Estimate hardcoded as character with prop.test() #1197
Description
When using model_parameters() with a prop.test() object, the reported Proportion column is hardcoded as a percentage and the underlying estimate is gone and un-extractable (unless I'm missing an argument that makes this work!)
This is happening because Proportion is explicitly built as a character string:
Lines 518 to 522 in 3a9c8f2
Here's a reproducible example:
library(parameters) table(penguins$sex) |> prop.test() #> #> 1-sample proportions test with continuity correction #> #> data: table(penguins$sex), null probability 0.5 #> X-squared = 0.012012, df = 1, p-value = 0.9127 #> alternative hypothesis: true p is not equal to 0.5 #> 95 percent confidence interval: #> 0.4406707 0.5504259 #> sample estimates: #> p #> 0.4954955That 0.4954955 (and other non-visible floating point digits) is the estimated proportion.
In model_parameters() it appears rounded, which is fine:
table(penguins$sex) |> prop.test() |> model_parameters() #> 1-sample proportions test #> #> Proportion | 95% CI | Chi2(1) | Null_value | p #> -------------------------------------------------------- #> 49.55% | [0.44, 0.55] | 0.01 | 0.50 | 0.913 #> #> Alternative hypothesis: true p is not equal to 0.5In the underlying data frame, the Proportion column is a character, not a double:
table(penguins$sex) |> prop.test() |> model_parameters() |> as.data.frame() #> Proportion CI CI_low CI_high Chi2 df Null_value p #> 1 49.55% 0.95 0.4406707 0.5504259 0.01201201 1 0.5 0.9127271 #> Method Alternative #> 1 1-sample proportions test two.sidedThis makes it tricky to extract that value and use it other places, like in plots:
library(tidyverse) table(penguins$sex) |> prop.test() |> model_parameters() |> mutate(estimate = parse_number(Proportion) / 100) |> ggplot(aes(x = estimate, y = "Proportion")) + geom_vline(xintercept = 0.5, color = "red") + geom_pointrange(aes(xmin = CI_low, xmax = CI_high))
↑ That works, but the value is actually only 0.4955, not the complete true value estimated by prop.test(), since parse_number() converted it from text to a number
It's possible to use broom::tidy() to extract values from htest objects like prop.test(), and it maintains the numeric-ness of the estimate column:
table(penguins$sex) |> prop.test() |> broom::tidy(conf.int = TRUE) |> ggplot(aes(x = estimate, y = "Proportion")) + geom_vline(xintercept = 0.5, color = "red") + geom_pointrange(aes(xmin = conf.low, xmax = conf.high))
But then it requires {broom} for this kind of plotting (while all other htest objects like t.test(), etc. can be extracted and plotted with CIs with model_parameters()—it's just prop.test() that has the character-only version of the estimate)