A peer cannot re-read every document the way the researcher did, so the
coding of any one observation can be disputed but not settled at a
distance. What a peer can compute is how many re-codings would change
the decision. Re-coding x observations relabels x observations that
supported the working theory as supporting the rival, moving the counts
from (y_W, y_R) to (y_W - x, y_R + x). The number of observations
does not change; only the split does.
Arguments
- y_W
Non-negative whole number. Count, or summed weight, of observations supporting the working theory.
- y_R
Non-negative whole number. Count, or summed weight, of observations supporting the rival.
- threshold_uniform
Positive number. The uniform-weights Bayes factor must be at least this large for the rule to set the rival aside. Default
20.- threshold_worst
Positive number. The worst-case Bayes factor must be above this value for the rule to set the rival aside. Default
1.
Value
A list with elements
bf_uniform,bf_worstThe two Bayes factors at the observed coding.
x_uniformThe smallest number of re-codings at which the uniform-weights value falls below
threshold_uniform.x_worstThe smallest number of re-codings at which the worst-case value falls to
threshold_worstor below.x_starThe smaller of the two.
A tipping point is NA when no re-coding of the y_W observations
makes its condition fail.
Details
The paper's rule sets the rival aside, at least provisionally, when the uniform-weights Bayes factor is 20 or more and the worst-case Bayes factor is above 1. This function returns the smallest number of re-codings at which each condition fails and the smaller of the two, which is the number the supplement reports. At nine observations for the working theory and three for the rival, one re-coding takes the uniform-weights value from 20.67 to 6.49, below 20, while the worst-case value falls to 1 or below only after two (1.10 at eight of twelve, 0.56 at seven). So one re-coding changes the decision. The conclusion tolerates so little re-coding because the uniform-weights value starts so close to 20.
Re-coding lowers both Bayes factors, so every re-coding past the first
one at which a condition fails also fails it. A tipping point of 0
means the condition already fails at the observed coding.
See also
sens_bias() for sensitivity to a biased search;
bf_uniform_weights() and bf_worst_case() for the values recomputed
here.
Examples
# The paper's running example.
sens_coding(9, 3)
#> $bf_uniform
#> [1] 20.67196
#>
#> $bf_worst
#> [1] 2.732168
#>
#> $x_uniform
#> [1] 1
#>
#> $x_worst
#> [1] 2
#>
#> $x_star
#> [1] 1
#>
# Winward's fourteen of seventeen: two re-codings change the decision.
sens_coding(14, 3)$x_star
#> [1] 2