POLS209 Democracy Dataset: Visualizing the Results Part II (Poverty and Democracy)

(Updated with latest data - 2nd update - 3d update)

(This is the second of two posts.)

(This is a bit difficult. If you are having trouble, PLEASE LET ME KNOW!!!!! I am happy to help).

In the post below, I described some of the general characteristics of the democracy data gathered by this class (it's great!), and created some graphs showing how the level of democracy has changed in different countries over time, according to this class. In this post, I will provide some analysis - as simple as possible, with only the minimum of necessary statistics - trying to figure out whether more democratic regimes are “good for the poor” - in the sense, at least, of reducing poverty more than non-democratic regimes. We will at best only be able to establish correlation not proof; but proof is basically impossible in the social sciences. If everything goes well, we should have (at the end of this post) some evidence (not proof! just a little bit of support!) for either the proposition that democracy is better for the poor than non-democracy, or the proposition that it is worse. But I'm not going to get ahead of myself.

Poverty is a pretty multidimensional concept, as difficult to measure as democracy. Moreover, to do a proper analysis we would need time-varying information of various dimensions on poverty over long periods for many countries, which may simply not exist. However, I think we can get a reasonable picture by using the following measures:

1. The level of per capita income in a country, per year (and its growth rate), taken from Angus Maddison's historical dataset of income statistics. This is your standard measure of how rich the average person is in a given country (and the growth rate tells us how fast the average person is getting richer), though of course it does not tell us much about how well distributed that income is. So we might also wish to look at…
2. The net Gini index of inequality (net of taxes and transfers), taken from the Standardized Worldwide Income Inequality Database. This is useful too, but it doesn't necessarily tell us much about how the poorest are doing. For that, we might look at…
3. The $2 a day international poverty line, compiled by the World Bank, where available. This is a standard measure of extreme “income” poverty. Nevertheless, it is missing for a lot of countries. So we might wish to look at more indirect measures of how the poor are doing, like…
4. The under 5 child mortality rate, compiled by the World Bank. We can also look (following Blaydes and Kayser) at a more specific measure of how the poorest in a society are doing, namely…
5. Total food availability, in kilocalories, from the Food and Agriculture Organization of the UN. This is available for most countries in most years, so it has good coverage, but it's probably only useful for the poorest countries, since poverty in “developed” countries is unlikely to be a matter of serious food deficits.

None of these measures are perfect (and you do not need to focus on all of them in your essay - choose the ones that you think are best. Can you suggest others?). Each gives only a partial picture. For example, you may have high average incomes, and high income growth, but high inequality (so the poor don't benefit much from economic growth). Other measures have their own problems. But why might these measures provide some insight into poverty in the countries in our dataset?

The first thing we do is to collect all this data and merge it with our dataset. The merged dataset is here.
We'll restrict our analysis to after 1972, where we have the most data. How might we go about figuring out whether democracy is correlated with (improvements in) poverty? The simplest thing we could do is simply to rank-order the contries in terms of these measures over the period of observation, and try to establish whether democracies “do better” than non-democracies in this ranking. Yet this is problematic.

Consider, for example, what happens if we order the countries in terms of average GDP per capita (average income):

Most democracies (with some exceptions) come out at the top: “being democratic” is (imperfectly) correlated with “being a wealthy country, in terms of having a high GDP per capita” - which is what the graph is telling us. You can get a better sense of this distribution with this table:

##                  income
## avg_simple_status High income: nonOECD High income: OECD Low income
##    Not Democratic                    2                 1         18
##    Hybrid                            3                 1         14
##    Democratic                        2                15          8
##                  income
## avg_simple_status Lower middle income Upper middle income
##    Not Democratic                  11                   6
##    Hybrid                          19                  16
##    Democratic                      14                  14

##                  income
## avg_simple_status High income: nonOECD High income: OECD Low income
##    Not Democratic                 0.05              0.03       0.47
##    Hybrid                         0.06              0.02       0.26
##    Democratic                     0.04              0.28       0.15
##                  income
## avg_simple_status Lower middle income Upper middle income
##    Not Democratic                0.29                0.16
##    Hybrid                        0.36                0.30
##    Democratic                    0.26                0.26

In other words, 28% of democratic periods in the data are high income OECD countries, while 47% of non-democratic periods are low income (look across the rows). Conversely,

##                  income
## avg_simple_status High income: nonOECD High income: OECD Low income
##    Not Democratic                 0.29              0.06       0.45
##    Hybrid                         0.43              0.06       0.35
##    Democratic                     0.29              0.88       0.20
##                  income
## avg_simple_status Lower middle income Upper middle income
##    Not Democratic                0.25                0.17
##    Hybrid                        0.43                0.44
##    Democratic                    0.32                0.39

88% of high income OECD countries in your data are democratic (look at the columns now).

