documenting return section for BayesFactor objects

[IndrajeetPatil]

Although most of the outputs from model_parameters for BayesFactor objects are clear, sometimes it's hard to figure out what some terms correspond to. And the documentation for this function doesn't have any information about this.

For example-
the first four rows here correspond to model-average posterior summary

    library(BayesFactor)
    library(magrittr)

    anovaBF(formula = Sepal.Length ~ Species, data = iris) %>%
      parameters::model_parameters(., priors = FALSE, centrality = "mean") %>%
      as.data.frame()
    #>            Parameter        Mean       CI_low    CI_high      pd
    #> 1                 mu  5.84403632  5.775817530  5.9097313 1.00000
    #> 2     Species-setosa -0.82709743 -0.917829717 -0.7252568 1.00000
    #> 3 Species-versicolor  0.09180601  0.002343341  0.1861748 0.93675
    #> 4  Species-virginica  0.73529142  0.640733329  0.8310193 1.00000
    #> 5               sig2  0.27070721  0.218365419  0.3194694 1.00000
    #> 6          g_Species  4.87206772  0.245442872  7.4836032 1.00000
    #>   ROPE_Percentage Effects   Component           BF
    #> 1        0.000000   fixed       extra 1.425293e+28
    #> 2        0.000000   fixed conditional 1.425293e+28
    #> 3        0.547599   fixed conditional 1.425293e+28
    #> 4        0.000000   fixed conditional 1.425293e+28
    #> 5        0.000000   fixed       extra 1.425293e+28
    #> 6        0.000000   fixed       extra 1.425293e+28

and agree with columns from JASP:

But, as a user, I may not know that mu corresponds to intercept and don't know what do sig2 and g_Species here refer to.

[strengejacke]

I think this is something for @mattansb. For Bayesian models, model_parameters() is not much more than a wrapper around bayestestR::describe_posterior() + bayestestR::diagnostic_posterior().

[mattansb]

I can do most of these, I think:

• ttestBF():
  • Difference: the raw difference between the
• correlationBF():
  • rho: linear correlation estimate (eqivilant to pearsons r)
• contingencyTableBF:
  • {none extracted - yet}
• proportionBF():
  • {none extracted - yet}
• lmBF() / generalTestBF() / regressionBF() / anovaBF()
  • whatever fixed / random effects parameters...
  • mu: mean-centered intercept? I think
  • sig2: model's sigma
  • g / g_*: these are the g parameters. See ANOVA BF paper.

Where should these definitions go? insight? parameters?

[strengejacke]

I'd say here:
https://easystats.github.io/parameters/reference/model_parameters.BFBayesFactor.html

[IndrajeetPatil]

@mattansb Maybe this is confusion on my part, but is this expected?

# data
set.seed(123)
df <- dplyr::filter(.data = gapminder::gapminder, continent == "Africa")

# one-sample t-test
mod <- BayesFactor::ttestBF(x = df$gdpPercap, mu = 10000) 
#> t is large; approximation invoked.

# median value
median(df$gdpPercap)
#> [1] 1192.138

# raw difference
10000 - median(df$gdpPercap)
#> [1] 8807.862

# extracting details
parameters::model_parameters(mod)
#> Parameter  |   Median |                   89% CI |   pd | % in ROPE |              Prior | Effects |   Component |     BF
#> -------------------------------------------------------------------------------------------------------------------------
#> Difference | 8.01e+06 | [  7.32e+06,   8.78e+06] | 100% |        0% | Cauchy (0 +- 0.71) |   fixed | conditional | > 1000

If the parameter here is the raw difference, I was expecting the estimate to be around ~ 8800, but it is actually 8010000.

[mattansb]

No, this is a gross little bug - fixed!

set.seed(123)

x <- rnorm(100)
mod <- BayesFactor::ttestBF(x = x, mu = 10) 
#> t is large; approximation invoked.

# raw difference
10 - median(x)
#> [1] 9.938244

# extracting details
parameters::model_parameters(mod, test = NULL)
#> Parameter  | Median |        89% CI |              Prior | Effects |   Component |     BF
#> -----------------------------------------------------------------------------------------
#> Difference |   9.91 | [9.78, 10.06] | Cauchy (0 +- 0.71) |   fixed | conditional | > 1000

^{Created on 2020-09-23 by the reprex package (v0.3.0)}

[IndrajeetPatil]

Thanks for fixing this so quickly!

@strengejacke Do you think there will be a new release of insight any time soon?
This bug fix is definitely going to lead to some failing tests in my packages and I would like to coordinate my releases accordingly.

[strengejacke]

easystats::on_CRAN()
#> insight     0.4 weeks

And just ~1 week before that update, there was the previous submission, so actually the next release is no planned before end of October.