Friday, October 31, 2014

annotated line graph from Uber

With the email that hit my inbox earlier this afternoon, Uber has impressed me twice in the past week. The first time was in response to a simple comment that accompanied my '3' numerical rating (the lowest I've ever given): "With the world series game today, should have avoided stadium area." I had an email in my inbox from Uber's customer service within the hour agreeing that was a silly route given the Giants' game and reducing the price to what it would have been without the crazy traffic. Amazing.

And now they've done it again, this time via effective data viz. The annotated line graph below shows expected Uber demand over the course of the evening and into the wee hours of morning. This is one of those rare cases where they can get away without showing the y-axis values at all, since the relative peaks and valleys are more interesting (and meaningful) than the absolute numbers.

Nice job Uber. Though I must say this makes me happy to report that kiddie Halloween in my neighborhood is on foot, so no need to even think about surge-pricing here!

Speaking of which, I find it impossible to publish a post on Halloween without couple pics of my superhero family.

Happy Halloween!

Tuesday, October 7, 2014

SF housing cycles visualized

If you've shopped for real estate in San Francisco recently, you've likely experienced the crazy world of multiple offers, waived contingencies, and all-cash deals well above asking price. We've been house shopping here for nearly two years, without much to show except a jaded view of the market and an ever-increasing pile of home-for-sale flyers. My husband and I joke that our toddler will grow up thinking that's what you do on the weekends: go look at other people's houses.

If you've been in this situation, or a similar one, you've perhaps also wondered (like us) whether prices will continue to increase at the rate they have been, or if there is an elusive bubble that is about to pop. To that end, we came across the visual below, which depicts a simplified view of San Francisco housing market cycles over the past few decades.

If you've followed this blog for long, you might expect that I will next proceed to rip the above visual apart. But I am not going to. 

I actually really like it. 

Sure, there are some minor things that could be changed. But let's focus instead on the good: it's well-labeled, both in terms of titles and text annotation on the graph itself. There is a clear narrative that calls out some interesting things in the data. For example, over the past 30+ years, the period between a recovery beginning and a bubble popping has been about 6 years.

According to the graph, the last recovery began in 2012, which would put the next bubble pop at approximately 2018.

Which means there's still time to buy before we hit the peak... 

Thursday, September 4, 2014

show the full picture!

There is still space available in my upcoming San Francisco public workshop; details and registration can be found here. Click here to suggest a location for a 2015 workshop.

I've posted a number of times about Pew Research articles. Well, not the articles exactly, but rather the visuals they contain. To be honest, it's rare that I read the actual article. I scan the headlines as they hit my inbox and if something piques my interest, I follow the link and scroll through the article, not reading, but taking a discerning look at the graphs.

This is what I was doing when the following visual caught my eye within an article titled, "Perceptions about women bosses improves, but gap remains." The full article can be found here.

They are nice looking graphs, as is the norm. But upon examination, for me both of these visuals leave out an important part of the picture. Let's examine them one at a time.

I'll start with the top chart titled "Boss Gender Preference." The first thing I want to do with a graph like this is add the percentages to understand what proportion of the overall population we're talking about. Everybody? No. In this case, a little math tells us that a lot is missing, especially in the more recent years. My "data-spidey sense" (as a former boss of mine used to call it) gives me suspicious pause. I go back and read the title, etc. to try to put what is missing (the piece or pieces that would enable the lines to add to 100%) into context. In this case, the lines represent the percent of those responding to the survey who said they prefer male bosses and the percent who prefer female bosses. So, I would assume that the remainder are those who have no stated preference (and only now as I write this do I see the footnote below the second graph confirming this; for sake of this discussion and my forthcoming makeover, I'll assume "no opinion" is the same as "no preference"). In recent years, this accounts for nearly half of the total. That seems important. And yet isn't shown explicitly.

Let's look at a different view of this data:

In this remake, I replaced the line graphs with stacked bars so that we can see the full 100%. Those preferring male bosses are along the bottom in blue, female along the top in orange so that we have a consistent baseline at both bottom and top to be able to compare what has happened over time for those groups. I only graphed the points that were labeled in the original graph, omitting the 2000 data point altogether so that I'd have a (roughly) consistent gap between the dates along the horizontal x-axis. That data point was strange to me in the original graph, anyway, since preference for both male and female bosses increased (while % indifferent went down), but then bounced back to the previous trend. It looks like something strange may have been going on there (was the question worded differently? had something recently happened in current events that influenced this? I'd want to better understand this, however since it doesn't seem critical to the overall story I'll simply omit).

With this view, you can visually lump together the grey and orange bars, or even focus on just the grey to see progress on a different level than was shown in the original graph - from an overwhelming preference for male bosses to lessened sensitivity when it comes to the gender of one's boss overall.

After plotting the data this way, I considered that perhaps a line graph would work better, but with an additional line for the indifferent portion. I drew it, but after seeing it, it reinforced for me that the stacked bars work better for being able to add different data points together, which I think is important here. In case you're curious, here's what the line version looks like:

Next, let's turn our attention to the second visual: Female CEOs. The graph as drawn by Pew Research looks like great success: up and to the right. In this case, frame of reference is critical. Yes, 4.8% is huge relative to zero. But it's really small compared to the potential (100%). Here's how I'd graph this one:

With this view, there has been progress - yes - but there is potential for much more.

As a side note, it also bothered me that the dates were totally different between the two graphs (it seems like you should be able to compare the data points across them, but if you follow 1975 in the original top graph down, this lines up with something like 2002 in the bottom graph). I attempted to address this through the titling of the graphs ("the past 60 years," "the past 20 years") but this isn't a perfect solution. Part of me wants to have a bunch of empty space on the left of the second graph and make it so the 20 years of data we have there lines up to the top graph along the same date scale, but I think this would squeeze the data we do have too much to be legible. So I settled with attempting to address through the titles.

Here is the side-by-side of the two visuals:

Note how in both cases above, the remake causes us to see a different story and perhaps even draw different conclusions than we might have with the original visuals. For me, the makeovers present a fuller picture.

It is possible that this fuller picture is all explained in the article. I really can't say, as I still haven't read it. I am probably not the only one who "reads" this way. Which is another lesson: design your visuals so they still work when your audience doesn't read the accompanying text!

In case you're interested, my other recent posts on Pew Research visuals can be found here and here. The Excel file with the makeovers from this post can be downloaded here.