5.3 Lab: Sampling & randomizing

Eventually, we might find ourselves in a situation in which we have a sampling frame, i.e. a list of all units within a population who can be sampled. For budget reasons we can’t run our experiment with all units. So we will need to draw a random sample. Subsequently, we want to randomly assign units in our sample to treatment/control.

Below we create an artifical population, i.e. a dataframe that contains all units in the population. We call this dataframe population.

# Create exemplary sampling frame
population <- data.frame(id = sample(x = 1:99999, 2000, replace = FALSE),
                            firstname = randomNames(2000, which.names="first"),
                            lastname = randomNames(2000, which.names="last"),
                            age = sample(x = 18:85, 2000, replace = TRUE))
population$email <- paste(population$firstname, ".", 
                             population$lastname, "@example", 
                             sample(1:9, 2000, replace = TRUE),".",  
                             sample(c("de", "com", "org", "es"), 2000, replace = TRUE), 
                             sep = "") 

The population, i.e., sampling frame contains 2000 units. First, we would like to draw a sample of 200 individuals among whom we conduct our experiment. Below we do so and have a look at the first six units in our sample.

sample <- sample_n(population, 200) # create new object containing sample
head(sample)

id	firstname	lastname	age	email
97102	Maaiz	Park	81	Maaiz.Park@example8.de
6365	Morganna	al-Pour	65	Morganna.al-Pour@example2.es
78890	Gedion	al-Fahmy	38	Gedion.al-Fahmy@example3.com
82562	Sirmichael	Braud	41	Sirmichael.Braud@example7.com
98332	Reanne	Rodriguez	35	Reanne.Rodriguez@example5.es
89963	Sareena	Saracay Pena	50	Sareena.Saracay Pena@example8.com

“Simple random assignment assigns all subjects to treatment with an equal probability by flipping a (weighted) coin for each subject.” (Source). To do so we need a variable that contains units’ assignment status. Accordingly, we create a variable treatment in which we store that information.

We can do that for two treatment groups.

sample$treatment <- simple_ra(N = nrow(sample))
head(sample)

id	firstname	lastname	age	email	treatment
97102	Maaiz	Park	81	Maaiz.Park@example8.de	1
6365	Morganna	al-Pour	65	Morganna.al-Pour@example2.es	0
78890	Gedion	al-Fahmy	38	Gedion.al-Fahmy@example3.com	1
82562	Sirmichael	Braud	41	Sirmichael.Braud@example7.com	1
98332	Reanne	Rodriguez	35	Reanne.Rodriguez@example5.es	0
89963	Sareena	Saracay Pena	50	Sareena.Saracay Pena@example8.com	1

…for three treatment groups (below names are assigned automatically).

sample$treatment <- simple_ra(N = nrow(sample), num_arms = 3)
head(sample)

id	firstname	lastname	age	email	treatment
97102	Maaiz	Park	81	Maaiz.Park@example8.de	T3
6365	Morganna	al-Pour	65	Morganna.al-Pour@example2.es	T3
78890	Gedion	al-Fahmy	38	Gedion.al-Fahmy@example3.com	T2
82562	Sirmichael	Braud	41	Sirmichael.Braud@example7.com	T1
98332	Reanne	Rodriguez	35	Reanne.Rodriguez@example5.es	T2
89963	Sareena	Saracay Pena	50	Sareena.Saracay Pena@example8.com	T3

Finally, an example with four treatment groups, where we decide about the assignment probabilities and name the values of the treatment variable (character vector).

sample$treatment <- simple_ra(N = nrow(sample),
                               prob_each = c(.25, .25, .25, .25),
                               condition =c("treatment1", 
                                            "treatment2",
                                            "treatment3",
                                            "control"))
head(sample)

id	firstname	lastname	age	email	treatment
97102	Maaiz	Park	81	Maaiz.Park@example8.de	control
6365	Morganna	al-Pour	65	Morganna.al-Pour@example2.es	treatment3
78890	Gedion	al-Fahmy	38	Gedion.al-Fahmy@example3.com	control
82562	Sirmichael	Braud	41	Sirmichael.Braud@example7.com	treatment1
98332	Reanne	Rodriguez	35	Reanne.Rodriguez@example5.es	control
89963	Sareena	Saracay Pena	50	Sareena.Saracay Pena@example8.com	control

Once, we have randomly assigned our sample units to treatment groups we can conduct our experiment and assign them the real treatments based on our dataframe. For instance, we might send them different versions of a questionnaire etc.

The randomizr package contains further useful functions arguments and there are other forms of random assignment (???). See the randomizr vignette.