]> code.communitydata.science - ml_measurement_error_public.git/blobdiff - simulations/simulation_base.R
changes from klone
[ml_measurement_error_public.git] / simulations / simulation_base.R
index ee46ec6e6d303462ff71c9b62c132e82752a76fb..af0340835ce26e91d296a47adda6e94acf58b155 100644 (file)
@@ -7,6 +7,7 @@ library(Zelig)
 library(bbmle)
 library(matrixStats) # for numerically stable logsumexps
 
+source("pl_methods.R")
 source("measerr_methods.R") ## for my more generic function.
 
 ## This uses the pseudolikelihood approach from Carroll page 349.
@@ -36,124 +37,6 @@ my.pseudo.mle <- function(df){
 
 }
 
-
-## model from Zhang's arxiv paper, with predictions for y
-## Zhang got this model from Hausman 1998
-### I think this is actually eqivalent to the pseudo.mle method
-zhang.mle.iv <- function(df){
-    df.obs <- df[!is.na(x.obs)]
-    df.unobs <- df[is.na(x.obs)]
-
-    tn <- df.obs[(w_pred == 0) & (x.obs == w_pred),.N]
-    pn <- df.obs[(w_pred==0), .N]
-    npv <- tn / pn
-
-    tp <- df.obs[(w_pred==1) & (x.obs == w_pred),.N]
-    pp <- df.obs[(w_pred==1),.N]
-    ppv <- tp / pp
-
-    nll <- function(B0=0, Bxy=0, Bzy=0, sigma_y=0.1){
-
-    ## fpr = 1 - TNR
-    ### Problem: accounting for uncertainty in ppv / npv
-
-    ## fnr = 1 - TPR
-    ll.y.obs <- with(df.obs, dnorm(y, B0 + Bxy * x + Bzy * z, sd=sigma_y,log=T))
-    ll <- sum(ll.y.obs)
-    
-    # unobserved case; integrate out x
-    ll.x.1 <- with(df.unobs, dnorm(y, B0 + Bxy + Bzy * z, sd = sigma_y, log=T))
-    ll.x.0 <- with(df.unobs, dnorm(y, B0 + Bzy * z, sd = sigma_y,log=T))
-
-    ## case x == 1
-    lls.x.1 <- colLogSumExps(rbind(log(ppv) + ll.x.1, log(1-ppv) + ll.x.0))
-    
-    ## case x == 0
-    lls.x.0 <- colLogSumExps(rbind(log(1-npv) + ll.x.1, log(npv) + ll.x.0))
-
-    lls <- colLogSumExps(rbind(df.unobs$w_pred * lls.x.1, (1-df.unobs$w_pred) * lls.x.0))
-    ll <- ll + sum(lls)
-    return(-ll)
-    }    
-    mlefit <- mle2(minuslogl = nll, control=list(maxit=1e6), lower=list(sigma_y=0.0001, B0=-Inf, Bxy=-Inf, Bzy=-Inf),
-                   upper=list(sigma_y=Inf, B0=Inf, Bxy=Inf, Bzy=Inf),method='L-BFGS-B')
-    return(mlefit)
-}
-
-## this is equivalent to the pseudo-liklihood model from Caroll
-## zhang.mle.dv <- function(df){
-
-##     nll <- function(B0=0, Bxy=0, Bzy=0, ppv=0.9, npv=0.9){
-##     df.obs <- df[!is.na(y.obs)]
-
-##     ## fpr = 1 - TNR
-##     ll.w0y0 <- with(df.obs[y.obs==0],dbinom(1-w_pred,1,npv,log=TRUE))
-##     ll.w1y1 <- with(df.obs[y.obs==1],dbinom(w_pred,1,ppv,log=TRUE))
-
-##     # observed case
-##     ll.y.obs <- vector(mode='numeric', length=nrow(df.obs))
