Do not remove levels only containing NA data if na.rm is set to FALSE

[MoREpro]

The option na.rm in layers, e.g. geom_boxplot or geom_point is not working as one would expect. Even if the option na.rm is set to FALSE, the removal of NA values occur and factors containing only NA values are dropped in the plot.
I propose to improve this option as follows:

• na.rm = NA (default): remove NAs with a warning (as now with na.rm=FALSE)
• na.rm = FALSE: do not remove NA values (respective factor levels only containing NA values)
• na.rm = TRUE: remove NAs without a warning (as now)

There are occasions where the current behaviour of removing NA values and associated factor levels containing only NA values is disadvantageous, e.g. if layers added later to the plot contain data for these factor levels. Then the current behaviour leads to misaligned graphical representation.

Code snippet demonstrating, how this plot was created:

p1 <- ggplot(data = plot_data_complete, aes(y = value_ros, x = sampling_site, fill = analysed_matrix)) +
    geom_boxplot(notch = TRUE) +
    scale_y_log10(breaks = log_breaks, minor_breaks = log_minor_breaks) +
    theme_bw()+
    facet_wrap(~ substance, ncol = 2, scales = "free_y")+
    geom_text(data = count_data, aes(y = max_y * 0.9, label=count),
              position = position_dodge(width = 0.75),
              size = 3, angle = 70, show.legend = FALSE)
# Adding an element using the x-positions from first ggplot-object:
p1 + geom_segment(data = ana_threshold,
                  aes(y = min_lod, yend = max_lod, x = ggplot_build(p1)$data[[2]]$xmin,# - 0.05,
                      xend = ggplot_build(p1)$data[[2]]$xmax, col = "LOD"),# + 0.05),
                  inherit.aes = FALSE, show.legend =TRUE) +
    scale_colour_manual("LOD shown where\nrelevant:", values = c("LOD" = "red"))+
    xlab("sampling site") +
    ylab("concentration in µg/l") +
    labs(fill = "analysed matrix:")

If it is necessary to provide a minimal example I can do so.

[teunbrand]

Thanks for the report! This is a duplicate of #3345.
To clarify, na.rm doesn't really remove or retain NA values, rather it warns about removing them or not.
The proposal wouldn't quite work, because stats like stat_boxplot() need finite non-missing values for their computation.