Nothing epitomizes the horse race nature of presidential politics quite like the conventions ratings war. For example, Matt Drudge and Rush Limbaugh were quick to dismiss the DNC’s higher tv ratings by noting that Honey Boo Boo tied Bill Clinton's much-anticipated speech. Democrats, on the other hand, were quick to proclaim victory and trumpet their superior performance. But do convention television ratings matter? What can they tell us about the looming presidential election? Nate Silver at fivethirtyeight.com examined television ratings as an indicator of the proverbial “convention bounce,” finding a strong historical relationship between the two factors. And while I agree with Jonathan Bernstein that patience is required in assessing these sorts of things, we might reasonably conclude that television ratings have some predictive validity in this context (though there are certainly some causality issues involved).
What I wanted to know is whether convention ratings have any predictive power with respect to election outcomes. So I calculated the difference in DNC and RNC ratings (X-axis). Higher values indicate the DNC drew the higher rating. The response (Y-axis) is the percentage of the vote received by the Democrat. Though we see a positive relationship in the chart to the right, as we would expect, the effect is small in magnitude (b=.27) and statistically insignificant (p=.597). Thus, it seems that victory in the ratings war is not a precursor of victory in the general election. As an important side note, I suspect this conclusion is especially true today as surveys have shown that fewer voters make electoral decisions based on party conventions compared to thirty years ago. The second chart (to the right) presents the percentage of ANES respondents who reported making their vote choice during the the DNC or RNC. In 2008, less than 15% of respondents made their decision during a convention while in 2000 it was fewer than 10%. So in this context I think convention television ratings are good for one thing: bragging rights.
I also examined whether convention ratings are predictive of voter turnout. I used the aggregate ratings for both party’s conventions (X-axis) and turnout in presidential elections (Y-axis). We see in the chart below a reasonably strong, positive relationship. In fact, the relationship is statistically significant with only 13 observations (p=.01). The magnitude of this effect is such that every 10 points in combined convention ratings adds about 3% to voter turnout. A bivariate regression indicates that summed convention ratings explain about 46% of the variation. If you look closely, you’ll see 2012 added to the chart based on the model’s predictions. The estimated turnout for this election is 53%. (disclaimer: yes, this is a very crude model of turnout). So in this context, convention ratings might give us a window into an important element of presidential elections. Notably, turnout matters in this election because higher turnout probably favors Obama. See Nate Silver's excellent post on this issue.
For those wondering, the chart below presents the raw DNC and RNC ratings from 1960 to 2012 (in the conventional red and blue lines). The data are available from Nielsen here. We know that the closing night of the DNC drew 35.7 million viewers while the RNC drew 30.3 million viewers. Both are down from 2008. Interestingly, we can see that fewer Americans are watching party conventions today compared to thirty years ago. One explanation for this may be greater electoral polarization. This dovetails nicely, I think, with the decline in the number of voters who indicate making their electoral decision during the party’s conventions (see the ANES data above). Another probable explanation is a general decline in network newscasts and a correspondent rise in electronic media. Finally, this trend may be due to the fact that Americans are inundated with so much political information today that there is less desire for watching party conventions as a source of political knowledge. (note: I’m not aware of any research in this area so I’m willing to concede these are simple conjectures. If anyone knows of a relevant study I’m happy to reference it.)