by Russ Roberts on October 9, 2006

in Standard of Living

In an earlier post, I wrote about the biases in the CPI. I focused on the difficulties in taking account of quality change and the failure to account for Wal-Mart’s impact on lowering food prices. Therefore inflation was overstated and improvements in standard of living, understated.

One reaction, (from Spencer, who is a reliable skeptic whose skepticism enhances these pages) was that these effects were relatively small potatoes. Food and consumer durables (the example I used of mismeasuring the impact of quality improvements) are too small a proportion of total spending to make up much of a difference. Besides he argued, while quality of consumer durables might have improved, the quality of service had declined.

I’m going to leave the latter point for another discussion. Here, I want to discuss the magnitude of my concerns about quality measurements and the surveys of inflation that miss the role of Wal-Mart and other discounters that make up an increasing portion of the marketplace.

How big are those effects?

So I poked around in the literature on CPI bias. Jerry Hausman at MIT has a nice paper
(this is the NBER version–it was published a year later in the Journal
of Economic Perspectives) that examines the theory behind the different
kinds of biases. Here is his summary of the magnitude of those errors:


Using data from 1972-1994 Costa finds that cumulate CPI  bias
during this period was 38.4% with an annual bias of 1.6% per year.
Hamilton  (2001) also estimates CPI bias to be 1.6% per year during
this period, using a similar  econometric approach on a different data
set. This sizeable estimate of bias demonstrates  how the BLS procedure
over estimates the COLI.  The actual bias would be even greater  if the
effect of new goods bias and quality change bias were included.   A
recent estimate of quality growth by Bils and Klenow (2001) finds a
significant  estimate of quality bias over the period 1980-1996. 66
They estimate that the BLS  understated quality improvement and
overstate inflation by 2.2% per year on products  that constituted over
80% of US spending on consumer durables.  These more aggregate  studies
along with my (and other) micro studies on particular goods demonstrate
that CPI  bias is likely to be substantial.

I should emphasize that these measures of the bias in the CPI are
measured in percentage points per year. The CPI isn’t off by 1.6%. It
is off by 1.6 percentage points a year. So that when measured
inflation is stated to be 3%, it is actualy 1.4%. This is a massive
error that when cumulated over even a short period of time grossly
understates the growth in real income and standards of living.

Robert Gordon at Northwestern comes to a similar conclusion in this paper:

Current upward bias in the CPI is
estimated to have declined from the revised 1.2 to 1.3 percent in the
Boskin era to about 0.8 percent today. Yet the Boskin report, like most
contemporary studies of quality change, failed to place sufficient
value on the value of new products and on increased longevity. Allowing
for these, today’s bias is at least 1.0 percent per year or perhaps
even higher.

David Leonardt has a nice article in the New York Times from a few weeks ago, on Gordon’s view:

A decade ago he served on a government-appointed group known as the
Boskin Commission. It argued, as Mr. Gordon still does, that the
government exaggerated inflation by more than one percentage point
every year.

Some other economists think the skew may be
somewhat smaller, but there’s broad agreement, even at the Bureau of
Labor Statistics, that the Consumer Price Index has its weaknesses.
”Use of the C.P.I. over extended periods of time,” Patrick C.
Jackman, a government economist, told me, ”is much more problematic
than using it over short periods of time.”


By overstating
inflation, the official numbers understate the country’s wealth, since
every meaningful calculation of income subtracts inflation. The Census
Bureau, for instance, says that the median-earning man who works full
time — the one in the dead middle of the earnings ladder — made
slightly less last year than in 1977. It is one of the main numbers
cited by those who say life isn’t much better now than 30 years ago.


Mr. Gordon’s adjustments show that men actually got a 27 percent raise
in this period and women 65 percent. The gains are not as big as those
of the 1950′s and 60′s, but they do sound far more realistic than the
official numbers. Think about it: we live longer than people did in the
1970′s, we’re healthier while alive, we graduate from college in much
greater numbers, we’re surrounded by new gadgets and we live in bigger
houses. Is it really plausible, as some Democrats claim, that the
middle class has made only marginal progress?

Leonhardt concludes by arguing that the Republicans are equally
crazy to argue that the last few years have been good for workers.

Forget Republicans and Democrats. The bottom line is that when
inflation is overstated by at least one percentage point a year at a
time when the overall rate of inflation is under 5%, the numbers are
broken. They are unreliable. They are not good for measuring changes in
standard of living over short periods of time or long periods of time.

Be Sociable, Share!



48 comments    Share Share    Print    Email


JohnDewey October 9, 2006 at 4:09 pm

I made this point before, but it may be worth repeating. If inflation has been overstated by 1% or 2% or whatever annually, then Social Security recipients have received much larger than warranted COLA raises. We don't have a snowball's chance in hell of changing the raises that seniors get, but perhaps economists should be reminding those seniors about how good they've got it.

