Week2_WorkData.rmd

---
title: "Working with Data"
author: "Joshua F. Wiley - last updated Michelle Byrne"
date: "`r Sys.Date()`"
output:
  tufte::tufte_html: 
    toc: true
    number_sections: true
---
```{r, include=FALSE, echo=FALSE}
#install.packages("tufte",repos = "http://cran.us.r-project.org")
library(tufte)
```

# Intro

`r margin_note("Core ideas and code are in the main document. Extra information that may help understanding but is not the big picture I will go in margin notes, like this.")`

Note from Michelle: Please make sure that you have downloaded the `R`markdown file for the lecture by pulling the most recent git repo
[https://github.com/michellebyrne1/MonashHonoursStatistics/blob/master/Week2_WorkData.rmd](https://github.com/michellebyrne1/MonashHonoursStatistics/blob/master/Week2_WorkData.rmd)
You can follow along with the pre-recorded lecture videos for Week 2, or you can read this markdown (in HTML format) independently, and follow along by executing the code in R Studio using the markdown (RMD) file.


# Basic `R` Use

As a brief refresher, we'll start just by using `R` as a basic calculator.

```{r calculator}

3 + 4 # addition

2 * 6 # multiplication

4 / 2 # division

3^2 # powers [2 squared]
```

`r margin_note("Objects in R can be almost anything. They can be a single number, a vector or set of numbers, the results from a regression, a graph, a dataset. You can have many objects at the same time in your worksapce, which can be a useful way to save results or store something for later use.")`
In `R` many things you do with an individual number can be done with a
vector, a set of numbers. Here's a quick refresher on creating
vectors, using `c()` and on storing them in an object, here I used
`x`.

Once you have created an object, it should show up in RStudio in the
"Environment". In addition, we can view or operate on all the elements
of an object by referring to it by name.

```{r vectors}
c(1, 5, 4) ## create a vector

x <- c(1, 3, 5) ## assign vector to x

x ## view x

x + 2 ## add 2 to every element of x
```

`r margin_note("Computers do not work quite like people or
regular math. Usually, numbers are represented using 
floating-point arithmetic (https://en.wikipedia.org/wiki/Floating-point_arithmetic)
often a binary representation is used as much in the computing world
is based on bits stored in 0s and 1s. For example, with a binary
based, 1/5 cannot be represented exactly. An implication of this is
that sometimes numbers are not represented in R quite like you might
think. For example: 1/5 == 0.2000000000000001 is evaluated as false
but 1/5 == 0.20000000000000001 is evaluated as true. The default
precision for floating points in R cannot distinguish those two
numbers.")`
Finally, here are a few quick tips on formatting `R` output.
Let's say we have a number with lots of decimal points, we might not
always need all that precision.

You can use `R` to round for you, using the `round()`
function. You can also use the `options()` function to
control the default options for how `R` prints output. This is helpful
if you want everything printed to your screen to be rounded more. One
difference is that `options()` actually uses 
[significant figures](https://en.wikipedia.org/wiki/Significant_figures)  
so you will not get a fixed number of decimal points, rather you get a
fixed number of significant figures.

`r margin_note("The round function has two main arguments (i.e., pieces
of input it needs from you). The first is the number or vector to be
rounded. It can be a single number or a whole vector of numbers. The
second is how many digits you want R to use in the rounding.
The results may not always print the number of digits you requested,
which may be confusing. This can happen when one of the digits rounds
to 0 (e.g., 1.199 rounds to 1.20 and the 0 would be dropped so
1.2).")`

```{r format}

1.214294254

round(1.214294254, digits = 2)

options(digits = 2) ## set default for the R session

1.214294254

```

# Working with Data

To continue on, we're going to open a few of the packages that you
should have already installed from Week 1.

```{r packages}

library(data.table)
library(reshape2)
library(JWileymisc)
library(extraoperators)

``` 

Next, we are going to load some sample data. This is simulated data
based on a 12-day study where people reported a few measures at
baseline along with stress and affect three times per day.
After loading it, we convert it to a data.table format, using the
`as.data.table()` function so that we can use many of the useful tools
from the `data.table` package for data management.

```{r data}

data(aces_daily)
d <- as.data.table(aces_daily)

```

## Operators 

`r margin_note("In software, logical values and logical operators are
often used to refer to things that return a boolean value. Booleans
are one of two values, conventionally: TRUE or FALSE. By
convention if you do any arithmetic on them, TRUE is treated as 1
and FALSE is treated as 0.")`

A lot of data management involves using logical operators. Operators
take data on the left hand side and a few options arguments on the
right hand side. Logical operators return `TRUE` or
`FALSE`.

