i love putting in hours of research work to at best get credited somewhere in prose

and i'm not talking the usual guiding nature of sourcing. people don't naturally know what info is needed. i'm talking about Doing Work For Someone, Unpaid And Uncredited

(all flying from New Orleans, FWIW)

other two are #2010 and #2002

had to do this in person once. hell

oh what i mean is a violin plot/heatmap per bin. the histogram used is largely problematic because it shows a single value, the mean, which is not a solid representation

hm, i can see that. what about something more distributional like a violin plot or 1D hexbin (heatmap?)

questions? hate my idea of how this would be better plotted, or have anything better? want to send death threats? ask away :P

of course: i'm assuming here that the sentiment analysis is solid, and that it handles well enough things like images, emojis, other languages, etc.

but i hope you've learned some key ideas in data viz! and apologies to @hailey.at for criticizing her graphs so harshly. was just teachable. 💜

this way, the dots both act to show d* instead of s* _and_ to show a sample size. if the scatter feels difficult to parse, add s* as a translucent set of lines, only to guide.

annotations are thus minimal — sample sizes as numbers, and a brief explanation of how your sentiment analysis works.

so what about a scatter plot?

keep the horizontal axis (you may be able to widen your bins), but make the y axis show absolute sentiment — the raw, non-summary value. then plot each post as a dot in the appropriate age bin. show examples of famous posts with their sentiments as a unit reference.

so how would this data be better plotted?

let's see: we want to show non-summaries of sentiment per group while also showing parts of D like sample sizes.

and we want to give a reference for what these sentiment values mean, and maybe leave an annotated note about how the sentiment analysis works.

hailey's graph shows _just_ s*, the mean sentiment of each age bin. this is not a visualization of data! it is a visualization of a summary of data!

to illustrate why that's bad, enter Ansombe's Quartet. four graphs, obviously different when plotted, but same mean, regression line, etc.

but this brings me to the third and final part of the sentence, and the second central thing hailey's plots fail at: D, and d*, are of infinitely greater importance vs. s*.

here, s* refers to a statistical summary — mean, median, regression, etc. — of d*.