Correction to: Mass Media and Electoral Preferences During the 2016 US Presidential Race

Correction to: Political Behavior https://doi.org/10.1007/s11109-018-9478-0

This erratum provides additional analysis and clarification relating to the original version of the article. Specifically, we provide further analysis of the cross-correlation function (CCF) between media tone and measured electoral preferences during the post-Labor Day general election campaign. The relevant results from Table 6 of the article are reported in the first column of Table 1 of this erratum.

Of special focus are the positive correlations between long leads of media tone and current values of pooled polls. Although we did not spend much time in the article on these results and their meaning, it is important to make clear that they appear to be spurious, the product of trends in both polls and media tone during the period the post-Labor Day general election campaign. Such trends are not evident over the long 200-day campaign period that is the focus of most of our analysis up to that point in the paper. The correlations for the fall campaign were estimated on the assumption that the same was true during the fall, which is not correct, as both preferences and tone trend upward slightly toward the end of the election cycle (also see Figures 2 and 4 in the original article).

Table 1 Cross-correlation function (CCF) between media tone and pooled polls, post-labor day, 2016

De-trending both variables and re-estimating the CCF substantially reduces the positive correlations, as can be seen in the second column of Table 1. Here the estimates are much smaller and none are significantly different from 0.0, thus precluding the inference of a positive effect. The negative correlations at long lags of media coverage now are larger, however, and some are statistically significant (p < .05). This implies that coverage not only did not positively influence preferences over the fall campaign but that it might have negatively influenced them, where good news for Clinton on one day was followed by a decline in her poll shares in the following week. Importantly, the correlations with short lags (and leads) are not significant. Also importantly, it may be that the de-trended polls variable is not stationary, i.e., that observed preferences during the fall campaign are not trend stationary. Dickey Fuller (DF) tests indicate that the coefficient for lagged polls (− 0.08) during the period is not significantly different from 0.0, on the basis of which we cannot reject the null of a unit root (MacKinnon p = 0.51).

Given the pooling used in the construction of our measure, these tests of (non-)stationarity seemingly may not be conclusive, as correlations on successive days will tend to be inflated. Also consider that the autocorrelation function for the post-Labor Day campaign (in the second column of Table 1 of the article) exhibits a pattern of decay that is much as we would expect for an autoregressive process. If we assume that preferences during the period are not stationary, however, we need to estimate correlations between de-trended media tone and first differences—instead of de-trended levels—of pooled polls. These are shown in the third column of Table 1. Here we see little pattern. Based on these results, there is not much evidence of a connection between media coverage and electoral sentiment during the post-Labor Day campaign in 2016. As shown in the article, there are clearer relationships over the longer campaign, and these are accentuated when focusing on the period leading up to the fall, underscoring that media-public dynamics can change over the election cycle. Describing and explaining this variation remains a seemingly important subject for future research.