We use logical operators when working with data for lots of
reasons. You might want to find outliers, values that are greater than
or less than a specific score, check if values fall within the
expected range, recode continuous variables into categorical variables
(low, medium, high, etc.). Much of this work is underpinned by logical
operators. Here is a list of some operators from `R` and expanded by
the `extraoperators` package.


| Operator         | What it does                                      |
|------------------|---------------------------------------------------|
| `<-`             | assign whatever is on right to the left in R      |
| `:=`             | assign right hand side to the left in data.table  |
| `==`             | Are values / vectors equal                        |
| `!=`             | Are values / vector NOT equal                     |
| `<` OR `%l%`     | Less than                                         |
| `<=` OR `%le%`   | Less than or equal                                |
| `>` OR `%g%`     | Greater than                                      |
| `>=` OR `%ge%`   | Greater than or equal                             |
| `%gl%`           | Greater than AND less than                        |
| `%gel%`          | Greater than or equal AND less than               |
| `%gle%`          | Greater than AND less than or equal               |
| `%gele%`         | Greater than or equal AND less than or equal      |
| `%in%`           | In                                                |
| `%!in%`OR `%nin%`| Not in                                            |
| `%c%`            | Chain operations on the RHS together              |
| `%e%`            | Set operator, to use set notation                 |


## Subsetting Data

A common task in analyses is to subset the data. There are many
reasons for this, from excluding outliers to selecting only
participants who meet some criteria you want to use.
Sometimes, the order you choose to subset matters too.

Most often, we subset data by using logical operators to pick specific
rows of a dataset or specific values from a single variable. Below, we
select cases/observations where Age = 18, this is done using the
logical operator `==`. `R` tests if the variable "Age" is equal to 18
and if it is returns TRUE and otherwise FALSE. Rows where the test
returns TRUE are returned. So we don't see a giant dataset, we just
pick two variables: UserID and Age.


```{r}

d[Age == 18, .(UserID, Age)]

```

Suppose we decide that under 20y or above 25y are extreme ages for our
data. We might want to run analyses in data excluding people outside
that range. We could create a copy of the dataset, `d.noout`, with
"outliers" removed where we only include cases that are between
20-25y. Here we use `%gele%` to capture people greater than or equal
to 20y and less than or equal to 25.

```{r}

d.noout <- d[Age %gele% c(20, 25), .(UserID, Age)]

## frequency table for ages
table(d.noout$Age)

## view a few rows of the data
head(d.noout)

``` 

Sometimes, a single variable is not sufficient. 
You can chain conditions together using the operators:
`&` and `|` which represent "and" and "or", respectively.

```{r}

## select only female participants who are 18y
d[Age == 18 & Female == 1, .(UserID, Age)]

## select anyone who is either a female participant OR 18y
d[Age == 18 | Female == 1, .(UserID, Age)]

```

If you want to select several values, the `%in%` operator is
helpful. It tests if the variable/data on the left is in the set on
the right hand side. This is much faster than writing many "ors"

```{r}

## select anyone whose age is in 18, 19, or 20
d[Age %in% c(18, 19, 20), .(UserID, Age)]

``` 

You can chain even more logical requirements together by using
parentheses. Operations occur within parentheses first.

```{r}

## 19 year old female participants or 18 year old male participants
d[(Age == 19 & Female == 1) | (Age == 18 & Female == 0),
  .(UserID, Age)]

``` 

Other common operators are greater than or less than, `>` and `<` or
greater than or equal to and less than or equal to, `>=` and `<=`.

```{r}

## anyone under age 20
d[Age < 20, .(UserID, Age)]

## anyone age 20 or under
d[Age <= 20, .(UserID, Age)]

## anyone who is not age 20
d[Age != 20, .(UserID, Age)]

## anyone who is not age 18, 19, or 20
d[Age %!in% c(18, 19, 20), .(UserID, Age)]

```

Finally, let us see why the order of operations may matter. Suppose
that we want to look at people who have at least 30 non-missing stress
values and look at just observations where STRESS > 3.
First, we'll remove cases that are missing on `STRESS`. We do this
using the `is.na()` function, which returns `TRUE` if a value is NA
(NA in `R` means not available, missing) and `FALSE` if a value is not
missing. Then we negate that using `!` to get "not missing".

```{r}

## remove missing stress observations
d2 <- d[!is.na(STRESS)]

```

Next, we can create a new variable in the dataset, which we'll call
"Count" and we'll assign this the number of observations that occur 
by ID.

