-
Notifications
You must be signed in to change notification settings - Fork 0
/
02_mle_example_v4.R
273 lines (194 loc) · 8 KB
/
02_mle_example_v4.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
#
# 02_mle_example.R
#
# Reed Sorensen
# November 2016
#
rm(list = ls())
library(foreign)
library(spatstat)
library(SpatialEpi)
library(dplyr)
#####
# read in individual and cluster data
df_individual <- read.csv("data/haitibirthclean3.csv") %>%
mutate(wealth = wealth / 10000)
df_cluster <- read.dbf("data/DHS GPS/HTGE61FL.dbf") %>%
dplyr::select(cluster_id = DHSCLUST, latitude = LATNUM, longitude = LONGNUM) %>%
filter(latitude != 0 & longitude != 0)
df_cluster[, c("xvar", "yvar")] <- latlong2grid( # convert lat/long to km
df_cluster[, c("longitude", "latitude")]
)
#####
# read in facility data
dat1 <- read.csv("data/facility_data_anc.csv")
dat2 <- read.csv("data/haitideliveryindicators.csv")
df <- left_join(dat1, dat2, by = c("facility_id" = "facil")) %>%
filter(!is.na(deliveryready))
df[, c("xvar", "yvar")] <- latlong2grid(df[, c("longitude", "latitude")])
# split data by facility type and urbanicity
# in order to estimate KDE sigma for each one
# # NOTE: there are only 18 rural hospitals, and
# # the model failed to give an estimate for sigma_hosp_rural
df_list <- list(
df_hosp_urban = subset(df, hospital == 1 & rural == 0),
df_hosp_rural = subset(df, hospital == 1 & rural == 1),
df_nonhosp_urban = subset(df, hospital == 0 & rural == 0),
df_nonhosp_rural = subset(df, hospital == 0 & rural == 1)
)
# # this version only differentiates between hospital and non-hospital
# df_list <- list(
# df_hosp = subset(df, hospital == 1),
# df_nonhosp = subset(df, hospital == 0)
# )
# this version makes no distinctions by facility type or urbanicity
# df_list <- list(
# df_alldat = df
# )
#####
# set pixel and grid size
pixel_length <- 0.5
xlo <- min(df$xvar) - pixel_length
xhi <- max(df$xvar) + pixel_length
ylo <- min(df$yvar) - pixel_length
yhi <- max(df$yvar) + pixel_length
x_points <- seq(xlo, xhi, by = pixel_length)
y_points <- seq(ylo, yhi, by = pixel_length)
window1 <- owin(
xrange = c(min(x_points), max(x_points)),
yrange = c(min(y_points), max(y_points)))
dat_grid <- expand.grid( # grid of zeros
xvar = x_points,
yvar = y_points,
deliveryready = 0
)
#####
# process facility-level data
df_list2 <- lapply(df_list, function(x) {
# x <- df_list[[1]] # dev variable
tmp <- x %>% # keep points within (potentially zoomed) area
filter(xvar >= xlo & xvar <= xhi) %>%
filter(yvar >= ylo & yvar <= yhi) %>%
dplyr::select(xvar, yvar, deliveryready)
# tmp$deliveryready <- 67 # counterfactual; max observed in dataset
tmp2 <- rbind(tmp, dat_grid) # add grid of zeros
tmp3 <- ppp( # create spatial data frame
x = tmp2$xvar,
y = tmp2$yvar,
window = window1
)
marks(tmp3) <- tmp2$deliveryready # add facility values
return(tmp3)
})
#####
# maximum likelihood function
# # version with 4 sigmas for facility type and urbanicity; wealth/educ PCA
# parameter_names <- c(
# "constant", "wealth_education_PCA", "surface_var",
# "sigma_hosp_urban", "sigma_hosp_rural", "sigma_nonhosp_urban", "sigma_nonhosp_rural"
# )
# parameter_stval <- c(-1.0043, 1.0019, 0.1063, 1.5, 3.0, 0.35, 0.7)
# PCA, age, surface_var and 1 KDE sigma
parameter_names <- c(
"constant", "age", "birthorder", "married", "pca",
# "sigma_hosp_urban", "sigma_hosp_rural", "sigma_clinic_urban", "sigma_clinic_rural"
# "sigma_hosp", "sigma_nonhosp"
"sigma_hosp_rural", "sigma_nonhosp_urban", "sigma_nonhosp_rural"
)
parameter_stval <- c(-2.1, 0.05, -0.4, 0.5, 0.65, 0.7, 0.7, 0.7)
# # version with 2 sigmas for facility type; birthorder, age, married, wealth/educ pca
# parameter_names <- c(
# "constant", "birthorder", "age", "married", "wealth_education_PCA", "surface_var",
# "sigma_hosp", "sigma_nonhosp"
# )
#
# parameter_stval <- c(-1.2591, -0.2200, 0.0325, 0.0157,
# 0.8903, 0.0119, 1.0975, 0.3562)
length(parameter_names) == length(parameter_stval)
llk.logit <- function(param) {
# param <- parameter_stval # dev variable
n_sigmas <- 3 # set this depending on the model
