trop_simulate() makes a balanced panel with unit effects, time effects, a
low-rank interactive term, and Gaussian noise. Each unit gets a score, which
is selection times its first factor loading plus random noise. The units
with the highest scores get the treatment.
With a positive value of selection, a unit with a larger loading has a
higher probability of treatment. With a negative value, a unit with a smaller
loading has a higher probability. With zero, the loading has no effect on
the selection. Imputation with two-way fixed effects (TWFE imputation) does
not model this factor, so its estimate for one panel can have a large error.
TROP can reduce such errors, but some error can stay.
Usage
trop_simulate(
n_units = 40L,
n_periods = 30L,
n_treated = 8L,
n_post = 5L,
effect = 1,
rank = 2L,
sigma = 0.5,
selection = 1,
design = c("block", "staggered"),
seed = NULL
)Arguments
- n_units, n_periods
The number of units and the number of periods.
- n_treated
The number of treated units.
- n_post
For a block design, the number of treated periods at the end of the panel. For a staggered design, the treatment starts in one of the last
n_postperiods.- effect
The constant treatment effect on each treated cell.
- rank
The rank of the interactive term.
- sigma
The standard deviation of the noise.
- selection
The weight of the first factor loading in the selection of the treated units. With zero, the selection is random.
- design
Use
"block"or"staggered".- seed
An optional seed. The function does not change the state of the random number generator in the R session.
Value
A data frame with the columns unit, time, y, d, and y0.
The column y0 contains the outcome without the treatment.
Examples
head(trop_simulate(seed = 1))
#> unit time y d y0
#> 1 1 1 0.2388666 0 0.2388666
#> 2 2 1 0.5934588 0 0.5934588
#> 3 3 1 -2.5737706 0 -2.5737706
#> 4 4 1 1.2415187 0 1.2415187
#> 5 5 1 0.4650530 0 0.4650530
#> 6 6 1 0.5049260 0 0.5049260