graphspergers - the graphs and quantitative visualization thread

i prefer to let the mysteries of science marinate in the sea of self-collected data for a while - eventually, the answers always rise to the top

it would be fun to fake your way onto rightwing AM radio as a "science expert"

I've been making graphs for my company for the past year or so, and with no formal training, I still feel like this newspaper is publishing the rantings of a crazy person.

For example, we used to measure net absorption rates in regular bar form, like this:

http://assets.inarkansas.com/32941/net-absorption-by-quarter.jpg

Net absorption is such a weird stat anyway - Basically how much square footage was gained or lost in a market between two quarters. It can be positive or negative. And if a shopping mall opens or a factory closes, the numbers can vary widely.

So I wanted to show how big of a difference those numbers can be sometimes and came up with this:

http://assets.inarkansas.com/49176/central-arkansas-industrial-real-estate-vacancy-553.jpg

The marks we would've used in bar formats are still there, but I represented the rates by size. The time used above was a good one since everything was positive, but if any were negative, I could've still used the space I've got and just put the zero line in the middle. Like this:

http://assets.inarkansas.com/49054/vacancy-rate-remains-flat-781.jpg

Are those too busy? Do they make any sense? What changes would you make?

I get a little lost in the woods some afternoons I'm putting these together. No one's complained yet, but hell, who knows if anyone's even looking.

Different sized circles are usually bad for data visualization, because they're easy to mess up and can be difficult to interpret.
http://blog.revolutionanalytics.com/2011/01/for-data-visualization-circles-dont-cut-it.html
And also see the discussion in Nathan Yau's book Visualize This.
But circles might be pretty good to use if you're using them to represent area (or change in area) like you are. Just make sure the circles actually represent area, and you're not accidentally sizing them by radius or diameter.

Hm, I think using both circle size and the Y axis to represent net absorption might be bad, because it makes the circle sizes more difficult to compare. If you want to represent net absorption by circle size, consider taking out the Y axis and just setting all the circles on the same horizontal line.

pplains, are you trying to demonstrate that the changes are mostly capricious/random or just that they can vary widely year-to-year? from looking at these i'd guess that something happened in 2013 4Q that lead to a huge boom in both commercial + industrial sectors?

I also think different sized squares would be a more easily interpretable indicator of area, because people aren't as good at perceiving that the outer parts of a circle contain more area (the famous "biggest pizza = best deal" thing)

(Also, Dr. Malone, get one information criterion!)

http://www.npr.org/blogs/money/2014/02/26/282132576/74-476-reasons-you-should-always-get-the-bigger-pizza

one problem with the circles (as used above) is that it can difficult to tell at a glance what the actual quantities are - is it the point at the middle of the circle? at the top edge? bottom edge? reasonable people could come to different conclusions, i think. it's really impossible to tell without labeling each of the individual circles, which you've done. but if you have to label each of the individual circles in order to communicate the quantities, then there's probably a better way to do it. also, Dan I otm about area vs radius vs diameter

And especially since some of the values are negative, it's better to just stick with the bar chart in your first image. Really feel like the human brain might have a problem interpreting the size of a thing on a plot as actually being the magnitude of the reduction in that thing.

you could implement a new system where you walk around the office giving tootsie rolls for every 1000 sq. ft. of net office space gained that quarter. if your company loses office space, then you take away an item on the person's desk for every 1000 sq. ft. lost. this simultaneously acts as an incentive system

You could change the unit from square feet to "# of john's houses" to point out to everyone how small your rival john's house is.

at the least there is a problem with the x positions of the circles. they are not regularly spaced in the x direction. is that real?

those plots are very hard to interpret imo. estimating areas of circles is something we are always terrible at. but in this case it's even harder because what you're trying to get across is conceptually complicated, and the range of point sizes you're using is colossal.

the bar chart in the first example is much clearer. i would stick with that tbh. if you want to explicitly include both the absolute value of SF and the change (i.e. net absorption) (which i don't think you need to, it's implicit, unless i've misunderstood net absorption), i would use two bar graph panels, one above the other, sharing an x axis

x = time
y1 = square feet
y2 = net absorption

These are really helpful, I mean it.

The circles do represent area, flat two-dimensional area. This is why I wanted to run with showing the different sizes.

I'm showing how little I know about algebra by not quite understanding the difference between diameter and area. I read the blog about the State of the Union address and didn't quite get what the fuss was about.

(Though if it means anything, I understand why there was a fuss and why I would want to avoid making that mistake even if I'm not sure what the mistake was. How's that for clarity?)

I based the circles off of the area in this way: I somehow did the calculations of what the square root of 448,568 would be. I then put those x,y coordinates into the "exact ratio" field of the circle selection tool, so it would be perfect circle. Then I based the ratios of the other circles on that. Is this voodoo economics?

