R/kling_as_function.R
In skirmer/skpack: Personal package of functions I use regularly.
Defines functions
itt_for_kling
itt_for_kling_p3
itt_for_kling_p4
itt_for_kling_p6
Documented in
itt_for_kling
itt_for_kling_p3
itt_for_kling_p4
itt_for_kling_p6
#' Basic Kling Model Function
#'
#' This function computes the kling regressions for dividing the effect size of two different years of treatment. Your block variable must be factor,
#' and your dataset must omit records with a missing outcome or a missing treatment assignment.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' School Level MTO
#' itt_for_kling(outcome = "mathxil_z_post2", participation1 = "treat_post1_single", participation2 = "treat_post2",
#' grouping = "schlid", sizefile = schoolsizes, weight = "mtoweight_calculation", number_to_crop = 12,
#' sourcefile = master_dataset_1_math15_no2531, weight_final = "stderr_scores_weight", "omits")
#'
#' # Block level MTO
#' itt_for_kling(outcome = "mathxil_z_post2", participation1 = "treat_post1_single", participation2 = "treat_post2",
#' grouping = "blocknum", sizefile = blocksizes, weight = "mtoweight_calculation", number_to_crop = 42,
#' sourcefile = master_dataset_1_math15_no2531, weight_final = "stderr_scores_weight", "omits")
itt_for_kling <- function(outcome, participation1, participation2, grouping, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
library(dplyr)
library(data.table)
library(reshape2)
library(sem)
library(doBy)
library(ggplot2)
require(sem)
require(AER)
require(ivpack)
require(lubridate)
library(mondate)
require("ggrepel")
#Assemble formulas for the various models
itt_step1_form <- formula(paste(outcome, "~", grouping, "+", grouping,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping, "+", grouping,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping, "+", grouping,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <<- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
a <<- nobs(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
b <<- nobs(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
ccc <<- nobs(partic2_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, stderr_scores, stderr_partic, stderr_partic2)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
d <<- nobs(ittmodel)
#Create the output with dynamic naming
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", grouping, "+", participation1, "+", participation2, "|", grouping, "+", grouping, ":dmatch"))
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
#ALTERNATIVE APPROACH - one formula, but the regressor is block, and the grouping variable is school.
#Build formula, uses inputs from the function arguments again
klingform2 <<- formula(paste("J~", "blocknum", "+", participation1, "+", participation2, "|", "blocknum", "+", "schlid", ":dmatch"))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small2 <- function(I) {
J <<- as.matrix(I)
require(sem)
model3 <<- ivreg(klingform2,
data=working_data_file,
weight=weight2)
totmod <<- summary(model3)
totmod
robustmod <<- robust.se(model3)
robustmod
}
print(tsls_fn_small2(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_mixed_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_mixed_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
#' Kling Model with Three Participation Variables
#'
#' This function is generally the same as itt_for_kling but it allows for three years and three participation vars.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' School Level MTO
#' itt_for_kling_p3(outcome = "mathxil_z_post2", participation1 = "treat_post1_single", participation2 = "treat_post2",
#' participation3 = "treat_post3", grouping = "schlid", sizefile = schoolsizes, weight = "mtoweight_calculation", number_to_crop = 12,
#' sourcefile = master_dataset_1_math15_no2531, weight_final = "stderr_scores_weight", "omits")
itt_for_kling_p3 <- function(outcome, participation1, participation2, participation3, grouping, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
#Assemble formulas for the various models
itt_step1_form <- formula(paste(outcome, "~", grouping, "+", grouping,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping, "+", grouping,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping, "+", grouping,":dmatch"))
itt_step4optional_form <- formula(paste(participation3, "~", grouping, "+", grouping,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2+coeflist_partic3-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <<- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <<- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
