“If Islamic people do something bad, you think, ‘Oh, it’s Muslims,’ ” she said. “But if a white Protestant does something bad, you just think he’s mad. That’s something we need to think about.”

– Sigrid Skeie Tjensvoll, Oslo resident.

A: Hats.

10. an inverse U-shaped relationship

9. abusing the notation

8. the exact channel through which these penile-effects take place

7. all evidence is suggestive

6. non-disposable groin-area endowments

5. male organ has not been touched in the growth literature before

4. quite penetrating an argument

3. male organs dwarf political institutions in importance

2. the `private sector’ deserves more credit for economic development than is typically acknowledged

1. The aim of this paper is to fill this scholarly gap with the male organ.

[Male Organ and Economic Growth: Does Size Matter?]

I used to get so mad when Sami would tell me I was wrong about something, especially when it was something that I study and talk about on a daily basis. After all, not only do I surround myself with academia about sociology and psychology, but I write about socialization and sexism all the time. How could I be wrong about something if I do all these things? How could I be wrong about feminism if I am active in the community and know a lot about it? How could I be wrong about what is sexist and what is not?

The answer is easy: I’m not a woman.

Or: maybe despite reading and studying a lot, you were just wrong. Most people don’t need a vagina and/or personal journey of discovery to know that ‘She was asking for it’ is not an acceptable response to a rape accusation.

Now, what do we blame when a woman ’surrounds herself with academia’ and still comes to wrong conclusions? (The patriarchy, I guess.)

[Laswonry]

In my last post, I introduced the slopegraph, a fun ‘new’ kind of chart from 1983. My first attempt didn’t quite live up to my expectations, so I’m trying again today with some more fresh and interesting data from Bill Easterly and Ariell Reshef’s working paper on African export successes. This data is actually so perfect for the slopegraph format (which focuses on both rank order and relative rate of change) that the original paper used a sort of proto-slopegraph to show how rankings of some African countries’ top export goods had been reordered. But it’s not only the characteristics of the data that make slopegraphs interesting in this context — it’s the role of chart-making in the construction of a narrative, which is the aspect I want to focus on today. To keep things simple, I’m going to use only one example, their export data and related chart for Tanzania:

Figure 1: Original ‘Tanzania Top Ten’ chart

No offense to whoever put these together — it is a working paper after all — but the charts in the original contain a lot of what data nerds would call ‘chartjunk,’ or unnecessary graphics and information. The preference for beautiful, minimalist charts is not just aesthetic chauvinism: while nice looking charts are, of course, nice to look at, there is a more important issue here. Chartjunk imposes a real cognitive burden on readers, which actually makes it harder for them to understand the chart’s message. Professional academics and researchers will be able to chew through the gristle and get at the meat, but of all the economic specialties, development economics probably has one of highest proportions of non-technical readers — notably development practitioners, policy makers, and advocacy groups. For this reason, minimizing unnecessary cognitive burden in development papers is especially important. This is equally true of working papers, since practitioners, politicians, and advocates will rarely wait for peer review to disseminate, digest, and discuss a new research finding.

I do think that the ‘throw everything in’ tendency that academics have when making charts arises from a laudable desire to fully disclose all the data on which they are basing their arguments, so that their peers can potentially spot inconsistencies or contradictions. I would say that this data should be included, but as an appendix. Charts form a part of the argument of the paper, and so should be tuned to focus on those aspects that the author considers important or notable.

(Read on to find out how I tried to ‘fix’ this funky-looking chart…)

Charts and Narrative

In order to determine what is ‘unnecessary’ in this chart, I needed to understand the argument that Easterly and Reshef were trying to make with this paper, and also the broader meta-narrative to which they are contributing. Let’s discuss the meta-narrative first. Easterly is a well-known proponent of the idea that aid cannot ‘fix’ poor countries: aid can help, but real economic development can only come from within, based on the many individual choices made by actors in a particular country, rather than the decisions of top-down planners. This is not just ideology: Easterly bases this argument partly on the fact that economic success is often idiosyncratic, which is well supported by the economics literature. It is this branch of the literature which Easterly and Reshef are contributing to here.

The Easterly anti-planning, pro-indigenous development meta-narrative also stands in (self-conscious) opposition to the mainstream view of the ‘development industry’, which is partly based on dependency arguments. A key aspect of the dependency meta-narrative is that the economies of countries which rely on a single commodity export are at the mercy of ‘the market,’ and furthermore that the value of their raw material exports are subject to a long-term, secular downward trend relative to the cost of manufactured imports. In the 1970s, this dependency meta-narrative was used as support for import-substituting industrialization policies. Today, the idea that commodity exporters are ‘dependent’ on the market prices of their main exports (while the type and quantity of commodities they export is seen as fixed for technological, climate, or sunk-cost reasons) is used to support notions that developed countries have a moral obligation to provide top-down aid to help them diversify, or that rich-country consumers have a moral obligation to buy fair trade goods or contribute money to INGOs.

