Perform a robustness analysis — bfrr • bayesplay

Perform a robustness analysis by systematically varying the values of the alternative prior

bfrr(
  likelihood,
  alternative_prior,
  null_prior,
  parameters,
  steps = 100L,
  cutoff = 3L,
  multicore = TRUE
)

Arguments

likelihood: likelihood object representing the data
alternative_prior: prior object for the alternative prior
null_prior: prior object for the null prior
parameters: list of min and max values for each parameter to vary over.
steps: numeric the number of steps each parameter is varied over. (Default: 100 steps)
cutoff: Minimum Bayes factor value for evidence for the hypothesis (Default: 3)
multicore: Run robustness analysis across multiple cores (Default: TRUE if available)

Value

A robustness object

Examples


# define a likelihood
data_model <- likelihood(family = "normal", mean = 1, sd = 5)

# define the alternative prior
alternative_prior <- prior(family = "normal", mean = 5, sd = 4)

# define the null prior
null_prior <- prior(family = "point", point = 0L)

# set the parameters to vary and the range to vary them across
# vary the mean from -10 to 5
# vary the sd from 1 to 5
parameters <- list(mean = c(-10, 5), sd = c(1, 5))

# mark all Bayes factors larger/smaller than 3/.3 as evidence for the
# alternative / null
cutoff <- 3
bfrr(data_model, alternative_prior, null_prior, parameters, steps = 10,
  cutoff,
  multicore = FALSE
)
#> Prior robustness analysis:
#> 
#>   Alternative prior
#>   -----------------
#>     Family
#>       normal
#>     Parameters
#>       mean: 5
#>       sd: 4
#> 
#>   Null prior
#>   -----------------
#>     Family
#>       point
#>     Parameters
#>       point: 0
#> 
#>   Likelihood
#>   -----------------
#>     Family
#>       normal
#>     Parameters
#>       mean: 1
#>       sd: 5
#> 
#>  The original parameters gave a Bayes factor of 0.66.
#> Using the cutoff value of 3, this result is inconclusive.
#> 
#> 2 parameters of the alternative prior were varied for the robutness analysis:
#>   The mean was varied from -10 to 5 (step size: 1.5)
#>   The sd was varied from 1 to 5 (step size: 0.4)
#> 
#> Outcome
#> -----------------
#> 25 of 100 (0.25) tested priors provided evidence for H0 (BF < 0.33)
#> 
#> 0 of 100 (0) tested priors provided evidence for H1 (BF > 3)
#> 
#> 75 of 100 (0.75) tested priors were inconclusive (0.33 < BF < 3)
#> 
#> The conclusion of 'Inconclusive' was robust over 0.75 tested priors
#>