Retesting the Marketplace Theory of Media Use—10

addition, it should be noted that control demographics are not depicted in the SEM figures. Their

effects, however, appear in tabular form.

Our approach to testing varies somewhat from that of Beaudoin and Thorson (2002).

With reliance on a modification index to determine relationships among the four main

determinants, Beaudoin and Thorson (2002) found that two different credibility measures

(minority coverage credibility and mainstream coverage credibility) were at the core of the

process. Because we see this approach, as well as that of MacKenzie and Lutz (1989), to be “data

driven,” we take a slightly different tack in the current study. We posit reciprocal relations among

the primary antecedent measures, with the effects of these measures on newspaper readership

mediated by attitude toward the newspaper (A

N

).

Finally, partial correlations were used as tests of mediation. This strategy is articulated by

Baron and Kenny (1986) and applied to mass communication by Eveland (2001, 2002).

Mediation can be examined by testing the relationship between an independent variable and a

dependent variable and then the independent variable and the dependent variable controlling for

the posited mediating variable. In our study, we test the partial correlation of newspaper

readership and a primary antecedent controlling for demographics. Then, we test the partial-order

correlation of newspaper readership and the same primary antecedent controlling for

demographics and A

N

.

The two structural equation models fit the data well. The model for Newspaper A has

excellent fit indices (NFI=.998, CFI=.999). Similarly, the model for Newspaper B has an NFI of

.992 and a CFI of .994. Each quotient is well above .995. Also important are the squared multiple

correlations. For the Newspaper A model, they are as follows:

A

N

(.182) and newspaper

readership (.194). For the Newspaper B model, they are as follows:

A

N

(.154) and newspaper

readership (.101).