Reusing statistics

2024-04-09

Besides using tidystats in combination with a text editor add-in to report statistics, you can also use tidystats to read and use the statistics for other purposes. For example, researchers can extract specific statistics and perform analyses such as meta-analyses or a p-curve analysis on the extracted statistics.

One particular useful function that was created for this purpose is tidy_stats_to_data_frame(). This function converts a tidystats list of statistics to a standard data frame. That means you can use common data manipulation functions such as filter() on the data to retrieve the statistics of interest.

An example

Below is an example of how to convert a list of statistics to a data frame and perform several simple operations.

In the example below we read the tidystats list and select all the p-values.

library(tidystats)
library(dplyr)

# Read the .json file containing the statistics and immediately convert it to
# a data frame
statistics <- read_stats("statistics.json") |>
  tidy_stats_to_data_frame()

# Extract all the p-values
p_values <- filter(statistics, statistic_name == "p")

p_values
identifier analysis_name group_name_1 group_name_2 statistic_name symbol subscript lower value upper interval level
sleep_t_test extra by group - - p - - - 0.002833 - - -
D9_lm weight ~ group Model - p - - - 0.249023 - - -
D9_lm weight ~ group Coefficients (Intercept) p - - - 0.000000 - - -
D9_lm weight ~ group Coefficients groupTrt p - - - 0.249023 - - -
npk_aov yield ~ block + N * P * K Terms block p - - - 0.015939 - - -
npk_aov yield ~ block + N * P * K Terms N p - - - 0.004372 - - -
npk_aov yield ~ block + N * P * K Terms P p - - - 0.474904 - - -
npk_aov yield ~ block + N * P * K Terms K p - - - 0.028795 - - -
npk_aov yield ~ block + N * P * K Terms N:P p - - - 0.263165 - - -
npk_aov yield ~ block + N * P * K Terms N:K p - - - 0.168648 - - -
npk_aov yield ~ block + N * P * K Terms P:K p - - - 0.862752 - - -

Alternatively, we can can also easily select all significant p-values.

sig_p_values <- filter(statistics, statistic_name == "p" & value < .05)
identifier analysis_name group_name_1 group_name_2 statistic_name symbol subscript lower value upper interval level
sleep_t_test extra by group - - p - - - 0.002833 - - -
D9_lm weight ~ group Coefficients (Intercept) p - - - 0.000000 - - -
npk_aov yield ~ block + N * P * K Terms block p - - - 0.015939 - - -
npk_aov yield ~ block + N * P * K Terms N p - - - 0.004372 - - -
npk_aov yield ~ block + N * P * K Terms K p - - - 0.028795 - - -

This could be useful if you want to conduct a p-curve analysis. Although do note that you should not blindly select all p-values. You should select only the p-values that are relevant to a particular hypothesis. If researchers provide the correct meta-information for each test (e.g., by indicating whether it is a primary analysis), this could help meta-researchers make correct decisions about which statistics to include in their analyses.

Summary

By importing a tidystats-produced file of statistics, you can convert the statistics to a data frame using the tidy_stats_to_data_frame function and apply common data transformation functions to extract specific statistics. These statistics can then be used in analyses such as meta-analyses, p-curve analyses, or other analyses.

References

Simonsohn, U., Nelson, L. D., & Simmons, J. P. (2014). p-curve and effect size: Correcting for publication bias using only significant results. Perspectives on Psychological Science, 9(6), 666-681. https://doi.org/10.1177/1745691614553988