But if instead we look at the rate of increase of income (“growth” rather than “wealth”), we see a different picture. The countries that most “improved” their income per capita are at the top, while the countries that did worse are at the bottom (in the period in which they have been classified):

##                  growth_cat
## avg_simple_status Bottom 20% Middle 60% Top 20%
##    Not Democratic         14         19       7
##    Hybrid                 12         29      13
##    Democratic              8         37      10

##                  growth_cat
## avg_simple_status Bottom 20% Middle 60% Top 20%
##    Not Democratic       0.35       0.48    0.18
##    Hybrid               0.22       0.54    0.24
##    Democratic           0.15       0.67    0.18

As we can see, Angola, Sudan, Equatorial Guinea, and Myanmar have the highest trend growth rates over the periods in question here; Libya, North Korea, Nicaragua, and Djibouti bring in the rear. Democracies seem to be scattered all over the place (though few democracies are at the bottom of the graph); the countries that are already wealthiest by the time we begin observing them are not necessarily those that have the highest growth rates, though we do observe that “not democratic” periods seem to have a disproportionate chance of ending in the “bottom 20%” of growth rates. (That is, 32% of non-democratic periods appear in the bottom 20% of growth rates, suggesting that non-democratic regimes are more likely to have had worse rates of growth than democratic regimes).

A similar thing happens if we look at inequality:

The least unequal countries tend to be consolidated democracies or former communist regimes (Czechoslovakia, Sweden, etc.), and though democracies are all over the place, non-democratic periods are more likely to becorrelated with high-inequality while democratic periods are more likely to be correlated with low inequality:

##                  inequality_cat
## avg_simple_status Low Medium High
##    Not Democratic  10     18    8
##    Hybrid          14     33   17
##    Democratic      23     37    7

##                  inequality_cat
## avg_simple_status  Low Medium High
##    Not Democratic 0.28   0.50 0.22
##    Hybrid         0.22   0.52 0.27
##    Democratic     0.34   0.55 0.10

But what if we looked not at the current level of inequality but at how fast countries improved their inequality (rates of decrease/increase in the gini coefficient) over the period in which you classified them? Here the picture looks slightly different:

##                  inequality_growth_cat
## avg_simple_status Top 20% improver Middle 60% Bottom 20% improver
##    Not Democratic                7         10                  10
##    Hybrid                       14         24                  10
##    Democratic                   11         29                  13

##                  inequality_growth_cat
## avg_simple_status Top 20% improver Middle 60% Bottom 20% improver
##    Not Democratic             0.26       0.37                0.37
##    Hybrid                     0.29       0.50                0.21
##    Democratic                 0.21       0.55                0.25

Democracies don't appear special here (in part because it's hard to improve when inequality is quite low to begin with), though we still see many non-democratic regimes overrepresented among the “low improvement” regimes.

What about simple income poverty, at the $2/day level that the World Bank gathers? This data is only gathered for very poor countries, and not always very regularly, so we lose a lot of observations, and it's highly correlated with the average GDP per capita, but it's worth looking at separately:

##                  pov_cat
## avg_simple_status Low Middle High
##    Not Democratic   4      3    8
##    Hybrid          18     15    7
##    Democratic      25      6    7

##                  pov_cat
## avg_simple_status  Low Middle High
##    Not Democratic 0.27   0.20 0.53
##    Hybrid         0.45   0.38 0.18
##    Democratic     0.66   0.16 0.18

Democratic countries are overrepresented among low poverty countries, and non-democratic ones among the high poverty ones; but this is partly an artifact that democratic countries are wealthier than non-democratic countries, and this data is only gathered for fairly poor countries. (I am not adding a graph with rates of improvement; there are too few datapoints for this to be meaningful). Finally, let's look at the measure of infant mortality:

