Replace GDP With Median Income

UPDATE: See the MELI FAQ for more up-to-date information about how and why we should reform the use of GDP.

Photo by Alex E. Proimos (CC BY-NC 2.0)

The reigning measure of economic progress and development is GDP per capita. Sure the United Nation’s Human Development Index (HDI) gets some use, but it only gets a tiny fraction of the love that GDP gets as you can see below.  One reason why people prefer GDP is that nobody can relate to the HDI.  We all know what income is and GDP is just a kind of measure of income, so people understand what it means if GDP per capita falls 17% from $35,000 to $30,000.  Ouch!  But if Argentina’s HDI of .811 falls 17%, nobody has the direct experience to know what that really means.  It turns out that that would be like going from Argentina’s HDI down to China’s level (.699).

Median income is a more useful statistic than either HDI or GDP, but it has been increasingly neglected in favor of the other two.  This is even more apparent if we add the term “GDP” to the above graph, because that familiar nickname is much more common than all the other terms combined.

One problem for median income is that it is hard to get a meaningful statistic for annual income unless one uses household income, but we care more about individuals than households, and it is hard to make the adjustment between the two (see below).  However, median hourly income in a given month would work well for macro purposes.  Europeans should love this statistic because it makes them look a lot better than the GDP statistic that Americans invented.  The US probably has lower median hourly earnings than much of Europe because Americans work a lot more hours per month and have higher inequality than Europeans.

Along with median hourly earnings, we should pay more attention to the total hours of work per month.  We measure annual hours worked, but it is a neglected statistic and it is not tracked monthly.  In some ways, total hours of work per month would be a better measure of recessions than GDP because even the crude-quality annual data that we have predicts recessions much better than GDP as you can see in my graph of hours worked per person.

Annual Hours Worked Per Capita

This is a pretty crude and inexact measure because of the poor-quality data I threw together to make it, but even so, it still predicts recessions better than many private companies that specialize in predicting recessions.

Hours worked per person helps demonstrate that people don’t “tighten their belts” during recessions and work harder when times get tough.  People work a lot less during recessions.  Recessions are defined as a decline in GDP, but the big problem during recessions is not the decline in GDP, but the decline in work.  A recession wastes the valuable labor of people who want to work.  Many are desperate for work.  It would also be interesting to see more information about seasonal variations in the hours of work too.  Research has shown that seasonal variations in GDP are typically bigger than the biggest recessions.  Fortunately, normal people don’t seem to notice this issue because seasonal variations are shorter than recessions, and well anticipated, so they are hardly problematic.  And anything that is impossible to avoid has no opportunity cost, so there is no problem to solve.  But it is interesting from a macro perspective.

Finally, in place of the HDI, we should use median lifetime income.  It would be calculated by estimating the median consumption for every age up to the median life expectancy and adding them all up.  I think we should call it median lifetime income rather than median lifetime consumption because people are more used to talking about income measures.  Median lifetime income has several advantages.  Unlike the HDI, median lifetime income is something that everyone would instantly understand.  It captures 2/3 of the variables in the HDI, lifespan and income.  And these two components are much more powerful than the third variable (education) which does not add much information because it is highly correlated with the other two anyhow.

Like the HDI, median lifetime income would show that economic development has been successful in Sub-Saharan Africa between 1960 and 1985.  Although per-capita GDP was stagnant, average lifespan increased almost 25% from about 40 years to about 50 years.  If the average person’s lifespan increases 25%, and annual incomes stay about the same, then median lifetime income rises about 25% despite the failure of GDP per capita to rise.   That is an economic development success.  At a conference presentation, someone in the audience responded to this point by suggesting that poor people in Africa are worse off if their suffering is prolonged 25%, but Africans aren’t committing suicide at higher rates than the rest of the world, so the poor people don’t seem to agree.  Of course, unhappy people could respond to their poverty by violently lashing out at others and Jared Diamond suggested that this was one reason for the genocide in Rwanda, but the genocide would have had a big effect on median lifetime income measures whereas suicide has been a relatively minor concern for all of the statistics mentioned here.

It might seem that median lifetime income could be calculated more easily by simply multiplying median individual income times median life expectancy. That is true in theory, but median individual income is easy for statisticians to fudge because most people live in households and share their income.  That makes it hard to determine how much of a fathers’ income should be apportioned to the wife and kids in his household.   That is a difficult decision for statisticians, but if we are adding up the median income for every age, then the decision becomes less important because a statisticians who splits the household income evenly between all members will raise the incomes of young people, but reduce the incomes of working age people by exactly the same amount.  So the total effect of the statistician’s decision on median lifetime income would be modest.

It might also seem that it would be a huge expense to calculate the median income for every age, but it is a lot cheaper than adding up the market values of all the final goods and services produced within a country.  That is a lot more difficult because everyone must be measured.  A median is much cheaper and easier to measure because nobody has to measure any of the people who are clearly above the median nor those who are clearly below it.  It is a much simpler statistical sample.

Speaking of statistical samples, why don’t economic statistics always have error estimates?  What is the error on a GDP measure? ±2% or ±5% or what?  If economics is going to be an empirical science, we should always include margins of error with our measures.  Why don’t we?  Are the margins of error so terribly embarrassing that nobody wants to try to estimate them?