Skip to contents

Per-group slopes for a continuous predictor

Source: R/marginal-estimates.R
group_slopes.Rd

group_slopes() returns the slope of a continuous predictor stratified by one or more factors (or by values of another continuous predictor), delegating to emmeans::emtrends(). Each row is one stratum with a point estimate, standard error, 95% confidence band, and an excludes_zero indicator.

Use this alongside parameter_interpretation() for any fit with a continuous-by-categorical interaction (or continuous-by-continuous): the coefficient table reports contrast slopes (differences from the reference group's slope); group_slopes() reports each group's slope directly.

Not supported on bivariate Gaussian (biv_gaussian) fits — see group_means() for the same limitation and the recommended alternatives.

Usage

group_slopes(
  x,
  continuous,
  at = NULL,
  scale = c("response", "link"),
  ci_method = NULL,
  ...
)

Arguments

x

A symbolized_model whose underlying fit is retained on x$metadata$fit.

continuous

Name of the continuous predictor whose slope is wanted.

at

How to stratify. Three accepted shapes:

  • NULL (default): use every factor in the model that interacts with continuous (determined from x$terms).

  • A character vector of factor names: stratify by those factors' levels.

  • A named list (e.g. list(z = c(-1, 0, 1))): for continuous-by- continuous interactions, get the slope at those values of z.

scale

One of "response" (default) or "link". A slope is the derivative of the linear predictor with respect to the continuous predictor, which emmeans::emtrends() reports on the link scale regardless of scale – so for group_slopes() the two scales coincide (unlike group_means(), where the response scale back-transforms). The argument is accepted and recorded on the scale column for consistency with the rest of the family; to read an effect on the response scale, compare group means with group_means() instead.

ci_method

Confidence-interval method. Defaults to x$metadata$ci_method. See group_means() for details.

...

Reserved for future use.

Value

A tibble (S3 class symbolizer_group_slopes) with one row per stratum. Columns: predictor, level_combo, one column per stratifying variable, estimate, std_error, confint_low, confint_high, excludes_zero, ci_method, scale.

See also

Other marginal estimates: group_contrasts(), group_means()