google_trends_UI_statepolicy.Rmd

---
title: "google_trends_UI_statepolicy"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)

# if (!require("devtools")) install.packages("devtools")
# devtools::install_github("paulgp/gtrendsR")

#install.packages("RApiDatetime")

library(gtrendsR)
library(tidyverse)
library(ggrepel)
library(RApiDatetime)
library(lubridate)
library(zoo)
library(knitr)
library(kableExtra)
library(readxl)
library(janitor)
library(RColorBrewer)
library(data.table)
library(plm)
library(directlabels)
```

# State Policy Actions

In recent weeks, Governors across the US have issued executive orders requiring that businesses change or limit their in-person operations. In most states, restaurants are only permitted to operate take-out and/or delivery services. While some restaurants have opted to stay open and attempt to weather the storm with reduced operations, others have shut down entirely. This has resulted in mass layoffs in the food service industry across the country, contributing to the skyrocketing UI claims numbers. 

Further, in some states, Governors have called for all non-essential businesses to close. Many shops and malls, theaters and museums, gyms and yoga studios, and concert halls and clubs have been ordered to cease operations. While some businesses can scale up their e-commerce operations or survive on take-out orders alone, many are forced to lay off their workers. 

The timing of states' closure orders have varied, so we've put together some maps to show the timing.

##Orders Affecting Business Operations and Closures
```{r, echo=FALSE, warning=FALSE, message=FALSE}

###### Liz doing state policy stuff #######

###Make maps of timing of state policies ###

#Set up data

##Import policy data
data_policy_states <- read_xlsx("data/covid_closures_by_state.xlsx") %>%
  mutate_at(c("partial", "school", "complete"), funs(ymd))%>%
  mutate(complete_week = epiweek(complete), partial_week = epiweek(partial), school_week = epiweek(school)) %>%
  mutate(complete_day = day(complete), partial_day = day(partial), school_day = day(school)) %>%
  mutate(region = tolower(state)) %>%
  rename("location" = "state_abbr")


##Add shape files
states <- map_data("state")
dates_map <- left_join(states, data_policy_states, by = "region")

#Map of complete closure orders
dates_map$complete <- format(dates_map$complete, format = "%m/%d")
complete_map <- 
  ggplot(dates_map, aes(long, lat, group = group)) + 
  geom_polygon(aes(fill = factor(complete)), color = "white") + 
  ggtitle("Non-Essential Business Closure Orders") +
  scale_fill_manual(values = colorRampPalette(brewer.pal(15, "BuPu"))(15), name = "Effective Date", na.value = "grey50", guide = "legend", drop = "TRUE") +
  theme(axis.text.x = element_blank(),
        axis.text.y = element_blank(),
        axis.ticks = element_blank(),
        rect = element_blank(),
        axis.title.y=element_blank(), 
        axis.title.x=element_blank(),
        plot.title = element_text(hjust = 0.5, size = 18)) 


#Map of partial closure orders
dates_map$partial <- format(dates_map$partial, format = "%m/%d")
takeout_map <- 
  ggplot(dates_map, aes(long, lat, group = group)) + 
  geom_polygon(aes(fill = factor(partial)), color = "white") + 
  ggtitle("Prohibition on Dining In and Partial Closure Orders") +
  scale_fill_manual(values = colorRampPalette(brewer.pal(15, "GnBu"))(15), name = "Effective Date", na.value = "grey50", guide = "legend", drop = "TRUE") +
  theme(axis.text.x = element_blank(),
        axis.text.y = element_blank(),
        axis.ticks = element_blank(),
        rect = element_blank(),
        axis.title.y=element_blank(), 
        axis.title.x=element_blank(),
        plot.title = element_text(hjust = 0.5, size = 18)) 

complete_map
takeout_map
```

We define partial closures as orders that prohibited restaurants from having patrons dine in. In many states, other provisions were included in these orders (e.g., limits on gathering sizes, requirements for bars/clubs to close). We use the switch to take-out only for consistency. 

We define complete closures as orders for non-essential businesses to close. States vary in their definitions of essential businesses, and some states also put stay at home or shelter in place orders. For consistency, we hone in specifically on the non-essential business closures. 

Both of these definitions are consistent with analyzing policy effects on UI claims and search activity -- as businesses close, layoffs increase. 


