A methodology for the empirical classification of newspaper content based on reader interests

The reader-interest and background data were gathered in personal interviews with an available, nonrandom sample of 280 adults. Each respondent rated 93 newspaper headlines using Raskins' (1960) 0 to 100 title-rating technique, a validated predictor of actual readership, and provided demographic and leisure-activity information. Although a random sample of respondents was not necessary for the methodological purposes of the study, a controlled effort was made to interview persons with a variety of background characteristics so that the relationship between background variables and reading interests could be adequately analyzed.

The results of the classification analyses revealed that the complete linkage cluster analysis and a selected factor analysis procedure (minimum residuals, orthogonal) each identified a set of reader-interest categories representing content likely to be read by a high proportion of the audience tested. Furthermore, the categories produced by each method were found to be highly interpretable.

Incremental analyses determined that the predicted audience coverage provided by representative content from the 16 factors produced and the predicted coverage achieved by representative content from the 16 clusters identified are 85.6 percent and 85.0 percent, respectively. It was therefore concluded that the two empirical classification techniques are equally useful in terms of specifying categories of content which, as a group, are likely to reach a wide audience.

Although the cluster analysis and the factor analysis each produced 16 reader-interest categories which cover the interests of about 85 percent of the audience, the composition of the clusters and that of the factors are not the same. Future research will have to determine whether the clusters or the factors are closer to actual, replicable categories of reader interest.

In the factor analysis of the data, the re-factoring of long, initial factors was employed. The usefulness of this novel process was tested by calculating the incremental predicted audience coverage provided by the new factors, and it was found that the new analyses revealed concise and interpretable factors of reader interest which substantially enhance predicted audience coverage.

The demographic and life-style characteristics of the respondents were found to be related to levels of interest in the various categories of news developed by the factor analysis. Overall, the demographics proved to be the dominant discriminators. Discriminant analyses determined the relative strength of the background variables in distinguishing among levels of interest in the various categories of news, and crosstabulation tables revealed specific differences suggested by the discriminant analyses.