Understanding Mental Models
What they are, what they are not, and how they are relevant for designers.
Designers frequently refer to the mental models of their users in casual and imprecise ways. For example, they frequently talk about supporting or designing for a user’s mental model, assuming that they fully understand it. It follows that this understanding will enable them to better align their designs with user expectations. Designers often obtain their understanding of mental models through user research that attempts to directly elicit model descriptions from users asked to verbally describe or draw their understanding of the inner workings of a design.
The mental model concept, which has existed in psychological theory for over 50 years, has become very popular and has been applied to a wide variety of fields. A quick Google search on Mental Models reveals a staggering 2,530,000,000 results. While some of these results accurately point to the earlier psychological research into mental models, many or even most are based on popularized notions of the mental model concept as synonymous with simple ideas, beliefs or theories. Furthermore, some make expansive and extravagant claims about how applying the concept will change your life through better insights and decision making.
We hold that, because many of these attempts to capitalize on the concept are based on slipshod thinking, they run the risk of polluting and poisoning some very valuable insights about our cognitive machinery, and more specifically, how we should design things to accommodate it. Clarity is needed.
Basics of mental models
For well more than half a century, cognitive psychologists have used a variety of concepts to explain behavioral phenomena that hint at hidden organizing structures used by the mind. These include mental representations, cognitive maps, schemas, frames of reference, analogical reasoning and mental models. These structures aren’t directly observable at present and may never be.
Mental model theory and research has engendered many controversies such as whether or not they are based on logical propositions or images and whether they reside in long-term or working memory. Furthermore, because different researchers have approached this problem from different theoretical orientations using different techniques we find ourselves in the familiar role of the blind men regarding different parts of an elephant. Or as Phillip Johnson-Laird (1989) said,
It is as if explorers keep reporting the existence of a hitherto unknown animal, but their fragmentary glimpses of it convince them that they are observing different creatures.
At the end of the day, these are likely different metaphors that capture some part of the evidence that suggest particular kinds of mental organization. We aim to add clarity rather than an additional metaphor. In short, Mental Models are structural analogues of our complex environment (Johnson-Laird 1983). They are thought to be cognitive structures (some have described these as internal representations of external reality) that form the basis of reasoning, decision making, and behavior. They are constructed by individuals based on their personal life experiences, perceptions, and understandings of the world. They provide the mechanism through which new information is filtered and stored. This definition builds upon decades of psychological research that started with Craik (1943) and was heavily influenced by Johnson-Laird (1983).
Don Norman in Psychology of Everyday Things (1986) gave an early and succinct description of mental models of devices that we use to complete tasks (We use “device” here as shorthand for things that a person can use to complete a task — devices can be physical such as a rock, a mechanical watch, or a car, digital such as a software application, or even procedural such a business process). He described two kinds of models: a design model — the designer’s or developer’s mental model of how a device works and a user model — the user’s mental model of how the thing works. The user model is a mental representation that the user creates in order to know or predict how to operate the device. The model provides a frame of reference for understanding interactions with technology and building expectations of machine behavior. Mental models set our expectations of experience for interacting with various types of devices and experiences.
While Norman uses the term mental representation, we’re not suggesting that all mental models are equivalent to pictures that users create in their minds while imagining something. A mental model can influence what you imagine, draw, and remember, yet mental models are more than simple memories.
They cannot be directly captured by verbalization or sketching. This is because like memories, they are fragmentary. When asked to describe the model being used in a given scenario or situation, a person will recall some parts and infer others on demand, typically to produce a cogent answer for the researcher. In short, they will come up with a model or story to explain their behavior as best they can recall it rather than describing every detail in their mind that led to each behavioral decision.
As an aside, there is plenty of psychological evidence of people being able to create mental representations of objects and spaces. For example, humans can imagine what an object looks like rotated in space or from a different point of view and mentally construct a map that lays out objects in a space. For clarity, we’re separating this work as mental imagery (and mental mapping) that you consciously create and think about to solve a problem. Regarding mental models, we’re referring more to the mental frameworks of procedural knowledge that enable us to understand the world around us as we interacct with it and predict the behaviors of animals, objects, and systems around us.
How mental models are created
Mental models need not be the product of conscious deliberation by users. People usually don’t go about consciously and deliberately building mental models of the world. Instead, they typically arise implicitly with minimal conscious engagement as a byproduct of interacting with our environments. They save us time and energy by enabling us to focus on the critical parts of an experience rather than needing to pay attention and remember all of the world around us. People may not be fully aware of content in their mental model until their expectations are violated.
Norman (1983) made some additional observations about mental models focusing on their parsimoniousness and fuzzy nature. They only provide the gist of understanding and are unstable. People tend to forget their details. Furthermore, people resist enriching them, preferring instead that they remain simple. This leads people to perform inefficiently and even superstitiously. Because they are incomplete they have predictive value and cause people to guess wrongly about device behavior. Finally, because they are so ephemeral, they have weak boundaries. People can mash up mental models of similar appearing devices and make invalid generalizations.
