set_cov: Set up covariates
In psborrow: Bayesian Dynamic Borrowing with Propensity Score

set_cov

R Documentation

Set up covariates

Description

This function saves the mean, variance and covariance among covariates. For technical details, see the vignette.

Usage

set_cov(n_cat, n_cont, mu_int, mu_ext, var, cov, prob_int, prob_ext)

Arguments

`n_cat`	Number of binary variable. See details
`n_cont`	Number of continuous variable
`mu_int`	Mean of covariates in the internal trial. All the covariates are simulated from a multivariate normal distribution. If left `NULL`, it uses default value 0 for all covariates. If provided one value, this value is used for all covariates
`mu_ext`	Mean of covariates in the external trial. If left `NULL`, it uses the same mean as `mu_int`
`var`	Variance of covariates. If left `NULL`, it uses default value 0 for all covariates. If provided one value, it uses this value for all covariates
`cov`	Covariance between each pair of covariates. Covariance needs to be provided in a certain order and users are encouraged to read the example provided in the vignette. If left `NULL`, it uses default value 0 for all covariates. If provided one value, it uses this value for every pair of covariates
`prob_int`	Probability of binary covariate equalling 1 in the internal trial. If left `NULL`, it uses default value 0.5 for all covariates. If provided one value, it uses this value for all covariates
`prob_ext`	Probability of binary covariate equalling 1 in the external trial. If left `NULL`, it uses the same probability as `prob_int`

Details

Categorical variables are created by sampling a continuous variable from the multivariate normal distribution (thus respecting the correlation to other covariates specified by cov) and then applying a cut point derived from the prob_int or prob_ext quantile of said distribution i.e. for a univariate variable it would be derived as:

binvar <- as.numeric(rnorm(n, mu, sqrt(var)) < qnorm(prob, mu, sqrt(var)))

Please note that this means that the value of mu_int & mu_ext has no impact on categorical covariates and thus can be set to any value.

As an example of how this process works assume n_cat=3 and n_cont=2. First 5 variables are sampled from the multivariate normal distribution as specified by mu_int/mu_ext, var & cov. Then, the first 3 of these variables are converted to binary based on the probabilities specified by prob_int and prob_ext. This means that that the 2 continuous variables will take their mean and sd from the last 2 entries in the vectors mu_int/mu_ext and var.