The Analysis Of The Article "The Timing of Partisan Media Effects" By Glen Smith

In The Timing of Partisan Media Effects during a Presidential Election author Glen Smith aims to expand on the pre-existing knowledge surrounding how partisan news media viewership can affect candidate favorability ratings. Specifically, Smith looks at levels of viewership of the Republican partisan news source Fox News and how it impacts the favorability of opposition candidate Barack Obama during the 2008 presidential election. Unique to this article is the addition of a time factor. Smith looks not just at how partisan news media can affect voter opinions during an election year, but at what differences may occur at different points in the year.

Thus, all models to be reviewed refer to an “early” period and a “late” period, with the exception of some included models which refer instead to specific month periods. The “early” period is workably synonymous to the real-world presidential primary elections; the “late” period is workably synonymous to the general election. By analyzing data in this way, Smith is expanding on previous research that did not consider the primaries and general elections separately and by doing so helps to distinguish effects that may have been overlooked by those studying only the period during the general election. This allows us to look at the effects more comprehensively, as close to the election it is possible opinions have long been formed.

The data for this analysis is drawn from National Annenberg Election Study surveys (NAES) conducted over the phone and Internet during the time period being studied (2008) and content analysis data from the News Coverage Index. These data were applied and/or manipulated across three separate models in order to most completely answer the question at hand while simultaneously leaving as few gaps as possible. Working with the limited available data, the author invokes three methods which, taken separately, each has weaknesses detrimental to their applicability. Taken together, these three approaches begin to give us a better and comprehensive understanding of the effects partisan media influence can have.

Method the first

Smith first looks at the amount a respondent viewed Fox News as the number of programs they viewed. This factor, or coefficient, is compared to an interval scale representative of favorability of candidate Obama, which is the dependent variable in this model. This is done for the “early” period, as well as the “late”, and it is done first for all respondents, then respondents are separated based on political leaning (Democrat, independent, Republican). Smith also makes use of several controls for comparison.

Through this method, it becomes evident that there is a more significant effect during the “early” portion of the year, and during the “late” the effect is more negligible. This will be a trend we continue to see throughout the article. As Smith points out, across all respondents you see the effect wherein “each additional Fox News program...predicted a 0.80 reduction in favorability toward Obama early in the campaign” (Smith, 2016). This is evident in the Ordinary Least Squares (OLS) regression model labeled “Table 1”. Out of a sample of 11,060 during the “early” phase” (represented by n in Table 1), you see the coefficient -0.800 (SE .093), reflecting what Smith noted about a change for each program watched. This does not continue for the category “late”. The negative effect is clear in each early state for all subgroups (democrats, -0.983; Independents, -1.26; Republicans -0.320). While the amount varies, each subgroup shows a statistically significant decrease during the “early” stage.

This statistical significance does not carry on into the “late” stage, where we see a much smaller effect size for all respondents and all subgroups. This suggests that partisan news media can have a significant effect on candidate favorability during the primary stages, but that effect decreases when the candidate pool is reduced. This supports the author’s hypotheses of this nature. This early/late difference is apparent through this first method and the related regression model and will continue to be the trend throughout the next two methods.

Method the Second

The second method employed by Smith could potentially be divided into two portions sharing a common factor. That common factor and the way in which this method differs from the first is that Smith does not use the variance of Fox News viewership in the same way as in the first method. Instead of analyzing only whether the respondent did or did not watch Fox News, it combines this data (through math) with the content of the program watched. It speaks more directly to whether or not what the viewer was seeing was related to Obama, and then goes through much the same steps as the first method. Again, the major difference being a comparison of what the viewer actually saw and how it affected their favorability thermometer (again used to form the dependent variable) the next week.

In this first regression model represented by Table 2, the independent variable is not the number of programs watched and is instead time (in minutes) Fox News devoted to Obama. The dependent variable, in the case of Table 2, remains the same as the first method. Another similarity to the first method is that a significant change can be noted across all respondents and subgroups in the “early” category, but the effect is less significant closer to the election. Out of a sample of 10,307, each unit of increased Fox News content viewing (I believe measured in increments of 30-minute intervals) decreased Obama’s favorability -0.108 points in the “early” phase, but only -0.005 points in the “late” phase. MSNBC, among others, is again used as a control. This model has mainly served to lend further support to the same hypotheses supported by method one.

