]> code.communitydata.science - ml_measurement_error_public.git/blobdiff - simulations/irr_simulation_base.R
Make summarize estimator group correctly for robustness checks.
[ml_measurement_error_public.git] / simulations / irr_simulation_base.R
index ee7112a233fcc9f72c185f991e320682891c62f1..f16c96b2594dfd7944a63c317dad7652e282b58f 100644 (file)
@@ -3,10 +3,10 @@ library(matrixStats) # for numerically stable logsumexps
 options(amelia.parallel="no",
         amelia.ncpus=1)
 library(Amelia)
+source("measerr_methods.R")
+source("pl_methods.R")
 
-source("measerr_methods.R") ## for my more generic function.
-
-run_simulation <- function(df, result, outcome_formula = y ~ x + z, proxy_formula = w_pred ~ x, truth_formula = x ~ z){
+run_simulation <- function(df, result, outcome_formula = y ~ x + z, proxy_formula = w_pred ~ x, coder_formulas=list(x.obs.1~x, x.obs.0~x), truth_formula = x ~ z){
 
     accuracy <- df[,mean(w_pred==x)]
     result <- append(result, list(accuracy=accuracy))
@@ -24,6 +24,8 @@ run_simulation <- function(df, result, outcome_formula = y ~ x + z, proxy_formul
 
 
 
+
+
     loa0.feasible <- lm(y ~ x.obs.0 + z, data = df[!(is.na(x.obs.1))])
 
     loa0.ci.Bxy <- confint(loa0.feasible)['x.obs.0',]
@@ -35,7 +37,7 @@ run_simulation <- function(df, result, outcome_formula = y ~ x + z, proxy_formul
                                   Bxy.ci.lower.loa0.feasible = loa0.ci.Bxy[1],
                                   Bzy.ci.upper.loa0.feasible = loa0.ci.Bzy[2],
                                   Bzy.ci.lower.loa0.feasible = loa0.ci.Bzy[1]))
-
+    print("fitting loa0 model")
 
     df.loa0.mle <- copy(df)
     df.loa0.mle[,x:=x.obs.0]
@@ -52,8 +54,11 @@ run_simulation <- function(df, result, outcome_formula = y ~ x + z, proxy_formul
                                   Bzy.ci.upper.loa0.mle = ci.upper['z'],
                                   Bzy.ci.lower.loa0.mle = ci.upper['z']))
 
+
+
     loco.feasible <- lm(y ~ x.obs.1 + z, data = df[(!is.na(x.obs.1)) & (x.obs.1 == x.obs.0)])
 
+
     loco.feasible.ci.Bxy <- confint(loco.feasible)['x.obs.1',]
     loco.feasible.ci.Bzy <- confint(loco.feasible)['z',]
 
@@ -65,41 +70,152 @@ run_simulation <- function(df, result, outcome_formula = y ~ x + z, proxy_formul
                                   Bzy.ci.lower.loco.feasible = loco.feasible.ci.Bzy[1]))
 
 
+    (model.naive <- lm(y~w_pred+z, data=df))
+    
+    naive.ci.Bxy <- confint(model.naive)['w_pred',]
+    naive.ci.Bzy <- confint(model.naive)['z',]
+
+    result <- append(result, list(Bxy.est.naive=coef(model.naive)['w_pred'],
+                                  Bzy.est.naive=coef(model.naive)['z'],
+                                  Bxy.ci.upper.naive = naive.ci.Bxy[2],
+                                  Bxy.ci.lower.naive = naive.ci.Bxy[1],
+                                  Bzy.ci.upper.naive = naive.ci.Bzy[2],
+                                  Bzy.ci.lower.naive = naive.ci.Bzy[1]))
+                                  
+    print("fitting loco model")
+
     df.loco.mle <- copy(df)
     df.loco.mle[,x.obs:=NA]
     df.loco.mle[(x.obs.0)==(x.obs.1),x.obs:=x.obs.0]
     df.loco.mle[,x.true:=x]
     df.loco.mle[,x:=x.obs]
     print(df.loco.mle[!is.na(x.obs.1),mean(x.true==x,na.rm=TRUE)])
+    loco.accuracy <- df.loco.mle[(!is.na(x.obs.1)) & (x.obs.1 == x.obs.0),mean(x.obs.1 == x.true)]    
     loco.mle <- measerr_mle(df.loco.mle, outcome_formula=outcome_formula, proxy_formula=proxy_formula, truth_formula=truth_formula)
     fisher.info <- solve(loco.mle$hessian)
     coef <- loco.mle$par
     ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96
     ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96
 
-    result <- append(result, list(Bxy.est.loco.mle=coef['x'],
+    result <- append(result, list(loco.accuracy=loco.accuracy,
+                                  Bxy.est.loco.mle=coef['x'],
                                   Bzy.est.loco.mle=coef['z'],
                                   Bxy.ci.upper.loco.mle = ci.upper['x'],
                                   Bxy.ci.lower.loco.mle = ci.lower['x'],
                                   Bzy.ci.upper.loco.mle = ci.upper['z'],
                                   Bzy.ci.lower.loco.mle = ci.lower['z']))
 
