-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathfunctions.r
341 lines (316 loc) · 15.3 KB
/
functions.r
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
display <- function(num, pct=FALSE){
if (!pct) return(sprintf("%.3f", num))
if (pct) return(sprintf("%.1f", num*100))
}
run_model <- function(dv, dv_pre = NULL, trt = "HuffPost", control=NULL,
data = svy, seed=123, blocks="W3_Browser_treatment_w3",
more_vars = NULL, verbose=TRUE) {
set.seed(seed)
if(is.null(control) & trt == "FoxNews") { droptrt <- "HuffPost" }
if(is.null(control) & trt == "HuffPost") { droptrt <- "FoxNews" }
if(!is.null(control) & trt %in% c("FoxNews", "HuffPost")) { droptrt <- "Control" }
svy_na <- na.omit(data[, c(vars, dv, dv_pre, more_vars)])
svy_model <- model.matrix(~., svy_na[, c(vars, dv_pre, more_vars)])
lasso_select <- cv.glmnet(x=svy_model,
y=as.vector(get(dv, svy_na)),
alpha=1)
coef.out <- coef(lasso_select, exact = TRUE)
inds <- which(coef.out != 0)
lin_formula <- formula(paste0(dv, " ~ W3_PATA306_treatment_w3"))
if(length(inds) == 1) {
if(!is.null(blocks)) {
lin_cov <- paste0(" ~ ", blocks)
} else {
lin_cov <- NULL
}
} else {
incl_vars <- grep("(Intercept)", rownames(coef.out)[inds], value = TRUE, invert = TRUE)
for(v in 1:length(incl_vars)) {
# if glmnet pulls out a single factor level from a factor var, throw in entire variable
if(!incl_vars[v] %in% names(data) & !grepl('state|raceeth', incl_vars[v])) {
c <- unlist(sapply(1:length(names(data)), function(x) if(grepl(names(data)[x], incl_vars[v])) x))
incl_vars[v] <- names(data)[tail(unique(c), 1)]
}
if(grepl('state', incl_vars[v])){
incl_vars[v] <- paste0("I(state==", gsub("state", "", incl_vars[v]), ")")
}
if(grepl('raceeth', incl_vars[v])){
incl_vars[v] <- paste0("I(raceeth==", gsub("raceeth", "'", incl_vars[v]), "')")
}
}
if(!is.null(blocks)){
lin_cov <- paste0(" ~ ", blocks, " + ", paste(incl_vars, collapse = " + "))
} else {
lin_cov <- paste0(" ~ ", paste(incl_vars, collapse = " + "))
}
}
if(!is.null(lin_cov)){
lin_model <- lm_lin(lin_formula, covariates = formula(lin_cov), data = dplyr::filter(data, W3_PATA306_treatment_w3 != droptrt))
} else{
lin_model <- lm_robust(lin_formula, data = dplyr::filter(data, W3_PATA306_treatment_w3 != droptrt))
}
sig <- ifelse(abs(lin_model$statistic[2]) >= 1.96, "STARS!", "")
if(verbose){
message("Estimate: ", display(coef(lin_model)[2]))
message("Std. Error: ", display(sqrt(vcov(lin_model)[2,2])))
message("CI Lower: ", display(confint(lin_model)[2,1]))
message("CI Upper: ", display(confint(lin_model)[2,2]))
}
# only pre-treatment DV
if (!is.null(dv_pre)){
lin_formula <- formula(paste0(dv, " ~ W3_PATA306_treatment_w3 + ",
dv_pre))
ss <- summary(lm(lin_formula, data=as.data.frame(data),
subset=W3_PATA306_treatment_w3 != droptrt))
message("Pre-treatment DV, Adj R2 = ", display(ss$adj.r.squared))
message("N = ", length(ss$residuals))
}
# Lasso covariates
if(length(inds) > 1){
lin_formula <- formula(paste0(dv, " ~ W3_PATA306_treatment_w3 + ",
paste(incl_vars, collapse = " + ")))
ss <- summary(lm(lin_formula, data=as.data.frame(data),
subset=W3_PATA306_treatment_w3 != droptrt))
message("Lasso covariates, Adj R2 = ", display(ss$adj.r.squared))
