AMA research shows how to cut noise, sharpen trend insights
New analysis techniques for deeper trending insights
Research in the American Marketing Association's (AMA) Aug. 2012 edition of the Journal of Marketing Research provides a template for marketers to better understand trend speed and acceleration while filtering out various sources of noise, says Professor Wagner Kamakura of Duke University, one of the study's coauthors.
The research enables marketers to go beyond identifying spikes in search traffic and social media buzz by applying structural dynamic factor analysis (SDFA) to complex, long-term and multi-dimensional Web data. The new techniques enable marketers to control for seasonal variation, identify the impact of both positive and negative shocks, and better understand competitive consequences—who gains and who loses as a result of changing trends.
The article describing this approach, "Quantitative Trendspotting," was written by Duke's Kamakura and Professor Rex Du of the University of Houston. "We looked at search patterns and social media patterns over time to identify trends behind the trends," Kamakura says. "This trendspotting technique can be used to understand how brand image shifts on a weekly basis, or find the reasons behind trends in consumer sentiment."
As a proof-of-concept, the authors analyzed Google search data, as well as sales figures for dozens of international automotive brands from January 2004 through September 2010. The automobile market provided rich fodder for analysis because of its myriad market segments and major systemic shocks in recent years, such as the bankruptcies of General Motors and Chrysler, the shutdown of GM brands Pontiac and Saturn, and the U.S. federal government's Cash for Clunkers stimulus program.
Among other findings in the scholarly work, the authors successfully controlled for seasonal fluctuations in brand interest, as well as the shocks of other key industry events such as Cash for Clunkers. With this data in hand, the pair presented models which could model and predict changes in search trends based on factors such as the unemployment rate and gas prices.
"This trendspotting approach can be applied to anything: sales, prices, or any other data you get on a daily, weekly, or monthly basis with multiple variables that are somehow related to each other," Kamakura says.
In work not published in the article, Kamakura and Du applied these techniques to supermarket sales data and wine trends as well. Substitution effects could be seen in their SDFA analysis on the U.S. wine market, which suggests that there is a concrete correlation between interest in Malbec and Tempranillo wines at the direct expense of interest in Shiraz varietals.
"In the raw data, that's hard to see, because the raw plots for searches on [each variety] show a lot of up-and-down, a lot of noise," Kamakura says. "But people can only drink so many bottles per week, and when we asked, 'who is losing interest?", we saw a negative impact on Shiraz."
If widely adopted, the SDFA approach could give marketers greater confidence to act on true variations in interest and demand, rather than responding to seasonal variations or one-time shocks. Because the models are predictive as well as backward-looking, they can be used to project consumer response to factors outside a brand's control as well.