Session 2 - Factors.Rmd

---
title: "Factors"
author: "Ashwin Malshe"
date: "06/01/2021"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

### Factors

Factors are objects that are build on top of a vector with defined, limited number of **levels**. Although I am covering them under vectors, they don't behave like vectors much.

```{r}
v1 <- c(1:10)
v2 <- c(rnorm(10)) # generate a sequence of 10 random numbers
```

```{r}
print(v2)
```



```{r}
v2_f <- factor(v2)

print(v2_f)
```

```{r}
attributes(v2_f)
```


The clearest distinction between a factor and a character vector is given by this example:

```{r}
# Start off with the frequency distribution
table(v2)
table(v2_f)
```

```{r}
# Now let's limit the frequency distribution to the first 5 values in each case
table(v2[1:5])
table(v2_f[1:5])

```


Notice that for the vector, we have no frequencies for the numbers 6 to 10 but for the factor we have explicitly 0 frequencies. What causes this difference?

Let's print the altered vector and the factor

```{r}
print(v2[1:5])

print(v2_f[1:5])

```

So although both the objects have the same elements, factors still have their original levels. Thus, you can subset a factor but that doesn't automatically subset the associated levels!

Let's create a factor from v1, which is an integer vector

```{r}
v1_f <- factor(v1)
v1_f
```

We can change the levels associated with factor easily
This will fail:

```{r}
attributes(v2_f)$levels <- c(1:10)
```

Why did we get a warning?


Do we know the class of vector `levels`?
Let's find it out.

```{r}
v2_f <- factor(v2)
class(attr(v2_f, 'levels'))
```

What if we assign 1-10 values but as character?

```{r}
attr(v2_f, 'levels') <- as.character(c(1:10))
```

Now we don't get a warning!

We learned a new function `as.character()`. You will come across several `as.something` functions in R. These functions coerce an object to change its class.

```{r}
attr(v2_f, 'levels')
```

In practice it's better to change the levels of a factor by using specific functions. `forcats` package in R helps with this.

What if we want to extract the factor levels as integers though?


```{r}
as.numeric(levels(v2_f))
```

What about the levels of `v1_f`? They looked integers!

```{r}
levels(v1_f)
```

Well, not so much in reality. We need to convert them to numbers too if you want to use them in mathematical operations.


```{r}
as.numeric(levels(v1_f))
```


### Example

Consider the following character vector with months of a year. We will plot the frequency distribution of this vector. As each months shows up once, all of them will have the same unit frequency.

```{r}
mons <- c("JAN","FEB","MAR","APR","MAY","JUN",
          "JUL","AUG","SEP","OCT","NOV","DEC")
plot(table(mons))
```

Can we change the labels on the X axis?

Yes, if you set the `levels` explicitly to the correct ordering.

```{r}
mons_f <- factor(mons,
                 levels = mons)
plot(table(mons_f))
```


#### Ordered factors

Ordered factors hold the information on the ordering of the factor levels.

```{r}
mons_o <- factor(mons,
                 levels = mons,
                 ordered = TRUE)

print(mons_o)
```
This can be useful when we want to order variable levels in a survey. A commonly used scale for survey data collection is a five-point Likert scale with varying levels of agreement with statement: "strongly agree", "agree", "neutral", "disagree" and "strongly disagree". We can create an ordered factor which will preserve the rank ordering of the levels.