DeclareDesign (version 1.1.0)

redesign: Redesign

Description

redesign quickly generates a design from an existing one by resetting symbols used in design handler parameters in a step's environment (Advanced).

Usage

redesign(design, ..., expand = TRUE)

Value

A design, or, in the case of multiple values being passed onto ..., a by-list of designs.

Arguments

design

An object of class design.

...

Arguments to redesign e.g., n = 100. If redesigning multiple arguments, they must be specified as a named list. lists should also be used if redesigning with respect to a dataset or with respect to a vector. For instance, redesign(design, df = list(new_df)).

expand

If TRUE, redesign using the crossproduct of ..., otherwise recycle them.

Details

If you attempt to change a parameter that is not saved into a design you will receive a message but not an error.

Examples

# Two-arm randomized experiment
n <- 500

design <-
  declare_model(
    N = 1000,
    gender = rbinom(N, 1, 0.5),
    X = rep(c(0, 1), each = N / 2),
    U = rnorm(N, sd = 0.25),
    potential_outcomes(Y ~ 0.2 * Z + X + U)
  ) +
  declare_inquiry(ATE = mean(Y_Z_1 - Y_Z_0)) +
  declare_sampling(S = complete_rs(N = N, n = n)) +
  declare_assignment(Z = complete_ra(N = N, m = n/2)) +
  declare_measurement(Y = reveal_outcomes(Y ~ Z)) +
  declare_estimator(Y ~ Z, inquiry = "ATE")

# Use redesign to return a single modified design
modified_design <- redesign(design, n = 200)

# Use redesign to return a series of modified designs
## Sample size is varied while the rest of the design remains
## constant
design_vary_N <- redesign(design, n = c(100, 500, 900))

if (FALSE) {
# redesign can be used in conjunction with diagnose_designs
# to optimize the design for specific diagnosands
diagnose_designs(design_vary_N)
}

# When redesigning with arguments that are vectors,
# use list() in redesign, with each list item
# representing a design you wish to create

prob_each <- c(.1, .5, .4)

population <- declare_model(N = 1000)
assignment <- declare_assignment(
  Z = complete_ra(prob_each = prob_each), 
  legacy = FALSE)

design <- population + assignment

## returns two designs

designs_vary_prob_each <- redesign(
  design,
  prob_each = list(c(.2, .5, .3), c(0, .5, .5)))

# To illustrate what does and does not get edited by redesign, 
# consider the following three designs. In the first two, argument
# X is called from the step's environment; in the third it is not.
# Using redesign will alter the role of X in the first two designs
# but not the third one.

X <- 3
f <- function(b, X) b*X
g <- function(b) b*X

design1 <- declare_model(N = 1, A = X)       + NULL
design2 <- declare_model(N = 1, A = f(2, X)) + NULL
design3 <- declare_model(N = 1, A = g(2))    + NULL

draw_data(design1)
draw_data(design2)
draw_data(design3)

draw_data(redesign(design1, X=0))
draw_data(redesign(design2, X=0))
draw_data(redesign(design3, X=0))