# Welcome to Development Research!

You will now learn how to write some code in R and to perform some basic statistics!

###################################################### Quantitative Practical 1 [START] ##################################################

############################################## Part 1 Importing and looking at your data #########################################

# Note that using a # allows you add comments to annotate your code. It's easy to write lots of lines of
# code and to loose track of why you did certain things. Keeping notes is generally a good idea.

# R works using commands. This means that you can write a line of code that tells R what to do.
# for example
1+2
# this tells R to calculate 1 + 2 (Yes! R will also do calculations like a calculator and this is a very useful feature)

# First, you will want to read in your data. R allows you to read and write many different data types.
# This includes files from EXCEL, data files commonly used in ArcGIS (e.g. .dbf), and other statistical,
# including SPSS and STATA. Some of these features are available in base, others have to be imported in separate packages
# We will use .csv files, which are easily transferable and which can be easily saved in and opened in EXCEL.

data <- read.csv(file.choose()) # this line of data includes two commands, read.csv() - which reads in the data,
# and file.choose() - which allows you to select your file using a browser.

# This command allows you to see the first few lines of data in your dataset
head(data)

# Looks at the variable "weight..kg.". It seems like R are doesn't like brackets ().
# Let's fix and rename the weight and height variables
colnames(data)[colnames(data) == "weight..kg."] <- c("weight")
colnames(data)[colnames(data) == "height..m."] <- c("height")

# Let's try and get some summary statistics

summary(data)

# note that the categorical variables have been imported as "characters" and we need to change these to be factors so that R can identify them as categorical variables
# let's re-assign them
data$pokemon <- as.factor(data$pokemon)
data$type <- as.factor(data$type)
data$sex <- as.factor(data$sex)
data$surface <- as.factor(data$surface)

# Let's run the summary function again to see what has happened
summary(data)
# let's check if what we have done has fixed the problem
names(data)

# now let's calculate the mean and the standard deviation for pokemon height and weight
# this command let's you calculate the mean for Pokemon weight in our sample (note that we specify the dataset [data] and the variable [weight]
# we and use $ to do this).
mean(data$weight)

# this command let's you calculate the standard deviation

sd(data$weight)

# let's calculate the standard error, which you'll remember is the standard deviation divided by the square root of the sample size
# Note the command nrow will take on the value of the number of rows of the data frame in brackets. It's your sample size or n
sd(data$weight)/sqrt(nrow(data))

# we can also create an object and assign a value or a calculation (this will also appear in the Environment pannel)
weight_sd <- sd(data$weight)

# if you run that command you'll get the value in the console
weight_sd

# we can also create an object for the square root of the sample size
sqrt_n <- sqrt(nrow(data))

# and we can now divide the standard deviation by the square root of n
weight_sd/sqrt_n

# We can also use a package with an in-built function

# first we need to install the package
install.packages("plotrix")

# Now we need to upload or attach the package

library(plotrix)

# now we can use the std.error() function

std.error(data$weight)

# now let's calculate the mean for Pokemons caught in the park
# note that the first part of the command is the same as above - the second part specifies the surface
mean(data$weight[which(data$surface == "natural") ], )

# now let's calculate the mean for Pokemons caught in the park and that are bug types
mean(data$weight[which(data$surface == "natural" & data$type == "bug") ], )

# What is the mean for bugn types on built surfaces?

# Generating a table to calculate all these sub classifications in individual steps is very time consuming
# Again, we can speed up the process by using a package
# Let's install another package
install.packages("doBy")

# and upload the package

library(doBy)

# let use the summaryBy() function to calculate the mean height and weight of Pokemons by surface type
# note that FUN = mean is telling summaryBy to calculate the mean, you could equally ask it
# to calculate the standard deviation
summaryBy(weight + height ~ surface + type, FUN = mean, data = data)

# we can also get summaryBy() to run several calculations using a function

# note here that we are also using std.error from the package plotrix
summaryBy(weight + height ~ surface,
data = data,
FUN = function(x) { c(mean = mean(x), sd = sd(x), sum = sum(x)) }
)

# we can also get summaryBy() to calculate statistics for both surface and pokemon type
# this time we are also going to assign it an object
# you could read what summaryBy does as: calculate weight and height as a function of surface and pokemon type
summary.table <- summaryBy(weight + height ~ surface + type,
data = data,
FUN = function(x) { c(mean = mean(x), sd = sd(x), se = std.error(x)) }
)
summary.table

# How come you are getting NA in some rows?

# You can check your data by double clicking on the data icon (the one with the little blue arrow) in the "Environment" tab in the upper
# right hand corner. How many poison types did we find in built areas?
# R cannot calculate standard deviations if there is only one row of data.