##                  mort_cat
## avg_simple_status Low Middle High
##    Not Democratic  25     21    8
##    Hybrid          47     26    3
##    Democratic      54     10    0

##                  mort_cat
## avg_simple_status  Low Middle High
##    Not Democratic 0.46   0.39 0.15
##    Hybrid         0.62   0.34 0.04
##    Democratic     0.84   0.16 0.00

##                  mort_growth_cat
## avg_simple_status Top 10% Middle 80% Bottom 10%
##    Not Democratic       5         33          2
##    Hybrid              10         45          3
##    Democratic           7         43          8

##                  mort_growth_cat
## avg_simple_status Top 10% Middle 80% Bottom 10%
##    Not Democratic    0.12       0.82       0.05
##    Hybrid            0.17       0.78       0.05
##    Democratic        0.12       0.74       0.14

This is most meaningful for very poor countries, but it's worth noting that mortality rates have been declining almost everywhere, and there is no obvious pattern to the reductions: in this dataset, reductions in infant mortality do not appear to be faster in democracies, or slower in non-democracies, even after we take out countries with high incomes:

##                  mort_cat
## avg_simple_status Low Middle High
##    Not Democratic  22     20    8
##    Hybrid          42     23    3
##    Democratic      40      9    0

##                  mort_cat
## avg_simple_status  Low Middle High
##    Not Democratic 0.44   0.40 0.16
##    Hybrid         0.62   0.34 0.04
##    Democratic     0.82   0.18 0.00

##                  mort_growth_cat
## avg_simple_status Top 10% Middle 80% Bottom 10%
##    Not Democratic       4         30          3
##    Hybrid               8         39          5
##    Democratic           6         31          6

##                  mort_growth_cat
## avg_simple_status Top 10% Middle 80% Bottom 10%
##    Not Democratic    0.11       0.81       0.08
##    Hybrid            0.15       0.75       0.10
##    Democratic        0.14       0.72       0.14

A similar picture appears if we look at increases in food availability (this just shows improvements in food availability for countries with GDP per capita of less than $10,000 per year, most rapid increases first):

##                  food_growth_cat
## avg_simple_status Bottom 10% Middle 80% Top 10%
##    Not Democratic          3         27       6
##    Hybrid                  5         40       6
##    Democratic              6         27      10

##                  food_growth_cat
## avg_simple_status Bottom 10% Middle 80% Top 10%
##    Not Democratic       0.08       0.75    0.17
##    Hybrid               0.10       0.78    0.12
##    Democratic           0.14       0.63    0.23

Here we see a slight overrepresentation of democratic regimes among poor countries where food availability increased fastest, but not much, and democracies are all over the place. In other words: the picture is all muddled!

Our problem is that we are trying to figure out a partial effect: the effect of democracy on poverty, net of other factors like the country's average income, etc. So the picture is going to be very unclear: many good things are easier for wealthier countries to achieve, regardless of regime type, and at the same time many fast improvements are possible for poorer countries, regardless of regime type. One way of getting around this is to look at specific cases - e.g., look at why Rwanda seems to have been so succesful at reducing infant mortality despite a non-democratic regime type (what is it about its political regime that let it achieve that?).

But we can also use a statistical analysis that would let us exclude those other factors - “control” for them, in statistical jargon - so we can see if democracy has some (small) effect, on top of a country's wealth and other country-specific factors (like culture, etc.). There are many sophisticated techniques one could use (most of which you cannot fully understand unless you are a working statistician, and most of which have pitfalls and problems of their own), but we can do a simple check using a basic version of the analysis in the Blaydes and Kayser piece. (For those of you keeping track, I'm going to run an extremely simple “fixed effects” model that basically tries to figure out if democracy at time t-1 shows some correlation with the value of our five measures of poverty at time t [four, really - there is too little data to use the poverty measure properly in this way])

I'm not going to bore you with the numerical details, but after controlling for the level of development (GDP per capita) and the unmeasured indiosyncratic characteristics of each country (“fixed effects”), I find that an increase in democracy seems to have some small positive independent impact on growth, some small positive impact on inequality reductions (so does being a hybrid regime), some small positive impact on mortality reductions (similarly, even being a hybrid regime helps), and some impact on food availability. (Do not put too much stock on this analysis, which has not been properly vetted).

How else should we approach this question?