Easterly and Reshef’s findings challenge this meta-narrative, and — interestingly, for a pair of economists — they do this using largely qualitative arguments to generate some ‘stylized facts’ (one of my favorite economics terms) about African export successes. This is another reason I think that a chart which integrates tightly with the narrative could be particularly helpful — and why a slopegraph is the right chart for the job. First, rank order is very important to their argument. If most African countries were ‘stuck’ with one or a few main commodities, we would expect these commodities to dominate their exports, at least over the medium term, with their total value varying in line with the market price. Instead, Easterly and Reshef found that the rank order of export commodities had undergone major turnover, even while the total value of each kind of export often went up in real terms. Their data analysis showed that these variations were primarily due to changes in quantity exported, not price, meaning that it was the African producers themselves who were effecting these shifts — not changes in world market prices.

These findings and the related argument are reasonably well reflected in the original chart. The chart shows three pieces of data for each export commodity in each time period: the name of the commodity, the total value of all exports of that commodity, and its rank relative to other commodities. Strangely, considering that much of their analysis is focused on the power law nature of commodity rank hierarchies (see here for a good run-down of what that means), the chart doesn’t report what proportion of total exports each grouping comprises in each year. As you can see from the table below, not only the rank order but also the rank structure of commodity rankings changed significantly from 1998 to 2007.

Figure 2: Top Tanzanian exports expressed as a percent of total merchandise exports (by value)

In both 1998 and 2007, the top commodity represented about 18% of total merchandise exports by value. But in 1998, the No. 2 export commodity (coffee) represented close to 13% of export value, while in 2007 this had fallen to 7% for precious metal ores, lower than the No. 4 commodity in 1998. Another way of expressing this is with ratios: in 1998, the ratio of the value of exports No. 2-10 to No. 1 was 2.9; in 2007, this had fallen to 1.4. In other words, Tanzania has become much more dependent on a single export (gold), an important change which you cannot easily see in the original chart. However, at the other end of the distribution, Tanzanians also seem to have diversified into a much broader range of exports: Top Ten commodities accounted for 70% of exports in 1998, and only 55% in 2007. The inability to show this is a major deficiency of both the original chart and my reworking of it (this could maybe be solved by adding an ‘All other commodities’ item).

In addition to lacking any data about export shares, the original also contains some redundant and unnecessary information, notably the numerical rank orders, which are already apparent from the position of each item on the list (except for items which fell out of the top 10). It is not really the numerical ranks themselves which are interesting, but the relative rank order, and particularly the change in the relative rank order. This is indicated on the original chart by a series of arrows, but the use of arrowheads and the narrow space into which they are placed makes them difficult to follow. A glance at this part of the chart tells you that the rank order has been rearranged, but it is not exactly clear how.

Cleaning Up the Chart

Once again in making this slopegraph, I came up against major scale issues (those of us born in the Excel era have no idea how lucky we are!). I think this is going to be a fundamental problem when using slopegraphs for economic data, since a lot of the economic data which might benefit from the slope treatment looks like this:

Figure 3: Some ‘stylized’ ficitional economic data.

In order to have items at Time 1 adequately spaced while still keeping the same scale at Time 5, you are often left with a huge amount of whitespace in the top left quadrant of the chart. This means readers must move down to the bottom of the page to start parsing the chart; not an ideal situation, since English speakers read from left to right and top to bottom. But if you use two different scales, even for a two period chart, the slopes become meaningless, and you are left with just a ‘chart with lines’ rather than a slopegraph. The problem can be rectified by reversing either the Time or Value axes, but I think we are so conditioned to reading charts with Time increasing along the X axis and Value increasing along the Y axis that this would be unnecessarily confusing for most readers.

My Attempt

Figure 4: My second ever slopegraph!

After some experimentation with this data set, I arrived at a compromise which might not please everyone: I scaled the vertical position of the items according to their proportion in total merchandise exports, while displaying their value in real dollar terms. By choosing not to also display the proportions for each item, I sacrificed precision for readability, while still giving the slopes themselves more meaning than they would have under an arbitrary Y scale (as in the original). Now each line encodes an interesting piece of information: the rate of change in that export’s proportion of total exports. However, I still had to do some ‘fudging’ so the scale is in fact a good bit looser around the 1% level than it is at the 17-18% level. This would probably need to be fixed somehow if this chart were used in an academic publication. I thought about grouping Wheat (1.4%), Cotton (1.4%), and Tea (1.3%) together in 2007 at their average dollar value ($41,061), but this would disguise the important fact that each of these commodities has increased in absolute value while still declining dramatically in share of total exports.

