Next, to test the ‘Collapse in Quality’ hypothesis, we
follow Khandelwal et al. (forthcoming), decomposing the
price of each variety into a quality-adjusted price and a
quality component. The quality-adjusted price is obtained
by subtracting from the log prices the estimated quality
(which is already in log).3 Then, using a single difference
approach, we regress, alternatively, the three variables
(price, quality and price-adjusted quality) on a dummy
indicator, which assumes the value of 0 for the 2007–2008
and the value of 1 for the 2008–2009 variation, yielding