In lhdjung/scrutiny: Error Detection in Science

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(scrutiny)

The descriptive binary test, or DEBIT, checks whether the reported mean, sample standard deviation (SD), and sample size of binary data are mutually consistent [@heathers_debit_2019]. Like GRIM, it tests if a given set of summary data can describe the same distribution.

This vignette covers scrutiny's implementation of DEBIT:

1. The basic, single-case debit() function.
2. A specialized mapping function, debit_map().
3. The audit() method for summarizing debit_map()'s results.
4. Finally, the visualization function debit_plot().

DEBIT basics

Consider these summary data for a binary distribution: a mean of 0.35, an SD of 0.18, and a sample size of 20. To test their consistency, run this:

debit(x = "0.35", sd = "0.18", n = 20)

As in grim(), the mean needs to be a string. (The same is true for the SD.) That is because strings preserve trailing zeros, which can be crucial for DEBIT. Numeric values don't, and even converting them to strings won't help. A workaround for larger numbers of such values, restore_zeros(), is discussed in the Data wrangling vignette.

Note that the SD is always assumed to be the sample SD, not the population SD. In some rare cases, this might possibly explain apparent inconsistencies.

debit() has some further arguments, but all of them can be used from within debit_map(). Since debit_map() is the more useful function in practice, the other arguments will be discussed in that context.

Testing multiple cases

Working with debit_map()

If you want to test more than a handful of cases, the recommended way is to enter them into a data frame and to run debit_map() on the data frame. Below are the example data from Heathers and Brown's (2019) Table 1. A useful way to enter such data is to copy them from a PDF file and paste them into tibble::tribble(), which is available via scrutiny:

flying_pigs <- tibble::tibble(
    x  = runif(5, 0.2, 1) %>% round(2) %>% restore_zeros(),
    sd = runif(5, 0, 0.3) %>% round(2) %>% restore_zeros(),
    n = 1000
)


flying_pigs

Now, simply run debit_map() on that data frame:

flying_pigs %>% 
  debit_map()

The x, sd, and n columns are the same as in the input. The main result, consistency, is the DEBIT consistency of the former three columns.

pigs3  # data saved within the package

pigs3 %>% 
  debit_map()

DEBIT only makes sense with binary means and SDs. Both debit() and debit_map() check if the inputs are such data, and fail if they are not:

pigs5  # no binary means / SDs!

pigs5 %>% 
  debit_map()

Compared to grim_map(), debit_map() is more straightforward. There is no percentage conversion or accounting for multiple scale items. The same is true when comparing the basic grim() and debit() functions. However, both implementations tap scrutiny's arsenal of rounding procedures, so that is discussed next.

Summarizing results with audit()

Following up on a call to debit_map(), the generic function audit() summarizes test results:

pigs3 %>% 
  debit_map() %>% 
  audit()

These columns are ---

1. incons_cases: number of inconsistent value sets.
2. all_cases: total number of value sets.
3. incons_rate: proportion of DEBIT-inconsistent value sets.
4. mean_x: average of binary distribution means.
5. mean_sd: average of binary distribution standard deviations.
6. distinct_n: number of different sample sizes.

Visualizing results with debit_plot()

There is a specialized visualization function for DEBIT results, debit_plot(). Only run it on debit_map()'s output. It will fail otherwise.

# Determine plot theme for the remaining session:
ggplot2::theme_minimal(base_size = 12) %>% 
  ggplot2::theme_set()

pigs3 %>% 
  debit_map() %>% 
  debit_plot()

DEBIT-consistent value pairs are blue, inconsistent ones red. The black arc is the DEBIT line: Given the sample size, the pairs of mean and SD values are DEBIT-consistent if and only if they cross this line. More precisely, the inner boxes must cross the line --- the outer boxes are just pointers to the inner ones, in case these are poorly visible. They have no inherent meaning.

Except for the colors, inner boxes must look exactly like they do here: Their sizes and shapes are completely determined by the mean and SD values. Since inner boxes cannot be enlarged at will, outer boxes might be helpful to spot them at first glance.

However, if outer boxes are not desired, they can be turned off like this:

pigs3 %>% 
  debit_map() %>% 
  debit_plot(show_outer_boxes = FALSE)

Color settings and other ggplot2-typical options are available via arguments, as are settings handed down to ggrepel::geom_text_repel(), which creates the labels. For more, see debit_plot()'s documentation.

Testing numeric sequences with debit_map_seq()

DEBIT analysts might be interested in a mean or percentage value's numeric neighborhood. Suppose you found DEBIT inconsistencies, as in out example pigs3 data. You might wonder whether they are due to small reporting or computing errors.

Use debit_map_seq() to use DEBIT for the values surrounding the reported means, SDs, and sample sizes:

out_seq1 <- debit_map_seq(pigs3)
out_seq1

Summaries with audit_seq()

As this output is a little unwieldy, run audit_seq() on the results:

audit_seq(out_seq1)

Here is what the output columns mean:

• x, sd, and n are the original inputs, reconstructed and tested for consistency here.

• The hits_* columns display is the number of DEBIT-consistent value combinations found within the specified dispersion range; either in total or by varying individual parameters.

• diff_x reports the absolute difference between x and the next consistent dispersed value (in dispersion steps, not the actual numeric difference). diff_x_up and diff_x_down report the difference to the next higher or lower consistent value, respectively.

• Accordingly with the diff_sd* and diff_n* columns.

The default for dispersion is 1:5, for five steps up and down. When the dispersion sequence gets longer, the number of hits tends to increase:

out_seq2 <- debit_map_seq(pigs3, dispersion = 1:7, include_consistent = TRUE)
audit_seq(out_seq2)

Visualizing DEBIT-checked sequences

Although it is possible in principle to visualize results of debit_map_seq() using debit_plot(), it's not recommended because the results don't currently look great. This issue might be fixed in a future version of debit_plot().

Handling unknown group sizes with debit_map_total_n()

Problems from underreporting

Unfortunately, some studies that report group averages don't report the corresponding group sizes --- only a total sample size. This makes any direct use of DEBIT impossible because only x and sd values are known, not n values. All that is feasible here in terms of DEBIT is to take a number around half the total sample size, go up and down from it, and check which hypothetical group sizes are consistent with the reported group means and SDs. debit_map_total_n() semi-automates this process, motivated by a recent GRIM analysis [@bauer_expression_2021].

Here is an example:

out_total_n <- tibble::tribble(
  ~x1,     ~x2,   ~sd1,   ~sd2,  ~n,
  "0.30", "0.28", "0.17", "0.10", 70,
  "0.41", "0.39", "0.09", "0.15", 65
)

out_total_n <- debit_map_total_n(out_total_n)
out_total_n

audit_total_n(out_total_n)

See the GRIM vignette, section Handling unknown group sizes with grim_map_total_n(), for a more comprehensive case study. It uses grim_map_total_n(), which is the same as debit_map_total_n() but only for GRIM.

References

lhdjung/scrutiny documentation built on Sept. 28, 2024, 12:14 a.m.