Econ paper

8.4 Nonlinear Effects on Test Scores of the Student-Teacher Ratio 289

Our regr 巳ssions therefor 巳 include two control vari- abies: the logarithm of ag巳 and the logarithm of the number of characters per year in the journa l.

τberegr 巳ssion results are summarized in Table S.2. 四ose results yield the following conclusions (see if you can find the basis for these conclusions in th 巳 table!):

1.Demand is less elastic for older than for newer journals.

2. The evidence supports a linear ，rath 巳r than a cubic，function of log price.

3. Demand is greater for journals with more charac ters，holding price and age constan t.

So what is the elasticity of demand for economics journals? Itdepends on th 巳 ag巳 of the journa l. Demand curves for an SO-year-old journal and a 5- year-old upstart are superimposed on the scatterplot

in Figure S.9c; th巳 old巳r journal's d巳mand elasticity is -0.2S (SE = 0.06)，while the younger journa l's is 一0.67 (SE = O.OS). This d巳mand is very inelastic: Demand is very

insensitive to price ，巳specially for old巳r journals. For librari 巳s，having the most recent research on hand is a nec 巳ssity，not a luxury. By way of comparison ， experts estimate the demand elasticity for cigar巳ttes to be in the range of -0.3 to -0 .5.Economics jour- nals are ，it seems ，as addictive as cigarett 巳s，but a lot better for your h巳alth!l

IThes巳 data were graciously provid巳d by Professor τneodore Bergstrom of the Department of Economics at the Universityof California，Santa Barbara.Ifyou are inter- ested in learning more about the economicsof economics journals，see Bergstrom (2001).

We answer these questions by considering nonlinear regression specifications of the type discussed in Sections 8.2and 8.3，extended to include two measures of the economic background of the students: the percentage of students eligible for a sub- sidized lunch and the logarithm of average district income. The logarithm of income is used because the empirical analysis of Section 8.2 suggests that this specification captures the nonlinear relationship between test scores and income. As in Section 7札 we do not include expenditures per pupil as a regressor and in so doing we are con- sidering the effect of decreasing the student-teacher ratio ，allowing expenditures per pupil to 血crease (that is，we are not holding expenditures per pupil constant).

Discussion of Regression Results The OLS r陀egre臼s岱ssi剑sio归on resu山its岱sa盯re summa盯n归zed in T:曰a抽ble 8.3.τη刊he columns labeled (ο1) t巾hro佣ug悼h(仍7η) ea叫chr即e叩po创r口旧ts臼ep阳a盯ra础t忧er时e吨gres臼创s岱ssior ficient岱s，standard errors ，certain F-statistics and their p-values ，and summary sta- tis tics ，as indicated by the description in each row. The first column of regression results ，labeled regression (1) in the table ，is

regression (3) in Table 7.1 repeated here for convenience. 四is regression does not