mance in markets with few suppliers, as in broadband. Yet, a policymaker cannot address questions about pricing without understanding whether the CPI measures prices accurately.

 

measuring technical improvement As it does elsewhere, BLS uses very cautious procedures for recording benefits from upgrading to broadband. There is a rationale behind that caution: Not all users experience the benefits of broadband in the same way. For example, capacity/ bandwidth differs between households and the location in the national grid matters. Accordingly, rather than merely assuming households benefit from the availability of a new service, such as broadband, the BLS waits for clear evidence that buyers like the improvements.

This approach leads to what is sometimes called a “transactional” index. Standard procedures do not measure technical progress at the frontier unless users transact for a better good or service. Moreover, the CPI focuses on the average experience, getting an average price by taking a weighted average over all transactions, including those not at the frontier.

That is why, for example, wireless broadband tended not to play a big role in the official price index until after 2006. Wireless broadband revenue was small in comparison to the revenue for wireline broadband. Incidentally, that is also why McDevitt and I focused our study on wireline broadband, the bulk of transactions for which there is public data up until 2006.

The CPI for Internet access is officially called Internet services and electronic information providers, and the

Broadband benchmarks.

Year internet hh ( 1) broadband hh ( 1) official index ( 2) Corrected index ( 3)

1997 19. 1 n.a.( 4) 100 100

1998 27. 2 n.a.( 4) 103. 3 103. 3

( 1) millions of u.s. households.

( 2) Quote from december of each year.

BLS began compiling data in December 1997. BLS deserves some credit for starting this index not long after the diffusion of the Internet began. At that point approximately 20% of U.S. households had adopted dial-up commercial Internet service, and less than 1% had adopted broadband. The second and third rows of the table illustrate this point.

In the fourth row is a monthly quote from the official price index, taken from December of each year, and normalized to 100 for the year in which the index began. It indicates that the CPI for Internet access in the U.S. went mildly down and up and down for eight years. Then, in late 2006, it broke with all prior patterns and declined more than 18% from its base (i.e., ( 94. 5 – 77. 2)/94.2 = 0.183). That drop continued and settled at a 23% from its base in January 2007 (i.e., ( 94. 5 – 73. 4)/94.5), staying near that level ever since.

Did that last drop reflect increasing competitiveness of broadband markets? In fact, it does not. The dramatic drop in the official price index for the U.S. primarily reflects the pricing of the declining good, dial-up service. An informed reading of the business news explains why. The largest U.S. dial-up provider, America Online (AOL), altered its pricing policies in the fall of 2006. AOL announced it was moving to advertising-supported service. On top of that, BLS used its standard cautious procedures. That gave AOL’s prices close to a quarter of the weight in the index (even though AOL’s market share was dropping).

 

Diffusion and Prices The official index is less informative about broadband prices for an addi-

tional but rather subtle reason. Standard price index survey procedures measure the price at which the new good transacted but not at the price that previously deterred the user from adoption. For example, in many neighborhoods broadband was not available in any form for some time until after 2000. Even when it became available, it may not have been reliable enough (initially) to spur many households to switch immediately from dial-up. These users waited until vendors improved the infrastructure or service arm of the organization. In short, users did not adopt until quality improved.

That common occurrence creates a problem for BLS. Household eventually converted from dial-up to broadband, as rows 2 and 3 of the table indicate. However, there was ( seemingly) no measured price change or qualitative improvement affiliated with the upgrade.

What would have happened to the broadband price index if it reflected unmeasured qualitative improvement? That is a succinct way to understand one aim of our study. McDevitt and I followed the standard recommendation for this problem. It goes back to work by Sir John Hicks, a Nobel Prize Laureate in Economics, and has been used many times in economic analysis. The price index should employ what a user would have been willing to expend to get the new service before it was available.

Many surveys tell us what that willingness was. Even the most cautious surveys from the time period suggest the average convert was willing to pay approximately $51.35 per month (on average), but actually had to pay less. If the actual price was $36, for example,

 

1999

35. 5

0.9

96

95. 4

2000

44

3. 2

95. 7

94. 1

2001

53. 8

9. 6 100. 3 95. 9

2002

56. 7

13

99. 6

93. 8

2003

59. 5

18. 5

97. 2

89. 5

2004

66

27. 5

94. 5

84. 1

2005

73. 3

41. 1

77. 2

65. 6

2006

81. 8

47

73. 1

61. 1

 

( 3) Applies estimates from “the broadband bonus: Accounting for broadband internet’s impact on u.s. GdP”, table 7, row 2.