In November, I published a column in the Washington Post‘s Monkey Cage blog about the media’s coverage of the Conservative party’s leadership race. In it, I argued that the they were granting disproportionate coverage to Kellie Leitch due to her embrace of a controversial policy of “values screening” for perspective immigrants. In doing so, the media elevated her from an also-ran to a contender in the span of a few short weeks. Why? In order to serve well documented biases in the media towards drama, conflict, and narrative. They really wanted to point to a “Trump of the North.” So many clicks were to be had.

This troubled me because there are good theoretical reasons to suspect that media attention to candidates in leadership races matter for the outcome. In general elections, voters can use relatively simple cues in their political environment (party labels) to sort themselves into one camp or another based on their interests and values. They vote accordingly. This choice is made even easier by the fact that the political environment is structured in a way that limits the number of choices for voters (especially with a FPTP system) and that these voters tend to have affect-anchored attachments to political parties. Media effects are likely to be at the margins – among those attentive to the campaign, but who are not strongly predisposed to any particular party.

Things are trickier in a leadership race. All candidates share the same party label, and usually there aren’t major policy gulfs between candidates. Things hinge a lot more on whether a candidate’s voice is even heard. This is particularly true in two contexts. First, as the candidate field widens (for obvious reasons). Second, as voting rules loosen (think of the move from delegates leadership races to one-member-one-vote, and now to primaries). Individual voters will have less prior knowledge of the candidates and will be less likely to be substantively contacted by leadership campaigns and volunteers (spam email aside) as the voting pool grows beyond party activists. The media has to make a choice as to who to cover and who not to cover, and this choice has greater implications when the choice context becomes more complex and voters become less able and motivated to sort through that complexity.

We observed the enormous implications of that choice in the Republican primaries. All registered Republicans and independents (depending on the state) can vote in a state’s primary (setting caucuses aside). This has been the case since the McGovern-Fraser reforms that were later adopted by the GOP. Early, prescient, critics of these reforms argued that they would strengthen candidate-centered campaigns and weaken party elites with the consequence of undermining their ability to vet candidates for quality and electability. But for a time it appeared the parties remained in control. The authors of the seminal work The Party Decides, argued that this was because the party machine generally decided who to throw their organizational weight behind in the Invisible Primary, well before votes are tallied in Iowa and New Hampshire. For a number of reasons I will not go into, GOP elites lacked the ability to decide in 2016, and voters were left with an enormous field of candidates and a whole lot of confusion.

As soon as Trump tossed his hat into the ring, the media decided they would turn the Republican primaries into the “Donald Trump Show”. They did this despite the fact he was simultaneously not a conservative, not really a Republican, and was initially badly trailing in the polls. The scale with which the media distorted coverage towards Trump is truly shocking, and it immediately led to skyrocketing poll numbers. Many observers at the time believed that Donald Trump’s courting of outrageous controversy was a masterstroke designed to suck the oxygen out of his opponents’ campaigns. Turns out it was just Donald being Donald. He accidentally manipulated the media to perfection, and it played a very big role in him winning the primaries. He cut through the noise of a crowded field with a campaign based almost exclusively on earned media. He was able to win despite trailing woefully behind his competitors in ground organization, fundraising, and ad spending – the traditional hallmarks of a good campaign.

Could this happen in Canada? Yes and no. On the one hand, there is no theoretical reason to expect the media in Canada would fail to jump at controversy for clicks. All it takes is a candidate with little shame to make full use of that media bias. On the other hand, the rules of the CPC leadership race guard against an out-sized media effect on the final outcome. Ground organization is paramount to sell memberships and maximize accumulation of the points available, which are spread out evenly across ridings. Parties are much stronger in Canada than in the United States.

So it is not surprising that certain leadership campaigns, Leitch only the most prominent among them, tried to ape Trump to cut through the noise of a similarly crowded leadership field. When I wrote my initial post in November of 2016, she had successfully crowded out most of the other candidates and their policy proposals. This led to a jump in her poll standing. The media appeared to be unwittingly placing their thumb on the scales of the leadership race.