In `R` `data.table`s have three main parts
    `DT i, j, by` the DT represents the name of the data table, the `i`
    represents which row(s) we want to select. If we leave it blank, that
    means select all rows. After the first comma, the `j` represents
    columns/variables. That could be selecting only certain
    columns/variables to display *or* creating / modifying a column or
    variable. The last part, the `by` represents a grouping variable or
    some way of organising our operations. For example, we might want to
    perform the same operation for each ID in the dataset. Lastly, if
    there are NO commas, that means just give me the
    rows/cases that match my criteria, but give me all columns/variables.

Here we take all cases (we aren't using any subsetting yet) and then
create a new variable Count, we use the assignment operator in
`data.table`, `:=` which means, assign to the thing on the left, in
this case a new variable, whatever the value on the right is. Here the
value on the right is `.N` that means the "n" or count / sample
size. Finally, we have `data.table` do this by ID, so that we don't
get the sample size of the entire dataset, we get the sample size, the
number of surveys, completed by each ID or person.

```{r}

d2[, Count := .N, by = UserID]

```

Now we can take people who have at least 30 non-missing stress values,
and look at high stress scores.

```{r}

d2 <- d2[Count >= 30]

d2 <- d2[STRESS > 3]

d2[, .(UserID, STRESS)] ## IDs and number of obs

```

This whole sequence could be collapsed or chained together to write
less code. In data tables in `R` you can chain operations by just
adding more brackets as below. This accomplishes the same as the above.

```{r}

d2 <- d[!is.na(STRESS)][, Count := .N, by = UserID][Count >= 30][STRESS > 3]

d2[, .(UserID, STRESS)] ## IDs and number of obs

```

`r margin_note("Here, the order is: first, remove missing stress
observations, second take only surveys/rows where stress scores > 3,
third, count how many observations are not missing by ID.")`

Now look what happens if we first take stress values greater than 3.
That is we change the order of the operations. We end up with much
fewer people. 

```{r}

d2 <- d[!is.na(STRESS)][STRESS > 3][, Count := .N, by = UserID][Count >= 30]

d2[, .(UserID, STRESS)] ## IDs and number of obs

```

In this new format, we are only taking people with 30
non missing stress values > 3 whereas before we took people with 30
non missing stress values (any value) and then only observations where
stress > 3.



## Finding Specific Cases or Outliers

Suppose that based on graphs or other information, we had decided that
negative affect scores above 4 were outliers and that participant 56
was an outlier overall.
We can exclude ID 56 and select only observations with negative affect
at or below 4 as below.

```{r}

d[UserID != 56 & NegAff <= 4, 
  .(UserID, NegAff)]

```

# Data Types

There are many types of data. In `R`, you can find what type of data a
particular variable is by asking for its `class()`. 

```{r}

class(d$Age)
class(d$BornAUS)
class(names(d))

``` 

- *Logical*: Used for logical data, which are either TRUE or FALSE. If data are logical, it is a very efficient format and useful for many cases. Logical variables can be compared. Arithmetic can be used for logical variables, in which case TRUE is treated as 1 and FALSE as 0.
- *Integer*: Used for integer type data, that is whole numbers like 0, 1, 2. For variables that are only whole numbers, integer format is more efficient than real numbers or numeric data (e.g., 1.4).
- *Numbers/real/numeric*: Used for real numbers, such as 1.1, 4.8. It also can be used for integer data (i.e., whole numbers only) but is a less efficient format. In `R` these are represented by the class `numeric` abbreviated `num`.
- *Text/character/string*: Used for text type data, such as names, qualitative data, etc. Also, any numbers can be stored as strings. In `R` these are represented by the class `character` abbreviated `chr`. Character data do not work with arithmetic operators, but can be sorted (e.g., alphabetically).

Although these are the basic building blocks of data, there also are special formats that can be built off of these, often off of numeric type data.

