For effect sizes encoded so that negative values correspond to therapeutic improvements, the one-sided p-values should be calculated for the null hypothesis $H_0: \mu \geq 0$ versus the alternative $H_A: \mu < 0$. We should include an option in selection_model() that allows the user to control the direction of the one-sided hypothesis. (Maya also suggested including a warning that gets thrown if the unadjusted average effect size is in the opposite direction of the specified alternative hypothesis, such as if alternative = "greater" but the average effect size is negative.)
Following metafor::selmodel(), we could also use this argument to allow for models based on two-sided p-values, although this would require some bigger changes in the code.
For effect sizes encoded so that negative values correspond to therapeutic improvements, the one-sided p-values should be calculated for the null hypothesis$H_0: \mu \geq 0$ versus the alternative $H_A: \mu < 0$ . We should include an option in
selection_model()that allows the user to control the direction of the one-sided hypothesis. (Maya also suggested including a warning that gets thrown if the unadjusted average effect size is in the opposite direction of the specified alternative hypothesis, such as ifalternative = "greater"but the average effect size is negative.)Following
metafor::selmodel(), we could also use this argument to allow for models based on two-sided p-values, although this would require some bigger changes in the code.