Econ paper

288 CHAPTER 8 Nonlinear Regression Functions

as the "price per citation" in the journa l.The pric 巳 Economics Association ，including th 巳American Eco- range IS enormo 叭的时�per citatio 州h巳American nomic Review! Economic Review) to 20¢ per citation or more. Some Because we are interested in estimating 巳lastici- journals are 巳xpensive per citation b巳cause th 巳yhav 巳 ties ，we use a log-log specification (Key Concept 8.2).

few citations ，others b巳caus 巳 th 巳ir library subscription The scatterplots in Figures 8.9a and 8.9b provide

price per year is very high. In 2010 ，a library print sub- 巳mpirical support for this transformation. Because

sαiption to the Journal of Econometrics cost $3264 ， som 巳 of the oldest and most prestigious journals are compared to only $455 for a bundled subscription the cheap 巳st p巳r citation ，a regression of log quantity

to all seven journals publish 巳d by the American against log price could have omitted variable bias.

··HE1 回事;Jva·Estimates of the Demand for Economics Journals 一一-一

Dependent variable: logarithm of subscriptions at U.S. libraries in the Year 2000; 180 observations.

Regressor (1) (2) (3) (4)

In(Price per citat， -0.533 料 一0.408时 一0.961 ** -0.899 忡 (0.034) (0.044) (0.160) (0.145)

[In(Price per citation) F 0.017 (0.025)

[In(Price per citation)p 0.0037 (0.0055)

In(Age) 0.424" 写 0.373 材 0.374** (0.119) (0.118) (0.118)

In(Age) X In(Price per c山tion) 0.156 忡 0.141 ** (0.052) (0.040)

In( Characters -71，000，000) 0.206* 0.235" 0.229'" (0.098) (0.098) (0.096)

Intercept 4.77** 3.21 ** 3.41** 3.43料 (0.055) (0.38) (0.38) (0.38)

F-Statistics and Summary Statistiα

F-statistic testing coeffici 巳nts on 0.25 quadratic and cubic terms (p-value) (0.779)

SER 0.750 0.705 0.691 0.688

IF 0.555 0.607 0.622 0.626 The F-statistictests the hypothesis that the ∞巳fficients on [In(Priceper c山tion)f and [In(Priceper citation)]3are both zero. Standard errors are given in parentheses under coefficients，and p-values ar巳given in parentheses under F-statistics Individual coefficients are statistically significant at the *5% level or **1% level.

J

continued