# Fourth ITT - Third Treatment Participation #
partic3_mod <- lm(itt_step4optional_form,weights=weight2, data=working_data_file)
coeflist_partic3 <- coef(partic3_mod)
stderr_partic3 <- coef(summary(partic3_mod))[,2]
resid_partic3 <- resid(partic3_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, coeflist_partic3, stderr_scores, stderr_partic, stderr_partic2, stderr_partic3)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2, resid_partic3)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic[,stderr_partic3_weight:=(1/(working_scorepartic$stderr_partic3^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
#Create the output with dynamic naming
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", grouping, "+", participation1, "+", participation2, "+", participation3, "|", grouping, "+", grouping, ":dmatch"))
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
#' Kling Model with Two Grouping Variables
#'
#' This function is generally the same as itt_for_kling but it allows for mixing block and school (or whatever) grouping variables.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' School Level
#' itt_for_kling_p4(outcome = "mathxil_z_post2", participation1 = "treat_post1", participation2 = "treat_post2",
#' grouping1 = "schlid", grouping2 = "blocknum", sizefile = schoolsizes, weight = "precision", number_to_crop = 42,
#' sourcefile = master_dataset_1_math15_allblocks, weight_final = "stderr_scores_weight", "noomits_nome")
itt_for_kling_p4 <- function(outcome, participation1, participation2, grouping1, grouping2, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
#Assemble formulas for the various models #
itt_step1_form <- formula(paste(outcome, "~", grouping2, "+", grouping1,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping2, "+", grouping1,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping2, "+", grouping1,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <<- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <<- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping1=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping1=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, stderr_scores, stderr_partic, stderr_partic2)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
#Create the output with dynamic naming
assign(paste0("ittmodel_", "mixed2","_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", "mixed2","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", participation1, "+", participation2, "+", grouping2,
"|", grouping1, ":dmatch", "+", grouping2))#, "+", grouping2, ":dmatch"))
#grouping1, ":dmatch", "+",
#grouping1, ":dmatch", "+" ,
print(klingform)
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", "mixed2","_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", "mixed2","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
#' Kling Model with Two Grouping Variables, Instrumenting Var 2
#'
#' This function is generally the same as itt_for_kling_p4 but it puts grouping variable 2 in the ITT models and instruments for it in the ivreg.
#' This variation seems to have produced good results for the Match/BAM project.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' Block Level
#' itt_for_kling_p6(outcome = "mathxil_z_post2", participation1 = "treat_post1", participation2 = "treat_post2",
#' grouping1 = "schlid", grouping2 = "blocknum", sizefile = schoolsizes, weight = "precision", number_to_crop = 42,
#' sourcefile = master_dataset_1_math15_allblocks, weight_final = "stderr_scores_weight", "noomits_nome")
itt_for_kling_p6 <- function(outcome, participation1, participation2, grouping1, grouping2, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
#Assemble formulas for the various models #
itt_step1_form <- formula(paste(outcome, "~", grouping2,":dmatch", "+", grouping2))#,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping2,":dmatch", "+", grouping2))#,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping2,":dmatch", "+", grouping2))#,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <<- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <<- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping1=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping1=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
print(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
print(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
print(partic2_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, stderr_scores, stderr_partic, stderr_partic2)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
#Create the output with dynamic naming
