
 end

test/ratio.jl

@@ -4,7 +4,7 @@
     ##############################
     # one-stage cluster sample
     apiclus1 = copy(apiclus1_original)
-    dclus1 = SurveyDesign(apiclus1; clusters = :dnum, popsize = :fpc)
-    @test ratio(:api00, :enroll, dclus1).SE[1] ≈ 0.151242 atol = 1e-4
-    @test ratio(:api00, :enroll, dclus1).Statistic[1] ≈ 1.17182 atol = 1e-4
+    dclus1 = SurveyDesign(apiclus1; clusters = :dnum, popsize = :fpc) |> bootweights
+    @test ratio(:api00, :enroll, dclus1).SE[1] ≈ 0.1275446 atol = 1e-1
+    @test ratio(:api00, :enroll, dclus1).ratio[1] ≈ 1.17182 atol = 1e-4
 end