Skip to contents

Add parameter formulas to a model formula

Source: R/bf.R
lf.Rd

brms's lf(): one or more two-sided formulas for distributional or nonlinear parameters, added to a bf() with +. It is sugar for passing the same formulas to bf() directly, and is useful when the parameter formulas are built somewhere else than the response formula.

Usage

lf(..., resp = NULL, center = NULL, cmc = NULL)

Arguments

...

Two-sided formulas naming the parameter on the left, e.g. sigma ~ x or (with nl = TRUE on the bf()) a nonlinear parameter's formula a ~ 1 + (1 | g), or one-sided formulas named by their parameter, sigma = ~ x. As in brms, lf() also takes what bf() takes besides formulas: a parameter fixed to a constant, sigma = 2, or equated to another, sigma1 = "sigma2".

resp

The response the formulas belong to, when the lf() is added to a multivariate formula. NULL (the default) adds them to the bf() on the left of the +, which must then be a single formula.

center

FALSE makes the intercept of each of these formulas an ordinary coefficient, class "b" with coef "Intercept", instead of brms's class "Intercept": the same as 0 + Intercept in the formula. See bf(). NULL, the default, leaves the intercept as class "Intercept".

cmc

FALSE turns off cell-mean coding in each of these formulas: without an intercept, a factor keeps its treatment contrasts, in the population-level and the group-level terms. See bf(). NULL, the default, keeps R's cell-mean coding.

Value

An object of class frmtmb_lf, to be added to a bf().

Details

bf(y ~ x) + lf(sigma ~ z) and bf(y ~ x, sigma ~ z) give the same model. In a multivariate model, add an lf() to the bf() of the response it belongs to, or name that response with resp =: bf(y1 ~ x) + bf(y2 ~ x) + bf(y3 ~ x) + lf(sigma ~ z, resp = "y3").

Examples

# the two spellings are the same model
bf(y ~ x) + lf(sigma ~ z)
#> y ~ x
#> sigma ~ z
bf(y ~ x, sigma ~ z)
#> y ~ x
#> sigma ~ z

# nonlinear parameter formulas can arrive the same way
bf(y ~ a * exp(-b * x), a ~ 1, nl = TRUE) + lf(b ~ 1 + (1 | g))
#> y ~ a * exp(-b * x) (nonlinear)
#> a ~ 1
#> b ~ 1 + (1 | g)

# in a multivariate formula, resp = says which response it modifies
bf(y1 ~ x) + bf(y2 ~ x) + bf(y3 ~ x) + lf(sigma ~ z, resp = "y3")
#> y1 ~ x
#> y2 ~ x
#> y3 ~ x
#> sigma ~ z
#> rescor: FALSE