Synthesising Data from Marginal Distributions

Data is synthesised by sampling from a multivariate cumulative distribution (Copula), using the simstudy package.

Without Correlations

Data can be synthesised from marginal distributions using the synthesise_data() function:

library(RESIDE)
marginals <- import_marginal_distributions()
simulated_data <- synthesise_data(marginals)

With correlations

User specified correlations can be added to the synthesised data by supplying a list of correlations to the correlations parameter. Each correlation is created with the correlation() function, giving the names of the two variables and the correlation (rho) between them:

library(RESIDE)
marginals <- import_marginal_distributions()
simulated_data <- synthesise_data(
  marginals,
  correlations = list(
    correlation("AGE", "RSBP", 0.3)
  )
)

Correlations should be between -1 and 1 and must be consistent with each other (the resulting correlation matrix must be positive semi definite), otherwise synthesise_data() will produce an error.

Categorical variables

Categorical variables are correlated using a single category, specified with the factor_name.x or factor_name.y parameters for the first or second variable respectively. For example to correlate male patients (SEX of M) with age:

simulated_data <- synthesise_data(
  marginals,
  correlations = list(
    correlation("SEX", "AGE", -0.2, factor_name.x = "M"),
    correlation("RATRIAL", "RSBP", 0.1, factor_name.x = "Y")
  )
)

A categorical variable without a factor name will produce an error.

Multiple tables

When the marginals were obtained from multiple data frames, variables that are present in more than one data frame (other than the common columns) must specify which data frame they belong to. Use df_name when both variables belong to the same data frame, or df_name.x and df_name.y when they belong to different data frames:

simulated_data <- synthesise_data(
  marginals,
  correlations = list(
    # Both variables in the same data frame
    correlation("DOMAIN", "AESTDY", 0.2, df_name = "ae", factor_name.x = "AE"),
    # Variables in different data frames
    correlation(
      "SEX",
      "AESEV",
      0.3,
      df_name.x = "dm",
      df_name.y = "ae",
      factor_name.x = "M",
      factor_name.y = "SEVERE"
    )
  )
)

A data frame name specified for a variable that only appears in a single data frame, but not that data frame, will be ignored with a warning.

Correlated variables are synthesised together, with one row per subject, and joined to each data frame by subject. Correlations between data frames are therefore between subjects, and a correlated variable takes a single value for each subject within a data frame.

NB The correlation specified is that of the underlying multivariate normal distribution (Copula), the correlation observed in the synthesised data will be weaker, particularly for binary and categorical variables. It is not possible to entirely maintain all the marginal distributions when specifying correlations, this is a known limitation and is not likely to change.