3.3 Recoding a variable
You might want to change the shape of a variable. Let’s look at two examples. One is taking a continuous variable and recoding it to be a categorical variable. There are sometimes reasons in doing analysis that we may want to do this. The second is recoding a categorical variable into a new categorical variable.
For this we are going to use another new package, sjmisc. You will need to both install and load this package. Details are in section 1.7 and section 1.8 if you need help.
Note to run the frequency you use the command freq.
3.3.1 Recode a continuous variable to a categorical variable
States and Gapminder
You might want a continuous variable to be in categories for analysis. The example uses the Gapminder dataset. We will divide the mortality rate into two categories: high mortality and low mortality. To ease interpretation, we add labels. We will use the mutate command from the dplyr package and the dicho and split_var commands from the sjMisc package. For each new variable, we will do the following:
- Filter data so we have only data for one year.
- Create a new variable with the categorical variable.
- Add value labels using the ordered command.
- After creating the new variable, we will look at the frequency distribution.
library(sjmisc)
my_gapminder_data_2003 <- my_gapminder_data_2003 %>%
mutate(mortality_rate_dicho = dicho(mortalityRate))
freq(my_gapminder_data_2003$mortality_rate_dicho)## Frequencies
## my_gapminder_data_2003$mortality_rate_dicho
## Type: Factor
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## ----------- ------ --------- -------------- --------- --------------
## 0 83 50.92 50.92 50.92 50.92
## 1 80 49.08 100.00 49.08 100.00
## <NA> 0 0.00 100.00
## Total 163 100.00 100.00 100.00 100.00my_gapminder_data_2003$mortality_rate_dicho <- ordered(my_gapminder_data_2003$mortality_rate_dicho,
levels = c(0,1),
labels = c("Low Mortality", "High Mortality"))
freq(my_gapminder_data_2003$mortality_rate_dicho)## Frequencies
## my_gapminder_data_2003$mortality_rate_dicho
## Type: Ordered Factor
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## -------------------- ------ --------- -------------- --------- --------------
## Low Mortality 83 50.92 50.92 50.92 50.92
## High Mortality 80 49.08 100.00 49.08 100.00
## <NA> 0 0.00 100.00
## Total 163 100.00 100.00 100.00 100.00The dicho command splits the variable in half. The countries with lower mortality rates are labeled “Low Mortality.” The upper half of countries are labeled “High Mortality.”
Below the split_var command splits the variable, the percentage of women in state legislatures in the States dataset, into thirds. We could change n = to a larger number to create more evenly sized groups. [Many more options are here.][https://strengejacke.github.io/sjmisc/articles/recodingvariables.html]
my_usstate_data_2011_small <- my_usstate_data %>%
filter(year == 2011 & st != "DC") %>%
select(state, gini_coef, incshare_top1, pctwomenearn, pctfemaleleg, evangelical_pop, region, povrate)
freq(my_usstate_data_2011_small$pctfemaleleg)## Frequencies
## my_usstate_data_2011_small$pctfemaleleg
## Type: Numeric
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## ----------- ------ --------- -------------- --------- --------------
## 9.4 1 2.00 2.00 2.00 2.00
## 12.8 1 2.00 4.00 2.00 4.00
## 13.6 1 2.00 6.00 2.00 6.00
## 14.4 1 2.00 8.00 2.00 8.00
## 14.9 2 4.00 12.00 4.00 12.00
## 16 1 2.00 14.00 2.00 14.00
## 17.3 1 2.00 16.00 2.00 16.00
## 17.4 1 2.00 18.00 2.00 18.00
## 17.9 1 2.00 20.00 2.00 20.00
## 18.2 1 2.00 22.00 2.00 22.00
## 18.8 1 2.00 24.00 2.00 24.00
## 19.3 1 2.00 26.00 2.00 26.00
## 20 1 2.00 28.00 2.00 28.00
## 20.9 1 2.00 30.00 2.00 30.00
## 21 1 2.00 32.00 2.00 32.00
## 21.2 1 2.00 34.00 2.00 34.00
## 21.3 2 4.00 38.00 4.00 38.00
## 22.2 1 2.00 40.00 2.00 40.00
## 22.4 2 4.00 44.00 4.00 44.00
## 22.7 1 2.00 46.00 2.00 46.00
## 23.3 1 2.00 48.00 2.00 48.00
## 23.7 1 2.00 50.00 2.00 50.00
## 23.9 1 2.00 52.00 2.00 52.00
## 24 1 2.00 54.00 2.00 54.00
## 24.5 1 2.00 56.00 2.00 56.00
## 24.8 1 2.00 58.00 2.00 58.00
## 25.6 1 2.00 60.00 2.00 60.00
## 25.7 1 2.00 62.00 2.00 62.00
## 25.8 2 4.00 66.00 4.00 66.00
## 26.8 1 2.00 68.00 2.00 68.00
## 27.3 1 2.00 70.00 2.00 70.00
## 27.6 1 2.00 72.00 2.00 72.00
## 27.8 1 2.00 74.00 2.00 74.00
## 28.3 2 4.00 78.00 4.00 78.00
## 28.6 1 2.00 80.00 2.00 80.00
## 29 1 2.00 82.00 2.00 82.00
## 29.9 1 2.00 84.00 2.00 84.00
## 30.9 1 2.00 86.00 2.00 86.00
## 31.1 1 2.00 88.00 2.00 88.00
## 31.3 1 2.00 90.00 2.00 90.00
## 32 1 2.00 92.00 2.00 92.00
## 34.2 1 2.00 94.00 2.00 94.00
## 34.4 1 2.00 96.00 2.00 96.00
## 38.3 1 2.00 98.00 2.00 98.00
## 41 1 2.00 100.00 2.00 100.00
## <NA> 0 0.00 100.00
## Total 50 100.00 100.00 100.00 100.00my_usstate_data_2011_small <- my_usstate_data_2011_small %>%
mutate(pctfemaleleg_3cat = split_var(pctfemaleleg, n = 3))
my_usstate_data_2011_small$pctfemaleleg_3cat <- ordered(my_usstate_data_2011_small$pctfemaleleg_3cat,
levels = c(1,2, 3),
labels = c("Low % Women", "Medium % Women", "High % Women"))
freq(my_usstate_data_2011_small$pctfemaleleg_3cat)## Frequencies
## my_usstate_data_2011_small$pctfemaleleg_3cat
## Type: Ordered Factor
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## -------------------- ------ --------- -------------- --------- --------------
## Low % Women 17 34.00 34.00 34.00 34.00
## Medium % Women 16 32.00 66.00 32.00 66.00
## High % Women 17 34.00 100.00 34.00 100.00
## <NA> 0 0.00 100.00
## Total 50 100.00 100.00 100.00 100.003.3.2 Recode a new categorical variable from an existing categorical variable
WVS
Sometimes you may want to recode your variable so that it has fewer categories. For example, look at the question below from the Ethiopia VWS dataset. It asks how democratically is this country being governed today. There are ten categories, ranging from “Not at all democratic” to “Completely Democratic.”
