Skip to contents

Regress Exogenous Variables on Latent Variables

Source: R/regress.R
regress.Rd

Performs regression analysis to examine the influence of exogenous (external) variables on latent class variables in an estimated slca model. The function uses logistic regression with a three-step approach to account for measurement error.

Usage

regress(object, ...)

# S3 method for class 'slcafit'
regress(
   object, formula, data = parent.frame(),
   imputation = c("modal", "prop"),
   method = c("naive", "BCH", "ML"), ...,
   align = c("rownames", "position")
)

# S3 method for class 'slcafit'
regress(
  object,
  formula,
  data = parent.frame(),
  imputation = c("modal", "prop"),
  method = c("naive", "BCH", "ML"),
  ...,
  align = c("rownames", "position")
)

Arguments

object

an object of class slcafit.

...

additional arguments.

formula

a formula specifying the regression model. The left-hand side must be a single latent class variable from the estimated model.

data

an optional data.frame containing the exogenous variables of interest. Variables absent from the fitted data and data are taken from the formula's environment and aligned before formula transformations are evaluated. Such vectors and factors need observation IDs in names(); matrices and data frames need observation IDs in rownames().

imputation

a character string specifying the imputation method for latent class assignment. Options include:

  • "modal": Assigns each individual to the latent class with the highest posterior probability.

  • "prop": Uses posterior probabilities as proportional assignment weights.

method

a character string specifying the method to adjust for bias in the three-step approach. Options include:

  • "naive": A simple approach without correction for classification error.

  • "BCH": The bias-adjusted Bolck, Croon, and Hagenaars method.

  • "ML": A maximum likelihood approach that accounts for classification error.

align

how supplied covariate rows are aligned with the fitted data. The default, "rownames", requires unique, nonempty row names and matches them to the fitted observations, regardless of row count or order. Extra rows are allowed, but every fitted observation must be present. Automatic row numbers work only if they still identify the original observations; do not reset them after subsetting or reordering. Use "position" explicitly for data already in fitted-observation order; it requires exactly the same number of rows and ignores row names. There is no automatic positional fallback. These rules also apply separately to each environment covariate, even when its length already equals the fitted sample size. Unnamed environment covariates require explicit align = "position" and fitted-row order. After observations have been excluded, an unnamed original-length vector cannot be aligned safely: supply data with preserved row IDs, or name the vector with its original observation IDs. Matrix covariates remain matrices and factor levels are retained. Unnamed scalar formula parameters are left unchanged, not recycled as covariates. Non-scalar transformation settings (such as spline knots or factor levels) must be written inline or captured inside a transformation function; otherwise they are treated as row inputs. Transformation functions must use their explicit covariate arguments rather than retrieve hidden observation vectors. Use explicit covariate columns instead of dynamic get() or eval() lookups with row-ID alignment.

Value

A list of class reg.slca with the following components:

coefficients

A matrix of regression coefficients representing the odds ratios for each latent class against the baseline class (the last class).

std.err

A matrix of standard errors corresponding to the regression coefficients.

vcov

The variance-covariance matrix of the regression coefficients.

dim

The dimensions of the coefficients matrix.

ll

The log-likelihood of the regression model.

The summary function can be used to display the regression coefficients, standard errors, Wald statistics, and p-values. The standard errors are derived by numerically calculated Hessian matrix from nlm function. A warning is issued if the regression information matrix is numerically rank deficient. Existing generalized-inverse estimates are retained, but their standard errors and Wald inference may be unreliable.

References

Vermunt, J. K. (2010). Latent Class Modeling with Covariates: Two Improved Three-Step Approaches. Political Analysis, 18(4), 450–469. http://www.jstor.org/stable/25792024

Examples

library(magrittr)
names(nlsy97)
#>  [1] "SEX"     "RACE"    "ESMK_98" "FSMK_98" "DSMK_98" "HSMK_98" "EDRK_98"
#>  [8] "CDRK_98" "WDRK_98" "BDRK_98" "EMRJ_98" "CMRJ_98" "OMRJ_98" "SMRJ_98"
#> [15] "ESMK_03" "FSMK_03" "DSMK_03" "HSMK_03" "EDRK_03" "CDRK_03" "WDRK_03"
#> [22] "BDRK_03" "EMRJ_03" "CMRJ_03" "OMRJ_03" "SMRJ_03" "ESMK_08" "FSMK_08"
#> [29] "DSMK_08" "HSMK_08" "EDRK_08" "CDRK_08" "WDRK_08" "BDRK_08" "EMRJ_08"
#> [36] "CMRJ_08" "OMRJ_08" "SMRJ_08"
nlsy_jlcpa %>% regress(SMK_98 ~ SEX, nlsy97)
#> Coefficients:     
#> class  (Intercept)  SEXFemale
#>   1/3   0.3983      -0.4445  
#>   2/3  -0.6614       0.0804  
# \donttest{
nlsy_jlcpa %>% regress(PROF ~ SEX, nlsy97)
#> Coefficients:     
#> class  (Intercept)  SEXFemale
#>   1/4   0.436       -0.106   
#>   2/4   0.157       -0.210   
#>   3/4   0.443       -0.221   
# }