-##     ll.y.obs[df.obs$y.obs==1] <- with(df.obs[y.obs==1], plogis(B0 + Bxy * x + Bzy * z,log=T))
-##     ll.y.obs[df.obs$y.obs==0] <- with(df.obs[y.obs==0], plogis(B0 + Bxy * x + Bzy * z,log=T,lower.tail=FALSE))
-
-##     ll <- sum(ll.y.obs) + sum(ll.w0y0) + sum(ll.w1y1)
-
-##     # unobserved case; integrate out y
-##     ## case y = 1
-##     ll.y.1 <- vector(mode='numeric', length=nrow(df))
-##     pi.y.1 <- with(df,plogis(B0 + Bxy * x + Bzy*z, log=T))
-##     ## P(w=1| y=1)P(y=1) + P(w=0|y=1)P(y=1) = P(w=1,y=1) + P(w=0,y=1)
-##     lls.y.1 <- colLogSumExps(rbind(log(ppv) + pi.y.1, log(1-ppv) + pi.y.1))
-    
-##     ## case y = 0
-##     ll.y.0 <- vector(mode='numeric', length=nrow(df))
-##     pi.y.0 <- with(df,plogis(B0 + Bxy * x + Bzy*z, log=T,lower.tail=FALSE))
-
-##     ## P(w=1 | y=0)P(y=0) + P(w=0|y=0)P(y=0) = P(w=1,y=0) + P(w=0,y=0)
-##     lls.y.0 <- colLogSumExps(rbind(log(npv) + pi.y.0, log(1-npv) + pi.y.0))
-
-##     lls <- colLogSumExps(rbind(lls.y.1, lls.y.0))
-##     ll <- ll + sum(lls)
-##     return(-ll)
-##     }    
-##     mlefit <- mle2(minuslogl = nll, control=list(maxit=1e6),method='L-BFGS-B',lower=list(B0=-Inf, Bxy=-Inf, Bzy=-Inf, ppv=0.001,npv=0.001),
-##                    upper=list(B0=Inf, Bxy=Inf, Bzy=Inf,ppv=0.999,npv=0.999))
-##     return(mlefit)
-## }
-
-zhang.mle.dv <- function(df){
-    df.obs <- df[!is.na(y.obs)]
-    df.unobs <- df[is.na(y.obs)]
-
-    fp <- df.obs[(w_pred==1) & (y.obs != w_pred),.N]
-    p <- df.obs[(w_pred==1),.N]
-    fpr <- fp / p
-    fn <- df.obs[(w_pred==0) & (y.obs != w_pred), .N]
-    n <- df.obs[(w_pred==0),.N]
-    fnr <- fn / n
-
-    nll <- function(B0=0, Bxy=0, Bzy=0){
-
-
-        ## observed case
-        ll.y.obs <- vector(mode='numeric', length=nrow(df.obs))
-        ll.y.obs[df.obs$y.obs==1] <- with(df.obs[y.obs==1], plogis(B0 + Bxy * x + Bzy * z,log=T))
-        ll.y.obs[df.obs$y.obs==0] <- with(df.obs[y.obs==0], plogis(B0 + Bxy * x + Bzy * z,log=T,lower.tail=FALSE))
-
-        ll <- sum(ll.y.obs)
-
-        pi.y.1 <- with(df,plogis(B0 + Bxy * x + Bzy*z, log=T))
-        pi.y.0 <- with(df,plogis(B0 + Bxy * x + Bzy*z, log=T,lower.tail=FALSE))
-
-        lls <- with(df.unobs, colLogSumExps(rbind(w_pred * colLogSumExps(rbind(log(fpr), log(1 - fnr - fpr)+pi.y.1)),
-        (1-w_pred) * colLogSumExps(rbind(log(1-fpr), log(1 - fnr - fpr)+pi.y.0)))))
-    
-        ll <- ll + sum(lls)
-        return(-ll)
-    }    
-    mlefit <- mle2(minuslogl = nll, control=list(maxit=1e6),method='L-BFGS-B',lower=c(B0=-Inf, Bxy=-Inf, Bzy=-Inf),
-                   upper=c(B0=Inf, Bxy=Inf, Bzy=Inf))
-    return(mlefit)
-}
  