But then Kevin O’Leary happened in January. If anyone was going to out-Trump Leitch, it would be him. He had the penchant for controversy and a B-list television star persona. However, to his credit, he refused to turn alt-right. O’Leary rocketed in the polls, and Leitch dropped. But then he unexpectedly dropped out of the race in late April and endorsed Maxime Bernier, perhaps realizing he needed to actually be in Canada in order to win a leadership race for a major Canadian political party – a cost he was unwilling to bear. It then went down to the wire in round 13 between Scheer and Bernier.

So what exactly happened in media coverage over this last half of the campaign? Did the media’s obsession with Leitch turn into a similar obsession with O’Leary? I replicated my analysis in the Monkey Cage piece with media data from the remainder of the campaign.

Data and Analysis

I gathered Canadian newspaper sources on LexisNexis and pulled any article that made reference to the Conservative party leadership and any one of the candidates.  I then used the automated content analysis software Lexicoder to count candidate last names and significant policies in news coverage (e.g. values screening, abortion, transgender rights, carbon taxes, supply management, etc.).  I coded an article as counting towards a candidate’s share of coverage if their last name was mentioned more than once to avoid including articles featured merely passing references to a candidate.

In Figure 1 below I plot the total amount of coverage over the course of the campaign (Panel A). Media coverage took off around September with Leitch’s announcement of values screening. We see another spike in the build up to, and aftermath of, O’Leary’s entrance into the race (January 17th), but coverage slid back to its equilibrium shortly thereafter.

Panel B provides a plot of candidate share of coverage, with Leitch in red, O’Leary in orange and an average of the top-two finalists, Scheer and Bernier in blue. I excluded lower-tier candidates. Much like in the leadership race itself, adding them in would be messy and ultimately pretty pointless (just kidding!…sort of). Kellie Leitch came to dominate leadership coverage after her values screening announcement in September, but by December, as rumblings about Kevin O’Leary’s possible entrance grew, she faded fast. Leitch’s coverage temporarily collapsed after he entered the race and did not recover. O’Leary came to dominate coverage, much like Leitch did before him. The media traded in their “Trump of the North” creation with perhaps one a bit more authentic. It is only after O’Leary dropped out of the race at the end of April when the media started paying attention to the two finalists on the last ballot, Scheer and Bernier.

Sadly, Leitch’s decline did not open up space for more policy coverage. Panel C plots policy share of coverage over the course of the campaign. Values screening decisively crowded out discussion of other policies, such as those advanced by policy wonks Maxime Bernier and Michael Chong. Policy coverage did not quite recover after the media’s obsession with Kellie Leitch subsided. The CPC leadership race became the “Kevin O’Leary Show” until he dropped out of the race at the end of April.

Even more concerning than the share of coverage received by Leitch, and later O’Leary, was the fact that the media’s obsession with Leitch was driving overall coverage of the race. The media gave her a greater share of coverage than her rivals and a megaphone with which to engage in Trump-ian politics. Interestingly enough, O’Leary’s entrance into the race did not change this dynamic. Total coverage of the race continued to ebb and flow with Leitch’s share of coverage with a notable exception around the period where O’Leary entered the race. This can be seen below in Figure 2, which plots the total amount of coverage of the leadership race (measured with # of daily articles) with each major candidate’s share of coverage (%), as well as noting the correlations between them. The inability of O’Leary to drive total coverage could be because of his refusal to endorse Trump-ian policy even if he did mimic Trump’s style. It is hard to say for sure.

What was the extent of the relationship between Leitch’s coverage share and the total volume of leadership coverage more concretely? I built a time series count model to answer this question using total daily coverage as the dependent variable and the share of coverage for Leitch, O’Leary, Scheer, and Bernier as the independent variables. More details can be found in the notes below. The results are shown below in Table 1. Regardless of model specification, coverage of Leitch stands alone as having a significant association with coverage volume. Moving between no daily coverage of Leitch to 100% of daily coverage is expected to increase the volume of coverage between 50 and 80%. This dwarfs similar estimates for O’Leary (10-16%), Scheer (8-20%), and Bernier (0-14%).