```{r, echo=FALSE, warning=FALSE, message=FALSE}

###Event Studies###

#Set up data for event studies

##Merge policy data with google search data 
policy_event_study <- left_join(data_states_short, data_policy_states, by = "location") %>%
  filter(date > as.Date("2020-03-01") ) %>%
  mutate(daystopartial = difftime(date, partial , units = c("days")))  %>%
  mutate(daystocomplete = difftime(date, complete , units = c("days"))) %>%
  group_by(location) %>%
  mutate(laghits=dplyr::lag(hits, order_by = location)) %>%
  ungroup %>%
  pivot_longer(
    cols = c(daystopartial, daystocomplete),
    names_to = "type",
    values_to = "days"
  ) 

##Define event study pre and post length
es_start = -7
es_end = 4
es_length = 11

## This sets the baseline period to t-1, change by making ref = 8 for t0 or ref = 6 for t-2
policy_event_study_releveled <- policy_event_study %>%
  filter(days %in% es_start:es_end) %>%
  mutate(days_factor = relevel(as.factor(days), ref = 7)) 

#Event study around partial closure policies
partial_es <- plm(hits ~ as.factor(days_factor), 
             data = subset(policy_event_study_releveled, type == "daystopartial"),
             index = "location", 
             effect = "individual",
             model = "within")

partial_coefs = as.data.frame(summary(partial_es)$coefficients[1:es_length]) %>%
  mutate(closuretype = "partial") 
colnames(partial_coefs)[1] <- "coeffecients"
partial_coefs$timefromorder <- es_start:(es_end-1)
partial_coefs$timefromorder <- ifelse(partial_coefs$timefromorder >=-1, partial_coefs$timefromorder + 1, partial_coefs$timefromorder)

#Plot the event study
partial_es_plot <- ggplot(data=partial_coefs, aes(x=timefromorder, y=coeffecients))+
  geom_line(,color = "darkgray") + 
  geom_vline(xintercept = 0, color = "mediumpurple2", size = 1.5, alpha = .75) +
  labs(y = "Change in Google Search Intensity",
       x = "Days from Partial Closure Order",
       title="Search Trends in Response to Partial Closures") +
  theme_classic(base_size=15) +
  scale_x_continuous(breaks = seq(es_start, es_end, by = 1))


#Event study around complete closure policies
complete_es <- plm(hits ~ as.factor(days_factor), 
             data = subset(policy_event_study_releveled, type == "daystocomplete"),
             index = "location", 
             effect = "individual",
             model = "within")

complete_coefs <- as.data.frame(summary(complete_es)$coefficients[1:es_length]) %>%
  mutate(closuretype = "complete")
colnames(complete_coefs)[1] <- "coeffecients"
complete_coefs$timefromorder <- es_start:(es_end-1)
complete_coefs$timefromorder <- ifelse(complete_coefs$timefromorder >=-1, complete_coefs$timefromorder + 1, complete_coefs$timefromorder)

#Plot the event study
complete_es_plot <- ggplot(data=complete_coefs, aes(x=timefromorder, y=coeffecients))+
  geom_line(,color = "darkgray") + 
  geom_vline(xintercept = 0, color = "cadetblue3", size = 1.5, alpha = .75) +
  labs(y = "Change in Google Search Intensity",
       x = "Days from Complete Closure Order",
       title="Search Trends in Response to Complete Closures") +
  theme_classic(base_size=15) +
  scale_x_continuous(breaks = seq(es_start, es_end, by = 1))


#Plot both event studies together 
both_es_coefficients <- complete_coefs %>%
  rbind(partial_coefs)
labs <- setNames(c(" Partial", " Complete"), c("partial", "complete"))
ggplot(both_es_coefficients, aes(x = timefromorder, y = coeffecients, group = closuretype, color = closuretype)) +
  geom_line() + 
  geom_vline(aes(xintercept = 0), alpha = .3, size = 1.5, color = "black") +
  theme_classic(base_size=15) + 
  labs(y = "Change in Google Search Intensity",
       x = "Days from Closure Order", 
       title = "Search Trends in Response to Closures",
       color = "Closure Type") +
  scale_color_manual(values = c("cadetblue3", "mediumpurple2"), guide = FALSE) + 
  scale_x_continuous(breaks = seq(es_start,es_end, by = 1)) + 
  geom_dl(aes(label=labs[closuretype]), method="last.points") +
  coord_cartesian(clip = "off") +
  theme(plot.margin = margin(1, 45, 1, 1))

partial_es_plot
complete_es_plot


```




```{r, echo=FALSE, warning=FALSE, message=FALSE}

###Graph average google search numbers over time (relative to policy enactment day), weighted by state labor force###