- *Date*: A special representation of numeric data when the data are dates. In `R` represented by multiple classes, including `Date`, `POSIX` and `chron`. The standard approach is to count the days since some origin. In `R`, the default origin in 1970-01-01. So 1970-01-01 is 0. 2 would indicate 1970-01-03. You can do some arthmetic on dates. For example adding or subtracting numbers, calculating the difference between dates.
- *Date and time*: A special representation of numeric data when the data are dates and times. In `R` represented by multiple classes, including `POSIX` and `chron`. The standard approach is to count the seconds since some origin. In `R`, the default origin in 1970-01-01 00:00:00. So 1970-01-01 00:00:00 is 0. 2 would indicate 1970-01-01 00:00:02. You can do some arthmetic on datetime objects. For example adding or subtracting numbers, calculating the differece between dates and times.
- *time*: A special representation of numeric data when the data are times. In `R` represented by class `chron`. 24 hours are formated so that they go from 0 to 1. For example, 3:31pm would be converted to 24 hour time: 15:30 and then to decimals for hours: 15.5 then converted ito range from 0 to 1: 15.5/24.
- *factor*: A special representation of numeric data when the data are fundamentally discrete. For example, if study condition is coded as 0 = control, 1 = medication, 2 = psychotherapy. The data could be stored numerically, or as text. If stored as a factor, R will do some convenient things automatically, such as dummy coding in regression models.


# Date Variables

Often when dates are read into R, they are character data.
We will work with some different character formats and 
convert them to date class in `R`.

```{r, error = TRUE}

## sample character data
x1 <- c("2019-03-12", "2017-09-15")
x2 <- c("2019-Mar-12", "2017-Sep-15")
x3 <- c("12/3/2019", "15/9/2017")
x4 <- c("12/3/19", "15/9/17")

# these are character strings
class(x1)


## convert character strings to Date classes
as.Date(x1, format = "%Y-%m-%d")
as.Date(x2, format = "%Y-%b-%d")
as.Date(x3, format = "%d/%m/%Y")
as.Date(x4, format = "%d/%m/%y")

## save results
d1 <- as.Date(x1, format = "%Y-%m-%d")

## look at the class once converted
class(d1)

## arithmtic on dates
d1 + 5 ## addition
d1 - 7 ## subtract a week
d1 * 5 ## multiplication does not work: error

## extract just one date
d1[1]
d1[2]

## compare two dates
d1[1] - d1[2]

## compare two dates and convert to a number
as.numeric(d1[1] - d1[2])

## compare two dates and convert to a number
## and then to years
as.numeric(d1[1] - d1[2])/365.25


``` 

To convert date into a date class in `R`, we need to tell it the
format the dates are written in. This is done by specifying a
"template". 

- "%Y" means 4 digit year [upper case]
- "%y" means 2 digit year [lower case]
- "%m" means digit month
- "%b" means abbreviated month (e.g., "Jan")
- "%d" means digit day

These pieces can be re-ordered an combined and use any separator. For
example "%Y-%m-%d" for YYYY-mm-dd or "%Y/%m/%d" for YYYY/mm/dd or "%Y
%m %d" for YYYY mm dd or "%d %m-%Y" for dd mm-YYYY. This flexibility
is needed as different people and countries commonly write dates in
many different formats. 


## Date Helper Functions

Dates have a variety of helpful functions that let you operate on or
learn more about them. We can use the `weekdays()`, `months()`, and
`quarters()` functions to find out the day of week, month or quarter a
particular date falls in. They return character strings. 

```{r}

weekdays(d1)
months(d1)
quarters(d1)

```

Dates also can be compared to determine if one is greater than or less
than another. This can be useful for selecting observations that meet
specific criteria. For example, suppose we wanted to exclude any dates
that occurred before the semester started. Not everyone in the daily
diary study is a student, but if they were, we might expect positive
affect to be different on break than during semester. Suppose semester
started 1 March 2017, we might want to only include observations that
fell on or after that date. 
Note here we do not need to specify a template because `R` will assume
YYYY-mm-dd when written this way. 

```{r}

## summary of dates
summary(d$SurveyDay)

## select surveys where the survey date 
## is greater than or equal to 2017-03-01
d2 <- d[SurveyDay >= as.Date("2017-03-01")]

## summary of dates
summary(d2$SurveyDay)

d2 <- d[weekdays(SurveyDay) %in% c("Monday")]

## frequency table of days of week
table(weekdays(d2$SurveyDay))

``` 

# Factor Variables

Factor variables can be made from numeric or 
integer or character data, but are converted into 
the factor format to tell `R` that they are discrete
variables. This is done using the `factor()` function.

```{r}

## some sample data
x <- c(0, 1, 2, 1, 1, 2, 0)

## convert to a factor in R
## adding specific labels
y <- factor(x, 
  levels = c(1, 0, 2),
  labels =c("Medication", "Control", "Psychotherapy"))
y

## if you do not specify levels or labels
## it will assume you want the levels 
## ordered numerically / alphabetically
## and the labels should be the same as the levels

## factor with labels, sorted according to 
## the level order (1, 0, 2)
table(y)

## default factor() approach, labels based on 
## levels and sorted numerically: 0, 1, 2
table(factor(x)) ## automatic conversion to factor 

```

# Merging Data

Often data come from multiple sources and are in separate files. For
example, a baseline and post intervention questionnaire may be
collected separately, or you may have questionnaire data and sleep
data from a FitBit. 

