visualize()

[MarHai]

Given the change to returning tibble-resolvable objects (rather than just tibbles), we could implement visualize() generic functions that allow to visualize wherever it makes sense. Here are some suggestions:

• describe(): horizontal box plot

• describe_cat(): horizontal stacked bar chart & unstacked (via argument)

• tab_frequencies(): 1 var: histogram, 2 var: facet_wrap, 3 var: facet_grid, 4 var: warning()

• crosstab(): horizontal stacked bar chart depending on absolutes / relative freq

• t_test(): points with 95% ci as pointrange/points with error lines

• unianova(): points with 95% ci as pointrange/points with error lines

• correlate(): scatter without line

• to_correlation_matrix(): corrgram

• add_index(): -

• get_reliability(): -

• test_icr(): -

• regress(): scatter + lm, qq plot (, which = ..., default ist ergebnis, also scatter/linie)

• plot

• save object to make further enhancements (e.g., coord_flip())

I would expect the produced plots to always represent ggplot visualizations (no plot() visualizations). Also, we could consider returning the object just as passed in (for chaining consistency) and just output the plots. This way, we could also output multiple plots, as could be the case for the visualization.regress.

Thoughts?

[joon-e]

Really like this idea! The implementation would be closely linked to #24, so we may want to decide here first. I'll also think more about potential visualizations.

Also making a mental note here that visualize() is one of the generics implemented by the generics package, so we should use this as well.

[LKobilke]

Just skimmed the ppts of our master students. They also learn how to interpret residual plots to check for homoscedasticity, so we might want to add this to regress() for upwards compatibility. + qq-plots or pp-plots.

See here:

[MarHai]

See #32

Updates:

• describe_cat only unstacked due to lack of meaningful generalizability
• crosstab absolute/relative switch implemented from crosstab(..., percentages = T/F) (i.e., no additional switch)
• correlate only scatters with two variables and otherwise internally switches to to_correlation_matrix
• to_correlation_matrix now uses GGally::ggpairs (instead of corrgram)
• regress allows to choose from a wide range of plots but, by default, plots the outcome; heavy documentation added