Don Lloyd October 9, 2006 at 4:50 pm


Assume the economy was completely unchanged every year with one exception. All food improves every year in terms of having a higher percentage of good cholesterol vs bad.

In your view, should the CPI be zero or negative? Should the COLA be zero or negative?

Regards, Don

joan October 9, 2006 at 4:58 pm

" The gains are not as big as those of the 1950's and 60's, but they do sound far more realistic than the official numbers."

There is aways a problem comparing the market baskets different time periods, determining the CPI. If instead you look at how the ratio of two nonminal values vary over time you can avoid the problem. When this is done for the income(from IRS data) of the bottom 90% of returns divided by the GDP. see graph at
It shows the bottom 90% share of gdp which was at 42% from the end of WWII to the mid 70’s, falling to a little over 30% today? Looking at this plot it is very hard to deny that their was a shift the economy in the 1970's, that accounts for part of the reason the gains we see today are not as big as those we saw in the 50's and 60's. One thing that occurs to me is that inflation targeting by the fed has suppressed wage growth. If any one has looked at this posibility I would like to know.

The explination that it is reflects higher reward for tech skills is weak because most high tech skill workers are in this group or only a little above. Since it is data from tax returns it does not include employer provided benefits, so the increasing cost of health insurance can account for part of the decrease ( gdp share of heath spending is up about 10% but gov pays about half and out of pocket expenses are about ¼ so that leaves an increase of about 2.5 percent of GDP for employer cost. This is probably an overestimate because the bottom 90% includes the uninsured and those covered by gov share. Nor is it the result of the fraction of joint returns that include 2 incomes. The change in the fraction of working wives is nearly balanced by the decrease in the fraction that is married. It may reflect the relative increase in women’s wages pushing some returns over the 90 percent mark. The influx of cheap immigrant labor also might change the cut off point distorting the data. But neither of these effects can account for more than about 1% of gdp since the cut off income is only around $90,000. 1 million x100, 000 = 100 billion which is about 1% of gdp.

spencer October 9, 2006 at 4:58 pm

If you look at a wide variety of real income measures what you find is that there was a sharp break in virtually every measure
of the growth in the well being of the US population around 1980 that has been made worse by the sharp increase in inequality.

Those who question this argue that the data is bad and it overstates the problem. Virtually all these arguments boil down to an argument that the cpi overstates inflation.

So this raises the question, did the cpi suddenly become more unreliable around 1980? What we are now looking at is data that says in the pre-1980 era real per capita income growth was about 3% and since 1980 it has been on the order of 1.7%.

So if we lower inflation a couple of percentage points we now say real per capita income growth has been almost 3% since 1980. But the same adjustment now says the growth rate was around 5% in the pre-1980 era.

Now, is that the argument you want to make?
Or do you want to argue that something happened to the inflation data around 1980 that made it much more inaccurate then it use to be?

Patrick R. Sullivan October 9, 2006 at 5:19 pm

I don't think spencer has thought through his argument above. It may well be that the CPI was more accurate prior to 1980, but because of the sea change in the economy it isn't NOW.

Also, in addition to overstating inflation the BLS is admitting that its income statistics have problems. They're going to start to wean us off them starting next year, because they're not measuring the same segment of the workforce they used to:

'Discontinuation of production and nonsupervisory worker hours and earnings series – BLS’s current plan is to phase out these series after the new all employee hours and earnings series are well established. The limited scope of the production and nonsupervisory worker series makes them of limited value in analyzing economic trends. Just as important to this decision, the production and nonsupervisory worker hours and payroll data have become increasingly difficult to collect, because these categorizations are not meaningful to survey respondents. Many survey respondents report that it is not possible to tabulate their payroll records based on the production/nonsupervisory definitions.'

And, Joan, wouldn't you expect the share of the top ten percent of income earners to increase rather drastically when we dropped the top marginal income tax rates from 90% and 70% to rates in the 30%s?

Patrick R. Sullivan October 9, 2006 at 5:23 pm

There was supposed to be a link to a BLS press release in the above:


joan October 9, 2006 at 6:07 pm

Patrick R. Sullivan, You did not look at the graph very carefully. The is a blip in the share of high income group in 1987 which was probably due to the tax change but the fall in the below 90 group started in the mid 70's. The effect of the cut in the marginal rate from 90% to 70% under Kennedy and from 70% to 50% in the early 80's cannot be seen.

alex October 9, 2006 at 6:48 pm

Overstating inflation, does not deal with the problem of relative inequality, however. Indeed, it has been shown that inflation hurts those at the top by far the most, whereas middle income and low income people are much more effected by GDP growth and uneployment rates. So, it turns out income inequality is even worse than we thought. How wonderful.

