# Load some useful libraries
library(ggplot2)
library(plyr)
library(dplyr)
library(tidyr)

# Fivethirtyeight is a company full of nerds (like us!) that have compiled lots of interesting data

# US births between 2000 - 2014
births <- read.csv('https://raw.githubusercontent.com/fivethirtyeight/data/master/births/US_births_2000-2014_SSA.csv', header=T)

# Take a look at the data
head(births)

# Add a column with a sequential index for plotting purposes
births$birthdate <- 1:nrow(births)

# Plot how birth number of births have changed over time
ggplot(births, aes(x=birthdate, y=births)) +
  geom_line()

# Funny periodicity! Let's split this by day of the week
ggplot(births, aes(x=birthdate, y=births, color=day_of_week)) +
  geom_point()

# Something funny is still going on here. What do day_of_week numbers correspond to?
# Hint: http://www.onthisday.com/date/2000/january/1
# Look at our data another way, and note that day_of_week is an int (makes more sense as factor)
tbl_df(births)

births$day_of_week <- factor(births$day_of_week)
levels(births$day_of_week) <- c('Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday', 'Sunday')

# Try plotting, now with factors
ggplot(births, aes(x=birthdate, y=births, color=day_of_week)) +
  geom_point()

# Looks like people don't like having babies on the weekends
# Let's look just at the year 2000 to try to figure out what the periodicity is all about
y2k <- subset(births, year==2000)

# Now you try: how many days are in the year 2000?


# Now you try: what is the average number of births in the year 2000?


# Look at some statistics by day of the week.
births %>%
  group_by(day_of_week) %>%
  mutate(mean_births=mean(births))

# Looks like weekends are an unpopular time to have a baby
t.test(subset(births, day_of_week %in% c('Saturday', 'Sunday'))$births, subset(births, !(day_of_week %in% c('Saturday', 'Sunday')))$births)

# Now you try: Are Saturday vs Sunday births significantly different?

# Plot single year's worth of births
ggplot(y2k, aes(x=birthdate, y=births, color=day_of_week)) +
  geom_point()

# Something funny going on at the end of the year. Same for 2001?
# Now you try: make a variable called year2001 with data only from the year 2001.


# Plot year 2001's data to see if end of the year also looks weird
ggplot(year2001, aes(x=birthdate, y=births, color=day_of_week)) +
  geom_point()

# Looks like people try to avoid having babies right around Christmas. 
# How many births were there over 15 years on Christmas?
subset(births, month==12&date_of_month==25)$births

# How many births were there over 15 years on June 25th?
subset(births, month==6&date_of_month==25)$births

# Let's see if there were more births on June 25 than December 25 (multiple criteria here!)
t.test(subset(births, month==6&date_of_month==25)$births, subset(births, month==12&date_of_month==25)$births)

# Now you try: are there fewer births on Christmas and Xmas eve than New Year's Eve and New Year's? 


# Silly superstition: Friday the 13th is unlucky. 
# Now you try: do people have fewer births on Friday the 13th than other random Fridays?


# Prep data for plotting
friday <- subset(births, day_of_week=='Friday') %>%
  mutate(the13 = ifelse(date_of_month==13, T, F))

# Look at how many Friday the 13th's there are in our data
table(friday$the13)

# Make a violin plot comparing Friday the 13th vs other Fridays
ggplot(friday, aes(x=the13, y=births)) +
  geom_violin()

# Extra super bonus
# Now you try: how many Friday dates aside from the 13th had fewer births than the average Friday the 13th?


# Now read a fun article reflecting some analyses here: 
# https://fivethirtyeight.com/features/some-people-are-too-superstitious-to-have-a-baby-on-friday-the-13th/