R

ETHINKING THE

D

IGITAL

D

IVIDE

11

gained. Then, we attempted to see whether the patterns of relationship between Internet use and

gratification gained are different depending on socio-economic status and age, the two most

important individual level factors in the presistent digital divide. Socio-economic status was

constructed by combining income and education, both of which were standardized from 0 to 1.

Socio-economic status and age were dichotomized at the median value. These two dichotomies

create four subgroups: the high SES-young (N=457), the high SES-old (N=602), the low SES-

young (N=675), and the low SES-old (N=439). Due to moderate correlation between socio-

economic status and age, sample sizes in these four subgroups are not equal. Then, structural

equation modeling techniques were applied again to all four subgroups. Taken as a whole, this

study presents five path models: one whole sample model and four subgroup models.

In order to rule out potential confounding variables, we employed a residualized covariance

matrix as input. To construct the residualized matrix, the six variables of our interest, Internet use

and gratifications, were regressed onto the sets of control variables. These regressions produce

residuals, the part not explained by the controls, with which we construct covariance matrix to be

analyzed. In so doing, we can except that our model would be controlling for the variables used to

create the residuals. We report outputs of the regression analyses in order to note how much

variances in the Internet use and gratification variables of our interest were explained by the

control variables (see Table 1-5). Since the variances explained by a set of control variables were

already taken out by using residualizing technique, the total variances reported in the following

path models can be interpreted as amount of variances uniquely explained by types of Internet use.