The New Science of Online Persuasion

Researchers are using Google Adwords to test the persuasive power of different messages.

The Web has fundamentally changed the business of advertising in just a few years. So it stands to reason that the process of creating ads is bound to change too. 

The persuasive power of a message is a crucial ingredient in any ad. But settling on the best combination of words is at best a black art and at worst, little more than guesswork.  

So advertisers often try to test their ads before letting them out into the wild.

The traditional ways to test the effectiveness of an advertising campaign are with a survey or a focus group. Surveys are shown to a carefully selected group of people who are asked to give their opinion about various different forms of words. A focus group is similar but uses a small group of people in more intimate setting, often recorded and watched from behind a one way mirror. 

There are clear disadvantages with both techniques. Subjects are difficult to recruit, hard to motivate (often requiring some kind of financial reward ) and the entire process is expensive and time consuming. 

What's more, the results are hard to analyse since any number of extraneous effects can influence them. Focus groups, for example, are notoriously susceptible to group dynamics in which the view of one individual can come to dominate. And there is a general question over whether recruited subjects can ever really measure the persuasiveness of anything.

Then there is the obvious conflict created by the fact that a subject is not evaluating the messages under the conditions in which they were designed to work ie to get the attention of an otherwise disinterested reader.

So there's obvious interest in finding a better way to test the value of persuasive messages. One approach is to use crowdsourcing services such as Mechanical Turk to generate an immediate readership willing to take part. 

But Turks are paid to take part. So the results are no better than those that conventional methods produce, although they are cheaper and quicker to collect.

Today, Marco Guerini at the Italian research organisation Trento-Rise and a couple of buddies say they've found an interesting way round this: to test messages on Google's AdWords service.

The idea here is to use Google Adwords to place many variations of a single message to see which generates the highest click through rate.  

That's a significant improvement over previous methods. The subjects are not paid and make their choice in the very conditions in which the message is designed to work. And the data is quick and relatively cheap to collect.

Google already has a rudimentary tool that can help with this task. The so-called AdWords Campaign Experiments (ACE) tool allows users to test two variations of an ad side-by-side.

But to really get to the heart of persuasiveness requires a much more rigorous approach. Guerini and co make some small steps in this direction by testing various adverts for medieval art at their local castle in Trento.

These guys used Google's ACE tool to test various pairs of adverts and achieved remarkable success with some of their ads. One ad, for example,  achieved a click through rate of over 6 per cent from just a few hundred impressions--that's an impressive statistic in an industry more used to measuring responses in fractions of a percent.

However, this click through rate  was not different in a statistically significant way from its variant so there's no way of knowing what it was about the message that generated the interest.

So while Guerini and co's experiments are interesting pilots they are not extensive enough to provide any insight into the nature of persuasive messaging. That will need testing on a much larger scale.

These will not be easy experiments to perform and present numerous challenges.  For example, the process of changing the wording of an advert is fraught with difficulty. Then there is the question of whether this method is able to test anything other adverts designed for AdWords. It might have limited utility for testing the messages in magazine adverts or billboard posters, for instance.

But the important point is that these kinds of experiments are possible at all. And it's not hard to imagine interesting scenarios for future research. For example, AdWords could be used as part of an evolutionary algorithm. This process might start with a 'population' of messages that are tested on Adwords. The  best performers are then selected to 'reproduce' with various random changes to form a new generation of messages that are again tested. And so on. 

Who knows what kind of insight these kinds of approaches might produce into the nature of persuasiveness and the human mind. But we appear to have a way to carry out these experiments for the first time.

Ref: arxiv.org/abs/1204.5369: Ecological Evaluation of Persuasive Messages Using Google AdWords

View the original article here

Network Science Reveals The Cities That Lead The World's Music Listening Habits

If you live in North America, you'll soon be listening to the music now playing in Atlanta whereas in Europe, Oslo leads the scene, according to a new analysis of global listening habits

The evidence that ideas and fashions spread through society like viruses or like wildfire is compelling. Numerous studies have examined the networks in which this spread takes place and with increasingly large data sets to work with, researchers have become increasingly confident in their network-centric view of the world. These tools are teasing apart the large scale behaviour of humanity in ever increasing resolution.

In the fashion world, London, New York and Paris are generally considered the leaders that everyone else follows. So an interesting question is whether network science can tell us which cities play a similar role for music.

That's exactly the question that  Conrad Lee and Pádraig Cunningham at the Clique Research Cluster in Ireland set out to answer by analysing data from Last.fm, an social website for music.  

Last.fm is interesting because it publishes lists of the most listened to artists divided geographically. So if you live in Seattle, for example, you can see what people in your area are listing to.

So Lee and Cunningham have studied the way these charts vary in time and looked to see whether some cities consistently lead others in terms of listening habits.

These guys studied the Last.fm data for 200 cities around the world dating back to 2003. This is compiled from some 60 billion pieces of data that Last.fm collects from its users.

This is a noisy data set. Some cities have so few listeners that their data is hard to distinguish from noise.  So Lee and Cunningham have to apply some fairly robust cleaning techniques to remove this noise. 

They then use recently developed statistical techniques to decide which cities lead others. They then construct  a network in which a link pointing from one city to another indicates that one follows the other.

The results are interesting. They show that certain cities appear to lead others for various genres of music. For example, Montreal seems to lead North American in indie music listening habits and the leader for hip hop is Atlanta. In Europe, Paris leads for indie music whereas Oslo leads for music as a whole. 

There are other interesting patterns too. For example, cities that have similar listening habits are not linked in this network. For example, Portland and San Francisco; Cracow and Warsaw; and Birmingham and Manchester.

Lee and Cunningham suggest that when two cities' listening habits are synchronised there is little to be gained from following the listening habits in the other city so residents look elsewhere.  

There's another interesting pattern. It's easy to imagine that the biggest cities ought to be those furthest ahead of the curve because they have biggest populations from which new and interesting bands can emerge. That doesn't seem to be the case in this data--big cities such as New York, LA and London do not lead. "We find only weak support for this hypothesis," say Lee and Cunningham.

That may cause some alarm bells to ring.  An interesting body of work has recently suggested that big cities benefit disproportionally for their size since qualities such as efficiency, productivity and innovation all scale super linearly with population.  

An important question for Lee and Cunningham is why that doesn't hold for music too.    

There is also a question over whether the trends that Cunningham and Lee have found really reflect their hypothesis that some cities' listening habits lead others. Humans are notoriously good at finding patterns in random data. 

The ultimate test, of course, is whether their discovery has any predictive value. For example, could they predict how listening habits will change in the near future? "We have not yet demonstrated that our models have this predictive power, although we plan to attempt this validation in future work," they say.

So we must wait and see. If they manage any kind prediction based on this work, it'll be an impressive feat.

In the meantime, if you want to know what you're likely to be listening to in the near future, cough, tune in to the music now playing in Atlanta and Oslo. 

Ref: arxiv.org/abs/1204.2677: The Geographic Flow of Music

View the original article here