Democracy and development since 1950 in one chart

My previous post, in one chart:

(Click here for larger size).

Bubbles to the right are more democratic countries, measured using the Unified Democracy Scores; bubbles to the left are more autocratic countries. Note how the left side of the graph has much more volatility, even within single countries, as expected; the bubbles move up and down like crazy. By contrast, on the right side of the graph everything oscillates more sedately. (Try following Lebanon, for example; you can also zoom in to particular regions of the graph). For another interesting view, change the color to "regime type" and set the Y axis to plot GDP per capita (rather than GDP per capita growth). (Change the scale to "log," too, for a clearer view).

(Data on economic growth from the Penn World Table.)

A Very Short Quantitative History of Political Regimes, Part 1.75: Democracy and Development

(Statistician General’s warning: no significance tests or confidence intervals were harmed in the writing of this post. Appropriate model assumptions not guaranteed. Do not use without consulting a trained and licensed statistician.)

I haven’t done one of these posts on the history of political regimes in a while, but I am preparing something for a class on the topic of the relationship between political regimes and economic development, and figured it’s a nice addition to the series. Besides, it is always a good time to put up pretty graphs on this blog.

What is the relationship between political regimes and economic development? The basics of this relationship in the post-WWII era seem pretty well understood: basically, the richer the country, the more likely it is to be “democratic” (in the sense I’ve discussed here and here, where democracy is conceived as a system of normatively regulated competition for control of states including the usual paraphernalia of elections, freedoms of speech and assembly, etc.), though the reasons for why this is the case remain disputed, and there are obvious and significant exceptions to this pattern. Conversely, the academic literature suggests that democratic regimes have a slight and indirect long-term development advantage, though the evidence for this claim is much more controversial, and there is no consensus on how this particular advantage operates, if it exists at all (for a meta-analysis of studies on this topic, see here).

Here I want to look at how the relationship between development and democracy has changed over the past 60 years. For our purposes in this post I am going to use the Unified Democracy Scores developed by Pemstein, Meserve, and Melton. These scores basically aggregate the information contained in nine other democracy scores – Polity IV, DD, Freedom House, and six other lesser known indexes – in such a way as to indicate the “uncertainty” associated with particular country ratings. The score for a given country ranges between around -2 and around 2, with higher numbers being “more democratic.” A score of “zero” can be interpreted as a cut-off between more or less “open” regimes and more or less “closed” regimes, though we could also use three breaks, with regimes at one end of the distribution being clearly “dictatorships,” regimes at the other end being clearly “democratic,” and regimes in the middle being “hybrid” regimes of various kinds – competitive authoritarian, tutelary democracies, democratizing monarchies, etc.  As an example, here are the scores for the year 2008, split into three quantiles to indicate 

these broad regime categories:

Unified Democracy Scores, 2008

(The length of each horizontal line is the 95% confidence interval for the score; in general, small democracies at the top of the scale have more uncertain ratings, while dictatorships at the bottom of the scale have narrower confidence intervals, indicating less disagreement about the classification of dictatorships than about the determination of the level of democracy beyond a certain point). A three way split works quite well: the countries in red are the more obvious dictatorships, while the countries in blue are more clearly democratic, and the countries in green have the sorts of problematic hybrid regimes that are difficult to classify with certainty as either wholly democratic or wholly undemocratic.

Using those breaks, here is what the distribution of regimes looks like since the 1950s:

Distribution of regime types since 1950

Fully closed regimes had their peak in the late 70s and early 80s, but the end of the Cold War pushed many of these to at least hold elections, as we saw in this post. But how has this distribution evolved in terms of income per capita? Consider this plot:

Income per capita per year and regime type

(GDP data from the Penn World Table). Each circle shows the median income of the particular regime type for a given year, where the size of the circle is proportional to the number of regimes that fall into that category. The lines show a lowess fit for the overall trend growth of income for each particular regime type. I interpret the story the graph tells as follows.

The median income of democratic regimes has been higher than the median income of both hybrid and fully authoritarian regimes since at least the 1950s, and the gap has in general widened, not narrowed, even as the number of democratic countries has increased. (From this graph we cannot tell, however, whether the gap has widened because democratic countries have grown faster, or because non-democratic countries that grew fast turned into democracies; from the graphs below, we may infer that it was a mixture of both). The gap was highest during “peak authoritarianism” in the late 1970s and early 1980s, when most poor and newly independent countries were either hybrid regimes or dictatorships, but it stopped growing after the end of the cold war, when a number of relatively poor countries became democratic.

