5 ms·

Methods of Comparison, Compared: A Visual Guide to Comparing Change

114 points by jashkenas 8y ago

eboyjr 8y ago
Useful for some politicians.
- coldacid 8y ago
  And statisticians. And readers of infographics.
- stareatgoats 8y ago
  Journalists. I'd say journalists must read this. Maybe every Monday or so.
kjeetgill 8y ago
Fantastic piece. I love the systematic approach to refine my intuitions. Not just how to reason about statistics, but how to effectively second guess myself. And the visualisations really help to reflect on that. Side note: I thought the visuals we're too impressive, turns out it's written by Mike Bostock of d3.js and bl.ocks.org/mbostock fame. His article "Visualizing Algorithms"[0] is a must read. [0]: https://bost.ocks.org/mike/algorithms/ https://bost.ocks.org/mike/algorithms/
- apendleton 8y ago
  Observable, the notebook platform hosting the piece, is his (along with Jeremy Ashkenas who posted this to HN, and Tom MacWright) new startup, and I think this post is meant in part to demonstrate its capabilities.
  - jacobolus 8y ago
    And for everyone reading: go play with Observable. It’s fantastic!
saagarjha 8y ago
If you take one thing away from this article, hopefully it's this: > No method is better universally, and none of them is “the best” even in the context of the dataset.
jzl 8y ago
Ha, I thought this was going to be an article comparing BeyondCompare, WinMerge, xxdiff, etc.. But it was interesting nonetheless!
jacobolus 8y ago
And this doesn’t even discuss the effect of changing the color scale on the visual appearance (linear vs. log vs. some binning algorithm etc.), measuring counts per region vs. per capita vs. per square mile, drawing a cartogram, etc. There are a lot of choices when displaying data on a map using colors. Mike & Jeremy: one that you might want to toss in here is normalizing so that regions matching the national average increase are colored neutrally (instead of no change being colored neutral). i.e. normalizing relative to the national trend instead of relative to the previous year’s value. Depending on the data, this can help separate national vs. regional trends, though it takes more careful explanation to avoid reader confusion.

114 points by jashkenas 8y ago

eboyjr 8y ago

Useful for some politicians.

coldacid 8y ago
And statisticians. And readers of infographics.
stareatgoats 8y ago
Journalists. I'd say journalists must read this. Maybe every Monday or so.

kjeetgill 8y ago

Fantastic piece. I love the systematic approach to refine my intuitions. Not just how to reason about statistics, but how to effectively second guess myself. And the visualisations really help to reflect on that. Side note: I thought the visuals we're too impressive, turns out it's written by Mike Bostock of d3.js and bl.ocks.org/mbostock fame. His article "Visualizing Algorithms"[0] is a must read. [0]: https://bost.ocks.org/mike/algorithms/ https://bost.ocks.org/mike/algorithms/

apendleton 8y ago
Observable, the notebook platform hosting the piece, is his (along with Jeremy Ashkenas who posted this to HN, and Tom MacWright) new startup, and I think this post is meant in part to demonstrate its capabilities.
- jacobolus 8y ago
  And for everyone reading: go play with Observable. It’s fantastic!

saagarjha 8y ago

If you take one thing away from this article, hopefully it's this: > No method is better universally, and none of them is “the best” even in the context of the dataset.

jzl 8y ago

Ha, I thought this was going to be an article comparing BeyondCompare, WinMerge, xxdiff, etc.. But it was interesting nonetheless!

jacobolus 8y ago

And this doesn’t even discuss the effect of changing the color scale on the visual appearance (linear vs. log vs. some binning algorithm etc.), measuring counts per region vs. per capita vs. per square mile, drawing a cartogram, etc. There are a lot of choices when displaying data on a map using colors. Mike & Jeremy: one that you might want to toss in here is normalizing so that regions matching the national average increase are colored neutrally (instead of no change being colored neutral). i.e. normalizing relative to the national trend instead of relative to the previous year’s value. Depending on the data, this can help separate national vs. regional trends, though it takes more careful explanation to avoid reader confusion.