The Relationship of Journal and Internet Use to JSTOR Use

Examination of the frequency of JSTOR use among faculty aware of JSTOR (n = 78) showed that 58% of the respondents had varying levels of use, while 42% reported no use. Using the frequency of JSTOR use as the dependent variable, the faculty who reported no use were censored on the dependent variable. The standard zero, lower-bound tobit model was designed for this circumstance (Tobin 1958). Most important, by adjusting for censoring, the tobit model allows inclusion of negative cases in the analysis of variation in frequency of use among positive cases, which greatly enhances degrees of freedom. Therefore, hierarchical tobit regression analyses were used to examine the influence of demographic characteristics, journal use, search preferences, Internet use, and attitude toward computing on the frequency of JSTOR use. Independent variables used in these analyses were selected on the basis of significance in univariate tobit regressions

on the frequency of use variable. Table 11.9 summarizes the independent variables used in the multiple tobit regression analyses.

Table 11.10 summarizes the results of the hierarchical tobit regression of demographic, journal use, search preference, Internet use, and computing attitude variables on frequency of JSTOR use. The line second from the bottom in Table 11.10 summarizes the log likelihood score for each model. Analysis of the change in log likelihood score between adjacent models gives a measure of the significance of independent variables added to the model. For example, in Model 1, the addition of the demographic variables failed to produce a significant change in the log likelihood score compared to the null model. By contrast, in Model 2, the addition of journal use variables produced a significant change in the log likelihood score compared to Model 1-suggesting that the addition of the journal

use variables improved the fit in Model 2 over Model 1. Similarly, the addition of search variables in Model 3 and of Internet use variables in Model 4 both produced significant improvements in fit, but the addition of the computer attitude variable in Model 5 did not. Therefore, Model 4 was selected as the best model. From Model 4, the coefficients for gender, article copying, abstract reading, and searching on-line catalogs are all positive and significant. These results suggest that, controlling for other factors, men were 0.77 points higher on frequency of JSTOR use than were women, that there was a 0.29-point increase in the frequency of JSTOR use for every point increase in the frequency of article copying, that faculty who read article abstracts were 0.82 points higher on frequency of JSTOR use than were faculty who didn't read abstracts, and that there was a 1.13point increase in the frequency of JSTOR use for every point increase in the frequency of on-line catalog searching. From Model 4, the coefficients for affiliation with an economics department and the number of paid journal subscriptions are both negative and significant. These results suggest that, controlling for other factors, economists were 0.88 points lower on frequency of JSTOR use than were historians and that there was a 0.18-point decrease in frequency of JSTOR use for every unit increase in the number of paid journal subscriptions.