Lecture 6 Factors.Rmd

---
title: "Lecture 6 Factors"
author: "David A McAllister"
date: "Monday 31st October, 2016"
output:
  slidy_presentation: default
  ioslides_presentation: default
---

## Objective of section
* To define a factor
* To describe the uses of a factor
* To create factors, add data to factors and change factor levels
* To combine factors together

## Factors

* R's way of handling labelled variables
* Useful in
    * regression modelling
    * tables
    * ggplot graphs
* Awful in
    * data scrubbing
    * merging

## Create factor
All factors are created from character vectors
```{r, echo = FALSE}
my_char <- sample(c("a", "b","c","e"), 20, replace = TRUE)
my_char
```
* Values include a,b,c and e
* Assume "d" was possible (just not in this dataset yet)
```{r}
my_factor <- factor(my_char, levels =c("a","b","c","d","e"), 
                    labels = c("A","B","C","D","E"))
str(my_factor)
```


## What happens
R-help "If x[i] equals levels[j], then the i-th element of the result is j."

```{r}
data.frame (original = my_char, levels = as.numeric (my_factor), 
            labels = as.character(my_factor))
```

## Modifying factors


```{r, eval =TRUE, warning = TRUE}
# Can't add new values
my_factor [3] <- "F" 
# Even if originally present when made factor
my_factor [3] <- "d"
# Only existing levels
my_factor [3] <- "D"
```

## Create a new factor with different coding
New metadata

* elephant, buffalo, antelope and crocodile
* No donkeys
```{r}
# create new factor
my_factor2 <- factor (my_char, levels = c("e", "b", "a", "c"), 
                      labels = paste ("new", c("E", "B","A","C")))
levels (my_factor2)
```

## Concatenating factors
In order to combine factors

* Must convert to character variables first

```{r}
naive <- c(my_factor, my_factor2)
correct <- c(as.character(my_factor), as.character(my_factor2))
#Tabulate - shows the problem
table(naive, correct)
```

## Changing reference level
Reference level

* Safe and easy, eg in regression modelling
```{r, echo = 2}
levels (my_factor)
my_factor <- relevel (my_factor, "C")
levels(my_factor)
```

## Collapsing

* but potentially dangerous
```{r, echo= 2}
my_vowels <- my_factor
levels (my_vowels) <- c("con", "vow", "con", "con", "vow")
table (my_vowels, my_factor)
as.numeric (my_factor) [1:10]
as.numeric (my_vowels) [1:10]
```
*Safest option is to use factor* levels and labels in same function

## Dropping levels
```{r}
my_factor <- factor (my_char)
table (my_factor)
table(my_factor[my_factor != "c"])
table(droplevels(my_factor[my_factor != "c"]))
```

## A factor is a numerical integer with an attribute

```{r}
attributes (my_factor)
attributes (matrix(1:9, nrow = 3))

```

```{r}
attributes (cars)
```

## Advice

* When importing data use `stringsAsFactors = FALSE`
* Used to have performance implications, not now
* Don't bother with ordered factors
* Use factors for regression modelling, explicitly set levels