message("Covariates: ", paste(incl_vars, collapse = " + "))
message("N = ", length(ss$residuals))
}
# all variables
lin_formula <- formula(paste0(dv, " ~ W3_PATA306_treatment_w3 + ",
paste(c(vars, dv_pre), collapse = " + ")))
ss <- summary(lm(lin_formula, data=as.data.frame(data),
subset=W3_PATA306_treatment_w3 != droptrt))
message("All covariates, Adj R2 = ", display(ss$adj.r.squared))
message("Covariates: ", paste(vars, collapse = " + "))
message("N = ", length(ss$residuals))
return(list(lin_model, sig, lin_cov))
}
extract_covariates <- function(mod){
coefs <- names(coef(mod[[1]]))
coefs <- gsub("_c$|Firefox_c", "", coefs)
coefs <- coefs[coefs %in% names(svy)]
return(coefs)
}
estimate_cace <- function(Y, D, Z, X, trt, control="Control", data = svy){
form <- formula(paste0(Y, " ~ ", D, " + ", paste(X, collapse=" + "),
" | ", Z, " + ", paste(X, collapse=" + ")))
dt <- filter(data, W3_PATA306_treatment_w3 %in% c(control, trt))
mod <- iv_robust(form, data=dt)
return(mod)
}
# to determine whether we need to run multiple imputation, compute the
# % of observations with at least one missing value in the covariates.
# Note that missing values in DV are not included in the estimation.
# If % of missing values is above 20%, then we will do multiple imputation.
compute_proportion_missing_covars <- function(mod, more_info=FALSE){
# mod = output from run_model()
# identify labels for covariates and dv
covars <- attr(mod[[1]]$terms, "term.labels")
dv <- mod[[1]]$outcome
# replacing state dummies with state factor variable
if (sum(grepl('state', covars))>0){
toreplace <- grep('state', covars)
covars[toreplace] <- 'state'
covars <- unique(covars)
}
# replacing race dummies with race factor variable
if (sum(grepl('raceeth', covars))>0){
toreplace <- grep('raceeth', covars)
covars[toreplace] <- 'raceeth'
covars <- unique(covars)
}
# extract relevant variables from svy file
dd <- svy[,c(dv, covars)]
# identify rows where DV is not missing
no_miss_dv <- which(!is.na(dd[,dv]))
# identify rows where at least one covariate has a missing value
miss_covars <- apply(svy[no_miss_dv,covars], 1, function(x) any(is.na(x)))
# print % of missing values
message(display(mean(miss_covars), pct=TRUE), "% missing")
if (more_info){
print(summary(svy[no_miss_dv,covars]))
}
return(invisible(mean(miss_covars)))
}
# Compute power analysis for DIM, ITT and CACE
power2 <- function(dv, dim, itt, cace, covariates = TRUE, cace_mde = FALSE, D, trt,
trt_var = "W3_PATA306_treatment_w3"){
message("MDE without covariate adjustment:")
pwr_dim <- power.t.test(n = dim$nobs,
sd = sd(unlist(svy[dv]), na.rm=TRUE),
sig.level = 0.05, power=0.80)
message(display(pwr_dim$delta), " (",
display(pwr_dim$delta / sd(unlist(svy[,dv]),
na.rm=TRUE), pct=TRUE),
"% of a 1-SD increase in DV)")
if (covariates == TRUE){
message("MDE with covariate adjustment:")
message("Covariates: ", itt[[3]])
reg <- summary(lm(paste0(dv, itt[[3]]), data=svy))
( pwr_itt <- power.t.test(n = itt[[1]]$nobs,
sd = reg$sigma,
sig.level = 0.05, power=0.80
) )
message(display(pwr_itt$delta), " (",
display(pwr_itt$delta / sd(unlist(svy[,dv]),
na.rm=TRUE), pct=TRUE),
