inferBF: Compute BFs for hypotheses
In mattansb/BFEffect: Analyse models and effects estimated with BayesFactor

Description Usage Arguments Details Value Author(s) References See Also Examples

Compute BFs for hypotheses. A kin to 'brms::hypothesis'.

1 2	inferBF(BF, hypothesis = NULL, hdi.level = 0.95, iterations = 10000, seed = NULL, index = 1, nice.names = TRUE, ...)

`BF`	an object returned by any of the BayesFactor modeling functions
`hypothesis`	specified hypotheses, returned from 'hyp'.
`hdi.level`	credible level
`iterations`	number of posterior iterations
`seed`	A single numeric value passed to set.seed to make results reproducible.
`index`	which BF mddel to test? (for `BFBayesFactor` objects)
`nice.names`	whould the coeff names be a little nicer? (for `BFBayesFactor` objects)
`...`	passed to posterior sampling function

Please note that 'inferBF' is in alpha stages of development, and though it is suited for estimating effects (contrasts / slopes), it should not be used for estimating predicted values (due to 'BayesFactor' currently not supporting "intercept" estimation).

For directed hypotheses:

BF.v.Prior - The change in probability of hypothesis from the prior distribution to the posterior. Can also be thought as the BF of the restricted model vs. the unrestricted model.
BF.v.Null - The evidence ratio between the denominator of the BF object and the restricted mode. Calculated as BF.v.Prior times the models BF.
BF.v.Opposite - The BF of 'a>b'/'b>a'
Post.Prob - Posterior probability

For undirected (point) hypotheses:

Estimate - The median of the posterior distribution
Est.Error - Standard error of the posterior distribution
HDI.Lower - upper bound of HDI credible interval
HDI.Upper - lower bound of HDI credible interval
BF - The Savage-Dickey density ratio

A list of data.frames, for each hypothesis, that prints nicely.

Mattan S. Ben-Shachar

Wagenmakers, E. J., Lodewyckx, T., Kuriyal, H., & Grasman, R. (2010). Bayesian hypothesis testing for psychologists: A tutorial on the Savage-Dickey method. Cognitive psychology, 60(3), 158-189.

Morey, R. D., & Wagenmakers, E. J. (2014). Simple relation between Bayesian order-restricted and point-null hypothesis tests. Statistics & Probability Letters, 92, 121-124.

Morey, R. D. (2015). Multiple Comparisons with BayesFactor, Part 2 - order restrictions [Blog post]. https://richarddmorey.org/2015/01/multiple-comparisons-with-bayesfactor-part-2-order-restrictions/

as.data.frame.inferBF

## Not run: 
# =======
# anovaBF
# =======
library(BayesFactor)
library(BFEffects)
data(puzzles)
BF <- anovaBF(RT ~ shape*color + ID, data=puzzles, whichRandom = "ID", progress = FALSE)

# If no hypotheses are passed, `inferBF` returns the paramater names:
inferBF(BF[4])
## [1] "mu"                     "round"                  "square"                 "color"                  "monochromatic"
## [6] "X1"                     "X2"                     "X3"                     "X4"                     "X5"
## [11] "X6"                     "X7"                     "X8"                     "X9"                     "X10"
## [16] "X11"                    "X12"                    "round...color"          "round...monochromatic"  "square...color"
## [21] "square...monochromatic" "sig2"                   "g_shape"                "g_color"                "g_ID"
## [26] "g_shape.color"
## Test a model that has two simple effects of color in each level of shape.

puzzle_hypotheses <- hyp(directed_h = (color + round...color < monochromatic + round...monochromatic) &
                           (color + square...color < monochromatic + square...monochromatic),
                         point_h = color - monochromatic)
inferBF(BF[4],puzzle_hypotheses)
## Hypotheses tested based on the model:
##         RT ~ shape + color + shape:color + ID
##
## Directional Tests:
##            BF.v.Prior BF.v.Null BF.v.Opposite Post.Prob
## directed_h      2.757    11.661        11.484      0.92
##
## Point Tests:
##         Estimate Est.Error HDI.Lower HDI.Upper    BF
## point_h    -0.86     0.383    -1.576    -0.079 3.128
## ---
## HDI level: 0.95
## Point BF calculated using the Savage-Dickey method


# ============
# regressionBF
# ============
data(attitude)
BF <- regressionBF(rating ~ ., data = attitude, progress = FALSE)
# test for a positive slope for complaints and a positive slope for learning.
attitude_hypo <- hyp(complaints = complaints > 0,
                     attitude   = learning > 0,
                     both       = learning > 0 & complaints > 0)
inferBF(head(BF)[2], attitude_hypo)
## Hypotheses tested based on the model:
##   rating ~ complaints + learning
##
## Directional Tests:
##            BF.v.Prior BF.v.Null BF.v.Opposite Post.Prob
## complaints      2.008  416208.6           Inf     1.000
## attitude        1.861  385670.5        12.699     0.927
## both            3.730  773203.4        12.699     0.927

# =================
# Advanced analyses
# =================
BF <- regressionBF(rating ~ ., data = attitude, progress = FALSE)

advance_hypo <- hyp(
  # privileges is not between [-0.2,0.2]
  # Post.Prob is the % of posterior mass outside the ROPE
  ROPE_privileges = !(privileges > -0.2 & privileges < 0.2),
  # Estimate privileges
  estimate_privileges = privileges,
  # Estimate privileges based on a one sided prior
  estimate_privileges2 = ifelse(privileges > 0,privileges,NA)
)

inferBF(BF[2], advance_hypo)
## Hypotheses tested based on the model:
##         rating ~ privileges
##
## Directional Tests:
##                 BF.v.Prior BF.v.Null BF.v.Opposite Post.Prob
## ROPE_privileges       1.23     3.907         4.118     0.805
##
## Point Tests:
##                      Estimate Est.Error HDI.Lower HDI.Upper    BF
## estimate_privileges     0.339     0.172     0.017     0.679 3.394
## estimate_privileges2    0.342     0.162     0.037     0.655 5.940
## ---
## HDI level: 0.95
## Point BF calculated using the Savage-Dickey method

## End(Not run)