Isaac Crawford October 9, 2006 at 11:46 pm

"So, it turns out income inequality is even worse than we thought. How wonderful."

Who's talking about inequality? I'm with Russ on this one, the "inequality" doesn't make any difference if everyone is (significantly) better off. These articles are trying to show that everyone is indeed better off. What the Bill gates and Warren Buffets of the world do is of no concern to me, they do not "capture" wealth, making it inaccessible to anyone else. What Russ is showing is that we are all better off, that's the sign of an economy that is working…


joan October 10, 2006 at 12:52 am

The fact that we are having a discussion about whether or not we are better off is proof the the economy has not been performing well. In 1980 no one questioned we were better off than in 1950, that 1950 was better than 1920 or 1920 was better than 1890.

Morgan October 10, 2006 at 1:23 am


I suspect that a large portion of the change in the incomes of the bottom 90% is attributable to two factors – welfare displacing work and an increasing number of retired people. There are probably also more zero-earner units (even excluding retirees) among the bottom 90% and more two-earner units among the top 90% than in 1970.

But just looking at transfer payments as an index of the first two factors is helpful. The numbers used to construct the graph you referenced exclude government transfers. So add the $1.3 trillion transferred through SSA and DHHS to the bottom 90% of incomes and voila – the share going to the bottom 90% is back above 40%.

I'm not sure about your last point. People have been nostalgic for the "good old days" for longer than either of us has been alive. But no one ever seems to go back.

Isaac Crawford October 10, 2006 at 1:48 am

"The fact that we are having a discussion about whether or not we are better off is proof the the economy has not been performing well."

Umm, no. Many of us are incredulous that this has been brought up at all. Don put it best, would you rather live in 1970's AMerica or today? I believe that the vast majority of people would prefer to live today even if they are unemployed right now. That fact is "proof" that the economy is doing just fine…


Russell Nelson October 10, 2006 at 3:01 am

Joan, no one who is paying attention questions whether we are beter off now than in 1980. So, your question boils down to: why are there so many idiots running around now? I notice that New York State used to have a School for Idiots in Newark, NY. Perhaps it's time to reintroduce such schools?

JohnDewey October 10, 2006 at 5:46 am

joan: "But neither of these effects can account for more than about 1% of gdp since the cut off income is only around $90,000. 1 million x100, 000 = 100 billion which is about 1% of gdp. "

I think we've added at least 12 million low-skilled immigrant workers since 1980. This probably represents 8 million households. Although some do not file tax returns, my guess is that half have become legal workers and do file tax returns. Shouldn't your calculation be more like 4 million 100,000 = approvimately 4% of GDP?

JohnDewey October 10, 2006 at 6:09 am

spencer: "growth in the well being of the US population around 1980 that has been made worse by the sharp increase in inequality."

I don't agree that the increase in inequality is responsible for the slower growth of wages. Three factors led to reduced demand for U.S. workers – especially blue collar workers:

1. globalization forced workers to compete with a much larger labor supply;

2. continued automation, much cheaper with the advent of personal computers, reduced demand for low skilled workers;

3. deregulation of important labor-intensive industries – including air travel, trucking, and telecommunications – sharply reduced the power of union workers in protected jobs.

All three factors led to lower prices for goods and services. Lost union wages weren't pocketed by the rich, but rather by average consumers who enjoyed those low prices.

Randy October 10, 2006 at 7:13 am


Re; "In 1980 no one questioned we were better off than in 1950."

Not true. In fact it was around 1980 that I first started hearing the talk about how we were going to be the first generation that had it worse than our parents. Japan was on the rise, Social Security appeared to be failing, unions were starting to lose their grip, etc.

The way I see it, my standard of living is very similar to my father's, but I was able to start working later (taking time off for education), I will probably stop working earlier (longer retirement), and I don't have to work anywhere near as hard as he did from day to day. Yes, the products I buy are generally nicer too, but these are a minor improvement in relation to the fact that my life is just plain easier. But that's what I realize now. Back in 1980, I was worried just like everyone else.

Aaron Krowne October 10, 2006 at 9:12 am

I examined CPI bias too, and I found the BLS CPI was off by almost 2% per year since 1983 — to the downside.


That means cumulative inflation is now 42% off (reported as too low).

If you think I'm wrong, then you are essentially arguing that the USPS is lying about its costs. If they're telling the truth, the BLS is lying.

Of course, what emerges is that M1 is basically real goods and wage inflation. Big surprise there.

I guess you can show anything you want with a clever research paper, but these a priori analyses of what inflation "really" is never seem to connect with people's and businesses costs in the real world.

Russ Roberts October 10, 2006 at 9:46 am


You can't look at one service, the cost of delivering mail, and use it as a proxy for the price of all services. Changes in U.S. Post Office costs or the rates it charges don't provide any information about the overall level of prices.

