There is a lot of hand wringing about the accuracy of the consumer price index (CPI) numbers published by Bureau of Labor Statistics (BLS).  This is not surprising since there are so many things that are dependent on these numbers (Social Security, government pensions, tax brackets, inflation protected securities, etc).  It is not uncommon to read about the latest inflation numbers, but to feel in your gut that you are experiencing inflation higher than what is reflected by the number.  As a result, any change to the way that the CPI is calculated leads to questions of whether the government is trying to reduce their long term liabilities.

One of the more shrill examples expressing this concern can be seen in the diatribe from John Williams’ Shadow Government Statistics where he claims that the CPI-U understates inflation by 7% a year.  Economists John Greenlees and Robert McCelland from the Bureau of Labor Statistics recently wrote an article(warning PDF) to debunk these complaints.  Below is a synopsis and interpretation of their findings.

The changes to the CPI-U that most people complain about are

  • Change from arithmetic weighting to geometric weighting
  • Quality adjustments (hedonic models)
  • Change to rental equivalence

Weighting Change

I won’t get into the math about this change because this one is easy to address.  The BLS has kept a parallel index that still uses arithmetic weighting (instead of geometric) and indeed when comparing the two, the geometric weighting did reduce inflation by an annualized rate of 0.28%.

Quality Adjustments

I will address the quality adjustments in two parts.  First, are quality adjustments necessary?  If the makers of ice cream reduce the amount of ice cream they put into a box (have you noticed those 1/2 gallon cartons of ice cream no longer have 1/2 gallon?) but keep the price the same, has inflation occurred?  Without quality adjustments the answer would be no, but clearly the answer should be yes.  So if the ice cream manufacturer puts in 20% less ice cream for the same price, the CPI will see that as a 25% rise in the price of ice cream.

Clearly, quality adjustments are needed and have been included in the CPI for a long time (decades), but in some areas it can get a bit murkier.  Imagine the average car cost $10K and the airbag option cost $1K.  And let’s further stipulate that in one year almost no one was choosing the airbag option so the average car was selling for $10K.  If the government now mandates that all cars must have airbags, and the following year the average car price is $11K, has there been inflation?  Because of the quality adjustment (i.e., you are getting a safer car with an airbag) and because the price for that safer car did not go up from the previous year, the CPI will not reflect the 10% increase in car price.  What has happened is that the government has in fact forced you to raise your standard of living whether you wanted to or not.

The type of products where this is most problematic is in technology.  You generally don’t have the option of buying a cell phone or computer circa 1990.  So technology puts you on the treadmill of forced upgrades to your standard of living.  And because these items get replaced often, those forced upgrades mean we can never realize the savings the quality adjustment says we are receiving.  In 1999, BLS added quality adjustments to electronic equipment which caused a lot of heartburn for people. 

So how much of a difference does it make?  If you look at all quality adjustments (up and down) historically they have been a wash, so simply eliminating quality adjustments would not have a huge impact on the CPI.  But as mentioned earlier, quality adjustments in some areas clearly make a tremendous amount of sense.  It is only when we have these forced upgrades that they sometimes feel “wrong”.  Focusing on just the items that started having quality adjustments in 1999 (mostly electronics), these had a combined weight of around 1% of the CPI makeup.  If we assume that technology doubles in power every three years for the same price, this would translate to around a 20% decrease in price every year due to quality adjustments.  This would translate to lowering the CPI by 0.2% a year.

Rental Equivalence

The last major complaint is with the use of rental equivalence for housing rather than actual housing prices.  Imagine for a moment that everyone owned a home, and then imagine that prices doubled.  Would that translate to inflation for people?  Since everyone already owns a home, their home would have doubled as well, so there was no net effect (ignoring effects on upgrading or downgrading).  Now if every one was renting instead and rents doubled, everyone would agree inflation had occurred.  What this illustrates is that home ownership has two aspects: giving you a place to live and as an investment.  By using rental equivalence the BLS is trying to remove the investment aspect of home ownership.

Over the long term, rents and prices are highly correlated to each other (i.e., move together), so this choice does not bias inflation in the long run, but it certainly does mean that if we are in a period of home pricing booming (like the recent past), the CPI will not accurately reflect the pain of a new home buyer.  On the flip side CPI will also not reflect the pain of home owners when home prices come crashing down.


The BLS works hard to get their inflation numbers right.  The process is much too transparent to be manipulated secretly for the sole benefit of the government.  I believe you could make a good case that changes to the CPI-U over the last decade have resulted in reducing CPI-U by as much as 0.5% per year, but would find it extraordinarily difficult to make a case for a number much larger than that.