Data can be combined by merging, and merging in `R` is typically done
using the `merge()` function.

`r margin_note("Merging in R is the same as joins which is the more
common term for databases, such as SQL.")`

There are four main types of merges in `R` using the `merge()`
function. Assume we are merging two datasets: `x` and
`y`. 

`r margin_note("Merging or joins always involves two datasets. Even if you need
to merge more than two datasets (e.g., you have baseline, post, and
follow-up data) you always merge / join them two at a time. For
example, first joining baseline and post into one bp dataset and
then joining bp and your follow up data. In R the x dataset is
always the left and the y dataset is always the right.")`
	
The type of merge performed is controlled by arguments to
the `merge()` function.

- **Natural join**: resulting data has only rows present in
  **both** `x` and `y`. Argument: `all = FALSE` .
- **Full outer join**: the data has all rows in `x` and all rows
  in `y`. Argument: `all = TRUE` .
- **Left outer join**: resulting data has all rows in `x`. Argument: `all.x
  = TRUE` .
- **Right outer join**: resulting data has all rows in `y`. Argument: `all.y
  = TRUE` .

`r margin_note("This small, hypothetical study collected data via surveys and
actigraphy. The surveys have peoples ages and the actigraphy has
their sleep duration, which we wish to join together for analysis.")`

We will explore joins using two, small datasets. `ID` is used
to indicate participant ID.

```{r}

surveys <- data.table(
  ID = c(1, 2, 3),
  Age = c(19, 18, 20))

acti <- data.table(
  ID = c(2, 3, 4),
  Sleep = c(8, 6, 7))

## view the two small datasets
print(surveys)
print(acti)

```

## Natural Join in `R`

`r margin_note("Natural joins have only the rows / observations that
are present in both datasets. In this example, the surveys and the
actigraphy. You can pick whichever dataset you want to be x and 
and y by which you list first.")`

**Only rows present in both datasets**. 

`r margin_note("When we write by = ID what is meant is that it should be
by the variable called ID. The word ID is not special, its just
that we happened to call the variable containing our IDs, ID. If
we had called the variable containings IDs Email or Name then
we would write: by = Email or by = Name. In other words,
match the text in quotes to an actual variable name in the data.")` 
	
Joining data requires specifying how `R` should know which rows match
each other. This is typically  done by an ID variable using the code:
`by = "ID"`. The argument, `all = FALSE` is what tells `R` this should
be a natural join.

```{r}

merge(
  x = surveys,
  y = acti,
  by = "ID",
  all = FALSE)

```

## Full Outer Join

`r margin_note("Full outer joins include cases that are present in either dataset. In this case, in either the surveys
or the actigraphy. You can pick whichever dataset you want to be x
and y by which you list first, although with a full outer join, it
does not make much difference, other than which columns/variables come
first and which come second. As before, we must specify what variable(s) to
merge by so R knows how to match rows together, accomplished using
the code: by = ID.
Note that R fills in missing values for rows /
observations and variables that are present in one dataset but not the
other. So if two datasets do not have identical rows / observations, a
full outer join will result in some missing values. These are shown in
the print out as NA")`

**All rows present in either dataset**. 

The argument, `all = TRUE` is what tells `R` this should be a full
outer join. Full outer joins include all rows or observations.	
	
```{r}

merge(
  x = surveys,
  y = acti,
  by = "ID",
  all = TRUE)

```


## Left Outer Join

`r margin_note("As before, we must specify what variable(s) to
merge by so R knows how to match rows together, accomplished using
the code: by = ID. Also as before, R fills in missing values for
the variables. In this case, missing values for the rows /
observations that are present in surveys, but not in acti (i.e., the
variable Sleep and ID = 1).")`

**All rows / observations that are present in left dataset
(surveys)**.

The argument, `all.x = TRUE` is what tells `R` this should be a left
outer join.

```{r}

merge(
  x = surveys,
  y = acti,
  by = "ID",
  all.x = TRUE)

