sim_ordbeta {ordbetareg}

Power Calculation via Simulation of the Ordered Beta Regression Model

Description

This function allows you to calculate power curves (or anything else) via simulating the ordered beta regression model.

Usage

sim_ordbeta(
  N = 1000,
  k = 5,
  iter = 1000,
  cores = 1,
  phi = 1,
  cutpoints = c(-1, 1),
  beta_coef = NULL,
  beta_type = "continuous",
  treat_assign = 0.5,
  return_data = FALSE,
  seed = as.numeric(Sys.time()),
  ...
)

Arguments

N

The sample size for the simulation. Include a vector of integers to examine power/results for multiple sample sizes.

k

The number of covariates/predictors.

iter

The number of simulations to run. For power calculation, should be at least 500 (yes, this will take some time).

cores

The number of cores to use to parallelize the simulation.

phi

Value of the dispersion parameter in the beta distribution.

cutpoints

Value of the two cutpoints for the ordered model. By default are the values -1 and +1 (these are interpreted in the logit scale and so should not be too large). The farther apart, the fewer degenerate (0 or 1) responses there will be in the distribution.

beta_coef

If not null, a vector of length k of the true predictor coefficients/treatment values to use for the simulation. Otherwise, coefficients are drawn from a random uniform distribution from -1 to 1 for each predictor.

beta_type

Can be either continuous or binary. Use the latter for conventional treatments with two values.

treat_assign

If beta_type is set to binary, you can use this parameter to set the proportion of N assigned to treatment. By default, the parameter is set to 0.5 for equal/balanced treatment control groups.

return_data

Whether to return the simulated dqta as a list in the data column of the returned data frame.

seed

The seed to use to make the results reproducible. Set automatically to a date-time stamp.

...

Any other arguments are passed on to the brms::brm function to control modeling options.

Details

This function implements the simulation found in Kubinec (2022). This simulation allows you to vary the sample size, number & type of predictors, values of the predictors (or treatment values), and the power to target. The function returns a data frame with one row per simulation draw and covariate k.

Value

a tibble data frame with columns of simulated and estimated values and rows for each simulation iteration X coefficient combination. I.e., if there are five predictors, and 1,000 iterations, the resulting data frame will have 1,000 rows. If there are multiple values for N, then each value of N will have its own set of iterations, making the final size of the data a multiple of the number of sample sizes to iterate over. The data frame will have the following columns: 1.

Examples

# This function takes a while to run as it has
# to fit an ordered beta regression to each
# draw. The package comes with a saved
# simulation dataset you can inspect to see what the
# result looks like

data("sim_data")

library(dplyr)

# will take a while to run this


  if(.Platform$OS.type!="windows") {

    sim_data <- sim_ordbeta(N=c(250,750),
    k=1,
    beta_coef = .5,
    iter=5,cores=2,
    beta_type="binary",
    treat_assign=0.3)

    }



# to get the power values by N, simply summarize/group
# by N with functions from the R package dplyr

sim_data %>%
  group_by(N) %>%
  summarize(mean_power=mean(power))