Major push of all files

R/data_vis.r

@@ -1,45 +1,94 @@
 library(ggplot2)
 library(fable)
 
+#Import data----
 # Import data from github repo.
 df_og <- read.csv("https://raw.githubusercontent.com/HPCurtis/causalcovidcattle/main/data/cattle_og.csv")
 df_sa <- read.csv("https://raw.githubusercontent.com/HPCurtis/causalcovidcattle/main/data/cattle_sa.csv")
 
-df_og$Date <- as.Date(df_og$Date)
-df_sa$Date <- as.Date(df_sa$Date)
+df_og$Date <- as.Date(df_og$Date) 
+df_sa$Date <- as.Date(df_sa$Date) 
 
-df_og <- df_og[,1:2] |> mutate(Date = yearmonth(Date))  |>
+#Pre/Post Covid----
+pre_covid <- df_sa %>%
+  drop_na() %>%
+  filter(Date < ymd("2020-03-01")) %>%
+  mutate(Date = yearquarter(Date)) %>%
   as_tsibble( index = Date)
 
-df_sa <- df_sa[,1:2] |> mutate(Date = yearmonth(Date))  |>
-  as_tsibble(index = Date)
+post_covid <- df_sa %>%
+  drop_na() %>%
+  filter(Date >=  ymd("2020-03-01") )%>%
+  mutate(Date = yearquarter(Date)) %>%
+  as_tsibble( index = Date)
 
 # Visualisations for the project
 
-# Visualise timeseries oforiginal total number of cattle slaughtered.
-ggplot(df_og, aes(x = Date, y = NumberSlaughteredCATTLEexclcalvesTotalState)) +
-  geom_line(color = "blue") +
-  labs(title = "Total number of slaughtered cattle (excl calves) across all Australian states",
-       x = "Date",
-       y = "Number of Slaughtered Cattle") +
-  theme_minimal()
+#Time series Decomposition------
+# Conduct STL decomposition.
+dcmp <- drop_na(df_sa) |>
+  model(stl = STL(NumberSlaughteredCATTLEexclcalvesTotalState))
+
+components(dcmp) |> autoplot()
 
-# Visualise timeseries of seasonally adjsuted total number of cattle slaughtered.
-ggplot(df_sa, aes(x = Date, y = NumberSlaughteredCATTLEexclcalvesTotalState)) +
-  geom_line(color = "blue") +
-  labs(title = "Total number of slaughtered cattle (excl calves) across all Australian states (seasonally adjusted",
-       x = "Date",
-       y = "Number of Slaughtered Cattle") +
-  theme_minimal()
 
-# Fit forecast model for seasonaly adjusted model.
-fit <- df_og |>
+# Fit forecast model for seasonal adjusted model.
+fitlinear <- pre_covid |>
   model(trend_model = TSLM(NumberSlaughteredCATTLEexclcalvesTotalState ~ trend()))
 
-fit |> forecast(h = 10) |> autoplot(df_og)
+fc <- fitlinear %>%
+  forecast(h = nrow(post_covid)) %>%
+  # Get forecast intervals
+  hilo()
 
