Another Measure of Human Development for US States

All right, forgive me, but it seems that I'm not quite done with this whole human development index thing. I've come across another effort, in addition to the American Human Development Project's, to apply a measure similar to the United Nations' HDI, to measure the human development of the US states. First, here is the AHDP's map of their version of HDI by state:


The other attempt to measure HDI by state is by Jeremy R. Porter of Rice University and Christopher W. Purser of Mississippi State in this 2008 paper. And they've come up with some slightly different results from the AHDP. Here is a map of HDI by state that I made based on data found in their paper, "Measuring Relative Sub-National Human Development: An Application of the United Nations' Human Development Index in the US".



The scales are different, obviously, but you can still see the general ordering of the states by HDI. And it's interesting to see what sort of differences can result from slightly different methodologies. The broad patterns here are about the same: in both studies the least developed states are all in the South, with the most developed states clustering in the Northeast, the Upper Midwest, and parts of the West. But there are some real differences. California is a top-ten state in the AHDP, but is 18th in the Porter and Purser paper. Wyoming is in the second-lowest quintile in the AHDP, but it shoots to #8 in Porter and Purser. Virginia is somewhere in the teens in the AHDP, but #33 in Porter and Purser. And Georgia, which is close to the middle in the AHDP, drops to 44th in Porter and Purser (out of 48: P & P don't list Alaska and Hawaii).

There are a couple of general trends that could describe some of the differences between the two studies: P & P's methodology seems to slightly favor the interior West; and the least developed region in the AHDP is the Upland South plus the lower Mississippi Valley states, whereas in P & P it's the Deep South that comes out at the bottom. (In both cases, Mississippi is the least developed state.)

So what are the differences in methodology that have led to these different outcomes? Well... I don't know. But you can compare the differences for yourself. Kristen Lewis, a Co-Director of the AHDP, told me about their methodology in general terms. It uses measures of income, education, and health, just like the UN HDI, but adapted for the US context:

Income: using the US GDP per capita would assign everyone the same income, which is far from reality and obviously not helpful for making comparisons among groups. A state GDP would not make that much sense, as economies are increasingly regional, with people living and working and selling their goods and services across state borders all the time (think of the NY-NJ-CT area, or DC-MD-VA-DE). Using personal earnings tied income to actual people and allowed us to disaggregate by state, congressional district, racial/ethnic group, and even gender (this is why we went with median personal earnings rather than household earnings). Thus we use median personal earnings for full and part time workers aged 16 and up.

Education: the global HDI uses literacy and school enrollment as their proxies for knowledge. In countries where one in five or one in three adults cannot read and where huge numbers of school-aged children are not in school, these are excellent proxies. In the US, however, we need a more demanding standard than literacy - just being able to read doesn't in and of itself allow for a life of freedom, opportunity, choices. Of course, functional illiteracy is still a problem here - but that group is well captured in our "less than high school degree" category. Also, data on literacy is not collected in the same way or down to the same level as the educational attainment data we have from ACS. THus we use a combination of degree attainment for the population of adults over 25 – HS, BA, and graduate degree – and school enrollment for the population aged 3-24.

Health: we also use life expectancy. We calculated life expectancy from mortality data, and ours is actually the first published source of state and congressional district-level life expectancy. We calculated LIEX from county death data collected by the Centers for Disease Control and Prevention.

As for Porter and Purser, here is part of the description of methodology in their paper:

Data for this project were acquired from a number of sources. First, data on literacy was obtained from the National Institute for Literacy (NIFL)1. As proposed by Lind (1992) there are five different types of literacy (NIFL 1998). The data used in this study reflect those in each county that are in the lowest literacy group (level 1 literacy). Those in this literacy category (level 1) would have minimal literacy skills and would be relatively disadvantaged in relation to the average individual in the U.S (NIFL 1998. For the creation of the final scale the variable was reverse coded so that a high score was desirable. Second, the data on those within the county with a bachelors degree was obtained from the census bureau2. This was substituted for the percent enrollment due to the low variation in enrollment rates at the county level (extremely high percent of students enrolled in high school in the U.S. with low variation) and is considered to be a proxy for the measure. In relation to the health component, data pertaining to the average life-expectancy of each county was obtained from the National Center for Health Statistics (NCHS)3.

Finally, data pertaining to the per capita personal income at the county level were obtained from the Bureau of Economic Analysis (BEA)4 via the CA 1-3 table at the county level and was used as a proxy for the GDP of the county. The per capita personal income variable was as a proxy for the county level GDP based on ancillary analyses that showed it to be an adequate substitute, as evidenced by the fact that, at the state level, the Gross State Product (GSP) and the State Personal Income per capita (SPI) correlate at the .001 significance level with a coefficient of .998. Region and metropolitan status were obtained from the U.S. Census Bureau and Economic Research Services (ERS), respectively as a classifiers for spatial description.

Make of that what you will. Personally, I find it counter-intuitive that New Mexico would be more highly ranked than Arizona, and that Georgia - anchored by the major metropolitan region of Atlanta - would rank in the bottom five and below states like Arkansas and Kentucky, which is what the Porter and Purser paper proposes. But then (with apologies to David Foster Wallace) the sum of my expertise in this area could be inscribed with a magic marker on the rim of a shot glass.

By the way, Porter and Purser have a couple of maps of their own, including this one of relative development at the county level:



Florida appears to be dripping.