Arnold Kling  

Income Mobility

PRINT
Wisdom from Ben Casnocha... Cato Policy Analysis...

Brink Lindsey writes,


OK, so how do experts measure intergenerational income mobility anyway? Using survey data that tracks the income of the same individuals and families over time, researchers compare the income of parents at some specific age or age range to incomes of children at the same or other ages. As I understand it, perfect correlation between parents’ and children’s incomes is measured as an “elasticity” of 1, whereas a lack of any correlation would correspond to an elasticity of 0.

So, if everybody's income is 50 percent higher than their parents', that shows up as zero income mobility. On the other hand, if average income stays constant, but a lot of low-earning parents have high-earning children, and vice-versa, that shows up as high income mobility.

The moral of the story to me is, "Be careful what you measure." My favorite issue is the treatment of immigrants, the life cycle, and changes in family structure. Do you measure family income, so that John's income is lower than his parents' because his parents stayed married and he is single? Or because you measure John's income at age 25 and his parents' income when they were 35? Do you measure John's income quintile relative to the incomes of other people born in the U.S. when John was born, or do you measure his income quintile relative to the incomes of the current U.S. population?

UPDATE: Here is the report that is getting all the play. It is not a longitudinal study at all. Instead,


We look at four generations of men born during different periods between 1925 and 1974, and focus on their individual incomes when they were in their thirties — thereby holding constant the point in their careers when measuring their economic status.

...Those in their thirties in 2004 had a median income of about $35,000 a year. Men in their fathers’ cohort, those who are now in their sixties, had a median income of about $40,000 when they were the same age in 1974


One concern is that the men in their thirties in 2004 are not necessarily descendants of men in their thirties in 1974. One factor that intervened is immigration.

Also, I would not be so cavalier to assert that income in one's thirties is a good proxy for lifetime income. Today, with so many occupations requiring post-graduate education, income in one's thirties may be lower relative to lifetime income than it was in the past.

On the other hand, 1974 was a bad recession year, and 2004 was not. I would think that would favor those in their thirties in 2004.

All in all, this is an example of how not to produce a convincing analysis of income mobility. What I want to see is research that (a) clearly distinguishes changes that occur within families from changes in the composition of the population and (b) offers a credible proxy for lifetime income.


Comments and Sharing


CATEGORIES: Income Distribution



COMMENTS (7 to date)
Tim Lundeen writes:

The other factor in this regression to the mean.

One issue is regression to the mean for the numbers themselves. There is a good article on this in Wikipedia at http://en.wikipedia.org/wiki/Regression_to_the_mean:

"Another way to understand regression towards the mean is the phrase "nowhere to go but up/down": If you look at any exceptional group, some related group will likely be less exceptional due to the effect of regression towards the mean. For example, if you looked at the families with below-average incomes today, and compared them to how they were doing 5 years ago, you would see that they were likely doing better earlier and conclude that things are getting worse, Also if you looked at the families with the below-average incomes 5 years ago, you would likely find that they are doing better now, and conclude that things are getting better. This seeming contradiction is an artifact of the regression towards the mean. Since below average families can only remain in the same below average group or change to the above average group, the effect of that change will be to pull the statistic towards the mean. Using a non-representative sample biases the results away from the mean."

But another factor in comparing families is regression towards the mean in abilities. Children of parents with much higher than average capability tend to be less capable, and children of parents with much less than average capability tend to be more capable. So you would *expect* that children would move towards average in terms of their abilities, and thus that children would tend to move towards average in their income.

spencer writes:

You are suppose to be a PhD economist at a major state university. Are you not at all aware of the complete revolution in economic research on the topic of income mobility that occurred over the last couple of decades.

Your explanation of how modern economist calculate intergenerational income mobility is completely wrong. They do not compare the income of someone versus their parents income at some point in time.
If anyone should know this it is you because you are always talking about how this is a poor measure.

They measure total lifetime income of a generation
against the total life time income of the parents generation.

Why don't you google Bhashkar Muazember and actually find out what is going on in the latest research in economic research on
intergenerational income.

In particular you should look at his paper at that great bastion of liberal economics known as the Chicago Federal Reserve. here is the link:
http://www.chicagofed.org/public...s/ wp2005_12.pdf

Arnold Kling writes:

Spencer,
I found a paper by Mazumder and Daniel Aaronson. It does not really measure lifetime income. It measures income at the time of a decennial census.

I don't mind if you disagree with me, and I appreciate references to research. However, I don't think much of your rhetorical style.

John Thacker writes:

They measure total lifetime income of a generation against the total life time income of the parents generation.

Which would mean that there is very little good data for recent generations-- especially if, as is true, people are spending on average more years in school and delaying their most productive earning years as a result.

Longitudinal studies are great for knowing things, but they take a long time to finish and thus aren't that useful for policy recommendations. By the time you know how one generation compared to their parents, it's almost too late to do anything-- and the situation might have change for *their* children anyway.

I certainly agree that lifetime earnings are an excellent "true" measure of inequality that eliminate certain potential biases, but they're very hard to measure. The research of Bhashkar Mazumder (note spelling) is mostly, from what I'm familiar with, concerned with measuring sibling differences, along with some research that uses the measures that Professor Kling has issues with. (See here for example.)

Note that that later paper uses 1984 data to produce a paper in 2001; longitudinal studies take a long time.

Matthew c writes:

Today, with so many occupations requiring post-graduate education

I think this deserves a huge amount of attention. Think about the collossal misallocation of resources this entails, if (as Bryan and many others suggest) education is mostly about signalling. Peacock's feathers, anyone?

Floccina writes:

'Which would mean that there is very little good data for recent generations-- especially if, as is true, people are spending on average more years in school and delaying their most productive earning years as a result.'

Add to that the fact the delay in income for college can be considered an economic good in it self (look at Richard Vedder's work on higher education). I for one enjoyed college. It is not so important how much we make but how much we consume.

Bill Conerly writes:

This study does NOT include non-cash benefits, such as the value of employer-provided health insurance. (See the report's footnote 15.) Thus it slants the conclusion against modern generations.

Also note that it's all pre-tax income. The social security and medicare tax hikes would slant the study's results in favor or modern generations. I'm not sure how personal income tax has changed in this time period for hte incomes being measured.

Comments for this entry have been closed
Return to top