Method the Second Continued

In an answer to the question of whether or not some of the negative “early” effects on democratic viewers might be a consequence of their never having been favorable of Obama, to begin with, the regression model in Table 3 is employed. Here, Smith looks specifically at democratic Fox News viewers and their original opinion on Obama at the start of the year. As we can see from the effect sizes in comparison to their standard errors, those who were already in support of Obama did not see a significant effect in relation to their Fox News viewership. Only those supporting a different candidate during the primaries saw significant effect (in this case, -1.05). Whereas Table 3 mirrors Table 1 in its inspection of Fox News in terms of the number of programs viewed, Table 4 mirrors Table 2 in its inspection of the minutes Fox News was viewed.

A secondary finding from this portion of the method does point a secondary effect. Democratic support did increase pretty significantly in the “late” phase, evident in both Tables 3 and 4. In Table 4 favorability among those that previously supported Obama from an effect of -0.051 during the primaries to +0.066 in the general election; for those supporting another candidate it jumped from -0.105 during the primaries (fairly low, as the model intended to point out) to 0.075 in the general election.

Method the Third

In the third method utilized in examining the effects of Fox News viewership on candidate favorability ratings, Smith brings into the fold the viewers’ general political knowledge. He does so by shifting the utilized data to that which includes the Fox News variable in a different way. Whereas the previous methods made some use of a binary variable, this method cannot avoid it. The method serves to answer the looming question of spurious relationships left unanswered by the first two methods. A different data set was used for the purposes of this method. The telephone survey utilized did include information about the respondents’ political interest, as well as their what their primary news source is.

This method looks closer at the idea that favorability may be affected by other factors and looks at how those factors may interact with Fox News viewership. Specifically, there is the addition of the independent variable “political knowledge”. As evidenced by the bar graph used in Figure 1, the higher the political knowledge the less effect partisan media will have, the lower political knowledge, the more effective. These effects are again confined to the “late” phase, where you can see almost no effect represented by the June-October bar in the “High Knowledge” category (-.01, as compared to -.13 for the “Low Knowledge” category during the same time period).

Advantages and weaknesses

All three of the methods used by Smith to determine the relationship between Fox News viewership and favorability of Obama support the hypotheses presented. Some hypotheses are supported by more than one method. The advantage of using these different angles is that it covers more bases and leaves fewer gaps in the research. Where one method may leave open the possibility of spurious relationships, another attempts to close that possibility. Each of the methods described fills in a part of this puzzle of how partisan media effects favorability. The use of the three together allows for a more comprehensive view of the effects with little gaps left in our understanding.

As in all articles of this kind, Dr. Smith addresses weaknesses in his models individually and of the research as a whole, addressing corrections or consequence along the way. As stated earlier, these three methods each have the weakness of not answering to the effect from every angle. It is the utilization of the three methods together than Smith uses to answer this weakness. Where one is lacking, the other picks up the pieces.

The first method employed left open the question of whether there might be other factors contributing to the effect that was not included in this study. This weakness is answered in the subsequent methods that attempt to account for some of the other factors that may be in play. For the second, and I believe parts of the third, call into question how much of the programming the viewer actually watched. Smith does address this as inconsequential; I agree with this assessment as it would be substantially disadvantageous to make any further attempts at correcting for this. Perhaps in future research, this disadvantage could be addressed, but for the purpose of the present, it is best to continue to view it as inconsequential.

Another common issue between the second and third model is the decreased sample size (as the threshold for who is or is not a viewer is higher in the second method and the latter parts of the second method look at just subsamples of the already decreased amount). In the case of the latter second, Smith does specify that the sample size for the specific subgroups is small enough that the model in question is more for consideration, not the conclusion (Smith, 2016).

Conclusion

All three methods used serve to point to the same general trend. When Fox News viewership increases, favorability of Obama decreases - but only in the “early” stage. The “late” stage saw a must narrower effect change. This is the major take away from the models presented, though other conclusions could be drawn, such as a reverse effect relationship between MSNBC viewership and/or McCain favorability ratings.

These conclusions were not drawn; this is not because they are not apparent, but because the evidence is not sufficient to support them. Which leads to another take away from this article. In fact, the results of the favorability test may have been a red herring for partisan media predictors including and predating this one. The general field study suffers from a lack of data, without which further conclusions could be drawn. Smith makes up for this where he can and does not balk when it is apparent the evidence is not sufficient. This article will serve well as a study of the conclusions that may be drawn from it, and also as a good jumping-off point for the apparent relationships that require more evidence before conclusions may be drawn.

Reference:

Smith, G. (2016). The timing of partisan media effects during a presidential election. Political Research Quarterly, 69(4), 655-666.