And fwiw, these graphs never appeared together. Even if they weren't part of the slide, it would still bug me that the circles would be the same size despite one being 448K and the other being -64K.

then there's probably a better way to do it.

oh most indeed.

at the least there is a problem with the x positions of the circles. they are not regularly spaced in the x direction. is that real?

The x positions, the number of square feet, are accurately pinpointed by the white squares. If it was a bar graph, the bars would rise and fall exactly to those spots on the graph.

The circles are illustrations only.

I have no idea if I answered that question or not.

standard advice is don't use the size or area of symbols on the page to represent any important data, because people can't "read" it

your first bar chart has the same data in it and is familiar and easy to read.

why aren't the centres of the circles on the white points?

oh wait. i had totally been misreading your graphs. they are very confusing!

my points still stand. i think you need to get rid of the circles.

And fwiw, these graphs never appeared together. Even if they weren't part of the slide, it would still bug me that the circles would be the same size despite one being 448K and the other being -64K.

this is not a minor thing. it's a huge flaw in the approach.

if area of the circle represents the data, and you have positive and negative data, then negative changes should have circles with negative areas. this is not possible.

things you can't do when you have positive and negative data:

log plots
area plots

you could compare the area to the peak area, which you could show for scale, and never go negative. but you shouldn't.

I'm starting to think I'm bombing on these.

Another stat that I thought wasn't being illustrated correctly has to do with unemployment figures. Usually, those are in bar graph form from month to month or by region, pretty straight forward.

http://assets.inarkansas.com/47290/arkansas-unemployment-rate-2013-4q-general.jpg

But there's a weird anomaly that happens from month to month where the number of jobs/number of people changes. So you might have a month where there are more people working, but the unemployment rate goes up the number of jobs go up too.

I did my circle thing again and second-guessed later that I should've made the red unemployed figure go around the circumference of the blue ball so that it would match the green total ball.

But now, I feel like I should be working for Fox News or the Enquirer.

http://assets.inarkansas.com/51231/may-2014-employment-in-arkansas.jpg

bar graphs aren't really performing any role when you can't visually tell the difference between them

like it is a fox news graph in a sense "look the economy hasn't changed at all!!"

https://www.youtube.com/watch?v=sz2mmM-kN1I

http://gecon.yale.edu/

cool data set with economic output on a 1 degree latitude by 1 degree longitude

http://oi58.tinypic.com/24en8kh.jpg
http://oi58.tinypic.com/inxu87.jpg
http://oi60.tinypic.com/k13skz.jpg

grid

https://pbs.twimg.com/media/D6PtWdLXsAAgIht.jpg

- "Maybe we should get another usual suspect in the lineup besides the Indian woman."

- "But from where? Estonia? Venezuela? There aren't many other countries to choose from!"

Look at the difference between 5’4” and 5’5” on the y-axis, compares to between 5’0” and 5’1”

Other than that, great chart design!!

The Latvian woman is huge and the woman from India is tiny to visually convey the fact that Latvia has a female population at least ten times larger than the female population of India.

The sum of the height of all the Indian women will be more though. Is there a graph of that?

https://pbs.twimg.com/media/D7K2CkyXkAAEJKT.jpg

that was so upsetting. Also needs the the n/2+7 line drawn on as well

is there a name for a visualization that would accomplish the following?

i want to compare two populations that each have two subsets--say one of them is 15M people total, then 4M of those people meet a specific condition, and 1.9M of those 4 meet a further specific condition. and the other population has the same conditions but completely different proportions.

so basically like a treemap but instead of the whole area adding up to the total it would have proportional smaller rectangles embedded within a big rectangle? is this even a thing?

would a simple stacked bar graph do the trick? here are two that meet your requirements:
https://i.imgur.com/me2obge.png
https://i.imgur.com/sW9m0c5.png

the first shows two populations of different sizes, the second shows two population of equal sizes.

or wait, i see what you're saying. subsets within subsets. if that's the case, you could just color code the results. 11M non-diarrhea, 6 million with diarrhea. non-diarrhea is a deep calm blue, diarrhea is an agitated warm color. 4.1 million of the 6 million have severe diarrhea, so make that deep red. the other 1.9 million have moderate diarrhea, so make that orange.

https://i.imgur.com/eZND3qv.png

or, to go to your op, a tree map, and just format the results to highlight the groupings you want

Karl I am gonna need seventeen more made up diarrhea graphs on my desk by COB tomorrow.

i'm glad i processed those extra participants' waivers during my lunch break yesterday, sunday, instead of eating

"Beginner's Diarhhea"!

What a wonderfully misleading diagram in the Times today 📈 pic.twitter.com/isQtZS6Mot

— Will Bailey-Watson (@mrwbw) June 27, 2022

loooooooool

An analysis looks at how defense spending among the nations with the highest expenditures has changed since 1992 and what may have driven the changes https://t.co/3ln08vOKAo pic.twitter.com/yqK6MqwQUm

— St. Louis Fed (@stlouisfed) January 22, 2023

https://i.redd.it/cxtoiiuy9l9b1.png

see you there in April!