-
Notifications
You must be signed in to change notification settings - Fork 0
/
multi_village_projects.R
308 lines (270 loc) · 18.1 KB
/
multi_village_projects.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
# multi-village projects make up ~20% of all PID projects
setwd("~/box sync/cambodia_eba_gie")
library(plyr)
library(dplyr)
library(readxl)
library(sf)
library(stringr)
library(sp)
library(spatialEco)
library(rlist)
# the commented out code merges the three PID datasets and assigns a common Activity Type index to the PID data
# it then writes out a complete PID dataset that can be read in instead of re-running this code each time
pid2008 <- read.csv("PID/completed_pid/pid_2008.csv", stringsAsFactors = F)
pid2012 <- read.csv("PID/completed_pid/pid_2012.csv", stringsAsFactors = F)
pid2016 <- read.csv("PID/completed_pid/pid_2016.csv", stringsAsFactors = F)
test2008 <- NULL
for(i in pid2008$project.id) {
temp <- pid2008[pid2008$project.id==i,]
if(length(unique(temp$vill.id)) >1) {test2008[length(test2008)+1] <- T}
else {test2008[length(test2008)+1] <- F}}
pid2008.sub <- pid2008[which(test2008),]
test2012 <- NULL
for(i in pid2012$project.id) {
temp <- pid2012[pid2012$project.id==i,]
if(length(unique(temp$vill.id)) >1) {test2012[length(test2012)+1] <- T}
else {test2012[length(test2012)+1] <- F}}
pid2012.sub <- pid2012[which(test2012),]
test2016 <- NULL
for(i in pid2016$project.id) {
temp <- pid2016[pid2016$project.id==i,]
if(length(unique(temp$vill.id)) >1) {test2016[length(test2016)+1] <- T}
else {test2016[length(test2016)+1] <- F}}
pid2016.sub <- pid2016[which(test2016),]
pid <- rbind.fill(pid2016.sub, pid2012.sub, pid2008.sub)
# multi-village projects make up ~20% of all PID projects
common.index <- as.data.frame(matrix(data = NA, nrow = 21, ncol = 0))
common.index[1:17,"2008"] <- sort(unique(pid2008$activity.type))
common.index[1:17,"2012"] <- sort(unique(pid2012$activity.type))[match(common.index[1:17,"2008"],
sort(unique(pid2012$activity.type)))]
common.index[1:17,"2016"] <- sort(unique(pid2016$activity.type))[match(common.index[1:17,"2008"],
sort(unique(pid2016$activity.type)))]
common.index[18:21, "2012"] <- unique(pid2012$activity.type)[!(unique(pid2012$activity.type) %in% common.index[1:17,"2012"])]
common.index[18:21, "2016"] <- unique(pid2016$activity.type)[match(common.index[18:21, "2012"], unique(pid2016$activity.type))]
common.index[,"type"] <- c(common.index[1:17,"2008"], common.index[18:21, "2012"])
common.index[,"id"] <- seq(1, 21, 1)
pid <- merge(pid[,!(names(pid)=="activity.type.num")], common.index[,c("type", "id")], by.x = "activity.type", by.y = "type")
names(pid)[names(pid)=="id"] <- "activity.type.num"
pid <- pid[pid$actual.end.yr!=1908 | is.na(pid$actual.end.yr),]
polygons <- readRDS("inputdata/gadm36_KHM_4_sp.rds")
shape <- as.data.frame(read.csv("inputdata/village_grid_files/village_data.csv", stringsAsFactors = F))
spatial.data <- SpatialPointsDataFrame(coords = shape[,c("longitude", "latitude")], data = shape,
proj4string = CRS("+proj=longlat +datum=WGS84"))
shape <- as.data.frame(point.in.poly(x=spatial.data, y=polygons))[,c("VILL_CODE", "VILL_NAME", "NAME_1", "NAME_2", "NAME_3")]
names(shape) <- c("village.code", "village.name", "province.name", "district.name", "commune.name")
# merging PID data with shape data based on Village ID
pid <- merge(shape, pid, by.x = "village.code", by.y = "vill.id", all.y = T)
# generating a variable that contains the length of each PID project
pid$project.length <- (((pid$actual.end.yr-pid$actual.start.yr)*12)+(pid$actual.end.mo-pid$actual.start.mo))
# some PID projects are missing an actual end date value. To avoid losing data, actual end date is estimated by determining the average
# project length of other projects in the same year and in the same province and treating that as the expected length of the project. This