#!! Before Paul's 4/6 email
#Set up data

##First pull in state labor force data 
labor_force_states = read_xlsx("data/labor_force_state.xlsx") 

##Merge with policy data
policy_search_lf <- left_join(policy_event_study, labor_force_states, by = "state") %>% 
  #Collapse hits for the days around policies, weighted by lf size
  group_by(type, days) %>% 
  summarize_at("hits", list(weighted.mean), weight = "lf_022020") %>%
  filter(days %in% -7:7) 

#Plot (first individviually, then together)
ggplot(data = policy_search_lf %>% filter(type == "daystopartial")) + 
  geom_line(aes(x = days, y = hits), color = "black") + 
  geom_vline(aes(xintercept = 0, show.guide = TRUE), alpha = .5, size = 1.5, color = "mediumpurple2") +
  theme_classic(base_size=10) + 
  labs(x = "Days from Partial Closure",
       y = "Google Search Intensity", 
       title = "Average Search Intensity Around Time of Partial Closure",
       subtitle = "Weighted by size of states' labor force in Feb. 2020") +
  scale_x_continuous(breaks = seq(-7, 7, by = 1))

ggplot(data = policy_search_lf %>% filter(type == "daystocomplete")) + 
  geom_line(aes(x = days, y = hits), color = "black") + 
  geom_vline(aes(xintercept = 0, show.guide = TRUE), alpha = .5, size = 1.5, color = "cadetblue3") +
  theme_classic(base_size=10) + 
  labs(x = "Days from Partial Closure",
       y = "Google Search Intensity", 
       title = "Average Search Intensity Around Time of Partial Closure",
       subtitle = "Weighted by size of states' labor force in Feb. 2020") +
  scale_x_continuous(breaks = seq(-7, 7, by = 1))

labs <- setNames(c(" Partial", " Complete"), c("daystopartial", "daystocomplete"))
ggplot(policy_search_lf, aes(x = days, y = hits, group = type, color = type)) +
  geom_line() + 
  geom_vline(aes(xintercept = 0), alpha = .3, size = 1.5, color = "black") +
  theme_classic(base_size=10) + 
  labs(x = "Days from Closure Effective Date",
       y = "Google Search Intensity", 
       title = "Average Search Intensity Around Time of Closures",
       subtitle = "Weighted by size of states' labor force in Feb. 2020", 
       color = "Closure Type") +
  scale_color_manual(values = c("cadetblue3", "mediumpurple2"), guide = FALSE) + 
  scale_x_continuous(breaks = seq(-7, 7, by = 1)) +
  geom_dl(aes(label=labs[type]), method="last.points") +
  coord_cartesian(clip = "off") +
  theme(plot.margin = margin(1, 45, 1, 1))

 

```


```{r, echo=FALSE, warning=FALSE, message=FALSE}

###Graph average google search numbers over time (relative to policy enactment day), weighted by state labor force###

#Based on Paul's 4/6 email: 

#Partial Treatment States:
data_policy_states %>% 
  filter(partial <= as.Date("2020-03-16")) %>%
  subset(select = state)
#Partial Control States:       
data_policy_states %>% 
  filter(is.na(partial) | partial >= as.Date("2020-03-24")) %>%
  subset(select = state)     
#Complete Treatment States:
data_policy_states %>% 
  filter(complete <= as.Date("2020-03-23")) %>%
  subset(select = state)
#Complete Control States:       
data_policy_states %>% 
  filter(is.na(complete) | complete >= as.Date("2020-03-31")) %>%
  subset(select = state)      