As you can see, I also chose to shorten the names of the export categories. This was a personal design decision intended to further reduce the cognitive burden on readers, who probably do not need to distinguish between ‘Cotton, not carded or combed’ and other types of cotton to understand the chart. There are two places where the full category names may provide a meaningful piece of information: the rise of ‘Precious metal ores’ alongside ‘Gold, unwrought, semi-manufactured, powder form’ (which suggests that maybe the growth of Tanzania’s gold mining is outpacing its processing capabilities), and the fact that ‘Mounted precious or semi-precious stones, not diamonds’ has displaced ‘Diamonds, not mounted or set’ in terms of relative share of exports (which suggests… I don’t know what?). These distinctions are probably better explained in the text, or a footnote, rather than forcing readers to guess whether the fact that Tanzania’s cotton exports are not ‘combed’ means something in the context of the chart.

Overall, I think my version presents the same data in a more easy-to-read way that focuses on the particular aspects which support the associated narrative. However, I am not sure if my attempts to reduce cognitive burden by removing ‘chartjunk’ were fully successful, since my version also imposes the new cognitive burden of figuring out the relationship between the vertical scale and the numbers reported next to each item (why is Gold in 2007 not four times higher up than Coffee in 1998?). Personally, I think this is a worthwhile trade-off, as the fact that certain commodities have declined dramatically in relative terms while still rising dramatically in absolute terms is essential to expressing the idea that these are African export ‘successes.’ I also think that despite its flaws this is a much better example of what a slopegraph can do, as compared to a table or line chart, than my previous example.

I’d be interested to hear from both academics and data geeks (I assume there is a pretty big overlap) on this one. Do you already think about these things when making charts? What would you do differently?

Big thanks to frequent blog target William Easterly for permission to rework his charts. You can find an ungated version of the original paper here.

Also, since I think it’s important to give proper respect to less ‘famous’ coauthors, you should really check out some of Ariell Reshef’s other work if you haven’t heard of him: I liked this paper on wages (and rents) in the U.S. financial industry from 1906-2006. 

What is a Slopegraph?

The slopegraph was invented by data visualization hero Edward Tufte back in 1983. Unlike Tufte’s other innovations, such as the sparkline, this one didn’t really catch on — in fact, Tufte didn’t even give it a name until 2002. Last week blogger Charlie Park was able to find only three examples on the entire Internet, most of which varied from the original concept in some way (such as comparing different types of data).

As Park explains, the slopegraph or table-graphic is basically a ‘super-close zoom in on a line chart.’ Why would this be useful? In classic Tuftian fashion, the slopegraph weeds out the junk to focus on specific aspects of the information being presented: the hierarchy or rank order at each period; the rate of change over each period; how the rates of change compare to each other; and any notable deviations from the trend (e.g. China above). Done correctly, the slopegraph has zero ‘non-data ink.’

Why Make One?

For me, it was mainly for the challenge, and to see what a slopegraph could do that a more traditional line chart couldn’t. It was an interesting experience: since there are no automated tools, making a slopegraph is hard work and requires careful planning and attention to detail. I spent an entire working day on this, and scrapped three previous versions before arriving at one I even sort of liked. This proves another point that Tufte, a notable critic of PowerPoint, has been making for years: modern data presentation tools often obfuscate as much as they reveal. As long as Excel has that big colorful pie chart button, people will continue filling their papers with worthless pie charts. And until there is a slopegraph tool available, I doubt most people will be making slopegraphs.

Still, while I can’t see the slopegraph becoming a hot new trend, I think they could be very effective in certain applications where both rank order and relative rate of change are important. I also like that they force you to really think about the story that you’re trying to tell with your data, rather than just slapping together a table or line chart because you have a bunch of numbers and you can. While some advocacy groups and media organizations (like Good magazine) have done some amazing things with visualizing development-related data, the quality of visual data presentation by development researchers is still generally abhorrent. If you are lucky enough to be paid to sit in a university office and think about things, you could find a worse way to spend a day than making a graph like this, especially if the experience encourages you to think more carefully in the future about what you are trying to say with your data.

What I Liked

As I said, the slopegraph makes you think hard about the story you are trying to tell. In this case, the basic story is one that anyone involved with development has already heard a million times: while some poor countries have ‘taken off,’ most remain poor. Another interesting aspect of this particular data set and choice of time periods is the lack of a consistent pattern in terms of region, culture, or type of government determining which country ‘takes off’ and which one stagnates: the three most successful countries, China, Morocco, and India, have very little in common except that they all started poor and became rich. Prior versions of the chart using slightly different time periods and GDP measures produced a list where Botswana and Equatorial Guinea were the big breakouts; another version starting in 1970 had Vietnam. Recognizing that my choices would reflect the story told by this chart was an important part of the process for me.

