Abstract
This study examines patterns of news consumption across multiple media platforms and relates them to civic participation. Analyzing a national sample of close to 25,000 respondents, nearly half the adult population in America is classified as news “Avoiders,” and the other half as “News-seekers.” Testing the relationship between civic participation and news consumption for each of 6 media platforms individually, and to an overall index combining those sources into 1 measure, the results show a positive relationship with civic participation, but the influence of Total News Consumption on civic participation is greater for Avoiders than for News-seekers.
Notes
The authors thank Simmons Market Research Bureau, Inc. for making the necessary data available. We also thank Markus Prior and Limor Peer for their comments on the manuscript.
Note: *p < .01.
Note. *p < .01.
1Additional details on the survey methodology are available from http://www.smrb.com/web/guest/core-solutions/national-consumer-study and http://www.smrb.com/web/guest/experian-simmons-methodology.
2The finite mixture model is necessary because five of the study's variables are roughly continuous, while local TV viewing is dichotomous and most clustering methods are not designed for a combination of categorical and numerical variables. A brief overview of the specific model is provided. Assume that the six variables have been sampled from a mixture of K distributions and that an observation x = (x 1, …, x 6) from group k has the following distribution
where x 1 is the log of time spent watching TV, …, x 5 is the log of the magnitude of internet use, x 6 = 1 if the respondent watches TV news and 0 otherwise, function g is a normal distribution with mean μ ik and standard deviation σ ik , and π k is the probability that someone in group k watches local TV news. Let η k be the (prior) probability of being a member of class k, where η1 + … + η K = 1. The unconditional distribution of x is then given by
The parameters of this model, η k , μ ik , σ ik , and π k are estimated using the maximum likelihood method in the Latent Gold software package. The posterior distributions give the probability of belonging to various groups, based on the observed media consumption, and will enable the researchers to assign respondents to news groups:
3Estimation using maximum likelihood or the squared-multiple-correlation methods produce a single eigenvalue greater than 1, suggesting a single factor. Using the principle components method of estimation, the second eigenvalue is 1.1, suggesting that there could be a second factor, but a case can also be made for a single factor from the scree plot. The two-factor solution has a TV-news dimension (local and network TV news) and an “other” news dimension (Web, magazines, and newspapers), with cable TV cross-loading. The two-factor model, however, is not ideal either because of the large cross-loading and questionable discriminant validity. Coefficient alpha for the 3 TV questions (putting cable news on this factor) is 0.45 and other is 0.34, both of which are lower than the alpha of the 6-item scale. The reliability is likely weakened by the dichotomous measure of local TV news consumption.
4A Likert-scale item was located in the Simmons data where respondents indicated an attitude toward being informed, although it is not specific to civic/public affairs and/or knowledge of those affairs. Still, an alternative model was tested with this variable entered as a control. It was not significant and had no impact on the coefficients reported in . Thus, the researchers chose to report the more parsimonious model with demographic controls.