-    ## print(rater_formula)
-    ## print(proxy_formula)
-    ## mle.irr <- measerr_irr_mle( df, outcome_formula = outcome_formula, rater_formula = rater_formula, proxy_formula=proxy_formula, truth_formula=truth_formula)
+    df.double.proxy.mle <- copy(df)
+    df.double.proxy.mle[,x.obs:=NA]
+    print("fitting double proxy model")
+
+    double.proxy.mle <- measerr_irr_mle(df.double.proxy.mle, outcome_formula=outcome_formula, proxy_formula=proxy_formula, coder_formulas=coder_formulas[1], truth_formula=truth_formula)
+    fisher.info <- solve(double.proxy.mle$hessian)
+    coef <- double.proxy.mle$par
+    ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96
+    ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96
+
+    result <- append(result, list(
+                                  Bxy.est.double.proxy=coef['x'],
+                                  Bzy.est.double.proxy=coef['z'],
+                                  Bxy.ci.upper.double.proxy = ci.upper['x'],
+                                  Bxy.ci.lower.double.proxy = ci.lower['x'],
+                                  Bzy.ci.upper.double.proxy = ci.upper['z'],
+                                  Bzy.ci.lower.double.proxy = ci.lower['z']))
+
+    df.triple.proxy.mle <- copy(df)
+    df.triple.proxy.mle[,x.obs:=NA]
+
+    print("fitting triple proxy model")
+    triple.proxy.mle <- measerr_irr_mle(df.triple.proxy.mle, outcome_formula=outcome_formula, proxy_formula=proxy_formula, coder_formulas=coder_formulas, truth_formula=truth_formula)
+    fisher.info <- solve(triple.proxy.mle$hessian)
+    coef <- triple.proxy.mle$par
+    ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96
+    ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96
+
+    result <- append(result, list(
+                                  Bxy.est.triple.proxy=coef['x'],
+                                  Bzy.est.triple.proxy=coef['z'],
+                                  Bxy.ci.upper.triple.proxy = ci.upper['x'],
+                                  Bxy.ci.lower.triple.proxy = ci.lower['x'],
+                                  Bzy.ci.upper.triple.proxy = ci.upper['z'],
+                                  Bzy.ci.lower.triple.proxy = ci.lower['z']))
+    tryCatch({
+    amelia.out.k <- amelia(df.loco.mle, m=200, p2s=0, idvars=c('x.true','w','x.obs.1','x.obs.0','x'))
+    mod.amelia.k <- zelig(y~x.obs+z, model='ls', data=amelia.out.k$imputations, cite=FALSE)
+    (coefse <- combine_coef_se(mod.amelia.k, messages=FALSE))
+
+    est.x.mi <- coefse['x.obs','Estimate']
+    est.x.se <- coefse['x.obs','Std.Error']
+    result <- append(result,
+                     list(Bxy.est.amelia.full = est.x.mi,
+                          Bxy.ci.upper.amelia.full = est.x.mi + 1.96 * est.x.se,
+                          Bxy.ci.lower.amelia.full = est.x.mi - 1.96 * est.x.se
+                          ))
+
+    est.z.mi <- coefse['z','Estimate']
+    est.z.se <- coefse['z','Std.Error']
+
+    result <- append(result,
+                     list(Bzy.est.amelia.full = est.z.mi,
+                          Bzy.ci.upper.amelia.full = est.z.mi + 1.96 * est.z.se,
+                          Bzy.ci.lower.amelia.full = est.z.mi - 1.96 * est.z.se
+                          ))
+
+    },
+    error = function(e){
+        message("An error occurred:\n",e)
+        result$error <-paste0(result$error,'\n', e)
+    }
+    )
+
+    tryCatch({
+
+        mod.zhang.lik <- zhang.mle.iv(df.loco.mle)
+        coef <- coef(mod.zhang.lik)
+        ci <- confint(mod.zhang.lik,method='quad')
+        result <- append(result,
+                         list(Bxy.est.zhang = coef['Bxy'],
+                              Bxy.ci.upper.zhang = ci['Bxy','97.5 %'],
+                              Bxy.ci.lower.zhang = ci['Bxy','2.5 %'],
+                              Bzy.est.zhang = coef['Bzy'],
+                              Bzy.ci.upper.zhang = ci['Bzy','97.5 %'],
+                              Bzy.ci.lower.zhang = ci['Bzy','2.5 %']))
+    },
+
+    error = function(e){
+        message("An error occurred:\n",e)
+        result$error <- paste0(result$error,'\n', e)
+    })
+
+    df <- df.loco.mle
+    N <- nrow(df)
+    m <- nrow(df[!is.na(x.obs)])
+    p <- v <- train <- rep(0,N)
+    M <- m
+    p[(M+1):(N)] <- 1
+    v[1:(M)] <- 1
+    df <- df[order(x.obs)]
+    y <- df[,y]
+    x <- df[,x.obs]
+    z <- df[,z]
+    w <- df[,w_pred]
+    # gmm gets pretty close
+    (gmm.res <- predicted_covariates(y, x, z, w, v, train, p, max_iter=100, verbose=TRUE))
+
+    result <- append(result,
+                     list(Bxy.est.gmm = gmm.res$beta[1,1],
+                          Bxy.ci.upper.gmm = gmm.res$confint[1,2],
+                          Bxy.ci.lower.gmm = gmm.res$confint[1,1],
+                          gmm.ER_pval = gmm.res$ER_pval
+                          ))
+
+    result <- append(result,
+                     list(Bzy.est.gmm = gmm.res$beta[2,1],
+                          Bzy.ci.upper.gmm = gmm.res$confint[2,2],
+                          Bzy.ci.lower.gmm = gmm.res$confint[2,1]))
+
 
-    ## fisher.info <- solve(mle.irr$hessian)
-    ## coef <- mle.irr$par
-    ## ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96
-    ## ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96
-    
-    ## result <- append(result,
-    ##                  list(Bxy.est.mle = coef['x'],
-    ##                       Bxy.ci.upper.mle = ci.upper['x'],
-    ##                       Bxy.ci.lower.mle = ci.lower['x'],
-    ##                       Bzy.est.mle = coef['z'],
-    ##                       Bzy.ci.upper.mle = ci.upper['z'],
-    ##                       Bzy.ci.lower.mle = ci.lower['z']))
 
     return(result)
 

Community Data Science Collective || Want to submit a patch?