Interestingly, whereas before the end of the cold war the median hybrid regime was also richer than the median dictatorship, this pattern is reversed after the end of the cold war. Full authoritarianism proved almost impossible to sustain in poor countries without the patronage of the major powers or natural resource wealth; for the most part only relatively rich dictatorships, or dictatorships that retained a special relationship with a major non-democratic country, survived, whereas most poor countries today have some sort of hybrid regime. Thus in 2008, the list of full dictatorships included Saudi Arabia, North Korea, Myanmar (Burma), Qatar, Libya, Swaziland, Uzbekistan, Turkmenistan, Laos, China, Eritrea, Cuba, UAE, Oman, Equatorial Guinea, Sudan, Iran, and Syria. (Not all of these had GDP data in the Penn World Table – North Korea, for example. Note also that some of these regimes have large levels of uncertainty associated with them – the list could very well have included Brunei, for example, instead of Iran; but the basic point does not appear to change if we switch some of these around. I haven’t really checked, since this is only a blog post and checking requires some actual programming, but one could check by drawing new samples of democracy scores from within the distribution Pemstein, Meserve, and Melton generate. That’s the fun thing about the UD scores. It is also worth noting that some of these regimes appear to have survived the end of the cold war only because they had unifying enemies – Eritrea, Cuba, and Iran come to mind; fear can be just as stabilizing as wealth.)

What about growth? Is any particular regime type consistently associated with economic growth? Here’s what it looks like when we plot the relative growth performance of different regimes in every year since 1950:

Economic growth by regime type per year

(Click for a larger version). Each circle represents the median growth rate for the regime type indicated by the color; the size of the circle is proportional to the number of regimes in that category for that year. The lines show a lowess fit of the trend growth rate of each regime type; the shaded areas represent automatically generated 95% confidence intervals. (Many caveats apply. See warning above). Any year where the blue circle is higher than the red or green circles is a year where the median democracy did better than the median hybrid regime or the median dictatorship.  

As indicated by the width of the confidence intervals around the red and green lines, dictatorships and hybrid regimes have had more variability in economic performance than democracies since at least the 1950s: more growth “miracles” and “disasters,” often in the very same country. (See this paper by William Easterly for the actual scholarly version of the argument). To the extent that we can ignore these confidence intervals and focus only on the trend performance, democracies have not always done better than these other regimes. In the early post-war era it seems that dictatorships did better (though most did about as well as democracies), but then decolonization came along and the growth performance of dictatorships basically cratered. Indeed, the 80s, when the so-called “third wave” of democratization began, was also (not coincidentally perhaps?) the time when the “growth gap” between democracies and hybrid and dictatorial regimes was at its widest. Ominously, the last decade has seen a reversal of this pattern, which explains much of the (not very well thought out) commentary about the rise of the “Chinese model.” (Democracies, in particular, seem to have been much more strongly affected by the financial crisis that started in 2008 than either dictatorships or hybrid regimes, though all regimes appear to have been affected to some extent). Yet we should not make too much of this; even in the last decade, democracies did basically the same as the other regimes, judging by the overlap in confidence intervals – their responses seem not to have been too obviously wrong relative to the responses of authoritarian  countries (many of which benefitted from high oil prices). And we should remember that the median per capita GDP of democracies is already much higher than that of dictatorships or hybrid regimes; if we ought to be worried about anything, it is about the effects of bad economic performance on hybrid regimes, which could potentially lead to a reversion to more authoritarian forms of government:

Economic growth by regime type per year

I have purposely refrained from making any big claims about a general relationship between regime type and economic performance. From the evidence above, it seems that there has not been a great deal of difference between different regimes, except in the 80s. But instead of throwing away information by breaking the democracy scores into categories, we could try to look at what the relationship between the raw scores and growth rates looks like in general. Here’s my idea (please tell me if this doesn’t work; there are probably a million problems with it I haven't thought about). For every year, we estimate a simple linear model of the form growth = a*democracy_score + b*log(gdp_per_capita) + c + error. We then plot the coefficient a for each year; when this coefficient is positive, the more democratic the country, the better its performance for that year, adjusting for its existing level of economic development, and when the coefficient is negative, the opposite is true, i.e., the less democratic the country the better its growth performance for that year (many caveats apply). When we do this, we find that more democratic countries had a clear growth advantage only between the mid-1970s and the mid-1990s, and that advantage seems to have been lost in the 1990s, even reversed:

Growth advantage of democracy per year

(An even simpler model, not controlling for the existing level of economic development, gives a broader democracy advantage, extending back to the 1960s, but not a radically different picture.) It is worth noting that these relationships are different in different regions: the “democracy advantage” calculated using this method has been negative in East Asia, South Asia, and Western Europe for the period 1950-2008, and zero or positive just about everywhere else:

Growth advantage of democracy per region

Interestingly, Polynesia, East Africa, and Central Asia have had some of the largest “democracy advantages”: the more democratic the regimes have been there, the better their growth performance. Looking at entire continents, the broadest democracy advantage by far is in Africa (and it is positive in all).(Many caveats apply: for example, there isn’t a lot of variability in the democracy score in some of these regions, like in Australia and New Zealand, and the calculated coefficients aren't always significant).

