improve documentation of gf_fitdistr()

[rpruim]

• add additional examples with distributions other than normal
• provide better documentation for ...
• explain use of inherit.

[jiyunson]

gf_fitdistr() has been great for normal distributions. How can we use it for t and F? I've tried putting in the argument dist = "df" or dist = "f" into the function...

SDoF <- do(1000) *summary(lm(length ~ shuffle(sex), data = KidsFeet))$fstatistic[1]
gf_dhistogram(~ value, data = SDoF) %>%
gf_fitdistr(dist="df")

[rpruim]

A few notes:

1. You can do this with gf_fitdistr() but it requires starting values for the parameter search.
2. Since there is a theoretical distribution to compare with, you could just use gf_dist() with the known paramters.
3. You can avoid the use of summary() since do() already records the F statistic.
4. The asymptote at 0 makes this F distribution a bit finicky -- typically you will need to do something to limit how close to 0 it tries to plot values. How close will depend on your bin widths. (As you make the bins smaller, the bar nearest 0 will also get taller.

library(ggformula)
library(mosaic, quietly = TRUE)
SDoF <- do(1000) * lm(length ~ shuffle(sex), data = KidsFeet)
gf_dhistogram(~ F, data = SDoF) %>%
  gf_fitdistr(dist = "df", 
              start = list(df1 = 11, df2 = 5), 
              xlim = c(0.1, 15),
              color = "navy") 

gf_dhistogram(~ F, data = SDoF) %>%
  gf_dist(dist = "f", df1 = 1, df2 = 37, color = "red", xlim = c(0.1, 10))

Created on 2018-06-25 by the reprex package (v0.2.0).