Decision-Making
Process and Normative Models
Understanding the decision-making process changes how we should make decisions.
Posted July 30, 2015
In many social sciences, such as psychology and economics, there is an ongoing debate between whether a model should be based on Normative or Process analyses. A normative model is one which asks what the answer to a problem should be, and a process model is one that asks how it is solved.
A normative model answers an optimization question – if you are trying to solve a problem, what’s the optimal way to solve it? For example, if you want to travel from one point to another, the fastest way is to take a straight line.
A process model answers a mechanistic question – how is the problem solved? So, for example, we might drive or walk from our house to the grocery store. Process models will come up with different answers depending on the how we are going to make our journey. If we are driving from our house to the grocery store, we can’t travel in a straight line because cars have to stay on roads. But if we’re walking, we can cut through the park. Both driving and walking can be viewed as approximations to the optimal path (a straight line – “as the crow flies”).
However, process and normative models are two sides of the same coin. One can ask What is the normative thing to do given that you will be using a given process? Thus, every process model implies a normative answer. But similarly, every normative model is an optimization under assumptions. Those assumptions are often not stated, but they still pertain. Thus, every normative model says something about the underlying process.
For example, normatively, the shortest distance between two points is a straight line, but if you are flying from one city to another, you actually don’t want to go in a straight line because the Earth is not flat. To get from one city to another, you actually fly in a great circle. Of course, if you are driving, it is rarely possible to travel along that great circle directly, so you have to take the roads that are available. (Because different roads have different travel speeds, the fastest way from one city to another might not actually be the direct route.) Many people would say that the route found by your favorite navigation app is an approximation to the optimal path. But the optimality of that path depends on the assumptions made – Where are the available roads? Is the Earth really a sphere?
As another example of normative, one can ask what’s the fastest way between two points in space (a straight line), but the fastest way from Earth to Pluto is not actually a straight line because it is possible to swing your spaceship around Jupiter (using what is known as the “slingshot effect”). The process of going past Jupiter changes the normative optimal path to Pluto. The original straight-line answer includes the hidden assumption that there are no planets to slingshot around.
What does this have to do with psychology or economics? There is a long and ongoing debate about whether we should be looking at questions of optimality (how should we do this?) or questions of description (how do we do this?) (In his new book Misbehaving, Thaler calls these models Econs and Humans – Econs are normative and optimal, while Humans do what humans really do.) But within the normative explanation of Econs, there lies a hidden process model.
For example, this model assumes that Econs have infinite knowledge and infinite processing power. (Which is of course not true. In economics and psychology, this is known as the "bounded rationality" problem.)
In fact, knowing the human decision-making processes creates a new normative model – what is the optimal decision-making given the algorithmic processes available? We now know a lot about the human decision-making algorithms – how there are four action-selection systems that interact, how humans form schemas to identify situations, etc. The optimal decisions given those processes is not necessarily the same as the optimal decisions given infinite search time.
The typical Econ normative model includes a hidden process model that humans are not part of a social network (which is, of course, wrong). The optimal decision given that a bad reputation as a jerk can limit your future interaction opportunities is fundamentally different than the optimal decision to maximize your income on a moment-by-moment basis.
In neuroscience, this debate is usually phrased in terms of Marr’s three levels – the computational level is the problem you are trying to solve (how do I get from Minneapolis to San Francisco?), the algorithmic level is the way that you are solving the problem (am I flying, walking, or driving?), and the implementation level is the way that you are implementing that algorithm (am I driving your car or mine?) While it is useful to think about all three of these levels for any question, the levels interact. The optimal path from Minneapolis to San Francisco depends on your algorithmic choice (flying will go in a straighter line than driving) as well as the implementation (your car can’t get up the mountains as easily so we need to take a less steep route across the Rockies). But the optimal path also depends on the assumptions of the situation (the Earth is a sphere ).
Process and normative are two sides of the same coin – What is the optimal choice given the constraints of the process? and What are the assumptions underlying a given optimization result?
Further Reading
Richard Thaler (2015) Misbehaving. WW Norton &co.
Herb Simon (1982) Models of Bounded Rationality. MIT Press.
David Marr (1982) Vision. MIT Press.
A David Redish (2013) The Mind within the Brain: How we make decisions and how those decisions go wrong. Oxford University Press.
