Is that Tracker Really Moving?

Brand trackers bounce around frustratingly from month to month. We discuss what causes these movements and detail strategies for seeing through this noise to the underlying image movements.

To Oscar Wilde, “anyone who lives within their means suffers from a lack of imagination.” But to most shoppers, weighing up a store’s prices against the width of their wallet, and against the prices charged by rival stores, is a key factor in where they shop.  That’s why we’ve never seen a brand tracker that didn’t include price image, typically measured using something like “they have great prices” and then reporting the number of people who “Agree” or who “Strongly Agree”.

But here’s the problem: such metrics bounce around, as illustrated by the figure’s dashed lines.  This often leaves insight teams in the awkward position of having to explain movements that are actually measurement noise.  By definition it’s hard to defend such a post-rationalisation and you can easily end up being contradicted by the next wave of research and looking like you made up the answer (which, to be fair, you did).  So how can you see through the noise and identify the underlying truth?

Cleaning Up Retail Brand Images

Part of the problem is that there isn’t a single cause.  For example, women and existing customers tend to provide kinder responses, so trackers will go up when that month’s sample includes more of these respondents and vice versa.  Likewise, trackers wobble with systemic so-called “sunny day” effects, where everyone’s responses to all the questions across all the competitors move gently up or down, like boats all rising or falling with the tide.  Whilst there are techniques to address these problems, such as re-weighting samples and three-month rolling averages, they are comparatively weak solutions.

The graphic shows an alternate approach in grocery.  Without going into details, techniques such as ipsatisation, ordered probit, kernel smoothing, and factor analysis can be used to extract the underlying tracker timeseries.  They can also be used to show that when it comes to coining head-scratching jargon, the world of statistics displays an inventiveness equalled only by the rock star-minted names of celebrity children.  Surely, Durbin-Levinson Recursion gives Princess Tiaamii Crystal Esther Price a run for her money.

Anyway, intuitively these techniques work by efficiently using all the response ratings to a question, contrasting information from multiple questions, controlling for both personality and demographics, optimally blending periods and so forth.  The graphic demonstrates the effectiveness of this approach.  Adopting this strategy you can then continue reporting an easy-to-communicate “top box agreement” measure to management whilst simultaneously knowing what’s really going on, adjusting your “voice-over” accordingly and sleeping soundly that night.