Navigating Norms: Making Sense of Products in Contested Markets

Much recent research investigates how firms’ actions are structured by the social factors – networks, institutions, and cultural schemas – that shape understandings of their products.  We contribute to this work by examining how cultural schemas – e.g., social norms, professional mores, and economic ideologies – influence how firms describe their products.  We argue that while objective constraints can influence product descriptions, that influence depends on cognitive schemas, which determine what product classifications and evaluations are available and acceptable.  This is especially true in legally and morally contested markets, where firms cannot simply “do business,” but rather risk being viewed as deviant – even evil – and must counter these views by working extra hard to legitimate their products.

We build on theories of norms and stigma to predict relationships among product descriptions, cognitive schemas, and objective product characteristics.  We argue that where contestation is intense, firms are highly attuned to local legitimacy concerns.  To counter local opponents, firms must describe their products in anodyne terms – terms that are congruent with broad (beyond local) social norms and connected to socially valued outcomes.  In contrast, firms in (relatively) uncontested markets have the leeway to describe their products in terms that are sui generis – terms that are derived from objective characteristics.

We study firms in rapidly growing but contested markets for marijuana in the United States, focusing on three states (Colorado, Oregon, and Washington) that legalized marijuana through ballot initiatives (Colorado and Washington 2012; Oregon 2014), where we can observe social norms in local detail.  Marijuana products have measurable objective characteristics:  each marijuana strain has a distinctive genetic lineage, which produces a distinctive morphology and physiological effects.  Yet because those physiological effects are multiple and ambiguous, users can interpret them in myriad ways.  Thus, perceptions of marjiuana also depend on what users learn from others about how to classify and evaluate effects.  In turn, which classifications and evaluations of marijuana are available and acceptable depend on social norms about this product.

We test our predictions using textual and pictoral data from industry websites, conducting dyadic analyses of product descriptions.  We measure product descriptions using (1) cosine similarity and (2) a combination of Word2Vec and Doc2Vec.  Cosine similarity calculates the overlap between words in a pair of texts, by turning each text into a vector of words and calculating the angle between the two vectors.  Word2Vec considers nearby words in its calculations of similarity between words; Doc2Vec extends Word2Vec to calculate similarity between entire documents.  We assess social norms at the local (county) level using behavioral data:  electoral support for legalizing marijuana.  We capture objective characteristics of each marijuana strain (among ~4,000) using genealogies (the generalized kinship coefficient, the probability that a random allele (a gene responsible for inherited characteristics) from each product will be identical because they share a common ancestor), chemical content (THC and CBD, the two main active ingredients), and morphology (features in photographs, using cosine similarity between vectors of features).  Preliminary results support our predictions.