Non-binary Treatment Specification
2024-10-29
Source:vignettes/non-binary-treatment.Rmd
non-binary-treatment.Rmd
Binary Treatment
When creating a Design, 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 _design()
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 Design
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 Design
, 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 <= 200
All 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 <= 200
Any 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 <= 100
In 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
des1 <- rct_design(dose ~ uoa(uoa1, uoa2), data = simdata)
summary(des1)
#> 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 `dtable` function to view full results.