consistent outputs across Bayesian and frequentist tests

[IndrajeetPatil]

Basically, we need to have one-row dataframe summaries for BFBayesFactor objects, the same way we have them for htest objects.

t-test

library(BayesFactor)
library(parameters)
data(raceDolls)

# Bayesian
ttestBF(mtcars$wt, mu = 3) |> parameters(cohens_d = TRUE)
#> # Fixed Effects
#> 
#> Parameter  | Median | Cohen's d |        95% CI |     pd | % in ROPE |              Prior |    BF |          Method
#> -------------------------------------------------------------------------------------------------------------------
#> Difference |   0.20 |           | [-0.16, 0.56] | 87.10% |    24.26% | Cauchy (0 +- 0.71) | 0.387 | Bayesian t-test
#>            |        |      0.19 | [-0.15, 0.53] |        |           |                    |       | Bayesian t-test

# frequentist 
t.test(mtcars$wt, mu = 3) |> parameters(standardized_d = TRUE)
#> One Sample t-test
#> 
#> Parameter |   mu | Difference |       95% CI | t(31) | Cohen's d |      d 95% CI |     p
#> ----------------------------------------------------------------------------------------
#> mtcars$wt | 3.00 |       0.22 | [2.86, 3.57] |  1.26 |      0.22 | [-0.13, 0.58] | 0.218
#> 
#> Alternative hypothesis: true mean is not equal to 3

contingency table

# Bayesian
contingencyTableBF(raceDolls, sampleType = "indepMulti", fixedMargin = "cols") |>
  parameters(cramers_v = TRUE)
#> Warning: Could not estimate a good default ROPE range. Using 'c(-0.1, 0.1)'.
#> # Fixed Effects
#> 
#> Parameter | Cramer's V |       95% CI |                            Prior |   BF |                              Method
#> ---------------------------------------------------------------------------------------------------------------------
#> Ratio     |            |              | Independent multinomial (0 +- 1) | 1.81 | Bayesian contingency table analysis
#>           |       0.16 | [0.01, 0.29] |                                  |      | Bayesian contingency table analysis

# frequentist 
chisq.test(raceDolls) |>
  parameters(cramers_v = TRUE)
#> Pearson's Chi-squared test with Yates' continuity correction
#> 
#> Chi2(1) | Cramer's V | Cramers 95% CI |     p
#> ---------------------------------------------
#> 3.86    |       0.17 |   [0.03, 1.00] | 0.050

^{Created on 2021-11-07 by the reprex package (v2.0.1)}

[IndrajeetPatil]

Another improvement we can do is display the Method column for Bayesian outputs as a table title, instead of "Fixed Effects", the same way we do it for frequentist outputs.

[strengejacke]

library(BayesFactor)
#> Loading required package: coda
#> Loading required package: Matrix
#> ************
#> Welcome to BayesFactor 0.9.12-4.2. If you have questions, please contact Richard Morey ([email protected]).
#> 
#> Type BFManual() to open the manual.
#> ************
library(parameters)
data(raceDolls)

# Bayesian
ttestBF(mtcars$wt, mu = 3) |> parameters(cohens_d = TRUE)
#> # Fixed Effects
#> 
#> Parameter  | Median |        95% CI | Cohen's d |         d  CI |     pd | % in ROPE |              Prior |    BF |          Method
#> -----------------------------------------------------------------------------------------------------------------------------------
#> Difference |   0.20 | [-0.18, 0.54] |      0.19 | [-0.16, 0.54] | 85.72% |    24.78% | Cauchy (0 +- 0.71) | 0.387 | Bayesian t-test

# frequentist 
t.test(mtcars$wt, mu = 3) |> parameters(standardized_d = TRUE)
#> One Sample t-test
#> 
#> Parameter |   mu | Difference |       95% CI | t(31) | Cohen's d |      d 95% CI |     p
#> ----------------------------------------------------------------------------------------
#> mtcars$wt | 3.00 |       0.22 | [2.86, 3.57] |  1.26 |      0.22 | [-0.13, 0.58] | 0.218
#> 
#> Alternative hypothesis: true mean is not equal to 3

# Bayesian
contingencyTableBF(raceDolls, sampleType = "indepMulti", fixedMargin = "cols") |>
  parameters(cramers_v = TRUE)
#> Warning: Could not estimate a good default ROPE range. Using 'c(-0.1, 0.1)'.
#> # Fixed Effects
#> 
#> Parameter | Cramer's V |  Cramers  CI |                            Prior |   BF |                              Method
#> ---------------------------------------------------------------------------------------------------------------------
#> Ratio     |       0.16 | [0.01, 0.29] | Independent multinomial (0 +- 1) | 1.81 | Bayesian contingency table analysis

# frequentist 
chisq.test(raceDolls) |>
  parameters(cramers_v = TRUE)
#> Pearson's Chi-squared test with Yates' continuity correction
#> 
#> Chi2(1) | Cramer's V | Cramers 95% CI |     p
#> ---------------------------------------------
#> 3.86    |       0.17 |   [0.03, 1.00] | 0.050

^{Created on 2021-11-07 by the reprex package (v2.0.1)}

[IndrajeetPatil]

Thanks! 🤩

This is another minor inconsistency we can get rid of:

Another improvement we can do is display the Method column for Bayesian outputs as a table title, instead of "Fixed Effects", the same way we do it for frequentist outputs.

[IndrajeetPatil]

Also, CI column is NA:

library(BayesFactor)
library(parameters)
library(statsExpressions)
data(raceDolls)
options(tibble.width = Inf)

contingencyTableBF(raceDolls, sampleType = "indepMulti", fixedMargin = "cols") |>
  parameters(cramers_v = TRUE) |>
  as.data.frame()

#>   Parameter Median CI CI_low CI_high Cramers_v Cramers_CI_low Cramers_CI_high
#> 1     Ratio     NA NA     NA      NA 0.1663267     0.01151098       0.2932553
#>   pd ROPE_Percentage      Prior_Distribution Prior_Location Prior_Scale
#> 1 NA              NA independent multinomial              0           1
#>         BF                              Method
#> 1 1.814856 Bayesian contingency table analysis

^{Created on 2021-11-07 by the reprex package (v2.0.1)}

This is present in {effectsize} output:

library(BayesFactor)
library(effectsize)
data(raceDolls)

contingencyTableBF(raceDolls, sampleType = "indepMulti", fixedMargin = "cols") |>
  effectsize() |>
  as.data.frame()
#>   Cramers_v   CI     CI_low   CI_high
#> 1 0.1620444 0.95 0.01022901 0.2918251

^{Created on 2021-11-07 by the reprex package (v2.0.1)}