A latent variable approach to investigate system 1/2 decision-making: evidence from a food choice and eye tracking experiment

20 August, 2019
Kobe, Japan

This research investigates whether and how different front-of-pack (FoP) nutrition labels contribute to individuals’ healthy food decisions by using evidence from an eye-tracking and discrete choice experiment. Understanding individuals’ choices and decision-making process play an important role in mitigating nutrition-related issues (e.g., obesity, chronic health conditions). Although FoP nutrition labels help consumers make more informed choices by giving them the opportunity to compare various food products with respect to their nutritional and calorie information, there are inconsistencies between different nutrition labelling formats used. Some retailers and manufacturers use traffic light colour coding to highlight the level of fat, saturates (saturated fat), sugars and salt, other use qualitative description of the level of nutrients, such as high, medium, low. Some retailers even use a mixture of the two formats. However, there is relatively little evidence on whether different FoP labelling formats influence consumers’ choices and decision-making process. This research sheds light on these issues by integrating discrete choice experiment (DCE) with an eye-tracking experiment.

Combining DCEs with an eye-tracking experiment provides richer insight into the process of decision-making. While DCE allows us to investigate the trade-off people make when choosing food products, the eye-tracking experiments shed light on the where and for how long individuals fixate their gaze on the areas of interests on FoP nutrition labels (e.g., energy information, fat content, and so on). More fixation can mean the area is more noticeable or more important to the viewer than other areas, or that they are finding the information difficult to process. Such information may also help understand consumers’ decision-making process, such as System 1 and System 2 thinking processes.

As part of the DCE exercise, respondents examine a number of ``experimentally designed’’ scenarios and are asked to identify their preferred scenario. In this research, each scenario depicts a packet of potato crisps with FoP nutrition labelling, which are presented to respondents using four different formats: colour codes (like traffic lights labelling) and descriptor texts (e.g., low, medium, high), colour codes with numbers (e.g. fat levels in grams), no colours with texts, and no colours with numbers. While respondents were making a series of choices, the eye-tracking device, which was attached to the bottom of the monitor, recorded fixation counts, sequence and dwell time.

The choice and eye tracking data, collected from student participants, is analysed using a novel modelling approach that investigates the tendency of system 2 decision-making (which is slow, deliberate, and logical), as opposed to system 1 decision-making (which is fast, instinctive and emotional). Our modelling approach examines what types of FoP labels lead to more system 2 decision-making, if any, are more successful. Obviously, we have no way to know if someone actually uses system 1 or 2 decision-making. So we treat this as a latent variable. For our measurement indicator, we use the length of time respondents looked at the choice alternatives (all else being equal, we can, reasonably, expect that someone who spent more time looking at an alternative is more likely to adopt a system 2 strategy, compared to someone who spends less time looking at an alternative). Therefore, we interact the latent variable into the measurement indicator function, where we use a two-part hurdle model that includes a binary logit for zero dwell times and a truncated negative binomial for positive dwell times. For the choice model, we consider a constrained multinomial logit (MNL) model. In this case, we are interested in finding how the latent variable influences the probability of considering the alternatives.

Overall, we find that our integrated method provides rich insights into explaining food choices and decision-making. The significant latent variable interaction in the choice model indicates that System 1/System 2 decision process explains the probability of considering an alternative. Specifically, we find that participants who are more inclined to have system 2 decision-making (as measured through the latent variable) are more likely to ignore alternatives. This latent variable is also significant in the measurement indicator function. In particular, as the magnitude of the latent variable increases, participants are less likely to not look at any of the alternative’s attributes. However, conditional on having a positive dwell time, the latent variable, while positive, is not significant, implying that system 2 decision-making explains if a participant looks at the alternative but not how long they look at the alternative. The latent variable aside, participants are more likely to look at a label if it is presented with colours. Our results also show that left/right positioning of a choice alternative and the number of choice tasks play an important role in whether, and for how long, a participant looks at the alternative. Earlier choice tasks and alternatives presented on the left side of the screen are more likely to be looked at. In regards to the influence of socio-demographic characteristics on the tendency of system~2 decision-making, we observe that male participants are more likely to adopt a system 2 decision rule.

The results have important implications for the food industry and the policy-makers regarding the use of colour-coded labels, like traffic light labelling. We find that labels with colour coding are more likely to help consumers process information through system 2 decision process as compared to other formats, such as no-colour coded numeric labels. This gives insights into other areas where communication is delivered via labels to encourage people to make informed choices.

Seda Erdem
Seda Erdem
Associate Professor of Economics
Danny Campbell
Danny Campbell
Professor of Economics