stack_bpg {multilevelmediation}

Stacks data in the style of Bauer-Preacher-Gil for multilevel mediation

Description

Stacks data in the style of Bauer-Preacher-Gil for multilevel mediation

Usage

stack_bpg(data, L2ID, X, Y, M,
  moderator = NULL,
  covars.m = NULL,
  covars.y = NULL)

Arguments

data

Data frame in long format.

L2ID

(String) Name of column that contains grouping variable in data (e.g., "SubjectID").

X

(String) Name of column that contains the X independent variable in data.

Y

(String) Name of column that contains the Y dependent variable in data.

M

(String) Name of column that contains the M mediating variable in data.

moderator

Optional Character that contains name of column that contains the moderator variable in data

covars.m

(Character vector) Optional covariates to include in the model for M.

covars.y

(Character vector) Optional covariates to include in the model for Y.

Details

This is a convenience function used primarily internally by the package to restructure data in the style of Bauer, Preacher, and Gil (2006). The point is to allow both Y and M to be outcomes in a single column ("Z"), so that both mediator and outcome models can be fit at the same time. This is necessary to estimate the covariance between "a" and "b" paths at the same time when both have random effects. Two selector variables, "Sy" and "Sm" toggle whether each row corresponds to the outcome or the mediator, respectively.

Value

An object that is a subclass of data.frame is returned. In particular a tbl_df or "tibble."

So that coefficients extracted later from modmed.mlm hopefully make more sense, the below lists all variables (i.e., typical column names) for the data frame.

• X Independent variable.

• L2id Level 2 ID variable.

• Md Value of the mediator (not necessarily used due to restructuring, however).

• Outcome Whether the row corresponds to M or Y as the outcome.

• Z Value of the outcome variable.

• Sy Indicator variable for Y as outcome. 0 if Y is not outcome for this row. 1 if Y is outcome for this row. In model output, this is the intercept for Y.

• Sm Indicator variable for M as outcome. 0 if M is not outcome for this row. 1 if M is outcome for this row. In model output, this is the intercept for M.

• SmX Value of X when M is the outcome (literally X times Sm); will be 0 when Y is outcome. In model output, this is the "a" path.

• SyX Value of X when Y is the outcome (literally X times Sy); will be 0 when M is outcome. In model output, this is the "cprime" path (direct effect).

• SyM Calue of M when Y is the outcome (literally M times Sy); will be 0 when M is the outcome. In model output, this is the "a" path.

• W Value of any moderating variable. This may show up in output later on as "SmX:W" (interaction with "a" path) or "SyM:W" (interaction with "b" path) or "SyX:W" (interaction with "cprime" path) depending on which path it moderates.

• If covars.m or covars.y are not null, any additional covariates will also be added to the data frame and their original names will be retained.

When modmed.mlm is used, any output with an "re" prefix will correspond to a random effect. Note that estimation with brms will result in slightly different output than listed here. Coefficients typically have a "b_" prefix, and random effects are parameterized not such that we end up with covariances, but using correlations and standard deviations for each effect.

References

Bauer, D. J., Preacher, K. J., & Gil, K. M. (2006). Conceptualizing and testing random indirect effects and moderated mediation in multilevel models: new procedures and recommendations. Psychological Methods, 11(2), 142-163. doi:10.1037/1082-989X.11.2.142

Examples

# restructure BPG data
data(BPG06dat)

dat <- stack_bpg(BPG06dat,
  "id", "x", "m", "y"
)

head(dat)

# restructure simulated data w/ moderator
data(simdat)
dat2 <- stack_bpg(simdat,
  "L2id", "X", "M", "Y",
  moderator = "mod"
)

head(dat2)