param_regression <- param[1:(length(parameter_stval)-n_sigmas)]
param_sigmas <- param[(length(param_regression)+1):(length(parameter_stval))]
if (any(param_sigmas <= 0.01)) return(-100000)
df_cluster_tmp <- df_cluster
get_surface_points <- function(surface, sigma_var, x_var, y_var) {
surface1 <- Smooth(surface, sigma = sigma_var)
get_pixel_value1 <- spatstat::as.function.im(surface1)
# get pixel values (and fix where the function returns an empty vector)
tmp <- mapply(get_pixel_value1, x = x_var, y = y_var)
tmp[unlist(lapply(tmp, function(x) length(x)==0))] <- NA
return(unlist(tmp))
}
surface_dat <- do.call("cbind", lapply(1:length(df_list2), function(i) {
sigma_var_tmp <- ifelse(i==1, 0.5, param_sigmas[(i-1)]) # changed from 1 to 0.5
tmp <- get_surface_points(
surface = df_list2[[i]],
# sigma_var = param_sigmas[i],
sigma_var = sigma_var_tmp,
x_var = df_cluster_tmp$xvar,
y_var = df_cluster_tmp$yvar)
tmp[is.na(tmp) | tmp < 0] <- 0
# # rescale to original
tmp * (max(df_list[[i]]$deliveryready, na.rm = TRUE) / max(tmp))
}))
# surface_tmp <- get_surface_points(
# surface = df_list2[[1]],
# sigma_var = param_sigmas[1],
# x_var = df_cluster_tmp$xvar,
# y_var = df_cluster_tmp$yvar,
# logged = TRUE )
#
# surface_tmp[is.na(surface_tmp) | surface_tmp < 0] <- 0
# surface_dat <- cbind(surface_tmp, rep(1, length(surface_tmp)))
# rescale to original
# surface_tmp <- apply(surface_dat, MARGIN = 1, max)
# df_cluster_tmp$surface_var <- surface_tmp * (max(df$deliveryready, na.rm = TRUE) / max(surface_tmp))
df_cluster_tmp$surface_var <- apply(surface_dat, MARGIN = 1, max) # this doesn't rescale
dat2 <- left_join(df_individual, df_cluster_tmp, by = c("clusterid" = "cluster_id")) %>%
filter(complete.cases(.)) # figure out why 25 out of 1991 are NA
# model_tmp <- facilitybirth ~ birthorder + age + married + pca + surface_var
# fit_tmp <- glm(model_tmp, data = dat2, family = binomial)
# tmp_sav <- predict(fit_tmp, newdata = dat2[, c("birthorder", "age", "married", "pca", "surface_var")], type = "response")
# # Not run: saveRDS(tmp_sav, "data/predict_counterfactual_fullmodel.RDS")
# # Not run: saveRDS(tmp_sav, "data/predict_original_fullmodel.RDS")
# dat2$cf <- readRDS("data/predict_counterfactual_fullmodel.RDS")
# dat2$pred <- readRDS("data/predict_original_fullmodel.RDS")
# dat2$diff <- dat2$cf - dat2$pred
# dat2$diff_births <- dat2$diff * dat2$svyweight
y <- dat2$facilitybirth
# x <- cbind(dat2$pca, dat2$surface_var)
x <- cbind(dat2$age, dat2$birthorder, dat2$married, dat2$pca, dat2$surface_var)
os <- rep(1,length(x[,1]))
x2 <- cbind(os,x)
# b <- param_regression[1:ncol(x2)]
b <- c(param_regression, 1)
xb <- x2 %*% b
sum(-1 * (y*log(1+exp(-xb)) + (1-y)*log(1+exp(xb))))
}
system.time(result1 <- optim(
par = parameter_stval, fn = llk.logit,
method = "BFGS", hessian = T, control = list(fnscale = -1)
))
matrix(
c(round(result1$par, digits = 4), round(sqrt(diag(solve(-1 * result1$hessian))), digits = 4)),
dimnames = list(parameter_names, c("Estimate", "Std. Error")),
ncol = 2
)
# saveRDS(result1, "tmp.RDS") # version with baseline sigma = 1 (urban hospital)
saveRDS(result1, "tmp2.RDS") # version with baseline sigma = 0.5 (urban hospital)
# tmp <- readRDS("data/results1.RDS")
# tmp2 <- readRDS("data/results_hosp_nonhosp.RDS") # covars: wealth and educat
# tmp3 <- readRDS("data/results_onesigma.RDS") # covars: wealth and educat
# tmp4 <- readRDS("data/results_hosp_nonhosp_wealthonly.RDS") # covars: wealth
# tmp5 <- readRDS("data/results_hosp_nonhosp_educonly.RDS") # covars: educ
# tmp6 <- readRDS("data/results_2sigmas_wealth_education.RDS")
# tmp7 <- readRDS("data/results_2sigmas_pca_v2.RDS")
# tmp8 <- readRDS("data/results_4sigmas_pca.RDS")
# tmp9 <- readRDS("data/results_2sigmas_pca_and_all_others_v2.RDS")
# tmp10 <- readRDS("data/new_results_2sigmas_age_birthorder_pca.RDS")
tmp11 <- readRDS("data/new_results_4sigmas_age_birthorder_pca.RDS")
#
result1 <- tmp11
#
saveRDS(result1, "data/new_results_4sigmas_age_birthorder_pca.RDS")