Saturday, April 10, 2010

Media’s mendacity should be news to all

Boston Herald article "Media’s mendacity should be news to all" (April 10th, 2010) by Bill O'Reilly discusses individual American's discovery that United States newspapers are politically biased, and appears surprised. Corrections suggests that far from being surprising, this is a natural occurrence, and we should see media bias in a different way.

While many Americans believe the national press is biased toward the left, a more damning charge is now being debated: Are U.S. media outlets actually corrupt? Those who believe they are point to the cheerleading during Barack Obama’s presidential campaign and to the recent reportage on the Tea Party movement.

As you may know, the Tea Party people have been branded in some media quarters as a bunch of racist, far-right loons. TV commentators on MSNBC and CNN have actually called the Tea Party folks dirty names on the air - all in an attempt to diminish the growing influence of the movement.

But a funny thing happened on the way to the gutter. Regular Americans have apparently opted to decide for themselves about the Tea Party, and the polling is interesting.

Corrections suggests that January 2010 Econometrica article "What Drives Media Slant? Evidence from U.S. Newspapers" by Jessie Shaprio and Matt Gentzkow has a more appropriate manner of examining media bias. In order to understand its relevance, we first note that in Shapiro and Gentzkow's 2006 Journal of Political Economy paper "Media Bias and Reputation", they posit consumers who do not know the quality of a news source with certainty. Their consumers have prior beliefs about the truth, read news articles, and sometimes the truth is revealed to them at a later time (so they can update, finding that the newspaper has deceived them, or not). The article finds that newspapers will optimally slant their news to their consumer bases's biases.

Returning to the original paper Corrections referred to, the authors find that, "consumer demand responds strongly to the fit between a newspaper's slant and the ideology of potential readers, implying an economic incentive for newspapers to tailor their slant to the ideological predispositions of consumers. We document such an effect and show that variation in consumer preferences accounts for roughly one-fifth of the variation in measured slant in our sample."

Corrections suggests that the cycle O'Reilly was referring to makes quite a bit of sense, in this light. First, individuals had some signal about Barack Obama as a Presidental candidate. News sources respond to that bias (and perhaps the two feed one another, though that conjecture is by no means clearly going to happen). Individuals vote for Obama, and perhaps discover, given a relatively monotonic downward trend, that they were deceived by media slant. Corrections offers Gallup Approval Rating Polling data below (click to enlarge). They bayesian update on the slant media stations have, just as the Tea Party, borne out of individual's discovery of deception by the media, occurs.

Corrections suggests that Mr. O'Reilly's article was not necessarily off-base, but was grasping at the model suggested by Shapiro and Gentzkow without explicitly mentioning it. It is in this clarifying manner that Corrections offers a clarification.


  1. this is tangential, but have you ever run the numbers on the presidential approval ratings to see what the typical trend is? (it appears to be negative, in that graph from that other post of yours).

    Not only do media fit their coverage to the consumers' biases, but a robust finding from psychology and behavioral economics is that consumers tend to fill in blanks with whatever makes them feel good. In particular, so little was actually known about Obama that it was easy to imagine he was God's gift to America. Sadly, people tend to vote for the candidate they most overestimated the quality of, so we tend to see a falling approval rating as information is revealed as to the true quality.

    I eagerly await the day when prediction market are big and approval ratings are subject to them. If we could bet on the 1-year-out approval rating, then the difference between that and the Gallup poll would serve as a rough "optimism index". (admittedly this is not perfect, intertemporal problems...)


  2. We have run that regression. We offer Presidents since 1941, and their average day-by-day trend (coefficients are percentage points approval per day):

    Clinton: .006
    Roosevelt: .004
    Reagan: .002
    Eisenhower: -0.004
    Ford: -.005
    Truman: -.015
    Nixon: -.016
    Kennedy: -.018
    George W. Bush: -.02
    George H. W. Bush: -.021
    Johnson: -.023
    Carter: -.023
    Obama: -.044

    However, this is, to a large degree, comparing apples to oranges. We are comparing tenures of different length. If instead we compare everyone on their first 443 days (or less), we have the coefficients:

    George W. Bush: .078
    George H. W. Bush: .037
    Roosevelt: .026
    Kennedy: .009
    Clinton: .002
    Eisenhower: -.014
    Nixon: -.016
    Johnson: -.018
    Ford: -.025
    Reagan: -.040
    Obama: -.046
    Carter: -.059
    Truman: -.137

    On average, Presidents lose -.016 points per day in their first 443 days.

    Also, in answer to your question, approval ratings by no means appear to be martingales. Regressing the level change today on the level change yesterday, we get a negative very significant coefficient (-.26).

    To interpret this regression in a straightforward manner, we would say that a 1% increase in approval between two polling terms ago and this term would result in, on average, a .26% decrease in approval this polling term.

    We note that this regression is slightly improperly done, as we treat all polling times equally (in the past, polling times were more distantly spaced). Nevertheless, significance at a p-value smaller than 0.1% is convincing prima facie evidence for Corrections that polling data is certainly not a martingale.

    This would give evidence that people, when responding to current polling questions, are not incorporating their future expectations (which should include significant variables such as this). That said, we shouldn't expect them to.

    We agree that if individuals fail to have rational expectations enter into their polling decisions, then the difference between prediction markets and polling data would provide information on expected disappointment, an expected residual we could term "optimism."

  3. thanks!