assign(paste0("ittmodel_", "mixed4","_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", "mixed4","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", participation1, "+", participation2, "+", grouping2,
"|", grouping2, "+", grouping2,":dmatch"))
#grouping1, ":dmatch", "+",
#grouping1, ":dmatch", "+" ,
print(klingform)
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", "mixed4","_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", "mixed4","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
skirmer/skpack documentation built on May 30, 2019, 1:06 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions itt_for_kling itt_for_kling_p3 itt_for_kling_p4 itt_for_kling_p6
Documented in itt_for_kling itt_for_kling_p3 itt_for_kling_p4 itt_for_kling_p6
#' Basic Kling Model Function
#'
#' This function computes the kling regressions for dividing the effect size of two different years of treatment. Your block variable must be factor,
#' and your dataset must omit records with a missing outcome or a missing treatment assignment.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' School Level MTO
#' itt_for_kling(outcome = "mathxil_z_post2", participation1 = "treat_post1_single", participation2 = "treat_post2",
#' grouping = "schlid", sizefile = schoolsizes, weight = "mtoweight_calculation", number_to_crop = 12,
#' sourcefile = master_dataset_1_math15_no2531, weight_final = "stderr_scores_weight", "omits")
#'
#' # Block level MTO
#' itt_for_kling(outcome = "mathxil_z_post2", participation1 = "treat_post1_single", participation2 = "treat_post2",
#' grouping = "blocknum", sizefile = blocksizes, weight = "mtoweight_calculation", number_to_crop = 42,
#' sourcefile = master_dataset_1_math15_no2531, weight_final = "stderr_scores_weight", "omits")
itt_for_kling <- function(outcome, participation1, participation2, grouping, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
library(dplyr)
library(data.table)
library(reshape2)
library(sem)
library(doBy)
library(ggplot2)
require(sem)
require(AER)
require(ivpack)
require(lubridate)
library(mondate)
require("ggrepel")
#Assemble formulas for the various models
itt_step1_form <- formula(paste(outcome, "~", grouping, "+", grouping,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping, "+", grouping,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping, "+", grouping,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <<- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
a <<- nobs(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
b <<- nobs(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
ccc <<- nobs(partic2_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, stderr_scores, stderr_partic, stderr_partic2)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
d <<- nobs(ittmodel)
#Create the output with dynamic naming
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", grouping, "+", participation1, "+", participation2, "|", grouping, "+", grouping, ":dmatch"))
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
#ALTERNATIVE APPROACH - one formula, but the regressor is block, and the grouping variable is school.
#Build formula, uses inputs from the function arguments again
klingform2 <<- formula(paste("J~", "blocknum", "+", participation1, "+", participation2, "|", "blocknum", "+", "schlid", ":dmatch"))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small2 <- function(I) {
J <<- as.matrix(I)
require(sem)
model3 <<- ivreg(klingform2,
data=working_data_file,
weight=weight2)
totmod <<- summary(model3)
totmod
robustmod <<- robust.se(model3)
robustmod
}
print(tsls_fn_small2(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_mixed_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_mixed_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
#' Kling Model with Three Participation Variables
#'
#' This function is generally the same as itt_for_kling but it allows for three years and three participation vars.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' School Level MTO
#' itt_for_kling_p3(outcome = "mathxil_z_post2", participation1 = "treat_post1_single", participation2 = "treat_post2",
#' participation3 = "treat_post3", grouping = "schlid", sizefile = schoolsizes, weight = "mtoweight_calculation", number_to_crop = 12,
#' sourcefile = master_dataset_1_math15_no2531, weight_final = "stderr_scores_weight", "omits")
itt_for_kling_p3 <- function(outcome, participation1, participation2, participation3, grouping, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
#Assemble formulas for the various models
