piecewiseSEM (version 1.2.1)

sem.coefs: Retrieves path coefficients for structural equation model

Description

Extracts path coefficients for a piecewise structural equation model (SEM).

Usage

sem.coefs(modelList, data, standardize = "none", corr.errors, intercept = FALSE)

Arguments

modelList

a list of regressions representing the structural equation model.

data

a data.frame used to construct the structured equations.

standardize

specifies the type of standardization to perform: "none", "scale", "range".

corr.errors

a vector of variables with correlated errors (separated by ~~).

intercept

whether values and significance tests for the intercept should be returned; default is FALSE.

Value

Returns a data.frame with the causal path (response, predictor), the coefficient estimate, the standard error, and associated p-value.

Details

Unstandardized coefficients are extracted directly from model output. For models of class lmerMod, denominator degrees of freedom and resulting P-values are calculated using the Kenward-Rogers approximation from the pbkrtest package.

When standardize = "scale", variables are scaled by mean and variance using the function scale. When standardize = "range", variables are scaled by subtracting the minimum and dividing by the difference of the range.

If predictors are factors or dummy variables, then they are left in their original units. If the response is not normally distributed, then it is left untransformed even when standardize = "scale" or standardize = "range". Predictors in these models are transformed, unless they meet the previous criteria.

Variables with correlated errors have no direct relationship but rather are hypothesized to be driven by the same underlying factor. This covariance should be reflected as correlated errors (double-headed arrow in a path diagram). Correlated errors are specified using the syntax from the lavaan package: var1 ~~ var2. If two exogenous variables are listed, then the partial Pearson correlation between the two is given, otherwise the correlation between the partial residuals is returned using partial.resid. Significance tests (i.e., P-values) for correlated errors are derived from a t-distribution using the transformation in Shipley, 2000 (p.76).

References

Shipley, Bill. Cause and correlation in biology: a user's guide to path analysis, structural equations and causal inference. Cambridge University Press, 2002.

Bollen, K. A. Structural equations with latent variables. New York: John Wiley & Sons, 1989.

See Also

scale

Examples

# NOT RUN {
# Load example data
data(shipley2009)

# Load model packages
library(lme4)
library(nlme)

# Create list of models 
shipley2009.modlist = list(
  
  lme(DD ~ lat, random = ~1|site/tree, na.action = na.omit, 
      data = shipley2009),
  
  lme(Date ~ DD, random = ~1|site/tree, na.action = na.omit, 
      data = shipley2009),
  
  lme(Growth ~ Date, random = ~1|site/tree, na.action = na.omit, 
      data = shipley2009),
  
  glmer(Live ~ Growth+(1|site)+(1|tree), 
        family=binomial(link = "logit"), data = shipley2009) 
  
)
  
# Get raw path coefficients
sem.coefs(shipley2009.modlist, shipley2009)

# Scale by mean and variance
sem.coefs(shipley2009.modlist, shipley2009, standardize = "scale")

# Scale by range
sem.coefs(shipley2009.modlist, shipley2009, standardize = "range")
# }