However, what I really like about this style of presentation is that it breaks the story down into discrete sub-narratives by country and decade. As I alluded to yesterday, I am somewhat skeptical of tendency in development economics to focus on large-scale trends — with the implication often being that this will allow us to devise ‘universal’ rules. Too often economists write off the outliers as uninteresting, unimportant, or even harmful to their analysis. But given the rarity of the outcome development economists are searching for — poor country becomes rich country — it seems like the outliers are exactly what they should be focusing on. Looking at this chart, the natural question for the uninitiated would be “What’s the deal with China?” not “Hey, what’s up with Burkina Faso, Ethiopia, and Malawi?”

This chart also helps to generate questions that would not be so obvious in a more granular presentation, especially to those not accustomed to reading charts, as the human brain is pretty good at recognizing and comparing the slopes of straight lines. Looking at the chart for a few minutes quickly reveals the important moments in the economic histories of these diverse countries: what happened in Uganda in the 70s? China in the 80s? Most countries in the 2000s? And what the hell is wrong with Zimbabwe? You can also discern other interesting pieces of data, such as the fact that economic dynamo India actually only surpassed economic basketcase Pakistan in per-capita GDP relatively recently. Some countries have been particularly volatile, like Malawi, while Morocco was a good, steady performer. You really lose surprisingly little meaning compared to the conventional line chart, especially considering the number of distractions which are eliminated.

What Needs Work

That said, the chart did not come out as ‘clean’ as I had hoped. The colors, for example, were a last-minute addition. I had hoped to get by with shades of gray, but that clearly wasn’t going to work for this data set. Getting the lines and especially the scale right was also painstaking and laborious, and still didn’t yield fully satisfying results, as I had to stick China in like an afterthought — for proper scale China would need to be almost a full chart length higher. That was particularly unsatisfying because the rise of China is one of the main takeaways of the chart. Thankfully, I settled on an initial country set that didn’t include Equatorial Guinea, which now has a PPP per-capita GDP of $31,000! This shows that it’s important to distinguish ‘interesting’ outliers from the ones that really are just aberrations (Equatorial Guinea has less than 700,000 people and discovered one of Sub-Saharan Africa’s largest reserves of oil in 1996).

I think that in addition to the lack of good automated tools, the use of slopegraphs will be mainly limited by the fact that it’s really only suited for certain types of data. To make full use of the slopegraph you need a dataset which is not too ‘extreme’ and also does not have too much overlap. This data was sub-optimal in both respects: many of the poor countries are clumped up in a confusing tangle at the bottom, while the successful countries shoot right off the top of the chart. I suspect a lot of the interesting data in development particularly contains these sorts of extremes.

This chart might have been better off with only two periods, 1960 and Today. This would certainly simplify the overall message, but would remove some of the interesting nuances. I even think the current version has a bit of ‘narrative tension’ as you follow each line across to see what happened to that country — particularly the countries which are grouped together at the beginning (the grouping also begs some important questions: why India and not Uganda? Why China, not Ethiopia?).

Future Prospects?

The extreme simplicity of the slopegraph gives the creator lots of room to add additional information without overloading the reader. For example, you could vary the color of the lines based on some third piece of information, like whether or not the country had experienced a civil war or structural adjustment program during that period. The slopegraph could help to make correlations much more intuitive to readers who are not accustomed to interpreting statistical data (untrained readers have a lot of problems interpreting scatter plots, for example). Slopegraphs are also great when the rank order is really important. For per-capita GDP, I don’t think most development economists care as much about the relative ranks of countries as they do why some countries grow and others don’t. But for data like the ranks of export goods*, and probably lots of other kinds of ‘power law’ data, such as city sizes, this could be a useful presentation option. Hopefully some tools will be developed to make creating them easier for non-Illustrator experts.

* I might actually try improving on the slopegraph-style charts used in this paper the next time I have a free day. Stay tuned!

Bent Flyvbjerg, 2006

So, I’ve signed up to yet another thing where I can write stupid stuff that I later regret (Twitter)! Follow me, and I promise I will try not to be boring!

(And yes, this officially makes me an Old; I had back pain yesterday so I figured, why not?)

Well whaddaya know, Bill Easterly be frontin’ again, sayin’ how all development practitioners secretly love/want to be dictators. Since I know all of you out in Aid Blogger Land do actually <3 unelected technocrats, why not buy some of these snazzy bumper stickers for your Land Cruiser? All proceeds will go towards new elevators for the beleaguered bureaucrats of Kabul, who are so knackered by the time they get to their third floor offices that they can’t even focus on running the country properly. (Yeah so I didn’t actually post these in my “store,” it’s all a joke see!)