```

## Right Outer Join

`r margin_note("As before, we must specify what variable(s) to merge by
so R knows how to match rows together, accomplished using the code:
by = ID. Note that R fills in missing values for the variables
in surveys that are not in acti and the rows / observations that are
present in acti, but not in surveys 
(i.e., the variable Age and ID = 4).")`

**All rows / observations that are present in right dataset
(acti)**.  The argument, `all.y = TRUE` is what tells `R` this should be a right
outer join. 

```{r}

merge(
  x = surveys,
  y = acti,
  by = "ID",
  all.y = TRUE)

```

## Rows and missing after joining


`r margin_note("Knowing how many rows you expect gives you a quick way to
check whether the join is working as you want or something
unexpected may be happening. Practical experience is that joins
can easily have some errors resulting in unintended consequences, 
like missing data or repetitions of rows that were not wanted.")`

It is helpful to know how many rows you expect in the 
results.


`r margin_note("This behaviour is sometimes referred to as cross or
cartesian joining. When there are multiple matches for a row/ID in
both datasets, all possible combinations are returned. In this
case, rows 2 and 3 are both ID 2 in surveys2 and rows 1 and 2 are
both ID 2 in acti2, so those are crossed so you have all possible
pairings of rows from each dataset associated with ID 2.")`

In each of these two new datasets, ID 2 is repeated twice. 
One might anticipate that a full outer join would return *5*
rows, (IDs 1, 2, 2, 3, 4). `R` actually returns **7** rows, with
four rows for ID 2, because all rows of ID 2 in `surveys2` are
repeated by all rows of ID 2 in `acti2`. 


**Merging with any duplicates can behave in unexpected ways**. 
In a small dataset, it is easy to see what has
happened. In a large dataset, checking whether the merge worked as
expected requires using summaries or other checks, such as evaluating
whether the number of rows matched what you wanted or needed.

```{r}

surveys2 <- data.table(
  ID = c(1, 2, 2, 3),
  Age = c(19, 18, 18, 20))

acti2 <- data.table(
  ID = c(2, 2, 3, 4),
  Sleep = c(8, 7, 6, 7))

## view small datasets
print(surveys2)
print(acti2)

merge(surveys2, acti2, by = "ID", all = TRUE)

```

## Merging Data by Different IDs

Often data are under our control, but sometimes, they are not or even
if they are ID variables are not named consistently across datasets.
As long as the actual ID values match, the variable names do not need
to be the same. For example, in Qualtrics, if you create custom links,
IDs may be called "RecipientFirstName" while on the actigraphy data,
you may have called IDs, "ID". `R` can handle this easily by using the
arguments, `by.x` and `by.y` to give the variable names in the `x` and
`y` datasets, respectively.


```{r}

surveys2 <- data.table(
  FirstName = c(1, 2, 2, 3),
  Age = c(19, 18, 18, 20))

acti2 <- data.table(
  ID = c(2, 2, 3, 4),
  Sleep = c(8, 7, 6, 5))

merge(
  x = surveys2,
  y = acti2,
  by.x = "FirstName",
  by.y = "ID",
  all = TRUE)

```

## Checking and Preparing Data for Merging

Because merging data can have unintended consequences if matched
incorrectly, often it is a good practice to check the data in advance
to make sure it is what you expect.

Generally, one-to-one merges, where each row of one dataset is matched
to one and only one row of another dataset is easiest. You can have
duplicates and these merge well in two cases: (1) the left, `x`
dataset has repeated measures but the right, `y` dataset only has one
row of data per ID or (2) the left, `x` dataset only has one row of
data per ID but the right, `y`, dataset has repeated measures per ID.
Many to many merges, involves *repeated IDs* in both datasets being
merged often is not what is intended. Note that repeated IDs does not
mean repeated measures. You can merge two repeated measures datasets
but not have repeated IDs by using two variables as your ID variable
or by creating unique IDs that combine ID and time, for example.

To check data, there are a few helpful functions in `R`.
The `anyDuplicated()` function returns 0 if no values are duplicated
and the position where the first duplicate occurs if any values are
duplicated.

```{r}

## there are duplicate IDs, starting in position 2
anyDuplicated(acti2$ID)

## there are no duplicated sleep values
anyDuplicated(acti2$Sleep)

## however if we paste / concatenate
## both ID and Sleep, there are no duplicates
anyDuplicated(paste0(acti2$ID, acti2$Sleep))

```

If you do not expect duplicates, it is a good idea to check for them
before merging, as `R` will not tell you if there are duplicates and
it impacted the merge.

