Non-Conservative Size-α Modified Fisher’s Exact Test

Installation

You can install the latest version of this modifiedfisher package by using devtools:

devtools::install_github("pvdmeulen/modifiedfisher")

Examples

The modified Fisher Exact Test (see more information below) can be called using the modified_fisher_exact_test() function:

library(modifiedfisher)
modified_fisher_exact_test(u = 5, m = 12, v = 7, n = 11, odds_ratio = 1, alpha = 0.05)
#> 
#>  Non-Conservative Size-α Modified Fisher's Exact Test
#> 
#> data:  u = 5, v = 7, m = 12, n = 11
#> p-value = 0.3208
#> alternative hypothesis: true odds ratio is not equal to 1
#> 95 percent confidence interval:
#>  0.072034 2.209549
#> sample estimates:
#> odds ratio 
#>  0.4081633

The above example uses $m = 12$ and $n = 11$, echoing Figure 1(a) and Table 2 in the paper linked below. As you can see, the p-value, estimated odds ratio, and confidence intervals match the paper’s. A similar size plot can be constructed by setting the local_size_data argument to TRUE, and using the resulting local size dataframe to construct the plot:

test_result <- modified_fisher_exact_test(u = 5, m = 12, v = 7, n = 11, odds_ratio = 1,
    alpha = 0.05, local_size_data = TRUE)

The data is now stored in the test_result$local.size.data object, with the first five rows given by:

pi1	size	method
0.00	0.0000000	zoom
0.01	0.0006922	zoom
0.02	0.0092791	zoom
0.03	0.0393075	zoom
0.04	0.1038314	zoom
0.05	0.2116501	zoom

Plotting this leads to a similar plot as Figure 1(a):

The Modified Fisher Exact Test

Read the article here

• Add brief rationale and explanation of theory

Next steps

• Implement checks and unit tests
• Document functions in a consistent way with article / SAS macro
• Add randomised Fisher Exact Test option