Anyway, more could be said, but it’s late and I need to move on. Any thoughts on how to further this sort of analysis? All code needed to replicate these plots is available here; you will need to download the Unified Democracy Scores and the Penn World Table separately. You also need this file of country and region codes.

[Update, 18 May: fixed a small mistake in the last two plots. Results don't change, though the Polynesian "democracy advantage" is now more pronounced]

[Update 2, shortly after: I take it back. It was fine the first time, though it doesn't make much difference.]

Educational software I wish existed: StackExchange for teaching in the Humanities and Social Sciences

Here’s an idea I’ve been mulling over for the past week. (Mulling over enough to have spent many hours in an abortive attempt to create a prototype. But even simple web application programming seems to be a nontrivial problem that requires more time investment than I can muster right now). I have a large bank of questions for some of the courses I teach. These are mostly essay questions (short and long) of varying degrees of complexity, that ask students to make arguments, provide evidence, point to real-world examples, etc. I usually select the questions for final exams or other forms of assessment from this pool; a student who could answer all of them well would have essentially mastered the content of the course.

I would like to have a web application that draws on this pool of questions to do this:

1. In “quiz mode” a student would either select a question or be served a random question from the pool, and then he/she would answer it. They could then move on to a different question, for as long as they wished. (Perhaps the questions could be served in such a way that students can vote for the questions they most want answered, as in Google Moderator). The software would also allow the students to rate both the quality of their own answers (how good they think their answer is) and/or their level of confidence in their answer (how certain they are that they have a good answer), as well as the level of difficulty of the question. Their answers and ratings would go into a database; as they accumulate, the instructor could see which questions are rated as “hard” by students, or display characteristic problems, and focus teaching efforts there.
2. In “rating mode” students would either select a question or be served a random question from the pool, which they would view along with any (anonymized) answers from themselves, other students in the course, or even the instructor. They could then vote on which answer is best (if there is more than one answer) or rate the quality of existing answers. Perhaps they could also comment or edit existing answers if they want to add something to them, or feel a correction is in order. As these ratings accumulate, students would get a better sense of what counts as a good or a bad answer (assuming the “wisdom of crowds” works its magic; the courses I have in mind for this sort of application have around 100 students, which seems like it would be enough).
3. In “asking mode,” the application would allow students to submit questions, which then would go into the pool and could be answered by other students in the class. (Administrators could edit the questions for clarity or reject questions that are not sufficiently related to the topic of the course). 
4. The application could even have something like the reputation management features of the StackExchange family of sites. Students who answer questions would gain reputation “points,” so long as their answers are rated relatively highly (fewer points for unrated or lower-rated answers); asking questions or rating answers would also get them some reputation points, though fewer. (For purely illustrative purposes, imagine that asking a question nets you 2 reputation points, so long as it is not rejected by the instructor, rating an answer nets you 1 point, and answering a question nets you between 5 and 10 points, depending on how highly rated your answers are). Perhaps these reputation points could be translated into actual grade points at the end of the term in accordance with some appropriate formula, though that would depend on the design of the course.

As I imagine it, an application like that would offer students extensive practice in writing, especially if combined with say, a requirement that they answer at least one question every week (in fact, this system could displace one of the traditional two essays we ask students to complete in many courses). It would also help them practice the entire material covered in the course: since the questions for the final exam would be drawn from the pool (or be very similar to some of the questions there) students who use the tool would be essentially studying for the final every time they use it. (“Quizzing” yourself is one of the most useful study techniques available, and the system would be designed so that you would get relatively quick, and eventually accurate, feedback on your answers, without the instructor having to grade hundreds of essays). And I (the instructor) would in turn get feedback on how well they understand the material, as well as on what aspects of the course they are having difficulty with. 

What do people think? What problems would you foresee emerging with a system like this?

As far as I can tell, nothing quite like that exists, though in some ways this would be like a private version of StackExchange, seeded with a pool of questions on some specific course topic and open only to people taking the course. Google Moderator has some useful features, and I’ve been thinking about using it as a study tool for students in my course this term, but it would not be fully integrated with the rest of the assessment in the way I would want. Or is there something out there that I’m missing? How difficult would it be to develop the system I've described above?