"% of a 1-SD increase in DV)")
}
if(cace_mde == TRUE){
message("MDE for CACE:") # following Bansak 2020
# compute proportion of variation in D/Y left unexplained by Z,
# that is explained by covariates W
if(covariates == TRUE){
d_z_df <- svy[as.data.frame(svy)[,trt_var] %in% c("Control", trt),]
d_z_df <- na.omit(d_z_df[,c(D, trt_var, extract_covariates(itt))])
d_y_df <- svy[as.data.frame(svy)[,trt_var] %in% c("Control", trt),]
d_y_df <- na.omit(d_y_df[,c(dv, trt_var, extract_covariates(itt))])
out_d_z <- lm(paste0(D, "~", trt_var), data = d_z_df)
out_d_y <- lm(paste0(dv, "~", trt_var), data = d_y_df)
out_d_zw <- lm(paste0(D, "~", paste0(c(trt_var, extract_covariates(itt)),
collapse = "+")), data = d_z_df)
out_d_yw <- lm(paste0(dv, "~", paste0(c(trt_var, extract_covariates(itt)),
collapse = "+")), data = d_y_df)
r2dw <- max(0, ((summary(out_d_z)$sigma)^2 -
(summary(out_d_zw)$sigma)^2)/((summary(out_d_z)$sigma)^2))
r2yw <- max(0, ((summary(out_d_y)$sigma)^2 -
(summary(out_d_yw)$sigma)^2)/((summary(out_d_y)$sigma)^2))
}
# compute compliance rate / average causal effect of treatment assignment
# on treatment uptake
d_z_df <- svy[as.data.frame(svy)[,trt_var] %in% c("Control", trt),]
out_d_z <- lm(paste0(D, "~", trt_var), data = d_z_df)
compliance_rate <- out_d_z$coefficients[2]
# compute conservative (upper) bound for minimum detectable effect size
if(covariates == TRUE) {
if(length(extract_covariates(itt)) > 1) {
pwr_cace <- try(powerLATE::powerLATE.cov(pZ = .5,
pi = compliance_rate,
N = cace$nobs,
sig.level = 0.05,
power = 0.8,
r2dw = r2dw,
r2yw = r2yw,
effect.size = TRUE,
assume.ord.means = TRUE,
verbose = FALSE)$output.parameter)
}else{
pwr_cace <- try(powerLATE::powerLATE(pZ = .5,
pi = compliance_rate,
N = cace$nobs,
sig.level = 0.05,
power = 0.8,
effect.size = TRUE,
assume.ord.means = TRUE,
verbose = FALSE)$output.parameter)
}
}else{
pwr_cace <- try(powerLATE::powerLATE(pZ = .5,
pi = compliance_rate,
N = cace$nobs,
sig.level = 0.05,
power = 0.8,
effect.size = TRUE,
assume.ord.means = TRUE,
verbose = FALSE)$output.parameter)
}
if(class(pwr_cace) == "try-error") {
pwr_cace <- NA
}
message(display(pwr_cace), " (",
display(pwr_cace / sd(unlist(svy[,dv]),
na.rm=TRUE), pct=TRUE),
"% of a 1-SD increase in DV)")
}
if(covariates == FALSE & cace_mde == TRUE){
data.frame(pwr_itt = c(display(pwr_dim$delta), display(pwr_dim$delta / sd(unlist(svy[,dv]),
na.rm=TRUE))),
pwr_cace = c(display(pwr_cace), display(pwr_cace / sd(unlist(svy[,dv]),
na.rm=TRUE))))
}
if(covariates == TRUE & cace_mde == TRUE){
data.frame(pwr_itt = c(display(pwr_itt$delta), display(pwr_itt$delta / sd(unlist(svy[,dv]),
na.rm=TRUE))),
pwr_cace = c(display(pwr_cace), display(pwr_cace / sd(unlist(svy[,dv]),
na.rm=TRUE))))
}
if(covariates == FALSE & cace_mde == FALSE){
data.frame(pwr_itt = c(display(pwr_dim$delta), display(pwr_dim$delta / sd(unlist(svy[,dv]),
na.rm=TRUE))))
}
if(covariates == TRUE & cace_mde == FALSE){
data.frame(pwr_itt = c(display(pwr_itt$delta), display(pwr_itt$delta / sd(unlist(svy[,dv]),
na.rm=TRUE))))
}
}
format_latex2 <- function(dim, itt, cace, trt, pwr, dv, cace_mde = FALSE){
if(cace_mde == TRUE){
ltx <- paste0(