# expected length, assuming we have an actual start date value, allows us to estimate the end date of the project.
pid$enddate.type <- NA
for(i in 1:nrow(pid)) {
if(is.na(pid$actual.end.yr[i])) {
if(is.na(pid$actual.start.yr[i])) {
# determining the mean project length for projects in the the same province and year as i
avg.length <- mean(pid$project.length[which(pid$planned.start.yr==pid$planned.start.yr[i] & pid$province.name==pid$province.name[i])], na.rm = T)
# for cases where the actual start date is missing, we use the planned start date as the reference point
# for estimating actual end date
pid$actual.end.yr[i] <- pid$planned.start.yr[i] + floor((pid$planned.start.mo[i]+avg.length)/12)
pid$actual.end.mo[i] <- round(((pid$planned.start.mo[i]+avg.length) %% 12), digits = 0)
# assigning end date estimation codes for robustness checks
pid$enddate.type[i] <- 2
} else if (!is.na(pid$actual.start.yr[i])) {
avg.length <- mean(pid$project.length[which(pid$actual.start.yr==pid$actual.start.yr[i] & pid$province.name==pid$province.name[i])], na.rm = T)
# expected end date is estimated based on mean project length and the actual start date value
pid$actual.end.yr[i] <- pid$actual.start.yr[i] + floor((pid$actual.start.mo[i]+avg.length)/12)
pid$actual.end.mo[i] <- round(((pid$actual.start.mo[i]+avg.length) %% 12), digits = 0)
pid$enddate.type[i] <- 1
} else {pid$enddate.type[i] <- 0}
} else {pid$enddate.type[i] <- 0}
}
pid <- pid[!is.na(pid$actual.end.yr),]
# ensuring all levels of the new/repair string variable are consistent
pid$new.repair[pid$new.repair=="Routinemaintenance"] <- "Routine maintenance"
pid$new.repair[pid$new.repair=="Repeatedservice"] <- "Repeated service"
# normalizing numeric new/repair variables
pid$new.repair.num[pid$new.repair.num==601] <- 2
pid$new.repair.num[pid$new.repair.num==602] <- 1
pid$new.repair.num[pid$new.repair.num==603] <- 3
pid$new.repair.num[pid$new.repair.num==604] <- 4
pid$new.repair.num[pid$new.repair.num==605] <- 5
# removing major outliers in the bidding variable
nrow(pid[(pid$n.bidders %in% c(2003, 3140)),])
pid <- pid[!(pid$n.bidders %in% c(2003, 3140)),] #May want to keep the rows with high n bidders
sum(pid$cs.fund>2e+8, na.rm = T)
sum(pid$cs.fund>1e+8, na.rm = T)
#hist(pid$cs.fund[pid$cs.fund<1e+8]) #do we want to remove major outliers?