##Set up data
policy_search_lf_filtered1 <- data_policy_states %>%
  #Filter to treatment and control
    mutate(complete_treatment = case_when(
                          complete <= as.Date("2020-03-23") ~ 1,
                          complete >= as.Date("2020-03-31") ~ 0,
                          complete = is.na(complete) ~ 0)) %>%
    mutate(partial_treatment = case_when(
                          partial <= as.Date("2020-03-16") ~ 1,
                          partial >= as.Date("2020-03-24") ~ 0,
                          partial = is.na(partial) ~ 0)) %>%
    filter(!is.na(partial_treatment) | !is.na(complete_treatment)) %>%
    mutate(partial_treatment = as.factor(partial_treatment)) %>%
    mutate(complete_treatment = as.factor(complete_treatment)) %>%
  #Add in google search data
    left_join(data_states_short, by = "location") %>%
  #Filter out lots of pre-data
    filter(date > as.Date("2020-03-10") ) %>%
  #Reshape for graphs later
    pivot_longer(
      cols = c(complete_treatment, partial_treatment),
      names_to = "group",
      values_to = "treatment",
      names_repair = "minimal") %>%
  #Merge in labor force data for weighting 
    left_join(labor_force_states, by = "state") %>% 
  #Collapse hits by date by group, weighted by lf size
    group_by(date, group, treatment) %>% 
    summarize_at("hits", list(weighted.mean), weight = "lf_022020") %>%
  #Remove lines that aren't treatment or control
    filter(!is.na(treatment))

##Plot (separately)
ggplot(data = policy_search_lf_filtered1 %>% filter(group == "partial_treatment")) + 
  geom_line(aes(x = date, y = hits, group = treatment, color = treatment)) + 
  annotate("rect", xmin = min(data_policy_states$partial, na.rm = T), xmax = as.Date("2020-03-16"), min = 0, ymax = Inf, alpha = .2, fill = "mediumpurple2") + 
  theme_classic(base_size=10) + 
  labs(x = "Date",
       y = "Google Search Intensity", 
       title = "Average Search Intensity Around Time of Partial Closure",
       subtitle = "Weighted by size of states' labor force in Feb. 2020") +
  scale_color_manual(values = c("black", "mediumpurple2"), guide = FALSE) +
  scale_x_date(date_breaks = "2 days", 
                 date_labels = "%m-%d",
                 limits = as.Date(c("2020-03-10","2020-03-24")))


ggplot(data = policy_search_lf_filtered1 %>% filter(group == "complete_treatment")) + 
  geom_line(aes(x = date, y = hits, group = treatment, color = treatment)) + 
  annotate("rect", xmin = min(data_policy_states$complete, na.rm = T), xmax = as.Date("2020-03-23"), ymin = 0, ymax = Inf, alpha = .2, fill = "cadetblue3") + 
  theme_classic(base_size=10) + 
  labs(x = "Date",
       y = "Google Search Intensity", 
       title = "Average Search Intensity Around Time of Complete Closure",
       subtitle = "Weighted by size of states' labor force in Feb. 2020") +
  scale_color_manual(values = c("black", "cadetblue3"), guide = FALSE) +
  scale_x_date(date_breaks = "2 days", 
                 date_labels = "%m-%d",
                 limits = as.Date(c("2020-03-10","2020-03-31")))


#This time, allow 3 treatment groups: 23rd, 24th, and 25th enactors, and control group is states who don't enact by 4/1  and only do for complete

##Set up data
policy_search_lf_filtered2 <- data_policy_states %>%
  #Filter to treatment and control
    mutate(complete_treatment = case_when(
                          complete == as.Date("2020-03-23") ~ "23rd",
                          complete == as.Date("2020-03-24") ~ "24th",
                          complete == as.Date("2020-03-25") ~ "25th",
                          complete >= as.Date("2020-04-02") ~ "Control",
                          complete = is.na(complete) ~ "Control")) %>%
    filter(!is.na(complete_treatment)) %>%
  #Add in google search data
    left_join(data_states_short, by = "location") %>%
  #Filter out lots of pre-data
    filter(date > as.Date("2020-03-10") ) %>%
  #Reshape for graphs later
    pivot_longer(
      cols = c(complete_treatment),
      names_to = "group",
      values_to = "treatment",
      names_repair = "minimal") %>%
  #Merge in labor force data for weighting 
    left_join(labor_force_states, by = "state") %>% 
  #Collapse hits by date by group, weighted by lf size
    group_by(date, group, treatment) %>% 
    summarize_at("hits", list(weighted.mean), weight = "lf_022020") %>%
  #Remove lines that aren't treatment or control
    filter(!is.na(treatment))

