Skip to contents

Summary of Parameter Estimates Across Models

Source: R/emp_eq_summary.R
est_summary.Rd

Summarize parameter estimates across a set of equivalent models, returning the requested aggregate statistics for each parameter.

Usage

est_summary(
  object,
  level = 0.95,
  standardized = TRUE,
  trim_par_label = TRUE,
  col = c("mean", "min", "max", "n")
)

# S3 method for class 'est_summary'
print(x, drop_constant = TRUE, constant_tolerance = 5e-05, nd = 3, ...)

Arguments

object

A partables object, typically the output of eq_models().

level

A confidence level passed to the functions that extract parameter estimates from each fitted model. Confidence intervals are not included in the summary, so this argument does not affect the returned aggregate estimates for now. It may be used in the future.

standardized

Logical; if TRUE (the default), summarize fully standardized (std.all) estimates; if FALSE, summarize unstandardized estimates. Standardized error variances are reported as R-squares.

trim_par_label

Logical; if TRUE (the default), remove the trailing .1.1 from par_label when all labels have that suffix (the single-block, single-group case). Set to FALSE to retain the block and group indices.

col

Character vector selecting which aggregate columns to return. Choose from "mean", "min", "max", and "n"; the default returns all four. At least one value is required. The identifier columns are always returned.

x

An est_summary object, which is the output of est_summary().

drop_constant

Logical. Whether parameters having the nearly the same estimates across models will be dropped in the printout.

constant_tolerance

The tolerance used to determine whether a parameter has nearly identical estimates across models. If the range is less than or equal to this value, the estimates are considered constant.

nd

The number of decimal places to print.

...

Optional parameters. Not used.

Value

A est_summary object (also a data frame and a lavaan.data.frame) with parameter identifiers and the requested aggregate columns:

  • par_label: A label based on the two variables, block, and group, independent of the operator. For example, x ~ y, y ~ x, and x ~~ y share the same label. The block and group suffix is omitted by default when all parameters are from block 1 and group 1.

  • lhs, op, rhs: The left-hand side, operator, and right-hand side of the parameter, as in lavaan model syntax. If op is "r2", lhs identifies the variable whose R-square is reported and rhs is empty.

  • mean, min, max: If requested, the arithmetic mean, minimum, and maximum of the available estimates for that parameter across models.

  • n: If requested, the number of non-missing estimates included in the aggregates. It can be smaller than the number of models when a parameter is absent or has no estimate in some models.

The print-method of est_summary object returns x invisibly. It is called for its side-effect.

Details

The input is usually the result of eq_models(), but any partables object containing fitted models can be used. Estimates are grouped by the parameter label and its lhs, op, and rhs values, then summarized across models. Parameters without an estimate in a model are omitted from that model's aggregates.

The function summarizes fully standardized (std.all) estimates. For standardized output, variances (~~ parameters with the same variable on both sides) are represented as R-squares (op = "r2"), calculated as one minus the standardized error variance. Set standardized = FALSE to summarize unstandardized estimates instead.

If no model is marked as the original model, the first model in object is used to establish the common parameter ordering.

The print-method of est_summary object print the results in a user-friendly way.

See also

eq_models() for the function to generate the models.

Examples


library(lavaan)

# For illustration, only a few models are generated below,
# using drop_k() and add_k manually.
# These two functions are usually not used directly.