sum(pid$local.cont>3e+7, na.rm = T)
sum(pid$local.cont>1e+7, na.rm = T)
#hist(pid$local.cont[pid$local.cont<2e+6])
# creating a dummy variable denoting whether there was competitive bidding for a contract based
# on the number of bidders variable
pid$bid.dummy <- ifelse(pid$n.bidders==0, 0, 1)
###################
# creating a skeleton dataset to store treatment data
treatment <- as.data.frame(matrix(NA, nrow = 1, ncol = 24))[0,]
names(treatment) <- c("village.code", "village.name", "province.name", "district.name", "commune.name", "earliest.end.date",
"enddate.type", "earliest.sector.num", "earliest.sector", paste0("count", 2003:2017))
# filling treatment dataset with necessary variables
for(i in 1:length(unique(pid$village.code))) {
temp <- pid[which(pid$village.code==unique(pid$village.code)[i]),]
row <- nrow(treatment)+1
# treatment dataset only stores one observations per village. For cases of multiple observations of the same village,
# the treatment data stores the village ID, province the village is in, and the end year and activity type information
# for the observation with the earliest end year
treatment[row, c("village.code", "village.name", "province.name", "district.name", "commune.name", "earliest.end.date",
"enddate.type", "earliest.sector.num", "earliest.sector")] <- c(temp$village.code[1],
temp$village.name[1],
temp$province.name[1],
temp$district.name[1],
temp$commune.name[1],
temp$actual.end.yr[which.min(temp$actual.end.yr)],
temp$enddate.type[which.min(temp$actual.end.yr)],
temp$activity.type.num[which.min(temp$actual.end.yr)],
temp$activity.type[which.min(temp$actual.end.yr)])
# storing the count of villages getting treated in each year in the treatment data
for(j in sort(unique(pid$actual.end.yr))) {
treatment[row, grep(paste0("count", j), names(treatment))] <- nrow(temp[temp$actual.end.yr==j,])
}
}
###################
# reading in and merging data from the GeoQuery extract
grid_1000_matched_data <- read.csv("inputdata/village_grid_files/grid_1000_matched_data.csv",
stringsAsFactors = F)
merge_grid_1000_lite <- read.csv("inputdata/village_grid_files/merge_grid_1000_lite.csv",
stringsAsFactors = F)
grid_1000_matched_data <- merge(grid_1000_matched_data, merge_grid_1000_lite, by = "cell_id")
grid_1000_matched_data <- grid_1000_matched_data[,-c(7:49, 72:221)]
#merging PID data with GeoQuery extract
id.list <- list()
id.list2 <- list()
for(i in 1:nrow(grid_1000_matched_data)) {
# creating lists storing the information of which villages lie in, and in the cells bordering, each grid cell from
# the GeoQuery data
point <- as.character(as.numeric(unlist(strsplit(grid_1000_matched_data$village_point_ids[i], split = "\\|"))))
box <- as.character(as.numeric(unlist(strsplit(grid_1000_matched_data$village_box_ids[i], split = "\\|"))))
if(grid_1000_matched_data$village_point_ids[i]=="") {
id.list[[i]] <- ""
} else {
# if grid cell i has a village/s within it, item i in the object id.list stores the village id/s
id.list[[i]] <- point
}
if(grid_1000_matched_data$village_box_ids[i]=="") {
id.list2[[i]] <- ""
} else {
if(length(setdiff(box, point)) > 0) {
# if grid cell i has a village/s in the cells bordering it, item i in the object id.list2 stores the village id/s of
# those villages but omits any villages within grid cell i
id.list2[[i]] <- setdiff(box, point)
} else {
id.list2[[i]] <- ""
}
}
}
# creating a skeleton dataset to store the merged PID/shape and grid cell data
pre.panel.names <- c("village.code", "village.name", "province.name", "district.name", "commune.name",
as.vector(outer(c("box", "point"), c("earliest.end.date", "enddate.type", "earliest.sector.num", "earliest.sector",
paste0("count", 1992:2017)), paste, sep=".")), names(grid_1000_matched_data))
pre.panel <- as.data.frame(matrix(NA, nrow = nrow(grid_1000_matched_data), ncol = length(pre.panel.names)))
names(pre.panel) <- pre.panel.names
# in the "pre.panel" data, there will be one observation per grid cell
for(i in 1:length(unique(grid_1000_matched_data$cell_id))) {
