The Media Is Lying: Trump is Actually Beating Clinton in Polls-Mostly Fiction!

The Media Is Lying: Trump is Actually Beating Clinton in Polls-Mostly Fiction!

Summary of eRumor:
The media is lying about the presidential election by oversampling Democrats in polls that show Hillary Clinton leading Donald Trump.
The Truth:
There are valid arguments to be made about the accuracy of polling data, but claims that “the media is lying” don’t seem to check out.
Rumors that the media was intentionally skewing the polls in Hillary Clinton’s favor cropped up after polls conducted by CBS News/New York Times and NBC News/Survey Monkey showed Clinton with a 6- and 7-point lead over Trump, respectively.
The conservative website American Thinker argued that CBS News/New York Time had skewed its polling data by oversampling Democrats to incorrectly inflate Clinton’s support:

However, based on the polling details, the final weighted sample of 976 registered voters is made up of just 28 percent Republicans and 35 percent Democrats.

In the unweighted sample of registered voters, the relative percentage by party was 29 percent Republicans and 35 percent Democrats.

So, during the weighting process, the poll increased the Democrat-Republican spread from 6 percent to 7 percent. This relative weighting should have been headed in the other direction.

The Daily Mail made essentially the same argument about an NBC News/Survey Monkey poll that showed Clinton had opened up a 7-point lead over Trump:

But the weekly tracking poll, from NBC News and the Surveymonkey company, included the opinions of 7 per cent more self-identified Democrats than Republicans – the same margin as the poll’s topline result.

Again, there are valid questions about the accuracy of these two polls. The specific argument that these polls are inaccurate because Democrats have been oversampled, however, doesn’t hold up. There are two main reasons for that.
First, polls with large sample sizes are adjusted, or “weighted,” to more accurately reflect the electorate. Given that more people identify as Democrats than Republicans, polls are intentionally weighted so more Democrats are included than Republicans to more accurately reflect the population.
Different polls show Democratic-leaning voters outweighing Republicans from anywhere from 5 to 11 percent, depending on timing and polling methods.
So, the Daily Mail argument that the poll is misleading because 7 percent more Democrats were polled is flat wrong. The sample is weighted that way intentionally to reflect national political affiliation trends.
And the American Thinker argument that the CBS News/New York Time poll is wrong because it was weighted 1 percent more in favor of Democrats than it should have been is misleading at best.
The issue is that a number of factors go into weighing for party affiliation, so a 1-point swing one way or another isn’t necessarily a red flag that a poll is wrong. It could simply mean that pollsters are taking into account national factors or trends that haven’t yet been reflected in party affiliation data.
PEW Research Center explains the concept of weighting for party affiliation like this:

While it would be easy to standardize the distribution of Democrats, Republicans and independents across all of these surveys, this would unquestionably be the wrong thing to do. While all of our surveys are statistically adjusted to represent the proper proportion of Americans in different regions of the country; younger and older Americans; whites, African Americans and Hispanics; and even the correct share of adults who rely on cell phones as opposed to landline phones, these are all known, and relatively stable, characteristics of the population that can be verified off of U.S. Census Bureau data or other high quality government data sources.

Party identification is another thing entirely. Most fundamentally, it is an attitude, not a demographic. To put it simply, party identification is one of the aspects of public opinion that our surveys are trying to measure, not something that we know ahead of time like the share of adults who are African American, female, or who live in the South. Particularly in an election cycle, the balance of party identification in surveys will ebb and flow with candidate fortunes, as it should, since the candidates themselves are the defining figureheads of those partisan labels. Thus there is no timely, independent measure of the partisan balance that polls could use for a baseline adjustment.

So, in conclusion, claims that the media is oversampling Democrats in polls to make it appear that Donald Trump is losing badly to Hillary Clinton just don’t check out.
But there are legitimate questions about these polls (and all national polls). Fivethirtyeight consistently grades various pollsters and scrutinizes their methods. Click here for a tool to help you see how different polls compare.