Extract fixed effects

Two shapes, and the difference is the NAMES. The default is brms's summary matrix: one row per population-level coefficient under brms's own name (Intercept, sigma_Intercept, x), and the columns Estimate, Est.Error, Q2.5 and Q97.5. A maximum-likelihood fit has no draws, so Est.Error is the standard error and the Q columns are the Wald interval at those probabilities, which is the interval confint() reports. flatten = TRUE is one vector of estimates in the INTERNAL parameter name, which is what confint(), par_template(), start and newparams use: dpar_column, with the location parameter's own coefficients left unprefixed (x, not mu_x), and the response prefixed ahead of that in a multivariate fit.

Usage

# S3 method for class 'frmtmb_fit'
fixef(
  object,
  summary = TRUE,
  robust = FALSE,
  probs = c(0.025, 0.975),
  pars = NULL,
  ...,
  flatten = FALSE
)

Arguments

object

A frmtmb_fit.

summary, robust

brms's arguments, in brms's positions so that a positional brms call asks the same question. brms answers summary = FALSE with the posterior draws and robust = TRUE with their median and MAD, and a maximum-likelihood fit has no draws, so both are refused by name with the reason. The default of each is accepted and changes nothing.

probs

Probabilities of the two quantile columns. brms takes the posterior quantiles there; here they are the ends of the Wald interval at those probabilities.

pars

Row names to keep, in the order given, as in brms. A name the fit does not have is an error listing the ones it does.

...

Refused: an argument the method does not have is an error naming it, rather than silently changing nothing.

flatten

If TRUE, one named vector of ESTIMATES in the confint() spelling instead of the summary matrix. Coefficients come in linear-predictor order, which need not be the matrix's row order; index by name. Distributional parameters held at a constant are included here and are absent from vcov() and from the matrix, which cover the ESTIMATED population-level coefficients only. The exception is a constant the family itself supplies and brms does not show, an ordinal family's disc at 1, which is left out.

Value

A coefficients-by-four matrix, or with flatten = TRUE a single named vector of estimates.

Details

A distributional parameter nobody wrote a formula for is not a population-level coefficient in brms and is not a row here: sigma of a plain gaussian fit is reported by summary() under Further Distributional Parameters, on its own natural scale.

hypothesis() and variables() use a THIRD vocabulary, brms's parameter names: b_Intercept, b_sigma_Intercept for a sigma formula, and sigma on its natural scale when no formula was written for it. hypothesis() puts class = "b"'s b_ in front of a bare name, so its default spelling of those is Intercept and sigma_Intercept.

unlist(fixef(fit, flatten = TRUE)) is none of them. Use flatten = TRUE to line coefficients up with a prior or with confint(), and the default matrix to read an estimate with its standard error.

See also

vcov.frmtmb_fit(), which names its rows as this matrix does, and confint.frmtmb_fit(), which names them as flatten = TRUE names its entries.

Examples

set.seed(1)
dd <- data.frame(x = rnorm(100), g = factor(rep(1:10, 10)))
dd$y <- rnorm(100, 1 + 0.5 * dd$x + rnorm(10, 0, 0.8)[dd$g], 1)

# brms's summary matrix, one row per population-level coefficient
fit <- frm(bf(y ~ x + (1 | g), sigma ~ x) + gaussian(), data = dd)
fixef(fit)
#>                   Estimate  Est.Error        Q2.5      Q97.5
#> Intercept        1.2368734 0.25319299  0.74062430 1.73312258
#> sigma_Intercept -0.0959985 0.07493824 -0.24287474 0.05087775
#> x                0.5943909 0.10979801  0.37919074 0.80959104
#> sigma_x          0.2082862 0.07165775  0.06783963 0.34873283
exp(fixef(fit)["sigma_Intercept", "Estimate"])  # sigma is on the log
#> [1] 0.9084654

# flatten to the vector confint() names its rows by
fixef(fit, flatten = TRUE)
#>       (Intercept)                 x sigma_(Intercept)           sigma_x 
#>         1.2368734         0.5943909        -0.0959985         0.2082862 
all(names(fixef(fit, flatten = TRUE)) %in% rownames(confint(fit)))
#> [1] TRUE

# the standard error column is vcov()'s diagonal
all.equal(fixef(fit)[, "Est.Error"], sqrt(diag(vcov(fit))))
#> [1] TRUE