Non-binary Treatment Specification
2025-01-17
Source:vignettes/non-binary-treatment.Rmd
non-binary-treatment.RmdBinary Treatment
When creating a StudySpecification, handling binary treatment
variables is straightforward. If the treatment variable is either
numeric with only values 0/1, or is logical,
then lmitt() will estimate a treatment effect of the
difference between the outcome in the treated group (1 or
TRUE) versus the control group (0 or
FALSE).
Missing treatment status
In all cases (binary and non-binary), missing values are allowed and
any units of assignment with missing treatment values are excluded from
models fit via lmitt().
Non-binary Treatment
However, the _spec()
functions can take in any (reasonable) form of treatment assignment.
If the treatment variable is a numeric with non-binary
values, it is treated as a continuous treatment effect and
lmitt(y ~ 1, ... will estimate a single coefficient on
treatment.
If the treatment variable is a character, it is treated
as a multi-level treatment variable and lmitt(y ~ 1, ...
will estimate treatment effects against a reference category. The
reference category is the first level defined according to R’s comparison of
characters.
factor and ordered objects are tricky to
deal with, so while a StudySpecification can be created
with factor or ordered treatment variables,
lmitt() will refuse to estimate a model unless it is also
provided a dichotomy
(see below).
Dichotomzing a Non-binary Treatment
Studies may offer treatment to units at different times or provide
treatment to units in varying intensities. Researchers may be interested
in estimating treatment effects at different times or given a certain
threshold of provided treatment, however. propertee
accommodates these wishes by storing the time or intensity of treatment
for treated units in the StudySpecification,
then offering a dichotomy= argument to the weights
calculation functions ett()/ate() and the
assginment creation function assigned() A
dichotomy is presented as a formula, where the left-hand
side is a logical statement defining inclusion in the treatment group,
and the right-hand side is a logical statement defining inclusion in the
control group. For example, if dose represents the
intensity of a given treatment, we could set a threshold of 200, say,
mg:
dose > 200 ~ dose <= 200All units of assignment with dose above 200 are treated
units, and all units of assignment with dose of 200 or
below are control units.
A . can be used to define either group as the inverse of
the other. For example, the above dichotomy could be defined as either
of
dose > 200 ~ .
. ~ dose <= 200Any units of assignment not assigned to either treatment or control
are assumed to have NA for a treatment status and will be
ignored in the estimation of treatment effects.
dose >= 300 ~ dose <= 100In this dichotomy, units of assignment in the range
(100,300) are ignored.
An Example
data(simdata)
table(simdata$dose)
#>
#> 50 100 200 250 300
#> 10 10 10 10 10
spec1 <- rct_spec(dose ~ uoa(uoa1, uoa2), data = simdata)
summary(spec1)
#> Randomized Control Trial
#>
#> Structure Variables
#> --------- ---------
#> Treatment dose
#> Unit of Assignment uoa1, uoa2
#>
#> Number of units per Treatment group:
#> Txt Grp Num Units
#> 50 2
#> 100 2
#> 200 2
#> ...
#> 2 smaller treatment groups excluded.
#> Use `specification_table` function to view full results.