## Frequencies
## ethiopia_small$Q251
## Label: How democratically is this country being governed today
## Type: Factor
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## --------------------------- ------ --------- -------------- --------- --------------
## Not at all democratic 218 17.96 17.96 17.72 17.72
## 2 54 4.45 22.41 4.39 22.11
## 3 46 3.79 26.19 3.74 25.85
## 4 73 6.01 32.21 5.93 31.79
## 5 282 23.23 55.44 22.93 54.72
## 6 123 10.13 65.57 10.00 64.72
## 7 94 7.74 73.31 7.64 72.36
## 8 89 7.33 80.64 7.24 79.59
## 9 37 3.05 83.69 3.01 82.60
## Completely democratic 198 16.31 100.00 16.10 98.70
## <NA> 16 1.30 100.00
## Total 1230 100.00 100.00 100.00 100.00This table is hard to interpret. To ease interpretation, we will recode the data into three groups:
- For people who answered 1 through 3. We’ll call this “Not Democratic.”
- For people who answered 4 through 7. We’ll call this “Somewhat Democratic.”
- For people who answered 8 through 10. We’ll call this “Democratic.”
We’ll use the fct_collapse command from forcats, part of the tidyverse here.
Note we gave the new variable a new name—Q251_recode—to distinguish it from Q251. This is good practice. We may want the original variable later!
Look at the frequency of the new variable. It gives us a better picture of how Ethiopians feel about the success of democracy in their country.
ethiopia_small$Q251_recode <- fct_collapse(ethiopia_small$Q251,
"Not Democratic" = c("Not at all democratic", "2", "3"),
"Somewhat Democratic" = c("4", "5", "6", "7"),
"Democratic" = c("8", "9", "Completely democratic"))
freq(ethiopia_small$Q251_recode)## Frequencies
## ethiopia_small$Q251_recode
## Label: How democratically is this country being governed today
## Type: Factor
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## ------------------------- ------ --------- -------------- --------- --------------
## Not Democratic 318 26.19 26.19 25.85 25.85
## Somewhat Democratic 572 47.12 73.31 46.50 72.36
## Democratic 324 26.69 100.00 26.34 98.70
## <NA> 16 1.30 100.00
## Total 1230 100.00 100.00 100.00 100.00A second example For this second example, we will turn a 5-point scale into a 3-point scale.
To simplify things, we will convert this to three responses: agree, neither agree or disagree, and disagree. We use the forcats fct_collapse command.
## Frequencies
## ethiopia_small$Q34
## Label: Jobs scarce: Employers should give priority to (nation) people than immigrants
## Type: Factor
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## -------------------------------- ------ --------- -------------- --------- --------------
## Agree strongly 541 44.09 44.09 43.98 43.98
## Agree 372 30.32 74.41 30.24 74.23
## Neither agree nor disagree 43 3.50 77.91 3.50 77.72
## Disagree 206 16.79 94.70 16.75 94.47
## Disagree strongly 65 5.30 100.00 5.28 99.76
## <NA> 3 0.24 100.00
## Total 1230 100.00 100.00 100.00 100.00ethiopia_small$Q34_recode <- fct_collapse(ethiopia_small$Q34,
"Agree" = c("Agree strongly", "Agree"),
"Neither agree nor disagree" = c("Neither agree nor disagree"),
"Disagree" = c("Disagree", "Disagree strongly"))
freq(ethiopia_small$Q34_recode)## Frequencies
## ethiopia_small$Q34_recode
## Label: Jobs scarce: Employers should give priority to (nation) people than immigrants
## Type: Factor
##
## Freq % Valid % Valid Cum. % Total % Total Cum.
## -------------------------------- ------ --------- -------------- --------- --------------
## Agree 913 74.41 74.41 74.23 74.23
## Neither agree nor disagree 43 3.50 77.91 3.50 77.72
## Disagree 271 22.09 100.00 22.03 99.76
## <NA> 3 0.24 100.00
## Total 1230 100.00 100.00 100.00 100.00