Analysis for a vector of values fails

[mattansb]

I'm trying to compute a power curve, and came across this rather off behavior - when providing d as a vector, the function fails, but supplying each element of d separately does not:

library(pwr)

d <- seq(0.05, 1.2, length.out = 25)
pwr.t.test(d = d, power = 0.8)$n
#> Warning in if (is.na(f.lower)) stop("f.lower = f(lower) is NA"): the condition
#> has length > 1 and only the first element will be used
#> Warning in if (is.na(f.upper)) stop("f.upper = f(upper) is NA"): the condition
#> has length > 1 and only the first element will be used
#> Error in uniroot(function(n) eval(p.body) - power, c(2 + 1e-10, 1e+09)): f() values at end points not of opposite sign

sapply(d, function(.d) pwr.t.test(d = .d, power = 0.8)$n)
#>  [1] 6280.04894 1638.24246  739.07484  419.13283  269.74754  188.15814
#>  [7]  138.77956  106.64471   84.56849   68.75267   57.03626   48.11621
#> [13]   41.16893   35.65311   31.20120   27.55646   24.53518   22.00310
#> [19]   19.86028   18.03104   16.45725   15.09369   13.90466   12.86180
#> [25]   11.94226

^{Created on 2020-11-22 by the reprex package (v0.3.0)}

[heliosdrm]

Thanks for your comment. As you say, the functions of the package are meant to return a single result. That's the clearest API, since there are many possible parameters that might be varied, so for instance: what output would you expect if you want to test different effect sizes, power values, etc.?

Using sapply or a loop, as you present, is the way to go if you want multiple outputs. You can also use plot to see the relationship of sample size and test power.

[mattansb]

the functions of the package are meant to return a single result

I can't seem to recreate it now, but I remember previously getting multiple results when passing a vector to a pwr.* function. But maybe it was just a dream 😅

I think it could be really useful to allow one of the inputs to be a vector, which in turn would result in an output where the argument that was left as NULL would be a corresponding vector.

e.g.:

pwr.t.test(d = 0.3, power = c(0.8, 0.9))
#> 
#>      Two-sample t test power calculation 
#> 
#>               n = 175.3847, 234.4627
#>               d = 0.3
#>       sig.level = 0.05
#>           power = 0.8, 0.9
#>     alternative = two.sided
#> 
#> NOTE: n is number in *each* group