Question on quantiles for "by" variable

[pbreheny]

Asked via email, but I'm replying here since others might have the same question:

I am writing to ask how the package sets the quantiles for a ‘by’ variable for plots of multiple regression output using a “by” variable.

The documentation at https://pbreheny.github.io/visreg/articles/web/cross.html#opt indicates that "visreg sets up three panels using the 10th, 50th, and 90th percentiles". However, the code below produces a plot that shows 3 panels with Wind values of 5.7, 9.7, and 14.9, whereas the 10th, 50th, and 90th percentiles for Wind in the dataset are 5.82, 9.70, and 14.90. I’m just curious as I find the plotting features of the package very helpful and am wanting to understand some of the details of how the functions options work.

[pbreheny]

The airquality data set has some missing data, which lm() throws out before fitting the model. The quantiles are calculated on the data that was actually used to fit the model:

airquality$Heat <- cut(airquality$Temp, 3, labels=c("Cool", "Mild", "Hot"))
fit <- lm(Ozone ~ Solar.R + Wind * Heat, data=airquality)
quantile(airquality$Wind, probs = c(.1, .5, .9))
#>   10%   50%   90% 
#>  5.82  9.70 14.90
quantile(model.frame(fit)$Wind, probs = c(.1, .5, .9))
#>  10%  50%  90% 
#>  5.7  9.7 14.9