Missing the Mark with Current Consumer Choice Models

In the mid 1500s the Polish cleric Nicholas Copernicus published a new model of the position of the sun and the planets in our solar system. In his view, the sun was more or less at the center of the solar system and the planets were assumed to orbit in epicycles. His theory was a major breakthrough in changing the Ptolemaic view that placed the earth at the center of the system.
However, the view was also a major failure because Copernicus refused to give up the idea that the planets’ orbits were perfect circles. In order to account for the seemingly complex pattern of positions of the planets, Copernicus and his contemporaries had to work with many circular paths.
The earth was conceived to rotate in a perfect circle around an imaginary point, which also was assumed to rotate in a perfect circle around another imaginary point, which itself revolved around the sun in another perfect circle. In combination, all of these circular paths (plus other assumptions about the periods of revolutions) provided a good description of the actual elliptical orbit of the earth around the sun (in Loftus, 1985).
Thus, an erroneous general conception of the solar system was nevertheless very descriptive of the behavior of the planets. Because the universe was conceived to have a very complex structure, sixteenth century astronomers seemed to abandon the idea of explaining any of it. A theory was not taken seriously as a reflection of truth, but simply as a way of providing a good fit of the data (Loftus, 1985).
The perfect circle orbits are reminiscent of the additive (linear) form assumed by many psychological models of [consumer] subjective value. Making further allowances via scaling functions, the [consumer decision-making] models may provide excellent descriptions of the observed measures, but it is not necessary that they reveal the true nature of the comparisons being made by individuals (González-Vallejo & Reid, 2006).

Challenges with Traditional Consumer Choice Models

The current dominant market research model of consumer trade-offs relies on the subjective expected utility model of choice (Savage, 1954), a model of the ‘economic human’, which has long been seen by decision scientists as an erroneous descriptive model of decision-making (Kahneman & Tversky, 1979). Consumers do not maximize subjective expected utility when making choices between products (Schoemaker, 1982).
Yet still, our industry has become quite adept at adopting elegant estimation methods for utilities that have improved the model’s predictive validity over time (e.g. Hierarchical Bayesian (HB) Utility Estimation). However, just like Copernicus, the choice model upon which the hierarchical bayesian estimation method has been applied is fundamentally flawed. Thus, regardless of how accurately we can estimate a utility for a product attribute, we are still fundamentally misunderstanding how consumers make trade-offs.

Part of the trouble here is the fact that people don’t think in absolute values. The human perceptual apparatus has evolved to be sensitive to change in the environment, rather than the absolute magnitude of a sensory stimulus (Kahneman & Tversky, 1979). For example, we are sensitive to changes in brightness, changes in volume and changes in temperature. That is, we are naturally programmed to detect relative differences between objects in our environment and this has an effect on how we make judgments. In fact, it has recently been shown to affect our preferences.
Consider the following research finding:
People prefer 4 oz. of ice cream to 5 oz. of ice cream if the 4 oz. of ice cream are in a container that makes it look like it is overflowing.

This effect remains true even if people are made cognitively aware of the actual difference between the amounts!
Intuitively this makes sense, doesn’t it? How do you know how much is a lot if you don’t have something to compare it to? How do you know if a man standing on a basketball court is tall unless you have other players standing next to him? How do you know that it is cold outside in the morning if you don’t have the vivid warmth of your bed as a comparison?  How do you know if one product is expensive if you don’t have another product to compare it to?

Proportional Difference Model & Proportion of Emotion M0del

Fortunately, the proportional difference model (Gonzalez-Vallejo, 2002) and the proportion of emotion model (Reid & Gonzalez-Vallejo, 2009) overcome this obstacle by providing a natural common currency for comparative evaluation and trading: the proportion.
A proportion is a measure that is necessarily comparative. The calculation of a proportion is dependent upon a reference value. The proportional difference model of choice recognizes this necessity of comparison, as well as the human propensity to be sensitive to change, and predicts choices based on relative differences in the magnitude of attribute values. The proportion of emotion model goes one step farther and incorporates emotion as an importance weighting mechanism to more accurately forecast consumer choices in the market.
We wanted to test the industry standard model with HB estimation against the proportional difference model to see which model more accurately accounted for consumer choice. So, we ran a test using a program that many industry choice modelers are familiar with: the Sawtooth Software CBC program for discrete choice studies using hierarchical bayesian estimation for utilities.
The study involved diamond ring choices. Participants were shown actual diamonds, all brilliant cuts, with equivalent color (D), and VVS clarity, set in a solitaire platinum ring. The diamonds varied along two attributes: price and carat size. The carat size was listed along with actual market prices obtained from bluenile.com <http://bluenile.com>.

The results were telling. The industry standard model (HB estimated random first choice multi-attribute utility model) performed well. The model’s root likelihood (RLH) was .775 or about 2.3 times chance level predictions.
However, the PD model’s predictions were significantly better. The PD model (RLH) was .935 or about 2.8 times chance level predictions.
These results indicate that there is an underlying truth of how consumers make trade-offs that is not captured by the current dominant model of choice. This has significant implications for product managers.
Let’s make a conservative assumption and say that the proportional difference model will give you, on average, 5% more accurate results per product pricing study. The key question quickly becomes: How long can product managers afford to be 5% less precise in the pricing of their products, particularly if that 5% difference equates to millions of dollars less for the bottom line?
The market research industry is driven by research advancements that improve the bottom line for companies. The proportional difference model represents one such advancement that will profit the research firms that adopt it early to improve the performance of their clients.