Noisy data smeared over a long period of time, which are then aggregated into a few key values that are not statistically significant.
Tell us more about the test of significance.
They started with the Survey of Income and Program Participation (SIPP) which is a large scale sample of households which are surveyed every 4 months for periods of 2 to 4 years. http://www.census.gov/sipp/
The SIPP data has short-term income changes, but to measure income mobility it was not sufficient (their measure of mobilities are based on income changes for the same individuals 10 years apart, roughly age 47 vs 37. So what they extracted from the SIPP data was social security numbers, age, and state of residence for a nationally representative sample.
The social security numbers were used to index into the SSA MEF data (Social Security Administration Master Earnings File), which are annual earnings (earned income) records recorded by the SSA. Presumably all this was done in a way that maintained confidentiality so all the Pew researchers got were records like:
State, Age, Income (age 35 ... 39, 45 ... 49).
They ended up with a sample of 64,686 individuals born between 1943 and 1958 (boomers roughly) with average annual income for ages 35-39 and ages 45-49. So for those born in 1943 they had earned income from 1978 to 1982 and 1988 to 1992; and those born in 1958 earned income from 1993 to 1997 and 2003 to 2007. They used 5-year averages to reduce volatility due to short-term changes, and dropped individuals with any 0 income years. If I understand correctly, some of their income data is subject to the FICA cap, and other does not include self-employed income. So the income data is from 1978 to 2007. It is not clear what they used to ensure correct distribution of individuals over time (ie the census bureau has estimates of the number of individuals who are 35 in each year, does their sample distribution match that?).
Absolute mobility was measured using the difference of the log of the income for the two five-year periods (for simplicity we can think of these as the age 37 and age 47 incomes).
ln x is approximately x - 1, for values of x near 1 (eg ln 1.200 is 0.182). I'm not sure why the log is used rather than the percentage change (maybe Muon can explain). I think it reduces the effect of extreme changes, persons doubling income and persons who income goes down (inflation adjusted income was used).
Now take a look at the data:http://www.pewstates.org/uploadedFiles/PCS_Assets/2012/MobilityofStates_Data.pdf.pdf
Absolute mobility is essentially average income growth between age 37 and 47.
The national level is 17%. Oregon grew at 20%, but it doesn't get a green box, because it is not statistically significant at the 95% level (20% is not two standard deviations above 17%).
New York and Pennsylvania that also grew at 20%, and they get green boxes because of their larger population (and sample size). Rhode Island has the largest increase, but also is not statistically significant above the 17% national level.
Given the overall sample size, a state sample is going to be about 150 per congressman. California will have a large sample size of several 1000; Delaware of perhaps 200, with a little more than 10 per year. Individual income change is probably pretty noisy, since it covers a wide range of income levels, and individual circumstance, layoffs, divorce, health issues, re-entry into the labor force by women after early child-rearing years. So a single number is not representative - and in the case of Rhode Island, we aren't even sure if 24% is greater than 17%.
Note that many of the states with higher income growth had lower population growth. (gross generalization ahead). People don't make interstate moves in their 30s and 40s, because they have families and mortgages. They move in their 20s, when they have fewer ties and don't have an established career. So in part, it might be the lack of jobs that causes younger people to move, but provide an opportunity for those who stay to advance.
Relative upward mobility is based on percentile ranking of income. Someone who is in the lower half of incomes at 37, who advances 10 percentile by 47 is considered to be upwardly mobile. The state measure is the percentage of those in the lower half at the start, who improve by 10 percentile.
This may be largely equivalent to absolute mobility (the two have a correlation of 0.76).
At a given percentile level at age 37, one could calculate the increased income to be at that percentage level plus 10 at age 47. From 1967 to 2010, real income for the middle quintile of households was 0.45% annually. http://www.businessinsider.com/us-household-incomes-a-42-year-perspective-2011-3
Much of this increase was before the period of the mobility study, so it would take perhaps 3% or so increase over the 10 years between age 37 and 47 to maintain their percentile rank. Middle-income percentile ranks are likely to be closer together. Going from the 45th percentile to 55th percentile requires a smaller increase in real income than going from the 5th percentile to 15th percentile. At the lowest levels, hours worked may have more of an impact than increase in wages. One can double their earnings by going from 1000 hours per year, to 2000 hours per year.
If there is a 20% income difference between the 45th percentile and the 55th percentile, then the difference between an average gain of 18% and one of 20% could be huge in the numbers of persons advancing 10 percentiles.
I think it might be more useful to see the distribution
of age-47 income vs age-37 income (and the similar comparison for percentile rank). Reducing upward mobility to a single number - percentage of persons who were below the median at age 37 who improved by 10 percentile rank by 47, and then further reducing that to red box or green box or no box simply eliminates too much information.
If you look at the regional upward mobility, where persons are compared based on their percentile rank in a regional income distribution, there is much less difference. That is, persons improve their lot relative to their neighbors at about the same rate between the ages of 37 and 47. But the correlation between absolute mobility and regional upward mobility is weak (0.32).
The downward mobility index is based on the percentage of individuals who have a percentile rank above the median at age 37, who drop 10 or more percentile by 47. Here, less is better. With perhaps 4 or 5% inflation-adjusted income needed to hold one's place between 37 and 47, it may require dislocation, such as losing a job, health issues, family issues, to actually drop 10 percentile.
If I had to hazard a guess, the regional difference is probably due to the importance of manufacturing and service jobs in the southeast vs. white collar jobs in the northeast (suburby states like Maryland and Connecticut did particularly well). Manufacturing wages did not increase as fast as white-collar wages, and there are fewer opportunities for career advancement for those holding manufacturing jobs.