Posted by on March 29, 2016

We have all heard it.  Hillary Clinton repeated it in the February 11, 2016, Democratic Party debate:

I know that a lot of Americans are angry about the economy, and for good cause.  Americans haven’t had a raise in 15 years.

This is one of the leading memes of the presidential campaigns and current economic discussions.  But what if it wasn’t true?

When people make this statement, they are usually thinking about real median household income (RMHI), as reported by the Bureau of the Census.  A chart of this statistic is included in the above-linked article.  It shows a steady increase from $48,664 in 1984 to $57,843 in 1999.  From that point onward, however, things look bad: RMHI falls to $53,657 in 2014, a decline of 7%.  Not only have Americans not had a pay increase, but it looks like they had to take a cut.

I have probably read this statistic, or something closely related, about a hundred times.  I consider myself to be fairly mathematically and economically sophisticated, but I have to confess that until I read an American Enterprise Institute article entitled “Update: How changing household composition, household work hours, and retirement explain median household income,” I never really thought seriously about these numbers.

And this is why the combination of statistics with 30-second soundbites is so deadly to rational political and economic debate.

Let’s unpack this metric.  It’s “real” because we want to reflect the impact of inflation – check.  It’s “median” (and not “average” or “mean”) because we want to reflect the experience of a typical household, without the distortions that outliers can produce in an average – check.  But what about that “household” word?  If we are trying to get a feel for how the typical American family is doing, this only makes sense if the composition of a “household” hasn’t changed much – if we are consistently talking about Ozzie, Harriet and their two boys.

Put it this way and the problem becomes obvious.  The US is undergoing enormous demographic changes, particularly with the retirement of the post-war baby boom.  This means, among other things, that:

  • the number of households with two earners went from 45% in 1999 (the peak year for RMHI) to 40% in 2014;
  • the number of households with no earners went from 20% in 1999 to 24% in 2014; and
  • the share of the retired US population, which was a consistent 15% from 1990 to 2007, rose to 17% in 2014.

Put all of this together, and the average number of hours worked per household declined by 8.8% between 1999 and 2014, slightly outpacing the decline in RMHI during that period.

All of these things would depress RMHI but would say basically nothing about the change in the standard of living of the typical American family.  Yet this is the way that someone like Hillary Clinton wants to use this statistic.  And she wants to be elected so that she can implement policies – like an increase in the minimum wage – designed to deal with a problem that might be a statistical illusion but which will certainly create problems that will be very far from illusory.

The article cites two other statistics that point in the opposite direction.  Real earnings per hour worked, for example, increased by 7.5% between 1999 and 2014.  And real compensation per hour worked, which includes the impact of benefits such as the skyrocketing cost of health insurance, has increased by 13.5% over the same period.[1]  Neither of these statistics are perfect, but they at least tell us that we should treat the RMHI numbers with far more caution than they are usually accorded.

There is another group of statistics that gets bandied about with reckless abandon and little understanding.  These are the statistics about inequality.  I guarantee that, when the average person hears statistics about how the upper 1% or 10% is making X% of the total income or Y% of the total increase in income over period Z, the average person thinks that these statistics relate to the same people.  But, of course, by no means need this be the case.

The problem with these statistics is that they are cross-sectional snapshots.  They take the upper X% of the population at one time and calculate how much income goes to this group.  They take the same upper X% – not the same people, just the same percentage – at another time and repeat the calculation.  But the members of this group may, and almost certainly have, materially changed in the interim.  This makes it very difficult to infer anything about inequality or changes in inequality simply by comparing these two numbers.  In order to do this, you have to look at longitudinal studies that track specific individuals over time.  But these studies are more difficult to conduct and, more importantly, much more difficult to paint onto an Occupy Wall Street poster.

We have all heard about the 1%.  But how many have heard that, based on one longitudinal study that tracked individuals over 44 years, fully 12% of the population will find itself in this lofty group at one point in their careers?  Or that the corresponding figure for the top 10% is 56% and a whopping 73% for the top 20%?  These figures certainly sound much less ominous than the ones generally propagated by the media.  They also probably go a long way to explaining something that always mystifies the Left: why voters are not more in favor of “soak the rich” tax schemes.   Like Pogo, the voters may have “seen the enemy and it is us.”[2]

To make this point in a different way, consider the following thought experiment.  Imagine an economy consisting of five people, each ten-year’s apart in age.  Each person has a 50-year employment career, starting at age 20.  Each starts off making an amount that puts him in the lowest quintile of earners.  Each ten years, each person moves up a quintile and his earnings double.  At the age of 70, each person retires and drops out of the calculation, to be replaced by someone just starting his career at the age of 20, earning in the lowest quintile.  At each point in time, there is one 20-something earning one unit of income in the lowest quintile, one 30-something earning two units of income in the second quintile, one 40-something earning four units of income in the third quintile, etc.