# creating temporary datasets containing the PID/shape data of the villages that lie within or border grid cell i. This
# matching relies on the list objects build previously to identify which villages are within/bordering each cell
temp.point <- treatment[which(treatment$village.code %in% as.character(id.list[[i]])),]
temp.box <- treatment[which(treatment$village.code %in% as.character(id.list2[[i]])),]
# if there are one or more villages within grid cell i, then we fill out the variables with the prefix "point." with
# the information frmo those villages
if(nrow(temp.point) > 0) {
pre.panel[i, "village.code"] <- temp.point$village.code[which.min(temp.point$earliest.end.date)]
pre.panel[i, "village.name"] <- temp.point$village.name[which.min(temp.point$earliest.end.date)]
pre.panel[i, "province.name"] <- temp.point$province.name[which.min(temp.point$earliest.end.date)]
pre.panel[i, "district.name"] <- temp.point$district.name[which.min(temp.point$earliest.end.date)]
pre.panel[i, "commune.name"] <- temp.point$commune.name[which.min(temp.point$earliest.end.date)]
pre.panel[i, "point.earliest.end.date"] <- temp.point$earliest.end.date[which.min(temp.point$earliest.end.date)]
pre.panel[i, "point.enddate.type"] <- temp.point$enddate.type[which.min(temp.point$enddate.type)]
pre.panel[i, "point.earliest.sector.num"] <- temp.point$earliest.sector.num[which.min(temp.point$earliest.end.date)]
pre.panel[i, "point.earliest.sector"] <- temp.point$earliest.sector[which.min(temp.point$earliest.end.date)]
# computing the total number of projects each year for the villages within grid cell i
for(j in sort(unique(treatment$earliest.end.date))) {
pre.panel[i, grep(paste0("point.count", j), (names(pre.panel)))] <-
as.data.frame(temp.point[, paste0("count", c(2003:2017)[2003:2017<=j])]) %>%
apply(., 2, sum, na.rm=T) %>%
sum()
}
}
# if there are one or more villages bordering grid cell i, then we fill out the variables with the prefix "box." with
# the information frmo those villages
if(nrow(temp.box) > 0) {
pre.panel[i, "village.code"] <- temp.box$village.code[which.min(temp.box$earliest.end.date)]
pre.panel[i, "village.name"] <- temp.box$village.name[which.min(temp.box$earliest.end.date)]
pre.panel[i, "province.name"] <- temp.box$province.name[which.min(temp.box$earliest.end.date)]
pre.panel[i, "district.name"] <- temp.box$district.name[which.min(temp.box$earliest.end.date)]
pre.panel[i, "commune.name"] <- temp.box$commune.name[which.min(temp.box$earliest.end.date)]
pre.panel[i, "box.earliest.end.date"] <- temp.box$earliest.end.date[which.min(temp.box$earliest.end.date)]
pre.panel[i, "box.enddate.type"] <- temp.box$enddate.type[which.min(temp.box$enddate.type)]
pre.panel[i, "box.earliest.sector.num"] <- temp.box$earliest.sector.num[which.min(temp.box$earliest.end.date)]
pre.panel[i, "box.earliest.sector"] <- temp.box$earliest.sector[which.min(temp.box$earliest.end.date)]
# computing the total number of projects each year for the villages bordering grid cell i
for(j in sort(unique(treatment$earliest.end.date))) {
pre.panel[i, grep(paste0("box.count", j), (names(pre.panel)))] <-
as.data.frame(temp.box[, paste0("count", c(2003:2017)[2003:2017<=j])]) %>%
apply(., 2, sum, na.rm=T) %>%
sum()
}
}
# merging the grid cell data for grid cell i with the pre.panel dataset for grid cell i
pre.panel[i, which(names(pre.panel) %in% names(grid_1000_matched_data))] <- grid_1000_matched_data[i,]
if(is.na(pre.panel$village.code[i])) {
villages <- c(id.list[[i]], id.list2[[i]]) %>% .[!.==""]
temp <- shape[shape$village.code==villages[1],]
if(nrow(temp)>0) {
pre.panel[i, "village.code"] <- temp$village.code
pre.panel[i, "village.name"] <- temp$village.name
pre.panel[i, "province.name"] <- temp$province.name
pre.panel[i, "district.name"] <- temp$district.name
pre.panel[i, "commune.name"] <- temp$commune.name
}
}
if(i %% 1000 == 0){cat(i, "of", nrow(grid_1000_matched_data), "\n")}
}
for(i in grep("count", names(pre.panel))) {pre.panel[which(is.na(pre.panel[,i])), i] <- 0}
pre.panel$unique.commune.name <- paste(pre.panel$province.name, pre.panel$district.name, pre.panel$commune.name)
# editing variable names to make data reshaping easier
names(pre.panel) <- gsub("v4composites_calibrated_201709.", "ntl_", names(pre.panel)) %>% gsub(".mean", "", .)
