Shows the formula behind every column of a tab_reg() table — exactly what reached
stats::glm(), survey::svyglm(), nnet::multinom() or MASS::polr(). Use it to check what a
shape =, a trials = or a model formula really built.
Arguments
- x
A table built by
tab_reg().
Value
A tibble with one row per model: model (its name in the table), outcome, family (the
outcome family), link (the measure that model estimates — the word link = takes), fit
(the R call it was fitted with) and formula. A svyglm() row also means robust
(Huber-White) standard errors: survey's design-based variance IS the sandwich, which is why a
ratio or a difference on a binary outcome is fitted through it.
Details
One row per model: several outcomes give one each, a predictors list one per model. Two things
the list does not repeat: under tab_vars the same formula is fitted within each group, and
color = "between_groups" (or stats = "group_interaction") fits one extra pooled model for the footer
test only.
A summed score (trials =) is fitted on a success / failure pair, so its formula names the two
internal columns tabxplor builds for it (.gb_succ, .gb_fail, and .gb_trials in the offset of
the risk-ratio link).
The formula names the columns as the user wrote them, but a continuous predictor is fitted
anchored at its ref (its mean by default), and a shape = may have recoded it — neither
changes any effect, only what the Constant row means.
See also
tab_reg(), reg_measures() (what an outcome can be modelled as).
Examples
# \donttest{
d <- forcats::gss_cat
d$married <- as.integer(d$marital == "Married")
reg_formulas(tab_reg(d, "married", c("race", "age"), family = "binomial"))
#> # A tibble: 1 × 6
#> model outcome family link fit formula
#> <chr> <chr> <chr> <chr> <chr> <chr>
#> 1 married: OR married binomial odds_ratio "glm(binomial(\"logit\"))" married ~ …
# }