If there are duplicates, you can count how many there are,
using the `duplicated()` function and making a frequency table using
`table()`. 

```{r}

table(duplicated(acti2$ID))

``` 

Another way to approach this is to create a frequency table of IDs and
then create a frequency table of the frequencies. If everyone only
shows up once, you should have many frequencies of 1. If any ID shows
up more than once, you will have non zero frequencies > 1.
The frequency table below, shows that two IDs showed up once and one
ID showed up two times.

```{r}

table(table(acti2$ID))

``` 

Other checks are the number of unique IDs in each dataset. This is
accomplished by using the `unique()` function to first exclude any
duplicated values and then finding the length of the IDs.

```{r}

length(unique(surveys2$FirstName))

length(unique(acti2$ID))

```

You can also check how many IDs from one dataset are in another
dataset by combining the `unique()` function with the `%in%` logical
operator to get a series of TRUE or FALSE values if an ID from one
dataset is in another dataset. In large datasets, this may be many
results, so instead, we can create a frequency table of the results
using `table()`. Note that which set of IDs is on the left or the
right can make a difference. If you expect one dataset to contain all
IDs (e.g., you are merging a new variable into a large "master"
dataset that you believe has every single one of your participants)
this is a very helpful way to check whether that is true.

```{r}

table(unique(surveys2$FirstName) %in% unique(acti$ID)) # why do you think it didn't say 2 are false?

table(unique(acti$ID) %in% unique(surveys2$FirstName))

``` 

# Reshaping Data

Another common data management task is reshaping data. This typically
occurs when data are stored in one format, such as different columns
for different time points, and you need data reshaped into another
format, such as multiple rows for each ID for mixed models.
What follows is a small sample of a "wide" dataset. That is, a dataset
where each timepoint is stored in a separate variable.

```{r}

dwide <- data.table(
  ID = c(1, 2, 3),
  STRESS1 = c(2, 1, 3),
  STRESS2 = c(4, 5, 6),
  STRESS3 = c(1, 1, 2),
  Happy1 = c(4, 5, 3),
  Happy2 = c(3, 2, 1),
  Happy3 = c(5, 5, 4))

``` 

```{r, echo = FALSE, results = "asis"}
knitr::kable(dwide, caption = "Example 'wide' dataset where different time points are stored in separate variables.")
```

`R` has a function, `reshape()` that allows reshaping data from wide
to long. It is quite complicated, so may not all be clear at first,
but its easiest to learn through examples. For now, we will focus on
five arguments to `reshape()`.

- `data` this is the dataset to reshape
- `varying` this is a list where each element is a vector of variable
  names that vary over time that belong to the same overall measure.
- `timevar` this is the name of the variable indicating to which time
  a measure belongs, it will be a new variable in a wide to long
  reshape.
- `idvar` this is the name of the ID variable in the data and is
  required.
- `direction` this is the direction to reshape ("long" or "wide").

```{r}

reshape(
  data = dwide,
  varying = list(
    STRESS = c("STRESS1", "STRESS2", "STRESS3")
  ),
  timevar = "time",
  idvar = "ID",
  direction = "long")

```

In this case, you can see that only STRESS was actually reshaped
long. To get Happy reshaped long as well, we need to list all the
variable names for happy in order in another vector using `c()` inside
the `varying = list()` section, like this:


```{r}

reshape(
  data = dwide,
  varying = list(
    STRESS = c("STRESS1", "STRESS2", "STRESS3"),
## extra line for happy
    Happy = c("Happy1", "Happy2", "Happy3") 
  ),
  timevar = "time",
  idvar = "ID",
  direction = "long")

```

Right now, `R` is just using the first variable name for each variable
in the long format, so we have "STRESS1" and "Happy1". If we want, we
can customise this, for example to get rid of the "1".

```{r}

reshape(
  data = dwide,
  varying = list(
    STRESS = c("STRESS1", "STRESS2", "STRESS3"),
    Happy = c("Happy1", "Happy2", "Happy3") 
  ),
## extra line to make it just STRESS and Happy without the "1"
  v.names = c("STRESS", "Happy"),
  timevar = "time",
  idvar = "ID",
  direction = "long")