##Plot
ggplot(data = policy_search_lf_filtered2) + 
  geom_line(aes(x = date, y = hits, group = treatment, color = treatment)) + 
  annotate("rect", xmin = as.Date("2020-03-23 00:00:00"), xmax = as.Date("2020-03-24 00:00:00"), min = 0, ymax = Inf, alpha = .2, fill = "palevioletred") + 
    annotate("rect", xmin = as.Date("2020-03-24 00:00:00"), xmax = as.Date("2020-03-25 00:00:00"), min = 0, ymax = Inf, alpha = .2, fill = "dodgerblue") + 
  annotate("rect", xmin = as.Date("2020-03-25 00:00:00"), xmax = as.Date("2020-03-26 00:00:00"), min = 0, ymax = Inf, alpha = .2, fill = "seagreen4") + 

  theme_classic(base_size=10) + 
  labs(x = "Date",
       y = "Google Search Intensity", 
       title = "Average Search Intensity Around Time of Complete Closure",
       subtitle = "Weighted by size of states' labor force in Feb. 2020") +
  scale_color_manual(values = c("palevioletred", "dodgerblue", "seagreen4", "black"), labels = c("23rd", "24th", "25th", "Control"), name = "Group") +
  scale_x_date(date_breaks = "2 days", 
                 date_labels = "%m-%d",
                 limits = as.Date(c("2020-03-10","2020-03-31")))
 


#This time, compare 3/23 complete enactments to states that hadn't enacted a complete closure by 3/31 (within 1 week). 

##Set up data
policy_search_lf_filtered3 <- data_policy_states %>%
  #Filter to treatment and control
    mutate(complete_treatment = case_when(
                          complete == as.Date("2020-03-23") ~ "Treatment",
                          complete >= as.Date("2020-03-31") ~ "Control",
                          complete = is.na(complete) ~ "Control")) %>%
    filter(!is.na(complete_treatment)) %>%
  #Add in google search data
    left_join(data_states_short, by = "location") %>%
  #Filter out lots of pre-data
    filter(date > as.Date("2020-03-10") ) %>%
  #Merge in labor force data for weighting 
    left_join(labor_force_states, by = "state") 
  #Collapse hits by date by group, weighted by lf size, for control group
  policy_search_lf_filtered3_c <- policy_search_lf_filtered3 %>%
    filter(complete_treatment == "Control") %>%
    mutate(location = "Control") %>%
      group_by(date, location, complete_treatment) %>% 
      summarize_at("hits", list(weighted.mean), weight = "lf_022020") 
  #Merge collapsed control group data back in 
  policy_search_lf_filtered3 <- policy_search_lf_filtered3 %>%
    filter(complete_treatment == "Treatment") %>% 
    subset(select = c("location", "hits", "date", "complete_treatment", "partial", "complete")) %>%
    full_join(policy_search_lf_filtered3_c)
  
##Plot
my_blue = brewer.pal(n = 9, "Blues")[3:9]

  ggplot() +
  geom_line(data = policy_search_lf_filtered3 %>% filter(location == "Control"), 
            aes(x = date, y = hits, group = location, color = location), lwd=1.5) + 
  geom_line(data = policy_search_lf_filtered3, 
            aes(x = date, y = hits, group = location, color = location)) + 
  annotate("rect", xmin = as.Date("2020-03-22"), xmax = as.Date("2020-03-24"), min = 0, ymax = Inf, alpha = .1, fill = "black") + 
  geom_point(data = policy_search_lf_filtered3 %>% filter(date == partial & location != "Control"),
             aes(x = partial, y = hits, group = location, color = location)) +
  theme_classic(base_size=10) + 
  labs(x = "Date",
       y = "Google Search Intensity", 
       title = "Search Intensity Around Time of Complete Closure",
       subtitle = "Weighted by size of states' labor force in Feb. 2020") +
  scale_colour_manual(values=c("black", my_blue), name = "State") +
  scale_x_date(date_breaks = "2 days", 
                 date_labels = "%m-%d",
                 limits = as.Date(c("2020-03-10","2020-03-31"))) + 
  labs(caption = "Note: Control group includes states that had not enacted a complete closure by 3/30. \nPoint indicates date when state enacted partial closure.") + 
  theme(plot.caption = element_text(hjust = 0, size = 9))


  
```

```{r, echo=FALSE, warning=FALSE, message=FALSE}

###Plot one state with both policies as an example (add daily UI first)###

#Set these to whatever you want! note to liz: make this a function later
  plotstate = "CA" #Change this to see another state
  cares = as.Date("2020-03-27")

