apabayes reports Bayesian and structural equation models in APA 7 form inside apaquarto documents: inline strings for the running text and APA tables that render through apa7::apa_flextable() to PDF, Word, HTML and Typst (Typst needs apaquarto 6.0.0 or newer).
It reports brmsfit objects (and therefore bmmfit), lavaan and blavaan fits, brms::hypothesis() output, loo::loo_compare() output, bayestestR::bayesfactor_models() output and emmeans posterior contrasts. Numbers come from the model’s own package and from the easystats packages (parameters, bayestestR, insight); apabayes selects, arranges and formats them, and every value it derives itself is documented on the route that reports it.
One shape is currently out of reach upstream: a brms model with a matrix response and a trials() term (family = multinomial(), which includes every bmm M3 fit) fails inside insight::get_data() on R 4.3 or newer, before apabayes sees a number. apa_tidy() says so and names the route that works, apa_tidy(brms::as_draws_df(fit)); convergence statements are unaffected.
The package is in development; 0.1.0 is the first planned release.
A worked apaquarto manuscript ships with the package, in inst/apaquarto-example/, and the vignette “Reporting brms models in apaquarto” walks through the whole API.
Installation
The development version installs from GitHub once the repository is public:
# install.packages("pak")
pak::pak("GidonFrischkorn/apabayes")Use
Every model class goes through apa_tidy() first, and everything after that takes its result rather than the fit:
The examples below use that table, stored with the package so that this page builds without a compiler:
posterior <- readRDS(
system.file("extdata", "parameters.rds", package = "apabayes")
)Inline, in an apaquarto document:
apa_inline(posterior, "wt")
#> *b* = −5.39, 95% CrI [−6.95, −3.78], *pd* > .999apa_value() returns the number rather than the string, for a value that goes into arithmetic or into a sentence you format yourself:
apa_value(posterior, "wt", column = "ci_low")
#> [1] -6.953618As a table, in a chunk with #| label: tbl-fit, #| tbl-cap: and #| apa-note:. apa_table() formats the columns and headers, and apa_note() defines whatever columns the table turned out to have:
tab <- apa_table(posterior)
apa7::apa_flextable(tab)A chunk option is evaluated before its own chunk runs, so tab has to be built in an earlier chunk than the one that prints it.
Reporting from stored summaries
A fit is not needed. apabayes_tidy() is a public constructor, so a table of stored posterior summaries — the shape most analysis pipelines save — is reported the same way, and a manuscript knits without the packages that made the fit:
stored <- data.frame(
lhs = c("visual", "visual"), rhs = c("x1", "x2"),
est = c(0.77, 0.42), lower = c(0.66, 0.29), upper = c(0.87, 0.55)
)
t <- apabayes_tidy(
data.frame(
term = paste0(stored$lhs, "=~", stored$rhs),
estimate = stored$est, ci_low = stored$lower,
ci_high = stored$upper, std = TRUE
),
type = "parameters", centrality = "median", ci_method = "eti",
ci_level = 0.95
)
apa_inline(t, "visual", "x2", op = "=~", symbol = "β")
#> *β* = .42, 95% CrI [.29, .55]apa_tidy() reads a plain data frame of numeric columns as posterior draws, one column per parameter, which is what makes a derived quantity reportable — but it means any numeric data frame is read that way, so pass draws, not raw data.