###################
# reshaping cross sectional data into a panel structure with time dimension being years 1992:2013 and the panel variable being
# cell id
merge_grid_1000_lite.uncalibrated <- read.csv("inputdata/village_grid_files/merge_grid_1000_lite_uncalibrated.csv",
stringsAsFactors = F)
pre.panel <- merge(pre.panel, merge_grid_1000_lite.uncalibrated, by = "cell_id")
names(pre.panel) <- gsub("v4composites.", "ntl_", names(pre.panel))
names(pre.panel) <- gsub("\\.mean", "_uncalibrated", names(pre.panel))
panel <- reshape(data = pre.panel, direction = "long", varying = list(paste0("ntl_", 1992:2013),
paste0("point.count", 1992:2013),
paste0("box.count", 1992:2013),
paste0("ntl_", 1992:2013, "_uncalibrated")),
idvar = "panel_id", sep = "_", timevar = "year")
panel <- panel[, !(names(panel) %in% c(paste0("point.count", 2014:2017), paste0("box.count", 2014:2017), "dist_to_water.na",
"dist_to_groads.na", "id", "panel_id", "village_box_ids", "village_point_ids"))]
names(panel)[names(panel)=="village.code"] <- "village_code"
names(panel)[names(panel)=="village.name"] <- "village_name"
names(panel)[names(panel)=="province.name"] <- "province_name"
names(panel)[names(panel)=="district.name"] <- "district_name"
names(panel)[names(panel)=="commune.name"] <- "commune_name"
names(panel)[names(panel)=="box.earliest.end.date"] <- "border_cell_earliest_enddate"
names(panel)[names(panel)=="point.earliest.end.date"] <- "intra_cell_earliest_enddate"
names(panel)[names(panel)=="box.enddate.type"] <- "border_cell_enddate_type"
names(panel)[names(panel)=="point.enddate.type"] <- "intra_cell_enddate_type"
names(panel)[names(panel)=="box.earliest.sector.num"] <- "border_cell_earliest_sector_num"
names(panel)[names(panel)=="point.earliest.sector.num"] <- "intra_cell_earliest_sector_num"
names(panel)[names(panel)=="box.earliest.sector"] <- "border_cell_earliest_sector"
names(panel)[names(panel)=="point.earliest.sector"] <- "intra_cell_earliest_sector"
names(panel)[names(panel)=="unique.commune.name"] <- "unique_commune_name"
names(panel)[names(panel)=="ntl_1992"] <- "ntl"
names(panel)[names(panel)=="point.count1992"] <- "intra_cell_count"
names(panel)[names(panel)=="box.count1992"] <- "border_cell_count"
names(panel)[names(panel)=="ntl_1992_uncalibrated"] <- "ntl_uncalibrated"
# write.csv(panel, file = "/Users/christianbaehr/Box Sync/cambodia_eba_gie/ProcessedData/panel_multi-village_projectsONLY.csv", row.names = F)