 ## This uses the likelihood approach from Carroll page 353.
 ## assumes that we have a good measurement error model
@@ -206,15 +89,27 @@ my.mle <- function(df){
     return(mlefit)
 }
 
-run_simulation_depvar <- function(df, result, outcome_formula=y~x+z, proxy_formula=w_pred~y){
+run_simulation_depvar <- function(df, result, outcome_formula=y~x+z, proxy_formula=w_pred~y, confint_method='quad'){
 
-    accuracy <- df[,mean(w_pred==y)]
+    (accuracy <- df[,mean(w_pred==y)])
     result <- append(result, list(accuracy=accuracy))
-
+    (error.cor.z <- cor(df$z, df$y - df$w_pred))
+    (error.cor.x <- cor(df$x, df$y - df$w_pred))
+    (error.cor.y <- cor(df$y, df$y - df$w_pred))
+    result <- append(result, list(error.cor.x = error.cor.x,
+                                  error.cor.z = error.cor.z,
+                                  error.cor.y = error.cor.y))
+
+    model.null <- glm(y~1, data=df,family=binomial(link='logit'))
     (model.true <- glm(y ~ x + z, data=df,family=binomial(link='logit')))
+    (lik.ratio <- exp(logLik(model.true) - logLik(model.null)))
+
     true.ci.Bxy <- confint(model.true)['x',]
     true.ci.Bzy <- confint(model.true)['z',]
 
+    result <- append(result, list(cor.xz=cor(df$x,df$z)))
+    result <- append(result, list(lik.ratio=lik.ratio))
+
     result <- append(result, list(Bxy.est.true=coef(model.true)['x'],
                                   Bzy.est.true=coef(model.true)['z'],
                                   Bxy.ci.upper.true = true.ci.Bxy[2],
@@ -255,11 +150,13 @@ run_simulation_depvar <- function(df, result, outcome_formula=y~x+z, proxy_formu
 
     temp.df <- copy(df)
     temp.df[,y:=y.obs]
+
     mod.caroll.lik <- measerr_mle_dv(temp.df, outcome_formula=outcome_formula, proxy_formula=proxy_formula)
-    fisher.info <- solve(mod.caroll.lik$hessian)
+    fischer.info <- solve(mod.caroll.lik$hessian)
     coef <- mod.caroll.lik$par
-    ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96
-    ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96
+    ci.upper <- coef + sqrt(diag(fischer.info)) * 1.96
+    ci.lower <- coef - sqrt(diag(fischer.info)) * 1.96
+
     result <- append(result,
                      list(Bxy.est.mle = coef['x'],
                           Bxy.ci.upper.mle = ci.upper['x'],
@@ -268,6 +165,19 @@ run_simulation_depvar <- function(df, result, outcome_formula=y~x+z, proxy_formu
                           Bzy.ci.upper.mle = ci.upper['z'],
                           Bzy.ci.lower.mle = ci.lower['z']))
 
+    mod.caroll.profile.lik <- measerr_mle_dv(temp.df, outcome_formula=outcome_formula, proxy_formula=proxy_formula, method='bbmle')
+    coef <- coef(mod.caroll.profile.lik)
+    ci <- confint(mod.caroll.profile.lik, method='spline')
+    ci.lower <- ci[,'2.5 %']
+    ci.upper <- ci[,'97.5 %']
+
+    result <- append(result,
+                     list(Bxy.est.mle.profile = coef['x'],
+                          Bxy.ci.upper.mle.profile = ci.upper['x'],
+                          Bxy.ci.lower.mle.profile = ci.lower['x'],
+                          Bzy.est.mle.profile = coef['z'],
+                          Bzy.ci.upper.mle.profile = ci.upper['z'],
+                          Bzy.ci.lower.mle.profile = ci.lower['z']))
 
     ## my implementatoin of liklihood based correction
     mod.zhang <- zhang.mle.dv(df)
@@ -285,33 +195,35 @@ run_simulation_depvar <- function(df, result, outcome_formula=y~x+z, proxy_formu
     
