Journalists cite the Real Clear Politics Average every-day in order to inform the public on the state of the presidential race. The idea is that, since the Average includes a whole variety of polls, merging them all together may provide an accurate snapshot of the race.
It’s fitting to ask: Does the RCP Average reflect the political diversity of the American people? I decided to go through the trouble of finding the breakdown for respondents’ political affiliation for all polls listed on the Real Clear Politics Average.
As of the afternoon of October 13th, there are 120 listed polls and 114 polls provide the relevant numbers. The findings are clear: Comparing the numbers to the Gallup poll, which recognizes the difficulties involved in horse race polling and hence is not doing any polling for the presidency in this cycle, the polls systematically over-sample Democrats (30% in Gallup vs. 36% in RCP) and under-sample independents (40% in Gallup vs. 30% in RCP). Republicans are also over-sampled, but not as much as Democrats (28% in Gallup vs. 30% in RCP).
What’s important to note is that these differences are “statistically significant.” This means that, compared to the venerated Gallup poll, the polls picked by RCP seem to amplify the voices of Democratic voters while diminishing the voices of independents. In other words, the voters who may be most interested in the non-establishment, populist campaigns of Trump, Johnson, and Stein are systematically underrepresented in the Average. I’ve posted the data for others to examine in whatever manner they choose.
Some readers will say that the issue of under-representation is “corrected” through a statistical means known as “weighting.” But this is not true – the vast majority of polls do not weigh results based on political affiliation. Moreover, the weighting of surveys allows considerable freedom to the data analyst to modify the data in a whole variety of ways. As Columbia Professor Andrew Gelman has noted, “survey weighting is a mess.”
The RCP Average is problematic in yet another manner: By aggregating data, RCP is doing a “meta-analysis” – a research method to pool data from different sources to estimate the true average. But, unlike scientific meta-analyses, RCP fails to disclose the criteria it uses for choosing which polls to include in the Average. This means that RCP can include whatever numbers it likes, thus allowing RCP unprecedented freedom to shape the perception of public opinion.
I contacted RCP elections staff as well as top executives at RCP for clarification. More than a week later, I have received no response.
Does RCP have a bias? Tom Bevan, Co-Founder and Publisher of RCP, has already written-off the candidacy of Governor Gary Johnson, as evidenced by his mocking of the “pop quiz” on Aleppo.
If he openly mocks and dismisses Governor Johnson, can Bevan be counted upon to include polls that would fairly represent the Governors’ core voters in the RCP Average?
Don’t trust the Real Clear Politics Average. It’s rigged in favor of the establishment, Democratic candidacy of Hillary Clinton.
* Rodolfo Cortes holds a Ph. D from Stanford University. He is committed to ensuring fairness in elections.