Table 1. Model Results

Graphical depiction of the result for the negative binomial model is shown below in Figure 3. Going from no daily coverage of Leitch to 100% of daily coverage is predicted to be associated with a rise in daily coverage from about 5.5 articles to just under 12. This increase of 6.5 articles per day represents about 32 percent of the effective range of the daily coverage variable (between 5th and 95th percentile). There is little doubt that coverage of Leitch was tied very closely with the attention paid by the press to the race as a whole. Adding interactions to the model for O’Leary’s presence in the race doesn’t appear to change this story. It was as true when he was in the race as it was when he was not.

In sum,

1. Leitch and her values screening policy dominated the middle stages of the CPC leadership race. She successfully crowded out substantive policy debate and coverage of her competitors.
2. Leitch’s share of coverage is closely tied with the total amount of coverage the media dedicated to the leadership race. The media were only attracted to the race by the controversy engendered by Leitch’s candidacy.
3. O’Leary’s entrance into the race appeared to crowd out Leitch, but for whatever reason he was not able to drive media interest in the leadership race – perhaps because of his lack of willingness to embrace controversial policies targeting immigrants.
4. Scheer and Bernier only began to attract significant media attention after O’Leary dropped out of the race.

Why does this matter?

In the end, Kellie Leitch was crushed in the election. Kevin O’Leary sucked the oxygen out of her campaign and she never recovered. O’Leary, for his part, dropped out. The candidate of the status quo, Andrew Scheer, wound up winning. He did this despite being virtually ignored by the media until the final few weeks of the campaign. In any direct sense, the media cannot be seen as responsible for the outcome.

However, we do not know how the race would have played out if the media had not spent the bulk of its time focusing on two fundamentally unserious candidates.  For instance, it is clear a part of Bernier’s strategy was to command media attention with the continuous roll out of rather bold and controversial economic policies. Before Leitch’s values screening announcement, he was rather successful. If the media had not entirely shifted its focus to Leitch, would that have given him the time he needed to consolidate enough support to beat Scheer? Hard to say. Would the tenor of the race have changed if the front-runners weren’t forces to respond to every crank comment made by Leitch? And, perhaps if they were instead made to respond to more constructive voices that were almost completely excluded from media coverage (sorry Deepak!)? Would this have affected the outcome?

Thinking through the counterfactual is difficult, if not impossible. But it is probably safe to say, at a minimum, that media coverage shaped the dynamics of the campaign. With a different race absent such media intervention, it is entirely possible there would have been a different outcome.

This problem isn’t disappearing anytime soon. The alt-right is not going anywhere. Ezra Levant will see to that. Trump and Leitch have provided a playbook for future leadership aspirants to play the media for national exposure and a campaign boost. The Canadian media, much like their American counterparts, need to grapple with the fact that there are real consequences in chasing clicks by giving finite air time to those with odious views at the expense of other, more constructive voices. As political parties in Canada move ever closer to the American primary model of leadership contests, the media’s role in the outcome of such races is liable to grow. It is essential that the media gets it right in the future.

Notes

1. There is evidence of over-dispersion, so I used a negative binomial count model as a first cut. However, there is memory in daily media coverage (AR(1)). Adding a lagged dependent variable is necessary, but not quite right. I also used a more conservative Poisson Autoregression (1) model as a robustness test. There is indeed some upward bias, but it does not change the story being told (i.e. Leitch drives coverage volume).
2. Other notes: robust standard errors were added to correct for heteroscedasticity; results are robust to the addition of liner and quadratic trends; F-tests ruled out the need for higher-order lags of the independent variable. There is no evidence to suggest the errors follow anything other than a white noise process.