How Peer Influence Affects Attribute Preferences: A Bayesian Updating Mechanism
Narayan, Rao, and Saunders examined the impact of peer influence on product attribute preference. It’s notable because this is a first.
The authors found that peer influence causes people to change their perspective on what’s important about a product.
The authors carried out two studies to generate a dataset to demonstrate this effect.
In the first, they asked 70 first year MBA students to fill out a survey about e-readers. They students were asked to assess e-readers attributes and their willingness to pay. Attributes included number of colors of gray, weight, book selection, and magazine availability. Later, they were shown a list of fellow students and asked to rate how likely they are to be influenced by them. ("This student’s choices will not influence my choices at all”; "this student’s choices will very strongly influence my choices.”). They were also asked if they follow each other on a social networking site.
Two weeks later, the students were asked to re-take the survey, but this time, they are shown samples of one another’s preferences.
A second study, executed on 140 students, was similar to first with few differences. They were asked of cell phones instead of e-readers. They were shown fictitious names of influencers instead of their real names. It is not possible to answer the question "did fake peers cause less influence”, as that would have required a different study design.
Because the paper appears in Marketing Science, the authors have to attempt a model that approximates these effects. They opt for a MCMC Chain (Markov Chain Monte Carlo Estimation Algorithm) that approximates how humans change their preferences. Their model is predictive.
There’s one important takeaway for practitioners. And it should be obvious, in hindsight, once you read it.
The extent that somebody will revise his or her opinion about an attribute in light of a peers’ opinion is linked to just how uncertain the person is about an attributes importance.
To expand on that, if John is uncertain about the importance of wireless download speed, he is more likely to be influenced by Jenny’s opinion on the subject. Moreover, the more certain Jenny is in her opinion, the greater John will be influenced by Jenny’s opinion.
Peer influence caused highly preferred attributes to become even more preferred, and least favored brands to become even less preferred. In other words, previously held opinions were reinforced by influence.
The authors are to be commended for breaking new ground and combining traditional conjoint analysis on product features with peer influence. Variation in influence does exist across attributes, and it’s moderated by both the decider’s knowledge and the influencer’s confidence.
The findings are obvious. Prior to this paper, there was no actual evidence linking product attributes to peer influence, and the interplay between the two.
How can you use this knowledge to make analytics better?
Much analysis has gone into what the brand should tell people to think. Comparatively, very little analysis has been done in the way of what people tell each other what’s important about a product’s attributes. Moreover, some market segments, which are groups of self-referential people, are likely to value different product attributes over others. This is particularly important in formulating new product strategy, and crossing the chasm to the early-majority.
The best way I can hammer this point home among the WAA readers is to talk about vendors. Person A may be confidently talking up that tag implementation efficiency is the most important feature of a web analytics vendor. Person B may be confidently pushing the point that variety in custom variables is the most important feature of a web analytics vendor. Somebody who markets web analytics solution, if executing a social strategy, would approach the person most confident in the attributes that is most beneficial to their product positioning. A new entrant may also want to align themselves with those that value the attributes they seek to disrupt upon.
In effect, it’s not enough to ask ‘who influences who’, but rather, ‘who influences who about what’.
The authors make no claims about discovering a natural law. They note the difficulties with their study; specifically that marketing wasn’t factored for over a two-week period. There’s plenty of room for incremental exploration.
Members of the Web Analytics Association who want to learn more about the techniques employed should read the paper.