# Model 1

mod1 <-
"
fx =~ x1 + x2 + x3
fm =~ m1 + m2 + m3
fy =~ y1 + y2 + y3
fm ~ fx
fy ~ fm + fx
"
fit1 <- sem(
          model = mod1,
          data = data_test_3_factor_3_item
        )
fit1_1_more <- drop_k(fit1)
fit1_1_more_1_less <- lapply(
  fit1_1_more,
  add_k
)

# All equivalent
partables1 <- combine_partables(fit1_1_more_1_less)

eq_out_1 <- eq_models(
          partables1,
          original_model = fit1,
          parallel = FALSE
        )

est_summary(eq_out_1)
#> Number of models: 4
#> Parameter Estimates: Standardized
#> Parameter estimates nearly identical across models are not printed.
#> - 24 such parameter(s).
#> 
#> Variables: fm fx 
#>         Mean    Min    Max N
#> fm~fx  0.207  0.206  0.208 2
#> 
#> Variables: fx fy 
#>         Mean    Min    Max N
#> fy~fx -0.041 -0.075 -0.007 4
#> 
#> R-Squares:
#>         Mean    Min    Max N
#> fm     0.094  0.043  0.145 2
#> fy     0.053  0.000  0.107 4
#> 
#> Notes:
#> - Mean: The mean of estimates.
#> - Min/Max: The minimum/maximum of estimates.
#> - N: The number of models with this parameter (not fixed).

# Select statistics and use unstandardized estimates
est_summary(
  eq_out_1,
  standardized = FALSE,
  col = c("mean", "n")
)
#> The max and min columns are absent drop_constant ignored.
#> Number of models: 4
#> Parameter Estimates: Not standardized
#> 
#> Variables: fm 
#>          Mean N
#> fm~~fm  1.252 4
#> 
#> Variables: fm fx 
#>          Mean N
#> fm~fx   0.284 2
#> fx~fm   0.151 1
#> fx~~fm  0.198 1
#> 
#> Variables: fm fy 
#>          Mean N
#> fm~fy   0.319 1
#> fy~fm   0.335 2
#> fm~~fy  0.422 1
#> 
#> Variables: fx 
#>          Mean N
#> fx~~fx  0.693 4
#> 
#> Variables: fx fy 
#>          Mean N
#> fy~fx  -0.056 4
#> 
#> Variables: fy 
#>          Mean N
#> fy~~fy  1.253 4
#> 
#> Variables: m1 
#>          Mean N
#> m1~~m1  0.434 4
#> 
#> Variables: m2 
#>          Mean N
#> m2~~m2  1.137 4
#> 
#> Variables: m3 
#>          Mean N
#> m3~~m3  0.848 4
#> 
#> Variables: x1 
#>          Mean N
#> x1~~x1  1.192 4
#> 
#> Variables: x2 
#>          Mean N
#> x2~~x2  0.683 4
#> 
#> Variables: x3 
#>          Mean N
#> x3~~x3  1.180 4
#> 
#> Variables: y1 
#>          Mean N
#> y1~~y1  0.676 4
#> 
#> Variables: y2 
#>          Mean N
#> y2~~y2  1.057 4
#> 
#> Variables: y3 
#>          Mean N
#> y3~~y3  1.003 4
#> 
#> Variables: fm m1 
#>          Mean N
#> fm=~m1  1.000 4
#> 
#> Variables: fm m2 
#>          Mean N
#> fm=~m2  0.519 4
#> 
#> Variables: fm m3 
#>          Mean N
#> fm=~m3  0.513 4
#> 
#> Variables: fx x1 
#>          Mean N
#> fx=~x1  1.000 4
#> 
#> Variables: fx x2 
#>          Mean N
#> fx=~x2  1.081 4
#> 
#> Variables: fx x3 
#>          Mean N
#> fx=~x3  0.733 4
#> 
#> Variables: fy y1 
#>          Mean N
#> fy=~y1  1.000 4
#> 
#> Variables: fy y2 
#>          Mean N
#> fy=~y2  0.512 4
#> 
#> Variables: fy y3 
#>          Mean N
#> fy=~y3  0.540 4
#> 
#> Notes:
#> - Mean: The mean of estimates.
#> - N: The number of models with this parameter (not fixed).

# The usual way to use eq_models:
# 'parallel' should be set to TRUE or omitted
# eq_out_all <- eq_models(
#           original_model = fit1
#         )
# eq_out_all