Opinion Leadership and Social Contagion in New Product Diffusion
The authors distinguish the difference between self-identified influencers, and sociometric influencers. As a result, Iyengar, Van Den Bulte, and Valente, hereafter referred to as IVV, further our knowledge about product adoption and influence.
A self-identified influencer is one who raises their hand and indicates that they consider themselves to be an influencer. IVV acknowledge the point that self-identification may be a reflection of self-confidence more than actual influence.
Sociometric influencers are those that are predicted to be influential based on observed features. IVV defined it as where a given person was at in the social referral graph relative to others – their indegree centricity. They predicted "better connected adopters exert more influence than less connected ones” (IVV, 2010).
The study examined drug adoption among physicians in San Francisco, Los Angeles, and New York City. They asked physicians to list 8 other physicians that they felt comfortable discussing new drugs with. From this data they constructed a social referral graph. They used this data, along with ordering information, to infer contagion. (Contagion, unpacked, loosely means ‘to adopt a product’. A lot of language in social analytics is adapted from epidemiology during the 1950’s – causing the terms ‘viral’, ‘contagion’ and ‘s-curve’ to enter.)
Their findings are best summarized by Godes, himself, in his commentary below.
"(1) Contagion exists even after controlling for many possible confounds including, significantly, marketing efforts; (2) contagion is moderated by the recipients’ self-reported opinion leadership: opinion leaders are less responsive to the actions of others; (3) this is not true of sociometric leadership (indegree); (4) contagion is moderated by usage: adoption by a heavy user has a bigger impact on her ties’ decision to adopt; and (5) time of adoption is increasing in both indegree and self-reported leadership: influentials adopt earlier.” (Godes, 2010).
Many may remember Jim Novo’s review of a major paper – Godes and Mayzlin (2009) Firm-Created Word-of-Mouth Communication: Evidence from a Field Test. Many of the themes in that paper are expanded upon in the IVV paper.
Godes was invited to comment on the IVV paper and he made several important statements.
First, Godes believes that we’re well past testing the assumption that contagion exists, and that we should be focusing on explaining the effects of awareness versus persuasion. Going one step further, he asks in which product domains does one dominate over the other – if at all. Second, he believes that we’re well past testing the assumption that influence varies among individuals. We should be aggressively inquiring as to why this is the case. He asks which customers are more susceptible to contagion and why. He believes that more work should be done around how and if firms are able to identify and target such influencers.
Web analysts are increasingly asked about influence and influencer identification – in no small part because of social media and the increasing interaction with the traditional bastion of influencer identification: Public Relations firms. There is a vital marketing analytics aspect to influencer identification that should be considered and web analysts need to pay attention.
Web analysts are typically concerned with observed behavior that occurs on the site. Many web analysts are concerned with the term ‘awareness’ which occurs, typically (and perhaps by some definitions) before the first click to the site. The domain of Influence, in many respects, is a form of pre-click analysis.
This study is the first to really make the differentiation between a self-reported opinion of being an influencer (which is a self-reported prediction and/or condition), and a prediction of influence based on what could ultimately be observational data (sociometric influence identification).
The IVV paper is very controlled and complete and their findings are important.
For one, self-identified people are harder to convince and convert. For two, they’re typically wrong about being actual influencers (correlation of 0.19 within the referral network).
This study examined product adoption over time, and adds to what Novo argued before. Novo’s emphasizes:
- That your most loyal customers are bad targets for social awareness programs because it is highly likely that their friends are already aware. You're preaching to the choir.
The IVV study emphasizes:
- Adoption by a heavy user has bigger impact on others decision to adopt than a light user.
- Influentials (both sociometric and self-reported) adopt earlier.
Whereas it may appear at first that the IVV findings are contradictory to Godes and Mayzlin, they’re quite complementary. The key variable is market saturation. I’ll repeat a common definition of ‘market’ as "a group of people that are self-referential in making a purchase decision”.
The strategy for introducing a new product into a market should differ from gaining market share in a mature market. In a mature market, chances are that influential early adopters have already championed the product. The influence, in a manner of speaking, has been exerted. That is not to say that they are worthless to the firm. They’re worth a lot. Their loyalty and affinity should be managed. They should not be targeted in a social exhortation campaign for the same product, because as Novo has pointed out, their ties are already likely aware.
Identifying influencers is important when introducing a new product, especially since influencers tend to be early adopters. Moreover, the marketing requires an investment, as the new product has not yet established a cash flow. This is an analytical domain where intelligence can make a difference.
The manner in which influence is assessed matters. Self-identified influencers are more stubborn and self-important. Those who are less self-confident might be likely to engage in social influence ponzi-schemes and game certain algorithms. Indeed, we see this all the time with the pathetic wholesale purchasing of blocks of 1000 twitter followers.
The attractiveness of sociometric algorithms is enhanced because they are less vulnerable to gaming and, as IVV’s findings suggest, more likely producing better results.
As adoption begins to pick up, the higher volume customers are likely to be more influential. On this point I would caution that their influence may be a function of multiple factors – for instance, a high frequency user of web analytics software at a large company is more likely to be sought out for advice, and referred to, than a low frequency user at a much smaller firm (though, this is not always the case!). IVV mention the same with respect to physicians and clinic intake volumes. In the meantime, until there is additional evidence, volume is an adequate proxy for generating influencer lists in maturing (not matured) markets.
As the market saturates, the role of Influencer Identification for tactical targeting and adoption ought to recede. The process may very well repeat itself as a new market is targeted, influencers identified, and saturation in that market begins anew.
Another consideration, emphasized by Novo, lies at the boundary between awareness and persuasion. I invite more research on this boundary – as the latter is more complex and then former.
I recommend the IVV paper, followed by the Godes review. I wrote the word ‘amazing’ multiple times in the margins, and was sure to say thanks to Godes for considering the practitioner in his review. The findings are important, and offer a springboard for web analysts to begin making contributions in their own right.