This document describes the implementation of the Vovk-Sellke MPR into each statistical procedure in JASP. Follow the steps to complete a procedure. The necessary global functions are in the file commonMPR.R in the existing JASP-Engine/JASP/R directory. At the bottom of this document is a to-do list for where this still needs to be implemented.
- In analysis R file
- In JSON file
- In AnalysisForms file
- In Markdown help file (if help file exists)
if (options$VovkSellkeMPR) {
.addFootnote(footnotes, symbol = "\u002A", text = "Vovk-Sellke Maximum
<em>p</em>-Ratio: Based the <em>p</em>-value, the maximum
possible odds in favor of H\u2081 over H\u2080 equals
1/(-e <em>p</em> log(<em>p</em>)) for <em>p</em> \u2264 .37
(Sellke, Bayarri, & Berger, 2001).")
fields[[length(fields) + 1]] <- list(name = "VovkSellkeMPR",
title = "VS-MPR\u002A",
type = "number",
format = "sf:4;dp:3")
}
if (options$VovkSellkeMPR){
res[["VovkSellkeMPR"]] <- .VovkSellkeMPR(p)
}
{
"name": "VovkSellkeMPR",
"type": "Boolean",
"default": false
},
Add a selection box under "additional options" with the text "Sellke Maxmum p-Ratio" and name
VovkSellkeMPR
- Vovk-Sellke Maximum *p*-Ratio: The bound 1/(-e *p* log(*p*)) is derived from the shape of the *p*-value distribution. Under the null hypothesis (H<sub>0</sub>) it is uniform(0,1), and under the alternative (H<sub>1</sub>) it is decreasing in *p*, e.g., a beta(α, 1) distribution, where 0 < α < 1. The Vovk-Sellke MPR is obtained by choosing the shape α of the distribution under H<sub>1</sub> such that the obtained *p*-value is *maximally diagnostic*. The value is then the ratio of the densities at point *p* under H<sub>0</sub> and H<sub>1</sub>.
For example, if the two-sided *p*-value equals .05, the Vovk-Sellke MPR equals 2.46, indicating that this *p*-value is at most 2.46 times more likely to occur under H<sub>1</sub> than under H<sub>0</sub>.
- Vovk-Sellke Maximum p-Ratio: The bound 1/(-e p log(p)) is derived from the shape of the p-value distribution. Under the null hypothesis (H0) it is uniform(0,1), and under the alternative (H1) it is decreasing in p, e.g., a beta(α, 1) distribution, where 0 < α < 1. The Vovk-Sellke MPR is obtained by choosing the shape α of the distribution under H1 such that the obtained p-value is maximally diagnostic. The value is then the ratio of the densities at point p under H0 and H1.
The one-sided transformation follows from Morey & Wagenmakers (2014).
For example, if the two-sided p-value equals .05, the Vovk-Sellke MPR equals 2.46, indicating that this p-value is at most 2.46 times more likely to occur under H1 than under H0.
- Sellke, T., Bayarri, M. J., & Berger, J. O. (2001). Calibration of *p* values for testing precise null hypotheses. *The American Statistician, 55*(1), 62-71.
- Sellke, T., Bayarri, M. J., & Berger, J. O. (2001). Calibration of p values for testing precise null hypotheses. The American Statistician, 55(1), 62-71.
Morey, R. D., & Wagenmakers, E.-J. (2014). Simple relation between Bayesian order-restricted and point-null hypothesis tests. Statistics and Probability Letters, 92, 121-124.
Independent Samples T-Test- Main results table
Paired Samples T-Test- Main results table
One Sample T-Test- Main results table
ANOVA- Main results table
- Levene's test (assumption check)
Repeated Measures ANOVA- Within Subjects effects table
- Between Subjects effects table
- Sphericity (assumption check)
- Levene's test (assumption check)
ANCOVA- Main results table
- Levene's test (assumption check)
Correlation Matrix? do this lastLinear Regression- ANOVA table
- Coefficients table
Binomial tests- Main results table
Contingency tablesNOTE: footnotes not workingChi-squared test table (for all three statistics!)Kendall's Tau-b
Log-Linear Regression- ANOVA table
- Coefficients table
- none!
Here is an example of the implementation