#Merge in daily UI data
  policy_event_study_withdailyui <- right_join(daily_UI_Claims, policy_event_study, by = c("location", "date"))
  policy_event_study_plotstate <- policy_event_study_withdailyui %>% 
    filter(location == plotstate)

#Plot daily trends data
dailygoogleplot_plotstate <- ggplot(data = policy_event_study_plotstate, aes(x=date, y=hits)) + 
  geom_line(,color = "darkgray") + 
  geom_vline(aes(xintercept = mean(partial), color = "Partial Closure"), show.guide = TRUE, alpha = .75, size = 1.5) + 
  geom_vline(aes(xintercept = mean(complete), color = "Complete Closure"), show.guide = TRUE, alpha = .75, size = 1.5) +
    geom_vline(aes(xintercept = cares, color = "CARES Act"), show.guide = TRUE, alpha = .75, size = 1.5) +
    scale_color_manual(name = "Policies", values = c("Partial Closure" = "cadetblue3", "Complete Closure" = "mediumpurple2", "CARES Act" = "palegreen4")) +
  theme_classic(base_size=10) + 
  theme(legend.position = "bottom") +
  labs(x = "Time",
       y = "Google Search Trends", 
       title = paste("Google Search Trends Over Time in", plotstate))

#Plot daily UI claims with policies (this is pretty useless right now, because we don't have enough daily data)
dailyuiplot_plotstate <- ggplot(data = policy_event_study_plotstate, aes(x=date, y=ui_claims_daily)) + 
  geom_line(,color = "darkgray") + 
  geom_vline(aes(xintercept = mean(partial), color = "Partial Closure"), show.guide = TRUE, alpha = .75, size = 1.5) + 
  geom_vline(aes(xintercept = mean(complete), color = "Complete Closure"), show.guide = TRUE, alpha = .75, size = 1.5) +
      geom_vline(aes(xintercept = cares, color = "CARES Act"), show.guide = TRUE, alpha = .75, size = 1.5) +
    scale_color_manual(name = "Policies", values = c("Partial Closure" = "cadetblue3", "Complete Closure" = "mediumpurple2", "CARES Act" = "palegreen4")) +
  theme_classic(base_size=10) + 
  theme(legend.position = "bottom") +
  labs(x = "Time",
       y = "Daily UI Claims", 
       title = paste("Daily UI Claims Over Time in", plotstate))

dailygoogleplot_plotstate
dailyuiplot_plotstate

```

```{r, echo=FALSE, warning=FALSE, message=FALSE}

###Map pct change in unemployment claims pre-covid to during-covid###

#Set up data

##First bring in weekly claims data prior to COVID 
weekly_UI_janthrumarch = as.data.frame(read_xlsx("data/UI_claims_jan1thrumarch21.xlsx")) %>%
  rename("state" = "State") %>%
  rename("week" = "Filed week ended") %>%
  rename("claims" = "Initial Claims") %>%
  mutate_at("week", funs(ymd)) %>%
  mutate(week = epiweek(week)) %>%
  subset(select = c("state", "week", "claims")) %>%
  left_join(data_policy_states[, c("state", "location")], by = "state")

##Clean data from week ending March 28 (4/2/20 news release) !! Will need to update weekly with new data pull
week13ui <- subset(UI_Claims_March28_True, select = c("location", "advance_0322")) %>%
  mutate(week = 13) %>%
  rename("claims" = "advance_0322") %>%
  left_join(data_policy_states[, c("state", "location")], by = "location")

##Merge together to get all weekly UI data available
weekly_UI_janthrumarch <- rbind(weekly_UI_janthrumarch, week13ui, by = "location")

##Merge with policy data at week-level
weekly_claims_policy <- left_join(weekly_UI_janthrumarch, data_policy_states[, c("partial_week", "complete_week", "location")], by = "location")

##Filter out data before mid-Feb
weekly_claims_policy <- weekly_claims_policy %>%
    group_by(location) %>%
    mutate(week = as.numeric(week)) %>%
    arrange(location ,week) %>%
    mutate(claims = as.numeric(claims))  %>%
    filter(week >= 8)   %>%
    filter(!is.na(location))