Morgan October 10, 2006 at 9:55 am

Aaron Krowne:

Nice linked post. I wonder if we could see the graph you present with values log-transformed. I ask because it's difficult to compare the relative discrepancies between postage rates, CPI, and M1 over time on an absolute scale.

Eyeballing, I think there may have been a larger relative discrepancy between postage rates and M1 around 1980 than the 42% difference between postage rates and CPI today. If so, that might indicate that postage rates are relatively independent of either measure.

JohnDewey October 10, 2006 at 10:23 am

Aaron, from your link: "The US Postal Service only raises rates when it has to–it doesn't structurally turn a profit."

That's an interesting study, but I have to agree with Russ. USPS rates are hardly representative of overall goods and services prices.

Even if USPS were representative, it's just about the worst company to analyze. It has a monopoly on slow mail, and has little incentive to hold down costs. While at Fedex, I remember reading several analyses explaining how USPS used its profits from slow mail to subsidize overnight and package delivery businesses. As an industrial engineer, I worked alongside former USPS engineers who related the wasteful practices at their former employer.

How could anyone know that USPS is cost-efficient if they operate a monopoly business? Do you – or anyone else – believe that the "efficiency" at the Post Office is in any way comparable to that of UPS or Fedex?

Kevin October 10, 2006 at 11:44 pm

Joan were you ALIVE in 1980? That was very possibly the single s—iest moment in US history. Hostages in Iran, continued *real* oil crises, the Soviets invading Afghanistan, Jimmy Carter's malaise, rampant inflation, double digit interest rates, vietnam hangover, watergate hangover. And unlike today's hyperbole, everyone then was correct in knowing that the worst president in history was in the White House.

Yes, if anyone had bothered to ask, plenty of people in 1980 would have insisted everyone was poorer than in 1950, moreover the world had gone to hell in a handbasket, but what did it matter since the world was about to be ended at any moment by nuclear annihilation anyway?

spencer October 11, 2006 at 10:08 am

Sorry I was not around yesterday to take
part in the discussion. But one though did occur.

After the Boskin Commission the BLS made several revisions to its methods that rougly reduced the reported inflation rate by about 0.5% annually. They also created a new series cpi-rs that applied the new methods to recalculate what the cpi would have been using this approach since 1980.
They did not go back past 1980.

Now many government agencies now use the
cpi-rs version of the cpi — that shows inflation after 1980 to be about 0.5%
lower then the headline ci — to deflate
the nominal data . For example, the real compensation data now uses the cpi-rs as does census for all of its real income series.

So for many series what we now have is a series that does assume that the inflation rate was higher prior to 1980 then it actually would have been reported using the new methodology. So my question about the sharp break in the series around 1980 is even more important then what I earier implied because the official data does assume there was a break in the rate of measured inflation in 1980.

John Dewey you are right about the causes of lower wages, but these same factors have also been major causes of the changes in inequality.

When I look at inequality I'm not concerned about the wealthy — I'm all for people becoming billionares because they create something new. What I'm more concerned with are the millions of manufacturing employees that have lost very good paying jobs and have had to settle for low wage retail, service sector jobs. For example, since 1980 manufacturing employment has fallen from 27% of employment to 12% of employment.

In absolute terms that is about 5 million jobs. I think that is the dominate reason
we are having a debate about the standard of living. What we have is most people are doing better and experiencing a rising standard of living and enjoying all these new products . But there is a very significant share of the population that is not sharing in this prosperity and that is what is dragging the averages down.

In recent years this has been supplemented by a drop in info tech jobs from 3.7 in 2000 million to 3.0 million now.

It is these types of structural changes away from high productivity, high wage jobs to low paying, low productivity jobs in retail, food service, leisure, etc.. that is the real problem, not minor problems with inflation measures.

Russ Roberts October 11, 2006 at 7:34 pm


Read the Gordon paper I cite in this post. He argues that even after the corrections made after the Boskin report, CPI is still underestimated by about one percentage point a year. Even so, the CPI-RS is not used throughout the government.

My understanding is that employment in the information technology sector isw HIGHER than it was in 2000 but maybe I'm remembering the date wrong. See Catherine Mann's work at Brookings.

Finally, if we're worrying about those 5 million lost manufacturing jobs, I'm with you. It's hard to rebuild your life when your skills are obsolete or greatly reduced in value. But you're suggesting that prosperity has left behind about 3% of the work force. Could be, but if that's really what you're worried about, we're both in the same half-full camp.

NKaster October 12, 2006 at 1:59 pm

John Cowperthwaite, former governor of Hong Kong when it was under British rule, "was so famously laissez-faire" Milton Friedman wrote recently, "that he refused to collect economic statistics for fear this would only give government officials an excuse for more meddling." Given the mischievous uses to which such statistics are put, I'd say Cowperthwaite had the right idea.

Previous post:

Next post: