Paired equivalence testing issue

[azh05]

Hi, thanks for creating the TOSTER package! It's been a great help.

I was running the t_TOST function with these parameters t_TOST(x = c(0, 0, 0, 0, 0), hypothesis = "EQU", y = c(1, 1, 1, 1, 0.5), paired = TRUE, eqb = c(-0.5, 0.5), alpha = 0.05) and I get this error:

Error in get_ncp_t2(t_stat, d_df, conf.level = 1 - alpha * 2) : object 'Best.Low' not found

It also occurs when I swap the values forx and y, but not when paired = FALSE. Changing the var.equal parameter also doesn't fix the bug. Am I using the function correctly?

[arcaldwell49]

You have an SD of zero in the "x" condition. The SMD cannot be estimated for paired samples in this case. You can get the results to run with smd_ci set to z or t.

Example:

# warning: bogus result interpretation
t_TOST(x = c(0, 0, 0, 0, 0), 
hypothesis = "EQU", 
y = c(1, 1, 1, 1, 0.5), paired = TRUE, eqb = c(-0.5, 0.5), alpha = 0.05, 
smd_ci = "z")

If this data is "real" however, I do not think I would recommend doing this. If you're trying to test whether the change from some baseline (x = 0) is greater than 0.5, I'd suggest just using a one-tailed Wilcoxon signed rank test (i'm making a lot of assumptions here).

wilcox.test(
  x =  c(1, 1, 1, 1, 0.5) - c(0, 0, 0, 0, 0),
  mu = 0.5, alternative = "greater"
)

Using the describe_htest function we can get the following description.

TOSTER::describe_htest(wilcox.test( + x = c(1, 1, 1, 1, 0.5) - c(0, 0, 0, 0, 0), + mu = 0.5, alternative = "greater" + )) No alpha level set or confidence interval provided. Defaulting to 0.05 [1]

"The Wilcoxon signed rank test with continuity correction is statistically significant (V = 10, p = 0.036) at a 0.05 alpha-level. The null hypothesis can be rejected. At the desired error rate, it can be stated that the true location is greater than 0.5."