-# Fit forecast model for seasonaly adjusted model.
-fitlinear <- df_sa |>
-  model(trend_model = TSLM(NumberSlaughteredCATTLEexclcalvesTotalState ~ trend()))
+fc$Date <- as_date(fc$Date)
+
+plinear <- fc %>% autoplot(drop_na(df_sa)) +
+  labs(y = "Number of Cattle Slaughtered", 
+       title = "Total number of cattle (excl calves) across all Australian States") +
+  theme(
+    plot.title = element_text(size = 10),      # Title font size
+    axis.title = element_text(size = 8),       # Axis titles font size
+    axis.text = element_text(size = 7)         # Axis text font size
+  ) + geom_vline(xintercept = as.Date("2020-03-01"))
+
+# Save plot out
+ggsave(filename = "/home/harrison/Desktop/gitHubRepos/cattlecovidcausal/img/linearforecast.png", plot = plinear, width = 6, height = 4, units = "in", dpi = 300)
+
+#Calculate Causal impact----
+yhat <- fc$.mean
+yhatupper <- fc$`95%`$upper
+yhatlower <- fc$`95%`$lower
+
+# Clauate mean and lower and upper bounds
+Totalslaughteredimpact <- post_covid$NumberSlaughteredCATTLEexclcalvesTotalState - yhat 
+Totalslaughteredimpactupper <- post_covid$NumberSlaughteredCATTLEexclcalvesTotalState - yhatupper 
+Totalslaughteredimpactlower <- post_covid$NumberSlaughteredCATTLEexclcalvesTotalState - yhatlower
+
+# Estimate impact of Covid over the forecast-able period.
+totalmean <- sum(Totalslaughteredimpact)
+totalupper <- sum(Totalslaughteredimpactupper)
+totallower <- sum(Totalslaughteredimpactlower)
+
+# PLot of causal impact 
+
+causal_impact_plot <- ggplot() +
+  geom_line(data = drop_na(df_sa), aes(x = Date, y = NumberSlaughteredCATTLEexclcalvesTotalState), color = "black") +
+  geom_ribbon(data = fc, 
+              aes(x = Date, ymin =  post_covid$NumberSlaughteredCATTLEexclcalvesTotalState),
+                  ymax = yhat, fill = "blue", alpha = 0.2) +
+  labs(y = "Number of Cattle Slaughtered", 
+       title = "Total number of cattle (excl calves) across all Australian States") +
+  theme(
+    plot.title = element_text(size = 10),      # Title font size
+    axis.title = element_text(size = 8),       # Axis titles font size
+    axis.text = element_text(size = 7)         # Axis text font size
+  ) +
+  geom_vline(xintercept = as.Date("2020-03-01"), color = "red", linetype = "dashed") +
+  theme_minimal()
 
-plinear <-fitlinear |> forecast(h = 10) |> autoplot(df_sa)
+# Save plots out-----
+ggsave(filename = "/home/harrison/Desktop/gitHubRepos/cattlecovidcausal/img/linearforecast.png",
+       plot = plinear, width = 6, height = 4, units = "in", dpi = 300)
+ggsave(filename = "/home/harrison/Desktop/gitHubRepos/cattlecovidcausal/img/causal_impact.png",
+       plot = causal_impact_plot, width = 6, height = 4, units = "in", dpi = 300)

README.md

@@ -1,8 +1,9 @@
 # Causal impact of Covid on Australian Cattle livestock slaughter numbers
 
-The following analysis is investigation into whether Covid had any potential causal impact of Covid on the Australian livestock market. Specifically, the analysis here uses an interupted time-series quasi-experimental methodology to whether the number of cattle slaughtered was impacted by the Covid-19 pandemic. The following work is highly inspired by the work of [Rami Kasparin](https://ramikrispin.github.io/2021/01/covid19-effect/), [Matheus Facure](https://matheusfacure.github.io/python-causality-handbook/landing-page.html) and the work of [CausalPy and all its developers](https://causalpy.readthedocs.io/en/stable/examples.html#interrupted-time-series) and generally [Hyndman & Athanasopoulos](https://otexts.com/fpp3/) incredible book and associated timeseries analysis packages that they developed thsat are simple brilliant.
+The following analysis is investigation into whether Covid had any potential causal impact of Covid on the Australian livestock market. Specifically, truethe analysis here uses an interupted time-series quasi-experimental methodology to whether the number of cattle slaughtered was impacted by the Covid-19 pandemic. The following work is highly inspired by the work of [Rami Kasparin](https://ramikrispin.github.io/2021/01/covid19-effect/), [Matheus Facure](https://matheusfacure.github.io/python-causality-handbook/landing-page.html) and the work of [CausalPy and all its developers](https://causalpy.readthedocs.io/en/stable/examples.html#interrupted-time-series) and generally [Hyndman & Athanasopoulos](https://otexts.com/fpp3/) incredible book and associated timeseries analysis packages that they developed thsat are simple brilliant.
 
-![Alt text](https://github.com/HPCurtis/causalcovidcattle/blob/main/img/linearforecast.png?raw=true)
+
+![Alt text](https://github.com/HPCurtis/causalcovidcattle/blob/main/img/causal_impact.png?raw=true)