itt_step1_form <- formula(paste(outcome, "~", grouping, "+", grouping,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping, "+", grouping,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping, "+", grouping,":dmatch"))
itt_step4optional_form <- formula(paste(participation3, "~", grouping, "+", grouping,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2+coeflist_partic3-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <<- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <<- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
# Fourth ITT - Third Treatment Participation #
partic3_mod <- lm(itt_step4optional_form,weights=weight2, data=working_data_file)
coeflist_partic3 <- coef(partic3_mod)
stderr_partic3 <- coef(summary(partic3_mod))[,2]
resid_partic3 <- resid(partic3_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, coeflist_partic3, stderr_scores, stderr_partic, stderr_partic2, stderr_partic3)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2, resid_partic3)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic[,stderr_partic3_weight:=(1/(working_scorepartic$stderr_partic3^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
#Create the output with dynamic naming
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", grouping, "+", participation1, "+", participation2, "+", participation3, "|", grouping, "+", grouping, ":dmatch"))
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", grouping,"_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
#' Kling Model with Two Grouping Variables
#'
#' This function is generally the same as itt_for_kling but it allows for mixing block and school (or whatever) grouping variables.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' School Level
#' itt_for_kling_p4(outcome = "mathxil_z_post2", participation1 = "treat_post1", participation2 = "treat_post2",
#' grouping1 = "schlid", grouping2 = "blocknum", sizefile = schoolsizes, weight = "precision", number_to_crop = 42,
#' sourcefile = master_dataset_1_math15_allblocks, weight_final = "stderr_scores_weight", "noomits_nome")
itt_for_kling_p4 <- function(outcome, participation1, participation2, grouping1, grouping2, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
#Assemble formulas for the various models #
itt_step1_form <- formula(paste(outcome, "~", grouping2, "+", grouping1,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping2, "+", grouping1,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping2, "+", grouping1,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <<- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <<- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping1=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping1=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, stderr_scores, stderr_partic, stderr_partic2)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
#Create the output with dynamic naming
assign(paste0("ittmodel_", "mixed2","_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", "mixed2","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", participation1, "+", participation2, "+", grouping2,
"|", grouping1, ":dmatch", "+", grouping2))#, "+", grouping2, ":dmatch"))
#grouping1, ":dmatch", "+",
#grouping1, ":dmatch", "+" ,
print(klingform)
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", "mixed2","_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", "mixed2","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
#' Kling Model with Two Grouping Variables, Instrumenting Var 2
#'
#' This function is generally the same as itt_for_kling_p4 but it puts grouping variable 2 in the ITT models and instruments for it in the ivreg.
#' This variation seems to have produced good results for the Match/BAM project.
#' @param number_to_crop should equal the number of grouping categories you have. It's for selecting the interactions only out of the model.
#' @export
#' @examples
#' Block Level
#' itt_for_kling_p6(outcome = "mathxil_z_post2", participation1 = "treat_post1", participation2 = "treat_post2",
#' grouping1 = "schlid", grouping2 = "blocknum", sizefile = schoolsizes, weight = "precision", number_to_crop = 42,
#' sourcefile = master_dataset_1_math15_allblocks, weight_final = "stderr_scores_weight", "noomits_nome")
itt_for_kling_p6 <- function(outcome, participation1, participation2, grouping1, grouping2, sizefile, weight, number_to_crop, sourcefile, weight_final, omit_type){
#Assemble formulas for the various models #
itt_step1_form <- formula(paste(outcome, "~", grouping2,":dmatch", "+", grouping2))#,":dmatch"))
itt_step2_form <- formula(paste(participation1, "~", grouping2,":dmatch", "+", grouping2))#,":dmatch"))