# now let's save this table so that we can use it later

# note that we first used read.csv; now we will use write.csv
write.csv(summary.table, "/Users/user/Desktop/PokemonTable.csv")
# One of the things that we might want to do is to also get some statistics for ranges of values
# First we will look at the range of values for weight in our data
# Notice that this function gives us the min. and the max. as well as the mean and the median, and the quartiles
summary(data$weight)

# Now let's plot some frequency distributions for HP

# Note that we have now generate a histogram, reflecting the values we just calculated
hist(data$HP)

# Now let's try and compare the frequency distributions of height between the two surfaces
# We are going to use a new package called "ggplot" which is the most powerful package for data visualisation
# Lets install and load the package first
install.packages("ggplot2")
library(ggplot2)

# now lets graph the distribution for HP

ggplot(data, aes(HP)) + # this part tells ggplot the data and variables you are selecting to plot, note that here you are only graphing on value on the x axis
geom_histogram(bins = 30) # this part tells ggplot what kind of plot you want to make and sets the number of bars you want to draw

# now lets graph the distribution for HP for the surfaces that we have
# note that in this instance, we have divided the dataset and drawing two separate histograms
# we can colour them and also change the transparency. In this case, we can make the green bars more transparent so that you can see the grey bars behind them
# we can also add the axis labels
ggplot() +
geom_histogram(data = data[which(data$surface == "built"),], aes(HP), bins = 6, fill = "grey") +
geom_histogram(data = data[which(data$surface == "natural"),], aes(HP), bins = 7, fill = "green", alpha = 0.3) +
xlab("Pokemon HP") +
ylab("Number of Pokemons") +
theme_bw()

# another way of graphing distributions is to use density plots

# let's consider variables
ggplot() +
geom_density(data = data[which(data$sex == "male"),], aes(CP), colour = "grey") +
geom_density(data = data[which(data$sex == "female"),], aes(CP), colour = "green") +
xlab("Pokemon HP") +
ylab("Density") +
theme_classic() # we can change the "theme" of graphs two
############################################## Part 2 Comparing means (t test) #########################################

# let's calculate the t-statistic

# now we can calculate the means for HP for the two surfaces
m_built <- mean(data$HP[which(data$surface == "built") ], )
m_natural <- mean(data$HP[which(data$surface == "natural") ], )

# and now the variance

s_built <- var(data$HP[which(data$surface == "built") ], )
s_natural<- var(data$HP[which(data$surface == "natural") ], )

# we can use the summary function to provide us with some summary statistics, including the number of datapoints (length) in each group
# how many data points do we have in each group?
summary(data)

# so according to the formula for the t-statistic

(m_built - m_natural) / sqrt((s_built/18)+(s_natural/40))

# now let's compare that with the built in t.test function in {base}
t.test(data$HP~data$surface)

# is there a statistical difference Pokemon HP between surfaces?

# can you run another t-test compare groups?
############################################## Compare frequencies (Chi Square test) #########################################

### Chi Square Analysis

# let's try and see whether there's a difference in the frequency of male and female Pokemon between surfaces
# let's generate a summary table for our analysis
summaryBy(type ~ surface + sex, FUN = sum, data = data)

# Now let's generate the contingency table

# the table is what we call a matrix
# f m
C<-matrix(c(85, 90, #built
141, 209), #natural
nrow=2)

# what does our table look like

C

# Now with a pen and paper let's calculate the chi-square statistic; you can use R as a calculator if you wish

# And now, let's run the Chi Square test running the test using R's function
chisq.test(C, correct = F)

###################################################### Quantitative Practical 1 [END] ##################################################