```

Finally, by default, `R` just makes the new "time" variable integers
from 1 to the maximum number of time points. Instead, we can set the
exact times. For example, suppose that the first assessment is week 0,
then week 6 and week 24. We might want to use 0, 6, 24 instead of the
default 1, 2, 3.

```{r}

dlong <- reshape(
  data = dwide,
  varying = list(
    STRESS = c("STRESS1", "STRESS2", "STRESS3"),
    Happy = c("Happy1", "Happy2", "Happy3") 
  ),
  v.names = c("STRESS", "Happy"),
  timevar = "weeks",
## extra line for customized times
  times = c(0, 6, 24),
  idvar = "ID",
  direction = "long")

print(dlong)

```

## Complications in Reshaping

Not all reshaping is straight forward. Here is another wide
dataset. Some variables are time varying, others (e.g., Age in yrs) are
not. Even the time varying variables are not measured at all time
points. In this case at the final time point, only Happy was
measured and coping only measured twice.

```{r}

dwide2 <- data.table(
  ID = 1:3,
  Age = c(18, 19, 22),
  STRESS1 = c(2, 1, 3),
  STRESS2 = c(4, 5, 6),
  STRESS3 = c(1, 1, 2),  
  COPE1 = c(1, 2, 3),
  COPE2 = c(6, 3, 4),
  Happy1 = c(4, 5, 3),
  Happy2 = c(3, 2, 1),
  Happy3 = c(5, 5, 4),
  Happy4 = c(1, 2, 3))

``` 

Following the previous template, we might try something like this:

```{r, error=TRUE}

reshape(
  data = dwide2,
  varying = list(
    STRESS = c("STRESS1", "STRESS2", "STRESS3"),
    COPE = c("COPE1", "COPE2"),    
    Happy = c("Happy1", "Happy2", "Happy3", "Happy4") 
  ),
  v.names = c("STRESS", "COPE", "Happy"),
  timevar = "weeks",
  times = c(0, 6, 12, 24),
  idvar = "ID",
  direction = "long")

```

This results in an error because we only have three stress and two coping
measures. To do the reshape, there must be an equal number of each
variable. One way to address this is to create a useless "Pad"
variable. We can repeat this "Pad" as much as needed (e.g., once for
stress, twice for coping). Age is not time varying so we can ignore
it. It will be used as a between person variable.

```{r}

dwide2[, Pad := NA]

reshape(
  data = dwide2,
  varying = list(
    STRESS = c("STRESS1", "STRESS2", "STRESS3", "Pad"),
    COPE = c("COPE1", "COPE2", "Pad", "Pad"),    
    Happy = c("Happy1", "Happy2", "Happy3", "Happy4") 
  ),
  v.names = c("STRESS", "COPE", "Happy"),
  timevar = "weeks",
  times = c(0, 6, 12, 24),
  idvar = "ID",
  direction = "long")


```


## Reshape Data Wide

We can also reshape data from long to wide. To do that, we will begin
with the long dataset we made earlier, `dlong`. Again we use the
`reshape()` function. This time, we do not need to indicate `varying`
and we set `direction = "wide"`.

```{r}

reshape(
  data = dlong,
  v.names = c("STRESS", "Happy"),
  timevar = "weeks",
  idvar = "ID",
  direction = "wide")

```

To create the different variables for each time point, `R` has used
the base variable name, a period, and then the time value (0, 6, 24).
We can customize the separator so that instead of a period, there is
nothing or an underscore, or whatever we wanted.


```{r}

reshape(
  data = dlong,
  v.names = c("STRESS", "Happy"),
  timevar = "weeks",
  idvar = "ID",
  direction = "wide",
  sep = "_T")

```

# Bonus Content - Merging by Multiple IDs

Sometimes a single variable is not enough to merge two datasets. This
is particularly true in longitudinal studies where there are repeated
measures. `R` can handle this by using multiple variables as IDs for
merging. Alternately, a second, unique ID can be created by combining
two or more pieces of information. The following shows an example.

```{r}

surveys <- data.table(
  ID = c(1, 1, 2, 2),
  Time = c(1, 2, 1, 2),
  Stress = c(6, 4, 3, 4))

acti <- data.table(
  ID = c(1, 1, 1, 3, 3),
  Time = c(1, 2, 3, 1, 2),
  Sleep = c(7, 8, 9, 8, 8))

## incorrect
merge(surveys, acti,
      by = "ID",
      all = TRUE)

## correct
merge(surveys, acti,
      by = c("ID", "Time"),
      all = TRUE)

```