itt_step3optional_form <- formula(paste(participation2, "~", grouping2,":dmatch", "+", grouping2))#,":dmatch"))
final_itt_form_1partic <- formula(paste("coeflist_scores~coeflist_partic-1")) #coeflist_partic2
final_itt_form <- formula(paste("coeflist_scores~coeflist_partic+coeflist_partic2-1")) #coeflist_partic2
#Get the counts together for weighting later
blocksizes <<- table(sourcefile[,"blocknum"])
blocksizes <- as.data.frame(blocksizes)
schoolsizes <<- table(sourcefile[,"schlid"])
schoolsizes <- as.data.frame(schoolsizes)
#adding block or school size for later weighting - block level excludes some blocks but school keeps all
master_dataset_1_forblocks <- merge(sourcefile, blocksizes, by.x="blocknum", by.y="Var1", all.x=T)
master_dataset_1_forschools <- merge(sourcefile, schoolsizes, by.x="schlid", by.y="Var1", all.x=T)
master_dataset_1_forschools$schlid <- as.factor(master_dataset_1_forschools$schlid)
master_dataset_1_forblocks$schlid <- as.factor(master_dataset_1_forblocks$schlid)
#Collect the Ns by school for using in the MTO weights
assign_block <- summarize(group_by(master_dataset_1_forschools, dmatch, blocknum),assign_block_count = n())
assign_school <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid),assign_school_count = n())
school_block <- summarize(group_by(master_dataset_1_forschools, schlid, blocknum),school_block_count = n())
assign_school_block <- summarize(group_by(master_dataset_1_forschools, dmatch, schlid, blocknum),assign_school_block_count = n())
assign <- summarize(group_by(master_dataset_1_forschools, dmatch),assign_count = n())
allcount <- summarize(group_by(master_dataset_1_forschools),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools, assign_block, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, school_block, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign_school_block, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forschools_withcounts <- merge(master_dataset_1_forschools_withcounts, assign, by=c("dmatch"), all.x=T)
master_dataset_1_forschools_withcounts$all_count <- allcount$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forschools_withcounts<-as.data.table(master_dataset_1_forschools_withcounts, keep.rownames=TRUE)
master_dataset_1_forschools_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forschools_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forschools_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forschools_withcounts <- as.data.frame(master_dataset_1_forschools_withcounts)
#Collect the Ns by block for using in the MTO weights
assign_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, blocknum),assign_block_count = n())
assign_school_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid),assign_school_count = n())
school_block_bk <- summarize(group_by(master_dataset_1_forblocks, schlid, blocknum),school_block_count = n())
assign_school_block_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch, schlid, blocknum),assign_school_block_count = n())
assign_bk <- summarize(group_by(master_dataset_1_forblocks, dmatch),assign_count = n())
allcount_bk <- summarize(group_by(master_dataset_1_forblocks),all_count = n())
#Add the various Ns to the existing dataset
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forschools, assign_block_bk, by=c("dmatch", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_bk, by=c("dmatch", "schlid"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, school_block_bk, by=c("schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_school_block_bk, by=c("dmatch", "schlid", "blocknum"), all.x=T)
master_dataset_1_forblocks_withcounts <- merge(master_dataset_1_forblocks_withcounts, assign_bk, by=c("dmatch"), all.x=T)
master_dataset_1_forblocks_withcounts$all_count <- allcount_bk$all_count
#Calculate the MTO weight for each student as appropriate
master_dataset_1_forblocks_withcounts<-as.data.table(master_dataset_1_forblocks_withcounts, keep.rownames=TRUE)
master_dataset_1_forblocks_withcounts[,mtoweight_calculation:=((assign_count/all_count)/(assign_school_block_count/school_block_count))]
master_dataset_1_forblocks_withcounts[,weight_calculation1:=(assign_count/all_count)]
master_dataset_1_forblocks_withcounts[,weight_calculation2:=(assign_school_block_count/school_block_count)]
master_dataset_1_forblocks_withcounts <<- as.data.frame(master_dataset_1_forblocks_withcounts)
#Conditional- if you're working based on schools, then your filename is this.
if(grouping1=="schlid"){
working_data_file <<- master_dataset_1_forschools_withcounts
file_string <<- "master_dataset_1_forschools_withcounts"
}
#Conditional- if you're working based on blocks, then your filename is this.
