Crypto-punditry and the media neutrality crisis

ABSTRACT

This article describes the corrosive practice of “crypto-punditry,” whereby subjective analysis is smuggled into media coverage under the guise of objective reporting. In search of a neutral basis for analyzing events, journalists latch onto public opinion. However, this opinion is rarely expressed directly. Instead, reporters engage in speculative assessment about what public opinion might be. Unfortunately, in an effort to shield themselves from charges of bias, journalists have triggered a counterproductive autoimmune practice which targets precisely those elements of descriptive reporting that give it strength: fairness, accountability, public interest. To make this argument, I first develop a theory of crypto-punditry and outline its effects. I then use content analysis to show that this practice has become far more common in political reporting over the past decade. I conclude by using the 2016 US presidential election as a case study to illustrate its effects.

Disclosure of potential conflicts of interest

No potential conflict of interest was reported by the author(s).

Coding Rubric for Crypto-Punditry Analysis

Issue 1: Predictive claim

Does the author state or imply a possible change in public opinion?

Note 1: Do not include predictions of likely outcomes based on the status quo (i.e. – statements about who is currently more likely to win an election).

Note 2: Only code YES if the predictive claim arises from the author. Many articles contain direct quotations from campaign affiliates attempting to spin the facts. Do not count these as authorial predictions unless the reporter provided additional commentary or framing to suggest that these claims possess independent objective validity.

Issue 2: Evidentiary support

If polling evidence is cited to supportive this predictive claim, how closely does the polling relate to the event being discussed:

STRONG: the polling cited is directly on-point (it concerns the specific event under discussion)

WEAK: the polling cited is related to the event but is not directly on-point (it concerns issues similar in nature to the event, but which are not identical

Example: An article suggests that a health care pledge which would require raising taxes would be unpopular, but only cites polling that voters dislike paying higher taxes, with no polling on the specific question of taxation to pay for health care.

If shoe-leather reporting evidence is cited to supportive this predictive claim, how comprehensive is the evidence:

STRONG: The claim is supported by substantive reporting—consulting more than one source, drawing from a non-biased sample, speaking to members of directly-affected constituencies.

WEAK: The claim arises from one source, or only one type of source and/or the sources cited are likely to be biased or appear to have been cherry-picked.

Example: an article suggests that raising emissions standards would be unpopular but only includes reporting from coal miners who would be directly harmed by the policy.

If expert opinion is cited to supportive this predictive claim, how strong is this evidence (check all that apply):

STRONG: The expert(s) 1) have genuine expertise in the specific area of the prediction (they are practitioners in this field, academics who study the issue, etc.) and 2) are neutral parties, or parties with no personal incentive to “spin” the facts in favor of a preferred outcome.

WEAK: The expert(s) only meet one of the two conditions described above.

Example: an article suggests that a campaign gaffe will hurt a candidate which only cites political strategists who are affiliated with the opposing party.

Issue 3: Inference requirement

If the article provides any kind of evidence to support the predictive claim, is further inference required to reach the predictive effect or is the evidence commensurate with the size of the predicted effect?

STRONG: The evidence closely matches the size of the predicted effect. No further inference is required.

WEAK: The evidence supports the prediction, but requires speculative inference to reach the conclusion.

Example: an article might offer strong polling evidence that a policy is popular, but then use that claim to speculate that adopting the policy might swing the election.

If the article contains multiple predictive claims, repeat this process for each separate claim.

Notes

¹ A fourth option was provided for evidence that did not fit into the three categories, but no claims depended on this categorization.

² To assess the intercoder reliability, 10 articles from each year were evaluated separately by two coders. When assessing whether a new article included a predictive claim, the unweighted Cohen’s Kappa was 0.928 (p < .001). Where both coders assessed Evidentiary support and Inference requirement, the unweighted Cohen’s Kappa is 0.922 (p < .001) and 0.928 (p < .001), respectively. These statistics indicate a high degree of intercoder reliability.

³ Results are robust to the use of logistic regression.

⁴ Results are robust to the use of logistic regression.

⁵ A myriad of contributing factors contributes to Trump voters’ unwillingness to accept the legitimacy of negative coverage, the study of which necessarily falls beyond the scope of this article. However, while crypto-punditry is not the cause of this effect, its inability to effectively respond to this set of facts constitutes an escalating factor.

⁶ David Fahrenthold at the Washington Post received a deserved Pulitzer for his dogged investigation of the Trump Foundation, while the New York Times released a bombshell report in early October 2016 about Trump’s tax record. Both were excellent examples of strong investigative journalism.