As marketers and market researchers, we’re constantly trying to understand consumer behaviour so we can better tailor products and services to their needs. Therefore, a thorough understanding of how people make choices is crucial to beating the competition. Although human decision-making behaviour is highly complex and can be influenced by a wide variety of factors, researchers typically try to simplify consumer preferences and choices. We try to replicate a complex reality using a model that captures the main drivers of choice.
Conjoint analysis is one very effective method for isolating drivers of choice. A product is broken down into its most basic characteristics or elements, allowing researchers to examine how each one affects choice. The Random Utility Model (RUM) is the most commonly used conjoint model, in which consumers select the product that offers them the best value (i.e. utility) based on a combination of product characteristics.
An alternative and lesser-known model is Random Regret Modeling (RRM), which assumes that, rather than going for the optimal product, consumers aim to minimise the potential regret of missing out on some product characteristic(s).
Both approaches have their pros and cons, and each is more or less suitable in different market situations. In an effort to improve the predictive validity of conjoint results, SKIM researchers compared and combined the utility model and the regret model by means of a case study and posed the question: would a hybrid model lead to more realistic results?
Limitations of the utility and regret models
The utility model (RUM) is widely used to estimate the preferences for different product characteristics, such as price, brand, size, and so on. It is based on the principle that a consumer chooses one product that has the best combination of characteristics, and this is dependent on which characteristics matter most to them. Chances are, if you’ve ever done a conjoint study, it was conducted using the utility approach.
This model works well, yet it has its limitations. For one, it assumes that choices are based solely on optimising the combination of characteristics of the product they select, independent of alternative products that are available on the market. In other words, it does not fully consider the context of the available products on the market.
Regret modeling (RRM) is one way to counteract the limitations of the utility model. The regret model assumes that, instead of choosing the option with the best combination of product characteristics, consumers make choices to avoid the potential regret of missing out on something else that’s available. The regret model assumes that as soon as people make trade-offs, they run the risk of regret. Usually there is at least one non-chosen alternative that outperforms the chosen product on one or more characteristics (e.g. it is cheaper, prettier, or performs better). Hence, the regret model considers external factors surrounding the selected product more accurately than the utility model.
As a simplified example, suppose you go to the cinema and would like to buy some popcorn. Let’s say that there’s a medium box available for $4, and a large box offered for $6. In this case you might choose the medium option. However, if the cinema were also to offer a jumbo option for $10, you might be more likely to choose the large box over the medium one. Even though nothing changed in the price and size of the medium and large boxes, knowing that there is an even larger, more expensive option available can have an impact on your choice between two other alternatives. It’s easy to imagine the same principles applying when other available products offer benefits different from the options you’re considering, whether it’s about available flavours, flashy pack designs, or any number of other characteristics.
Nevertheless, the regret model also has clear drawbacks. For example, to properly analyse the context of a choice, one must test for different types of context. In the popcorn example, if we want to know the effect of either two or three available products, we need to present choice tasks with both two and three available options. However, if we want to understand the regret effects that might take place with four available options, these options need to be shown explicitly as well. This makes the setup of a regret model a lot more complex than the more commonly used utility model, in which we can predict choice regardless of the number of products available. In practice, this means that the regret model is very difficult to apply to product categories with many available choices.
The hybrid solution
As described, both models have their shortcomings. The question is, which model brings us closest to reality? Do people base their choices more on selecting the optimal combination of product characteristics, or do they choose based on the desire to minimise the risk of regret? What if we combine both methodologies? Would a hybrid model yield results greater than the sum of its parts?
Case study: tablets
To answer this question, researchers at SKIM set up a study in which they examined the choice behaviour of consumers buying tablets. The utility, regret, and hybrid models were applied to this same dataset to determine which model best fits the data and which model has the highest ‘hit rate’ – a measure of predictive accuracy.
As we reviewed the results of the three different models, we saw that the difference in both data fit and predictive validity between the hybrid model and the utility model was rather small. Only the pure regret model performed less effectively. Moreover, the average correlation across the three methods was 0.95. In other words, none of the models clearly outperformed the others in these terms.
Even though the accuracy of predicting choices was similar, the predicted shares for hypothetical scenarios in simulation models based on each approach were not. A closer look reveals that large differences exist in the predicted shares. In the example scenario below, all three methods predict limited shares for Product 2. However, using the utility model, Product 1 clearly performs better than Product 3. While using the regret model, Product 3 performs better. The hybrid approach predicts that both Products 1 and 3 will perform similarly. Therefore, the same data could potentially lead to different recommendations (Figure: Test Scenario).
The hybrid approach
The result of the hybrid approach has a similar predictive validity to the other two models, but takes into account how some consumers choose to maximise utility, while others choose to avoid regret. The main advantage of combining both methods is that, without knowing the underlying decision-making tactics of consumers, the results are more actionable. Assuming that both utility maximising and regret minimising are equally likely in a market, the hybrid model is the safest bet because its output is less likely to be skewed to choices in either direction.
In terms of statistical predictive validity, the most often-used utility model still performs well. As a simplified model of complex choice behaviour, it’s still a solidly reliable tool. That said, the utility model is more likely to overestimate the ‘rational’ utility-maximising decision behaviour of consumers, potentially leading to skewed insights. Whenever possible, a hybrid approach is more likely to correctly take into account both utility-maximising and regret-minimising choice behaviour for more reliable predictions of choice. After all, the goal of modeling is to more accurately replicate the complex process of real-world decision making. ■