 
     # amelia says use normal distribution for binary variables.
+    amelia_result <- list(Bxy.est.amelia.full = NA,
+                          Bxy.ci.upper.amelia.full = NA,
+                          Bxy.ci.lower.amelia.full = NA,
+                          Bzy.est.amelia.full = NA,
+                          Bzy.ci.upper.amelia.full = NA,
+                          Bzy.ci.lower.amelia.full = NA
+                          )
+
     tryCatch({
-        amelia.out.k <- amelia(df, m=200, p2s=0, idvars=c('y','ystar','w'))
-        mod.amelia.k <- zelig(y.obs~x+z, model='ls', data=amelia.out.k$imputations, cite=FALSE)
+        amelia.out.k <- amelia(df, m=200, p2s=0, idvars=c('y','ystar','w'),ords="y.obs")
+        mod.amelia.k <- zelig(y.obs~x+z, model='logit', data=amelia.out.k$imputations, cite=FALSE)
         (coefse <- combine_coef_se(mod.amelia.k, messages=FALSE))
         est.x.mi <- coefse['x','Estimate']
         est.x.se <- coefse['x','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
-                              ))
-
+        amelia_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,
+                          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)
-    })
-
+    result[['error']] <- e}
+    )
+    result <- append(result,amelia_result)
 
     return(result)
 
@@ -319,11 +231,36 @@ run_simulation_depvar <- function(df, result, outcome_formula=y~x+z, proxy_formu
 
 
 ## outcome_formula, proxy_formula, and truth_formula are passed to measerr_mle 
-run_simulation <-  function(df, result, outcome_formula=y~x+z, proxy_formula=NULL, truth_formula=NULL){
+run_simulation <-  function(df, result, outcome_formula=y~x+z, proxy_formula=NULL, truth_formula=NULL, confint_method='quad'){
 
     accuracy <- df[,mean(w_pred==x)]
-    result <- append(result, list(accuracy=accuracy))
-
+    accuracy.y0 <- df[y<=0,mean(w_pred==x)]
+    accuracy.y1 <- df[y>=0,mean(w_pred==x)]
+    cor.y.xi <- cor(df$x - df$w_pred, df$y)
+
+    fnr <- df[w_pred==0,mean(w_pred!=x)]
+    fnr.y0 <- df[(w_pred==0) & (y<=0),mean(w_pred!=x)]
+    fnr.y1 <- df[(w_pred==0) & (y>=0),mean(w_pred!=x)]
+
+    fpr <- df[w_pred==1,mean(w_pred!=x)]
+    fpr.y0 <- df[(w_pred==1) & (y<=0),mean(w_pred!=x)]
+    fpr.y1 <- df[(w_pred==1) & (y>=0),mean(w_pred!=x)]
+    cor.resid.w_pred <- cor(resid(lm(y~x+z,df)),df$w_pred)
+
+    result <- append(result, list(accuracy=accuracy,
+                                  accuracy.y0=accuracy.y0,
+                                  accuracy.y1=accuracy.y1,
+                                  cor.y.xi=cor.y.xi,
+                                  fnr=fnr,
+                                  fnr.y0=fnr.y0,
+                                  fnr.y1=fnr.y1,
+                                  fpr=fpr,
+                                  fpr.y0=fpr.y0,
+                                  fpr.y1=fpr.y1,
+                                  cor.resid.w_pred=cor.resid.w_pred
+                                  ))
+
+    result <- append(result, list(cor.xz=cor(df$x,df$z)))
     (model.true <- lm(y ~ x + z, data=df))
     true.ci.Bxy <- confint(model.true)['x',]
     true.ci.Bzy <- confint(model.true)['z',]
@@ -358,79 +295,120 @@ run_simulation <-  function(df, result, outcome_formula=y~x+z, proxy_formula=NUL
                                   Bxy.ci.lower.naive = naive.ci.Bxy[1],
                                   Bzy.ci.upper.naive = naive.ci.Bzy[2],
                                   Bzy.ci.lower.naive = naive.ci.Bzy[1]))
-                                  
 