This economy demonstrates perfect equality.  Over each worker’s life, each one earns exactly the same amount in exactly the same pattern.  However, a cross-sectional analysis would show, at any point in time, a very unequal result.  The upper quintile would capture 52% of total earnings.  The two upper quintiles would get 77% of total earnings.  The suffering souls in the lowest quintile, conversely, would only get 3% of the total pie.[3]

This is not meant to be an accurate picture of the US economy, although the phenomenon of rising incomes with work experience is certainly correct.  But, like with the RMHI statistics, it is easy to see how Brandolini’s Law[4] would certainly allow unscrupulous, or just plain dumb, politicians to misuse these statistics.  Not that we have any of those.

Finally, there is the statistic on the gender pay gap.  I have commented on this completely bogus figure before (in footnote 2 of this blog), but it is worth mentioning again because if Clinton becomes President, we are virtually certain to see legislation to solve this non-problem.

All of this, frankly, is very depressing.  Although all the examples I have chosen above come from the Left, I am pretty sure that distorting statistics is a bipartisan effort.  The biggest problem isn’t the abuse of mathematics and logic by one faction or the other, but rather the limitations of the political process and the attention spans of voters.  Unfortunately, this means that it isn’t just a problem of throwing the rascals out.

As Senator Patrick Moynihan once famously said: “Everyone is entitled to his own opinion, but not to his own facts.”  If we are unable to discuss intelligently the basic facts, then there is little hope that we will agree on anything else.

Dog Whistling with Caitlyn Jenner

I suspect that everyone has heard the expression “dog whistling” at this point.  This is a reference to an alleged tendency of Republican Party candidates to engage in coded communication with their racist, sexist, homophobic, etc., constituencies in ways that cannot be discerned by the uninitiated. (Although it is perfectly understandable to the finely tuned ears of left-wing commentators.)

Here’s what I don’t understand.  An alleged hidden appeal to a voting constituency via dog whistling is reprehensible.  But explicit appeals to a voting constituency based on “identity politics” is perfectly all right.  In other words, when Hillary Clinton stands before a national debate audience and identifies one of her distinguishing qualifications for the office of the Presidency as her genitalia, this is perfectly legitimate.   Whereas anything that can possibly be construed as in the interests of a “stale, male and pale” audience is a priori illegitimate dog whistling.

Caitlyn Jenner recently caused a storm when she said that she gets a lot more flack for being a life-long conservative than she has ever gotten for being transgender.  This means that Jenner reveals the complexity of the politically correct code of behavior.  Explicitly appealing to her as a transgender person is perfectly acceptable, but dog whistling to her as a Republican is a major no-no.

It is getting harder and harder to keep up with the rules as the PC revolution eats its own children.

Roger Barris

Weybridge, United Kingdom

I Wish That I Had Said That…

“Science doesn’t make it impossible to believe in God, it just makes it possible not to believe in God,” by Steven Weinberg, astrophysicist, echoing my belief that “Religion exists because of what we don’t know…yet.”


[1] I have discussed before (in footnote 4 to this blog) how an exclusive focus on earnings can distort the economic picture, particularly in an environment where politicians (particularly of the Democratic stripe) are heaping additional costs on employers, such as the Obamacare mandates.  These costs depress earnings since businesses care about the total cost of employment and not just take-home pay.

[2] For those who have a spare 30 minutes, I recommend a recent episode of Contra Krugman for a fuller discussion of this issue.  In addition to providing some good statistics and analysis, this episode does a great job tearing down Krugman, which is always a useful and pleasurable exercise.

[3] A standard measure of inequality, the Gini Coefficient, would also show a very unequal economy since it is also cross-sectional.

[4] “The amount of energy needed to refute BS is an order of magnitude bigger than to produce it”

Read Offline:
Posted in: Economics, Politics

Leave a Reply

4 Comments on "Lies, Damned Lies and Statistics"

newest oldest most voted
Notify of
Thank you for the interesting post. I am from an odd, snowy country north of the US so I may be out to lunch. However, for what it is worth, here is my view from afar. I fully agree that statistics are often misused and misinterpreted. Stats on the gender gap or the famous 1% are perhaps the best example. And by the way, the well-meaning American left winger using these statistics seems to be unaware that on a worldwide basis, almost all Americans are members of the 1% club. However, I am not sure exactly what your conclusion is.… Read more »
Bernard: Thank you for your comment. I am sorry that I have not replied sooner, but I am having some technical difficulties with this feature of the blog. I certainly think that the gap between the rich and the poor is widening, largely because of monetary policy which is inflating asset values (held predominately by the rich) and widening the gap. I don’t think that there is any dispute about this, but this to me doesn’t seem to be the major problem, since this will disappear once the aberrational monetary policies cease. The bigger issue is whether the poor, and… Read more »

[…] within High Income Countries such as the US or the UK. However, as Roger Barris discusses here, these statistics are cross-sectional and thus, give no notion of how wealth moves over time. With […]


[…] have already written about how many people use statistics the way a drunkard uses a light pole: more for support than […]