A Call for Context-Aware Measurement

Context awareness seems to be everywhere, and everyone seems to be saying that context matters.  Gartner foresees "a game-changing opportunity" in what it calls context-aware computing.  The title of their report states it best, "Context Shapes Demand at Moments of Truth."  Their reasoning is straightforward.  They assume that what you want (your preferences) are dependent on the consumption context.  Once they know your context (e.g., from your mobile devices or on-site surveillance), they can feed you information, advertising, promotions and anything else that would help sell their products or services.  This is another sign of the growing acceptance that context plays a determinant role in consumer decision making. If who, what, where, when and why are so important, how do we explain context-free measurement in marketing research?  For example, "What Do Customers Really Want?" is a Harvard Business Review article outlining "best practices" for the measurement of preference.  According to these Bain and Company researchers, rating scales are "blunt instruments" unable to distinguish between "nice to have" and "gotta have."  Thus, instead of rating scales, customers are asked to make a series of trade-off among different sets of four restaurant attributes, such as:

• Food served hot and on time,
• New specials weekly,
• Healthful menu options, and
• Portions are just right.

The authors came to the conclusion that the most important attribute for their restaurant client was food served hot and on time.  Obviously, no one was ordering a cool crisp salad for lunch or dinner.  In fact, the meal occasion was never mentioned.  Shouldn't it make a difference whether the meal is breakfast, lunch, dinner, or a late night snack?  Does it matter if the meal is with family, friends, or business clients?  What if the meal was a date, a birthday, or just because you have nothing to eat in the house or don't feel like cooking?  How can our measurements be context aware when our questions contain no contextual information? Our response scale is no better.  There is no contextual information in an importance rating.  Nor is there any contextual information in the MaxDiff or best-worst scaling that was advocated in the above article.  Let us start by adding context to the response scale.  What would be the impact of serving "portions that were just right"?  In order for a respondent to be able to answer this question, we need to add a context and make the attribute more specific so that the respondent has something concrete to consider.  For instance, what customer behaviors do we seek to change by offering small portion sizes at cheaper prices?  Is it something that customers would not care about, would be nice to have, or would it break the tie between two otherwise equal restaurants and get you to select one restaurant over another?  This is an example of context-aware measurement.  We have a response scale with several marketplace alternatives that encourage our respondents to think about their likely response within a realistic purchase context. It's Simpler If You Make It More Complex Before I move on to the measurement model and its implementation in R, I recommend that you visit the Hartman Group's website called the World of Occasions.  Here you will find restaurants as one of many possible eating occasions.  Each occasion is defined by who, when, where, what and why.  You should note what might seem to be a paradoxical effect.  The Hartman Group appears to have made things more complex with their multifaceted world of occasions.  However, once an occasion has been identified, the consumer uses the context to help them reach their purchase decisions.  This is a form of situated cognition.  Consumers simplify their decision tasks by using the occasion to structure the problem space.  Instead of needing to reassess preference anew in each situation, preference is tied to the occasion so that identifying the context activates the preference structure and simplifies the choice process.  In fact, once the context is seen as a particular type of occasion, it evokes its own action possibilities, which is a property of all perceived affordances. The Rating Scale Isn't Broken - You just don't know how to use it We are ready for the last component, the context-aware measurement model, which we will estimate using the R statistical programming language. Context-free measurement provides few specifics and asks the respondent to fill in the blanks.  The respondent is free to use any criteria they wish to decide which feature is the most or least important.  Similarly, although the endpoints may be marked with words such as "not at all" or "extremely," importance itself is left undefined when a rating scale is used.   As a result, respondents will give us responses when we ask them to select the best and the worst among alternatives with no clear external referents.  What are "healthful menu options" or "new specials weekly"?  My favorite is "portions are just right."  Who doesn't want portions to be just right?  It is not uncommon for respondents to report that attributes such as color or price are unimportant until they are shown specific colors ("I will not buy it in that color") or actual price points ("I'm not paying that much").  Without a realistic context, the entire data collection process becomes an exercise in semantic reasoning yielding little or no actionable findings. Context-aware measurement resolves the ambiguity by using behavioral anchors.  But we lose the "flexibility" of context-free measurement.  Context is added by using feature descriptions with high imagery and specificity.  As we saw earlier, "portions are just right" leaves too much to the imagination.  We need to get much more specific in order for respondents to tell us how they are likely to behave in a realistic context, which means we need a lot more items and some way to manage respondent burden. We are using the respondent as the measuring instrument: input = product or service features and output = behavioral response.  Each contextualized feature or service is evaluated by the respondent and assigned a score using the behavioral anchors.  We believe that the "pull" or "attractiveness" of the features and services can be measured along a continuum with thresholds separating the different scale values.  For example, a respondent is shown Attribute A and imagines their likely response using their thresholds:  Do I care about Attribute A?  Would I pick one over the other because of Attribute A?  Would I pay more to get Attribute A?  All this is shown in the figure below with the change in color indicating the threshold that needs to be exceeded before the next response category is picked. The theoretical foundation for this measurement procedure is the belief that it reflects the simplification processes that consumer actually use to make purchases.  That is, consumers compare very different features and services by placing them all on the same preference scale.  In order to decide that "you want a fine dining experience, but it costs too much," you must be able to place your desire for a fine dining experience on the same scale as the cost of the meal.  It is essential that we provide a marketplace justification for our measurement technique.  Otherwise, we find ourselves playing games with attribute comparisons that have little or no real world implications. However, we still have a self-report.  The customer is not actually experiencing the contextualized feature or service.  As Gilbert and Wilson note, "Just as retrospection refers to our ability to re-experience the past, prospection refers to our ability to 'pre-experience' the future by simulating it in our minds."  Thus, respondents read "seldom more than one or two cars ahead of you in the drive-thru" and note their reactions upon imagining themselves in that situation.  Then, they translate that reaction into one of the behavioral categories.  Gilbert and Wilson list all the errors associated with this form of mental simulation.  "The devil is in the details" is one way to summarize their more comprehensive discussion.  For instance, it makes a difference if the two cars ahead of you are filled with people who seem to be having some difficulty deciding what they want to eat.  But it is unlikely that this is what you thought about when you first read the description.