-    tryCatch({
-    amelia.out.k <- amelia(df, m=200, p2s=0, idvars=c('x','w'))
-    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))
+    amelia_result <- list(
+        Bxy.est.amelia.full = NULL,
+        Bxy.ci.upper.amelia.full = NULL,
+        Bxy.ci.lower.amelia.full = NULL,
+        Bzy.est.amelia.full = NULL,
+        Bzy.ci.upper.amelia.full = NULL,
+        Bzy.ci.lower.amelia.full = NULL
+        )
 
-    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
-                          ))
+    tryCatch({
+        amelia.out.k <- amelia(df, m=200, p2s=0, idvars=c('x','w'))
+        mod.amelia.k <- zelig(y~x.obs+z, model='ls', data=amelia.out.k$imputations, cite=FALSE)
+        (coefse <- combine_coef_se(mod.amelia.k))
 
-    est.z.mi <- coefse['z','Estimate']
-    est.z.se <- coefse['z','Std.Error']
+        est.x.mi <- coefse['x.obs','Estimate']
+        est.x.se <- coefse['x.obs','Std.Error']
+        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
-                          ))
+        amelia_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,
+                              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)
-    }
+        result[['error']] <- e}
     )
 
+
+    result <- append(result, amelia_result)
+
+
+   mle_result <- list(Bxy.est.mle = NULL,
+                      Bxy.ci.upper.mle = NULL,
+                      Bxy.ci.lower.mle = NULL,
+                      Bzy.est.mle = NULL,
+                      Bzy.ci.upper.mle = NULL,
+                      Bzy.ci.lower.mle = NULL)
+
     tryCatch({
         temp.df <- copy(df)
         temp.df <- temp.df[,x:=x.obs]
-        mod.caroll.lik <- measerr_mle(temp.df, outcome_formula=outcome_formula, proxy_formula=proxy_formula, truth_formula=truth_formula)
-        fisher.info <- solve(mod.caroll.lik$hessian)
+        mod.caroll.lik <- measerr_mle(temp.df, outcome_formula=outcome_formula, proxy_formula=proxy_formula, truth_formula=truth_formula, method='optim')
+        fischer.info <- solve(mod.caroll.lik$hessian)
         coef <- mod.caroll.lik$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']))
+        ci.upper <- coef + sqrt(diag(fischer.info)) * 1.96
+        ci.lower <- coef - sqrt(diag(fischer.info)) * 1.96
+
+        mle_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'])
+        },
+        error=function(e) {result[['error']] <- as.character(e)
+        })
+
+    result <- append(result, mle_result)
+    mle_result_proflik <- list(Bxy.est.mle.profile = NULL,
+                               Bxy.ci.upper.mle.profile = NULL,
+                               Bxy.ci.lower.mle.profile = NULL,
+                               Bzy.est.mle.profile = NULL,
+                               Bzy.ci.upper.mle.profile = NULL,
+                               Bzy.ci.lower.mle.profile = NULL)
+
+    tryCatch({
+        ## confint_method == 'bbmle'
+        mod.caroll.lik <- measerr_mle(temp.df, outcome_formula=outcome_formula, proxy_formula=proxy_formula, truth_formula=truth_formula, method='bbmle')
+        coef <- coef(mod.caroll.lik)
+        ci <- confint(mod.caroll.lik, method='spline')
+        ci.lower <- ci[,'2.5 %']
+        ci.upper <- ci[,'97.5 %']
+
+        mle_result_proflik <- list(Bxy.est.mle.profile = coef['x'],
+                                   Bxy.ci.upper.mle.profile = ci.upper['x'],
+                                   Bxy.ci.lower.mle.profile = ci.lower['x'],
+                                   Bzy.est.mle.profile = coef['z'],
+                                   Bzy.ci.upper.mle.profile = ci.upper['z'],
+                                   Bzy.ci.lower.mle.profile = ci.lower['z'])
     },
 