if(grouping1=="blocknum"){
working_data_file <<- master_dataset_1_forblocks
file_string <<- "master_dataset_1_forblocks"
}
#Creates school level summary variables
sample <- summarize(group_by(master_dataset_1_forschools_withcounts, blocknum, dmatch, schlid, assign_block_count, assign_school_count, school_block_count, assign_school_block_count, assign_count, all_count, mtoweight_calculation),
count = n(),
tx_yr1 = sum(treat_post1, na.rm=T),
tx_yr2 = sum(treat_post2, na.rm=T),
pct_tx_yr1 = tx_yr1/count,
pct_tx_yr2 = tx_yr2/count,
math_2015_avg = mean(mathxil_z_post2)
)
# Layered ITT Work ####
#Build the weight variable out of the function arguments
weight2 <<- eval(parse(text=paste0(file_string, "$", weight)))
# First ITT - Math Score #
scoremod <- lm(itt_step1_form, weights=weight2, data=working_data_file)
coeflist_scores <- coef(scoremod)
stderr_scores <- coef(summary(scoremod))[,2]
resid_scores <- resid(scoremod)
print(scoremod)
# Second ITT - Year 1 Treatment Participation #
partic_mod <- lm(itt_step2_form, weights=weight2, data=working_data_file)
coeflist_partic <- coef(partic_mod)
stderr_partic <- coef(summary(partic_mod))[,2]
resid_partic <- resid(partic_mod)
print(partic_mod)
# Third ITT - Year 2 Treatment Participation #
partic2_mod <- lm(itt_step3optional_form,weights=weight2, data=working_data_file)
coeflist_partic2 <- coef(partic2_mod)
stderr_partic2 <- coef(summary(partic2_mod))[,2]
resid_partic2 <- resid(partic2_mod)
print(partic2_mod)
# Combine ITT output into a new data frame ####
scorepartic <- cbind(coeflist_scores, coeflist_partic, coeflist_partic2, stderr_scores, stderr_partic, stderr_partic2)
scorepartic_resids <- cbind(resid_scores, resid_partic, resid_partic2)
#Drop the estimates for the individual blocks- we just want the interactions here#
working_scorepartic <- scorepartic[-c(1:number_to_crop),]
working_scorepartic <- cbind(sizefile, working_scorepartic)
working_scorepartic<-as.data.frame(working_scorepartic)
#Calculate the precision weights- 1/standard error squared
working_scorepartic<-as.data.table(working_scorepartic, keep.rownames=TRUE)
working_scorepartic[,stderr_scores_weight:=(1/(working_scorepartic$stderr_scores^2))]
working_scorepartic[,stderr_partic_weight:=(1/(working_scorepartic$stderr_partic^2))]
working_scorepartic[,stderr_partic2_weight:=(1/(working_scorepartic$stderr_partic2^2))]
working_scorepartic <- as.data.frame(working_scorepartic)
#Create the weight variable for the final model
weight_final2 <- eval(parse(text=paste0("working_scorepartic", "$", weight_final)))
#Run the final ITT model
summary(lm(final_itt_form,weights=weight_final2,data=working_scorepartic))
ittmodel <- lm(final_itt_form,weights=weight_final2,data=working_scorepartic)
ittmodel_robust <- robust.se(ittmodel)
print(ittmodel_robust)
#Create the output with dynamic naming
assign(paste0("ittmodel_", "mixed4","_", weight, "_", omit_type), envir=.GlobalEnv, ittmodel)
assign(paste0("ittmodel_", "mixed4","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, ittmodel_robust)
#ALTERNATIVE APPROACH - uses a different strategy with just one formula.
#Build formula, uses inputs from the function arguments again
klingform <- formula(paste("J~", participation1, "+", participation2, "+", grouping2,
"|", grouping2, "+", grouping2,":dmatch"))
#grouping1, ":dmatch", "+",
#grouping1, ":dmatch", "+" ,
print(klingform)
suppressMessages(library(formattable))
suppressMessages(library(knitr))
suppressMessages(library(data.table))
suppressMessages(library(dplyr))
suppressMessages(library(xtable))
suppressMessages(library(sem))
suppressMessages(library(AER))
suppressMessages(library(ivpack))
#Run an IV Reg with the formula designed above.
fulloutcome <- eval(parse(text=paste0(file_string, "$", outcome)))
tsls_fn_small <- function(I) {
J <<- as.matrix(I)
require(sem)
model <<- ivreg(klingform,
data=working_data_file,
weight=weight2)
totmod <<- summary(model)
totmod
robustmod <<- robust.se(model)
robustmod
}
print(tsls_fn_small(fulloutcome))
#Create the output with dynamic naming
assign(paste0("klingmodel_", "mixed4","_", weight, "_", omit_type), envir=.GlobalEnv, model)
assign(paste0("klingmodel_", "mixed4","_", weight, "_", omit_type, "_robustse"), envir=.GlobalEnv, robustmod)
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.