Factors

Exercises

1. How are factors stored internally by R? To explore this, use the str function on a factor variable and see what it looks like?
2. To further this idea from #1, what happens when you do each of the following commands? Why is this happening?

as.numeric(resp)
as.numeric(resp_fact)

Reorder Factor Variables

To show examples of this operation, suppose we calculated the proportion of respondents that checked the weather daily by region of the country. We could use dplyr for this:

prop_check_weather <- weather_check %>%
  group_by(region) %>%
  summarise(prop = mean(ck_weather))
prop_check_weather

## # A tibble: 10 × 2
##    region              prop
##    <chr>              <dbl>
##  1 East North Central 0.858
##  2 East South Central 0.927
##  3 Middle Atlantic    0.885
##  4 Mountain           0.792
##  5 New England        0.942
##  6 Pacific            0.697
##  7 South Atlantic     0.740
##  8 West North Central 0.815
##  9 West South Central 0.904
## 10 <NA>               0.548

This would be a bit easier to view if we plotted this data:

ggplot(prop_check_weather, aes(prop, region)) + 
  geom_point()

This plot is difficult to read, primarily due to the way the points are ordered. Showing the regions in alphabetical order makes it more difficult to discern the trend. Instead, we would likely wish to reorder this variable by the ascending order of the proportion that check the weather. We will use the fct_reorder function from the forcats package. Note, I also omit the NA category here.

ggplot(na.omit(prop_check_weather), 
       aes(prop, fct_reorder(region, prop))) + 
  geom_point()

Need to be a bit careful with this operation however. For example:

weather_check %>%
  group_by(hhold_income) %>%
  summarise(prop = mean(ck_weather)) %>%
  na.omit() %>%
  ggplot(aes(prop, fct_reorder(hhold_income, prop))) + 
  geom_point()

Instead, this is the proper way to show this relationship:

weather_check %>%
  group_by(hhold_income) %>%
  summarise(prop = mean(ck_weather)) %>%
  na.omit() %>%
  ggplot(aes(prop, hhold_income)) + 
  geom_point()

Rename Factor Levels

These operations are useful to collapse categories or rename levels for publication. The primary function we will use for this operation is fct_recode from the forcats package.

Again, using the weather_check data, suppose we wished to change the levels of the age variable. The levels currently are:

levels(weather_check$age)

## [1] "18 - 29" "30 - 44" "45 - 59" "60+"

Suppose we wished to better represent these as words. We can use this with mutate from dplyr combined with fct_recode:

weather_check %>%
  mutate(age_recode = fct_recode(age,
      '18 to 29' = '18 - 29',
      '30 to 44' = '30 - 44',
      '45 to 59' = '45 - 59'
  )) %>%
  count(age_recode)

## # A tibble: 5 × 2
##   age_recode     n
##   <fct>      <int>
## 1 18 to 29     176
## 2 30 to 44     204
## 3 45 to 59     278
## 4 60+          258
## 5 <NA>          12

We could also collapse categories by assigning many levels to the same new level. For example, suppose we wished to collapse the ck_weather_watch variable to unlikely and likely instead of the very unlikely to very likely.

levels(weather_check$ck_weather_watch)

## [1] "Very unlikely"     "Somewhat unlikely" "Somewhat likely"  
## [4] "Very likely"

weather_check %>%
  mutate(watch_recode = fct_recode(ck_weather_watch,
         'Unlikely' = 'Very unlikely',
         'Unlikely' = 'Somewhat unlikely',
         'Likely' = 'Somewhat likely',
         'Likely' = 'Very likely'
  )) %>%
  count(watch_recode)

## # A tibble: 3 × 2
##   watch_recode     n
##   <ord>        <int>
## 1 Unlikely       281
## 2 Likely         636
## 3 <NA>            11

Finally, one last option that may be useful is to lump together categories that are too small to report independently. This functionality is implemented with the function fct_lump. For example, suppose we want to lump the region variable together to have only 5 regions.

weather_check %>%
  mutate(region = fct_lump(region, n = 5)) %>%
  count(region, sort = TRUE)

## # A tibble: 7 × 2
##   region                 n
##   <fct>              <int>
## 1 Other                219
## 2 Pacific              185
## 3 South Atlantic       154
## 4 East North Central   141
## 5 Middle Atlantic      104
## 6 West South Central    94
## 7 <NA>                  31