We create and refine our mental models as we experience the world. They are dynamically reinforced through direct experience with the world. For example, consider purchasing gasoline. Many years ago, these purchases were handled through a process called “full-service”. The steps consisted of pulling up to a pump and waiting for an attendant to ask you which kind and amount of gasoline you wanted. He (usually it was a male) then checked your oil and told you whether or not you needed more and cleaned your windshield. He then orchestrated the payment process and you drove off.
When the full-service model was replaced by “self-service”, there was a model mismatch. Now, no one comes out to your car when you pull up to the pump. Initially, this new process resulted in confusion until drivers learned and adjusted their mental models to account for the new procedural knowledge and tasks demanded from a different type of purchase experience. This same scenario can operate in both directions. Now imagine a person that has only experienced self-service stations pulls into a full-service station. They will experience confusion that requires adjusting their mental model accordingly.
Assessing a person’s mental model
It’s important for designers and user researchers to realize that mental models are not directly observable. As noted earlier, asking a person to describe their mental model or what they were thinking during a task may provide you with a story or description of the device or its behaviors, but researchers need to realize that this output is post hoc — rather than describing the non-conscious intent driving their actual behavior, users are consciously and purposely generating stories afterwards to explain their behavior along with the devices actions. While these descriptions and stories can produce some insights into confusing areas of their experience, they should not be taken as the default rules or structures inherent to a person’s mental model.
Additionally, asking a person how they think something works will typically provide interesting (and possibly useful) stories, yet these don’t necessarily reflect how a person behaves with the product or what triggers their behaviors during an experience. People can generate explanations on the fly regarding how they think something works under the hood, but these are stories made to answer the researcher’s questions rather than conscious intent driving their behavior. Researchers have repeatedly demonstrated the inaccuracies between what people say they do and actions they actually take. While these inaccuracies typically reflect people not accurately remembering all the actions they completed, it’s also natural for people to reflect more positively on their behavior to build a better story for the researcher.
Similarly, think aloud protocols (TAP) are useful but also incomplete. TAP are best if the user is echoing what they are doing as they do it as opposed to being asked after task completion (when they have a chance to generate a self-consistent narrative explanation of their recalled behavior). TAP can provide an idea of the why behind some of the behaviors witnessed, yet this includes people reflecting on what they are seeing and what they think it means to them and to the researcher. While useful for understanding confusing areas of an experience, TAP is not reflecting the procedural knowledge that builds a mental model. It will help identify scenarios in which people become confused, but additional testing is needed to better understand one’s mental model.
To get at the procedural knowledge in a mental model, use a method or set of methods that have users predicting device behavior to answer questions. For example, have users select and describe a device action that will result from the question posed by the researcher. Providing opportunities for people to make predictions or estimations of how something will unfold and why gives some deeper understanding of how they are framing the device behaviors along with the user behaviors that trigger and result from the device. This will help frame what device assumptions are being made and what behaviors are expected as a result.
Conclusions
Mental models have been researched and debated for decades, yet a lack of clarity still surrounds the topic. Nevertheless, it is popular again in user experience design as designers focus on and provide design rationale tied to a user’s mental model.
To provide clarity, we emphasize the function of mental models as a framework for structuring procedural knowledge to set expectations and guide behavior. With this perspective, mental models enable us to quickly capture and interpret what we experience in the world to guide our understanding and behavior.
It’s important to realize that capturing a person’s mental model is not as simple as asking them to describe it or draw it. Having users draw a picture of how a device works, by itself, is a self-reflection technique used to generate answers to a researcher’s question. While this will provide an answer for the researchers, the result may not necessarily reflect the underlying mental structures and procedures actually used when interacting with the device. Many user research techniques, like TAP and post-task interviews, are also not exposing the users mental model either.
Gaining an understanding of a person’s mental model requires several techniques used to converge on assumptions being made about a device and its behaviors. We suggest starting with tasks that require participants to predict actions based on certain conditions while explaining rationale along the way. While not perfect, this will start to surface nuggets of procedural knowledge that are structuring the participants responses and actions.
At the end of the day, the researcher and designer should be modest about their understanding of their users’ mental models. At best, they will have a mental model of their users’ mental models — a working hypothesis that is good enough for design but also always subject to revision.
Hopefully we’ve provided some useful insights on mental models in design. However, we’ve only scratched the surface. There are many other questions related to the concept and how it is used, especially from the perspective of User Experience research and design. We will be exploring these topics, including designing to support a person’s mental model, in a series of upcoming posts that provide other interesting and potentially useful details.
In this thread
References
Craik, K. J. W. (1943). The nature of explanation. Cambridge University Press, Cambridge, UK.
Johnson-Laird, P. N. (1983). Mental models. Cambridge University Press, Cambridge, UK.
P.N. Johnson-Laird, Mental Models (1989), in Michael L. Posner [Ed.] Foundations of Cognitive Science, MIT Press.
Donald Norman (1983) Some observations about mental models. In Dedre Gentner and Albert Stephens [Eds.], Mental Models, Erlbaum.
Donald Norman (1986) The Psychology of Everyday Things. Basic Books.
