Making the Lead: Article and Contextual Correlates of Front-Page Coverage within the Immigration-Crime Debate

Abstract

While an abundance of literature demonstrates that immigration is associated with lower rates of crime, public opinion expects the opposite. There remain many reasons for such incongruity, including crystalized political ideologies and structural barriers to assimilation. The current study examines another important dimension: media narratives among the most prominent immigration and crime news articles. Specifically, we explore how news outlets narratively describe or “frame” the link between immigration and crime and, in turn, how such frames influence the prominence of news stories. Using content analysis of over 3,800 articles from 2008 to 2012 geo-located and paired with a host of macro-level data, multi-level models reveal that (1) one-third of local stories describe immigration as crime-increasing; (2) articles that link immigration to rising rates of crime are more likely to appear on the front page of newspapers, as are stories describing immigration’s impact on the justice system or the rights of immigrants within it; and (3) articles published in places with lower rates of crime, higher median household incomes, and smaller foreign-born populations are more likely to feature on the front page. We conclude with implications for ongoing public policy debates and research on immigration and crime.

Notes

1 Alternative search strategies using other words and phrases produced nearly identical results (between 90 and 95% overlap).

2 A review of our data reveals that (a) less than 5% of our papers have what might be considered large online circulations (see also our supplemental models that control for circulation in multiple ways), while (b) there appears to be no specific partisan lean of those papers that have larger online circulations when we examined media bias rankings (see https://www.allsides.com/media-bias/media-bias-ratings). While it is beyond the scope of the current study to fully disentangle such differences across papers, this issue warrants attention in future research that could better articulate how local papers might actually reflect non-local conditions in news decision making.

3 These 220 counties represent every state but one (Arkansas). While states like New York (14%), California (13%), Texas (7%), and Arizona (6%) that have large immigrant populations provide the largest number of unique counties and articles in our data, sizeable shares of articles also appeared in counties, for example, in Georgia and Massachusetts with somewhat smaller immigrant populations. Overall, our sample is dispersed and not disproportionately composed of articles from a single geographic area.

4 The coding of this frame occurred when an article did not clearly employ one of the frames described above. For example, some articles described immigration and crime alongside a host of other potential social forces (e.g., concern over immigration, crime, and rising healthcare costs) without pointing to a specific direction of association (e.g., criminogenic, protective). When we could not (with a strong degree of certainty) code an article as employing one of our six primary frames – or when new frames were not clearly defined – we coded them as ambiguous. In this manner, we view our content analysis as conservative because we used strict criteria for inclusion into each the most substantively meaningful framing categories.

5 The majority of articles employing multiple frames include the “civil/legal rights” frame in combination with another of our substantive frames (approximately 92% of the articles with multiple frames). In nearly every instance, the civil/legal rights narrative is juxtaposed against the other frame (e.g., immigrants are victims of crime but the justice system has yet to settle how specific rights of the foreign born are to be handled).

6 In these multi-level models, we center both level-1 and level-2 variables on the grand mean. This technique transforms the model intercept into the expected value of the dependent variable when the predictors included are set to the sample mean (rather than zero, as is the case in a model with un-centered variables). Centering on the grand mean also adjusts for compositional differences within individual counties by recognizing that how level-1 variables influence the outcome may be contingent upon the context in which they appear. In other words, these grand mean centered models allow us to interpret level-2 effects as controlling for level-1 variables, and thus as true “contextual” effects.

7 Given the rarity of our dependent variable, we caution that our models may not contain enough variation to leverage or adjudicate each specific relationship (i.e., other correlates might emerge as statistically significant, as well). Perhaps more importantly, those cases providing the rare outcome (i.e., front page articles) could unduly influence the model if they disproportionately share underlying characteristics. To address this, we estimated a series of penalized logistic models using Stata’s firthlogit command. Except for small changes to the magnitude of the relationships for several control variable, the general pattern of our findings remained remarkably stable: compared to the “protective” frame, articles using the criminogenic and civil/legal rights frames were more likely to appear on the front page. Finally, for parsimony, we estimated multi-level logistic regression models that omitted correlates that were non-significant across our models and for which there were closely correlated and theoretically related measures already in the model. Again, we found substantively similar results as those displayed in our tables.

8 We caution that the AAM data suffer some degree of missingness in that the full count of all available publications in each county and the readership/circulation of papers appears to be biased by the relative size of the county itself. For example, we found circulation and publication data to be (a) non-existent for many rural locales, (b) outdated in many smaller communities, or (c) missing papers even in some of the more metropolitan counties.