The purpose of this in-class lab is to practice using dummy variables in R. The lab should be completed in your group. To get credit, upload your .R script to the appropriate place on Canvas.

For starters

Open up a new R script (named ICL6_XYZ.R, where XYZ are your initials) and add the usual “preamble” to the top:

# Add names of group members HERE
library(tidyverse)
library(broom)
library(wooldridge)
library(modelsummary)

Also install the package magrittr by typing in the console:

install.packages("magrittr", repos='http://cran.us.r-project.org')

and then add to the preamble of your script

library(magrittr)

The magrittr package contains extra features for writing even more expressive code.

Load the data

We’ll use a new data set on extramarital affairs, called affairs.

df <- as_tibble(affairs)

Check out what’s in the data by typing

datasummary_skim(df,histogram=FALSE)

You’ll notice that there are a number of variables that only take on 0/1 values: male, kids, affair, hapavg, vryrel, etc. There are also variables that take on a few different values: relig, occup, and ratemarr.

Creating factor variables

Let’s convert our 0/1 numeric variable male to a factor with levels "yes" and "no":

df %<>% mutate(male = factor(male), male = fct_recode(male, yes = "1", no = "0"))

The df %<>% mutate(...) uses the %<>% operator which is shorthand for df <- df %>% mutate(...). In other words, %<>% pipes forwards and then pipes everything backwards. This saves a little bit on syntax if you want to overwrite your data frame.

The first part of the mutate() function converts the 0/1 values to categories named "0" and "1". The second part gives the categories more descriptive labels ("yes" and "no").

Let’s repeat this for some of the other variables: ratemarr, relig, kids, and affair:

df %<>% mutate(ratemarr = factor(ratemarr), 
               ratemarr = fct_recode(ratemarr, very_happy = "5", happy = "4", average = "3", 
                                     unhappy = "2", very_unhappy = "1")) %>%
        mutate(relig = factor(relig), 
               relig = fct_recode(relig, very_relig = "5", relig = "4", average = "3", 
                                  not_relig = "2", not_at_all_relig = "1")) %>%
        mutate(kids  = factor(kids), kids = fct_recode(kids, yes = "1", no = "0")) %>%
        mutate(affair = factor(affair), affair = fct_recode(affair, yes = "1", no = "0"))

where we used multiple pipe operators (%>%) to do each factor recode in a separate step. (We could have also put them all into one giant mutate() statement.)

Do another datasummary_skim(df,histogram=FALSE) to make sure the code worked in the way I told you it would. You’ll notice that the factor variables now disappear from the datasummary_skim() output. To have them show up separately, follow the steps in the next section.

Summary stats of factor variables

You can look at the frequency of factor variables using the datasummary_skim() function from the modelsummary package, or by using the table() function:

datasummary_skim(df, type="categorical", histogram=FALSE)
table(df$relig)
table(df$ratemarr,df$kids)

You can also use the prop.table() function to get shares within-row (margin=1) or within-column (margin=2):

table(df$ratemarr) %>% prop.table()
table(df$ratemarr,df$kids) %>% prop.table(margin=1)
table(df$ratemarr,df$kids) %>% prop.table(margin=2)

You can also create a histogram of a factor variable in ggplot() as follows:

ggplot(df,aes(x=ratemarr)) + geom_bar()

This helps you visualize what share of the data falls into which category.

Multiple regression with factor variables

Let’s run a regression with naffairs as the dependent variable and male, yrsmarr, kids, and ratemarr as the covariates.

est1 <- lm(naffairs ~ male + yrsmarr + kids + ratemarr, data=df)

Interpret the coefficient on ratemarrvery_happy.

Linear Probability Model

Let’s run the same regression as before, but this time use affair as the dependent variable. What happens when you run the following code?

est2 <- lm(affair ~ male + yrsmarr + kids + ratemarr, data=df)

R doesn’t want you to run a LPM because R was designed by statisticians who focus more on the “cons” of LPMs than on the “pros.”

To run the LPM, adjust the code by using as.numeric(affair) as the dependent variable. Interpret the coefficients on ratemarraverage and kidsyes.

Interaction terms

Finally, let’s run a more flexible model where we allow the effect of fathers and mothers to be different. The way to do this in lm() is as follows:

est3 <- lm(as.numeric(affair) ~ male*kids + yrsmarr + ratemarr, data=df)
print(tidy(est3))

The coefficient on the interaction term is labeled maleyes:kidsyes. Do fathers have a differential rate of extramarital affairs compared to mothers?