##Reshape to wide
weekly_claims_policy_wide <- weekly_claims_policy %>% 
    spread(week, claims) %>%
    rename("week13" = "13") %>%
    rename("week12" = "12") %>%
    rename("week11" = "11") %>%
    rename("week10" = "10") %>%
    rename("week9" = "9") %>%
    rename("week8" = "8") 

#Replace OR data manually (Oregon DOL economist asked me to run this map with updated data, commenting out)
  #weekly_claims_policy_wide2[38, "week13"] <- 92700
  #weekly_claims_policy_wide2[38, "week12"] <- 76500
  #weekly_claims_policy_wide2[38, "week11"] <- 4900

## Generate percent change in claims (3 week moving average)
weekly_claims_policy_wide <- weekly_claims_policy_wide %>% 
    mutate(claims_postcovid = as.numeric(week11)+as.numeric(week12)+as.numeric(week13)) %>%
    mutate(claims_precovid = as.numeric(week10)+as.numeric(week9)+as.numeric(week8)) %>%
    mutate(claims_pchg = 100*((claims_postcovid - claims_precovid)/claims_precovid))  %>% 
    mutate(region = tolower(state))

##Set up shape files
states <- map_data("state")
claims_chg_map_data <- left_join(states, weekly_claims_policy_wide, by = "region")

#Do some stuff to get pattern overlaid on map for complete closures by 3/2
nonessential_closure_states <- data_policy_states %>%
  filter(!is.na(complete)) %>%
  filter(complete <= "2020-03-28") %>%
  mutate(region = tolower(state)) %>%
  subset(select = region) 
nonessential_closure_states <- as.data.frame(nonessential_closure_states)
source("https://raw.githubusercontent.com/imaddowzimet/drawcrosshatch/master/draw_crosshatch.R")
claims_chg_map_data_completeclose <- claims_chg_map_data %>%
  filter(region %in% nonessential_closure_states$region) %>%
  group_by(group) %>% 
  nest() 
lines <- map_df(claims_chg_map_data_completeclose$data, draw.crosshatch, width = .3, pattern= "horizontal")

#Make map
ui_pctchg_map <- ggplot(claims_chg_map_data, aes(long, lat, group = group)) + 
  geom_polygon(aes(fill = claims_pchg), color = "white") + 
  geom_segment(data=lines, aes(x= x, y = y , xend = xend, yend = yend), 
                    inherit.aes = F, color = "white", alpha = .6) +
  labs(title="Percent Change in Initial UI Claims", 
       subtitle = "Between 2/16-3/7 to 3/8-3/28") +
  scale_fill_gradient2( 
      high = "orangered3", na.value = "gray50",  
      breaks = c(round(min(claims_chg_map_data$claims_pchg)),
                round(max(claims_chg_map_data$claims_pchg)))) +
  labs(caption = "Note: States with lines adopted a non-essential business closure order by 3/28.") +
  theme(axis.text.x = element_blank(),
        axis.text.y = element_blank(),
        axis.ticks = element_blank(),
        rect = element_blank(),
        axis.title.y=element_blank(), 
        axis.title.x=element_blank(), 
        legend.title=element_blank(),
        legend.position = "bottom",
        plot.title = element_text(hjust = 0.5, size = 18),
        plot.subtitle = element_text(hjust = 0.5, size = 14),
        plot.caption = element_text(hjust = 0.5, size = 12))

ui_pctchg_map

```

```{r, echo=FALSE, warning=FALSE, message=FALSE}

#This all kind of ends up being useless-- should play around with more 
#Now look at weekly search trends and UI data by week of policy adoption 

#There's really nothing interesting here. A few late-adopting states seem to have big increases in week 13, but I think this is more to do with technology issues (so claims were late). 

#ggplot(subset(weekly_claims_policy, complete_week == 13), aes(x = week, y = claims_pchg, color = state)) + geom_path()

#ggplot(subset(weekly_claims_policy, complete_week == 12), aes(x = week, y = claims_pchg, color = state)) + geom_path()

#ggplot(subset(weekly_claims_policy, partial_week == 12), aes(x = week, y = claims_pchg, color = state)) + geom_path()

#ggplot(subset(weekly_claims_policy, partial_week == 13), aes(x = week, y = claims_pchg, color = state)) + geom_path()

#ggplot(weekly_claims_policy, aes(x = week, y = claims_pchg, color = partial_week)) + geom_path()


```