variation (EV) between each adjacent pair of quarters.

The EV between period

t

1 and t is the change in spending required to achieve physicians’ utility in

period t relative to t

1, taking into consideration that both the quality of the

drug regimens on the market and the prices of those regimens may have changed between these two periods. Positive values imply that the value of the drugs is increasing over time by more than their prices, whereas negative values imply the opposite. The third step is to translate the EV into a quality-adjusted price index by calculating the change in drug prices that is consistent with the welfare effect

captured by the EV.

We generate three additional indices for purposes of comparison. First, we estimate a “naive” price index that merely reports the mean price of colon can- cer regimens in each quarter, relative to the first quarter of 1993, without any adjustments for regimen attributes. The naive index will be based on the price physicians pay to acquire each regimen, and the market share of each regimen. Comparing the quality-adjusted price index to the naive index illustrates the importance of accounting for the changing quality of pharmaceutical products. Second, we estimate a hedonic price index that is computed by regressing prices on product characteristics and quarter indicator variables. This regression controls to control for changing product attributes by means of a reduced form projection of markups on the characteristic space, and it has been the method tradition- ally used to account for changing attributes and the introduction of new goods. Finally, to assess the implication of the logit assumption we estimate a vertical model in the spirit of Bresnahan (1987).

The naive price index, which increased by 2600 percent between 1993 and 2005, greatly overestimates the price increase. The hedonic price index and the quality-adjusted price index show that prices have actually decreased slightly over the 13-year period we study. The hedonic and quality-adjusted price index are similar during the period without innovation, and both exhibit a decrease when the first branded product is introduced in the market. For the following period of

5