## The inutility of stacking columns

##### Sep 03, 2013

When I see a stacked column chart like this:

(via Andrew Sullivan)

I immediately want to remake it as a line chart, like this:

The decline in households with two parents and children appears more dramatic in the line chart, primarily because it has the right spacing of time.

You can follow this conversation by subscribing to the comment feed for this post.

Ummm….no. Sorry.

It appears more dramatic for primilarly because the verticle difference between 40% and 20% is three times as large in the line graph as the stacked columns (relative to the width).

It is secondarily because time is properly spaced.

While I agree that stacked column charts are less than useful, the inutility of the scale on the axes is completely incidental and could have been reproduced in the line chart, too. The only objective advantage of the line chart is that you can reduce the maximum value of the Y axis to 40%, enhancing the change in values by increasing the scale.

Indeed, since you choose the bottom set of data for your comparison, you're choosing the set of data that is *least* differently represented in the two charts. If you look at elsewhere (e.g. the "Married couples w/o Children"), the change in values is more clearly conveyed in the line chart than the stacked column.

I like the stacked charts better, because you can look at each bar and perceive the percentages within that bar. When you convert it to a line chart, you can no longer see how much each thing is a proportion of the total.

Spacing the time axis correctly is certainly better, though. It should be stacked lines with colored areas rather than stacked bars, and the categories should be ordered to minimize their shift in position over time.

It depends on what you're displaying. For percentage charts, absolutely; it makes it much easier to see what's happening to each segment.

But if, instead of percentages, they'd been displaying absolute numbers of households, then stacked columns lets you show relative percentages, changes, and growth of the total. Since the number of US households in 2012 is substantially different than in 1970, the stacked column has definite advantages.

And if you're using percentages, why aren't you using a pie chart?

I can see only one case for stacked bars. It is when you have only two parts of a whole. In all other cases it is a mess and there are much better ways eg small multiples or linecharts like here.
It would be nice if stacked bar chart could indeed for absolute numbers show relative percentages, changes and growth of the total. In reality it only shows growth of the total and growth of the bottom series. Relative values and changes are next to imossible to see. sorry

Stacked columns are also better than line graphs when the categories are related and can be rank ordered. Graphing likert-scale-style results with line charts doesn't help much because you have to be able to add up the "very" and "somewhat" in your head. Not so with stacked bars. Of course if the lines represent "very" and then "very and somewhat", you can get the same effect but it's a bit jarring.

The time spacing excepted, the stacked bar shows me more than the line chart. Specifically if I want to scan for the changing impact of 'living alone' - the proportion of the combined boxes for men and women, is easier to see. It's also immediately clear that the total is a constant 100%.

Wow, these comments are like Bizzarro land! The stacked chart is almost unreadable to me -- ignoring the inconsitant x-axis, I find it really difficult to gain any insights when the base of each category moves over time. The line graph, conversely, immediately displayed its message. I even gained an insight that, even now that I know it, I can’t easily see on the stacked column chart: the percentage of married couples without children has remained steady over the years, so the decline in child-bearing married couples comes entirely from the increase in other living situations and not an overall decline in marriage (despite that marriage, overall, is declining – something still easy to see from the line chart).

But, if you really need to know that the proportions of a whole always add up to 100%, I guess the stacked chart *does* work better…

A dead discussion now, obviously - I'm only here because I read through the archives on my lunch breaks to get helpful tips for my afternoons of graphing... But I agree with Bryan! "The changing impact of 'living alone'" is a useful thing to know about, but that's not enough of an argument for the use of unreadable stacked bars; if anything that's an argument to say that the "Men living alone" and "Women living alone" figures should have been aggregated in the original graph, but that's a separate issue. Just throwing in my two cents for the edification of anyone reading this during their lunch break a year from now...

The comments to this entry are closed.