-    error = function(e){
-        message("An error occurred:\n",e)
-        result$error <- paste0(result$error,'\n', e)
+    error=function(e) {result[['error']] <- as.character(e)
     })
+        
+    result <- append(result, mle_result_proflik)
 
-    tryCatch({
+    zhang_result <- list(Bxy.est.mle.zhang = NULL,
+                       Bxy.ci.upper.mle.zhang = NULL,
+                       Bxy.ci.lower.mle.zhang = NULL,
+                       Bzy.est.mle.zhang = NULL,
+                       Bzy.ci.upper.mle.zhang = NULL,
+                       Bzy.ci.lower.mle.zhang = NULL)
 
-        mod.zhang.lik <- zhang.mle.iv(df)
-        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 %']))
+    tryCatch({
+    mod.zhang.lik <- zhang.mle.iv(df)
+    coef <- coef(mod.zhang.lik)
+    ci <- confint(mod.zhang.lik,method='quad')
+    zhang_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)
+    error=function(e) {result[['error']] <- as.character(e)
     })
+    result <- append(result, zhang_result)
 
     ## What if we can't observe k -- most realistic scenario. We can't include all the ML features in a model.
     ## amelia.out.nok <- amelia(df, m=200, p2s=0, idvars=c("x","w_pred"), noms=noms)
@@ -481,29 +459,29 @@ run_simulation <-  function(df, result, outcome_formula=y~x+z, proxy_formula=NUL
                           Bzy.ci.lower.gmm = gmm.res$confint[2,1]))
 
 
-    tryCatch({
-    mod.calibrated.mle <- mecor(y ~ MeasError(w_pred, reference = x.obs) + z, df, B=400, method='efficient')
-    (mod.calibrated.mle)
-    (mecor.ci <- summary(mod.calibrated.mle)$c$ci['x.obs',])
-    result <- append(result, list(
-                                 Bxy.est.mecor = mecor.ci['Estimate'],
-                                 Bxy.ci.upper.mecor = mecor.ci['UCI'],
-                                 Bxy.ci.lower.mecor = mecor.ci['LCI'])
-                     )
-
-    (mecor.ci <- summary(mod.calibrated.mle)$c$ci['z',])
-
-    result <- append(result, list(
-                                 Bzy.est.mecor = mecor.ci['Estimate'],
-                                 Bzy.ci.upper.mecor = mecor.ci['UCI'],
-                                 Bzy.ci.lower.mecor = mecor.ci['LCI'])
-                     )
-    },
-    error = function(e){
-        message("An error occurred:\n",e)
-        result$error <- paste0(result$error, '\n', e)
-    }
-    )
+    ## tryCatch({
+    ## mod.calibrated.mle <- mecor(y ~ MeasError(w_pred, reference = x.obs) + z, df, B=400, method='efficient')
+    ## (mod.calibrated.mle)
+    ## (mecor.ci <- summary(mod.calibrated.mle)$c$ci['x.obs',])
+    ## result <- append(result, list(
+    ##                              Bxy.est.mecor = mecor.ci['Estimate'],
+    ##                              Bxy.ci.upper.mecor = mecor.ci['UCI'],
+    ##                              Bxy.ci.lower.mecor = mecor.ci['LCI'])
+    ##                  )
+
+    ## (mecor.ci <- summary(mod.calibrated.mle)$c$ci['z',])
+
+    ## result <- append(result, list(
+    ##                              Bzy.est.mecor = mecor.ci['Estimate'],
+    ##                              Bzy.ci.upper.mecor = mecor.ci['UCI'],
+    ##                              Bzy.ci.lower.mecor = mecor.ci['LCI'])
+    ##                  )
+    ## },
+    ## error = function(e){
+    ##     message("An error occurred:\n",e)
+    ##     result$error <- paste0(result$error, '\n', e)
+    ## }
+    ## )
 ##    clean up memory
 ##    rm(list=c("df","y","x","g","w","v","train","p","amelia.out.k","amelia.out.nok", "mod.calibrated.mle","gmm.res","mod.amelia.k","mod.amelia.nok", "model.true","model.naive","model.feasible"))
     

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