Baseline model specification

[Leslie Rutkowski]

Hi all,

I'm wondering if I've run into a possible bug in how the baseline model is specified in lavaan. Or I don't understand the rationale for one option vs. the other. I include reproducible code below, however the fundamental issue is this:

Model 1: use "fixed.x = FALSE" command in sem() and only regression terms in model object gives:

Model Test User Model:

                                                      

  Test statistic                                11.107

  Degrees of freedom                             5

  P-value (Chi-square)                     0.049

Model Test Baseline Model:

  Test statistic                               165.608

  Degrees of freedom                               9

  P-value                                              0.000

model1@baseline$partable$ustart includes the exogenous covariance

Model 2: identical to Model 1 except explicitly include exogenous variance/covariance terms in model object gives: 

Model Test User Model:

                                                      

  Test statistic                                11.107

  Degrees of freedom                                 5

  P-value (Chi-square)                           0.049

Model Test Baseline Model:

  Test statistic                               165.944

  Degrees of freedom                                10

  P-value                                        0.000

model2@baseline$partable$ustart does not include the exogenous covariance

So, why is the exogenous covariance part of the baseline model in Model 1 but not in Model 2?

Thanks,
Leslie

Reproducible code follows:

# principles and practice of sem (4th ed.), rex kline

# recursive path model of illness, figure 7.5 (p. 159), table 4.2

library(lavaan)

# input the correlations in lower diagonal form

rothLower.cor <- '

   1.00

    -.03 1.00

     .39  .07 1.00

    -.05 -.23 -.13 1.00

    -.08 -.16 -.29  .34 1.00 '

# name the variables and convert to full correlation matrix

rothFull.cor <- getCov(rothLower.cor, names = c("exercise","hardy","fitness",

                                                "stress","illness"))

# display the correlations

rothFull.cor

# add the standard deviations and convert to covariances

rothFull.cov <- cor2cov(rothFull.cor, sds = c(66.50,38.00,18.40,33.50,62.48))

# path model 1 - fixed.x = FALSE

roth.model1 <- '

  # regressions

  illness ~ fitness + stress

  fitness ~ exercise

  stress ~ hardy'

reduced1 <- sem(roth.model1,

               sample.cov=rothFull.cov,

               sample.nobs=373, fixed.x = FALSE, sample.cov.rescale = FALSE)

summary(reduced1, fit.measures = TRUE, rsquare = TRUE)

parameterestimates(reduced1)

fitmeasures(reduced1, fit.measures = c("chisq", "df"))

reduced1@baseline$test$standard$stat

# path model 2 - fixed.x = FALSE & variance terms specified

roth.model2 <- '

  # regressions

  illness ~ fitness + stress

  fitness ~ exercise

  stress ~ hardy

  exercise ~~ exercise

  hardy ~~ hardy

  exercise ~~ hardy'

reduced2 <- sem(roth.model2,

               sample.cov=rothFull.cov,

               sample.nobs=373, fixed.x = FALSE, sample.cov.rescale = FALSE)

summary(reduced2, fit.measures = TRUE, rsquare = TRUE)

parameterestimates(reduced2)

fitmeasures(reduced2, fit.measures = c("chisq", "df"))

reduced2@baseline$test$standard$stat