dv, " & ", display(dim$coefficients), " (", display(dim$std.error), ")",
" & ", display(coef(itt[[1]])[2]), " (", display(sqrt(vcov(itt[[1]])[2,2])), ")",
" & ", display(cace$coefficients[2]), " (", display(cace$std.error[2]), ")",
" & ", pwr$pwr_itt[1],
" & ", pwr$pwr_cace[1], "\\\\ \n",
"(", trt, ") & [", display(dim$conf.low), ", ", display(dim$conf.high), "]",
" & [", display(confint(itt[[1]])[2,1]), ", ", display(confint(itt[[1]])[2,2]), "]",
" & [", display(cace$conf.low[2]), ", ", display(cace$conf.high[2]), "]",
" & ($d$=", pwr$pwr_itt[2], ")",
" & ($d$=", pwr$pwr_cace[2], ")\\\\"
)
}else{
ltx <- paste0(
dv, " & ", display(dim$coefficients), " (", display(dim$std.error), ")",
" & ", display(coef(itt[[1]])[2]), " (", display(sqrt(vcov(itt[[1]])[2,2])), ")",
" & ", display(cace$coefficients[2]), " (", display(cace$std.error[2]), ")",
" & ", pwr$pwr_itt[1], "\\\\ \n",
"(", trt, ") & [", display(dim$conf.low), ", ", display(dim$conf.high), "]",
" & [", display(confint(itt[[1]])[2,1]), ", ", display(confint(itt[[1]])[2,2]), "]",
" & [", display(cace$conf.low[2]), ", ", display(cace$conf.high[2]), "]",
" & ($d$=", pwr$pwr_itt[2], ")\\\\"
)
}
message(gsub('_', '\\\\_', ltx))
}
format_latex3 <- function(dim, itt, dv){
ltx <- paste0(
dv, " & ", display(dim$coefficients), " (", display(dim$std.error), ")",
" & ", display(coef(itt[[1]])[2]), " (", display(sqrt(vcov(itt[[1]])[2,2])), ")","\\\\ \n",
"(", trt, " vs. ", control,") & [", display(dim$conf.low), ", ", display(dim$conf.high), "]",
" & [", display(confint(itt[[1]])[2,1]), ", ", display(confint(itt[[1]])[2,2]), "]\\\\"
)
message(gsub('_', '\\\\_', ltx))
} # print stats: treatment vs. control
heterogeneous_effect <- function(dv, dv_pre=NULL, trt, moderator){
# compute standard itt
itt <- run_model(dv = dv, dv_pre = dv_pre, trt = trt, verbose=FALSE)
vars <- unique(c(extract_covariates(itt), moderator))
# now compute itt but adding moderator (if it wasn't included)
lin_formula <- formula(paste0(dv, " ~ W3_PATA306_treatment_w3"))
lin_covars <- formula(paste0(" ~ ", paste(vars, collapse = " + ")))
res <- lm_lin(lin_formula, covariates = lin_covars,
data=svy[svy$W3_PATA306_treatment_w3 %in% c("Control", trt),])
# extract t stat for interaction
interaction_t <- res$statistic[grep(paste0(":", moderator), names(res$statistic))]
sig <- ifelse(
interaction_t > 1.96, "+",
ifelse(interaction_t < (-1.96), "-", "n.s.")
)
return(sig)
}
heterogeneous_effect2 <- function(dv, dv_pre=NULL, trt, control, moderator){
# compute standard itt
itt <- run_model(dv = dv, dv_pre = dv_pre, trt = trt, verbose=FALSE)
vars <- unique(c(extract_covariates(itt), moderator))
# now compute itt but adding moderator (if it wasn't included)
lin_formula <- formula(paste0(dv, " ~ W3_PATA306_treatment_w3"))
lin_covars <- formula(paste0(" ~ ", paste(vars, collapse = " + ")))
res <- lm_lin(lin_formula, covariates = lin_covars,
data=svy[svy$W3_PATA306_treatment_w3 %in% c(control, trt),])
# extract t stat for interaction
interaction_t <- res$statistic[grep(paste0(":", moderator), names(res$statistic))]
sig <- ifelse(
interaction_t > 1.96, "+",
ifelse(interaction_t < (-1.96), "-", "n.s.")
)
return(sig)
}