How to recognize a decision

Rahul Saxena
6 min readFeb 12, 2022

--

Analytics is used to support decisions. Sadly, students of analytics and its avatars (data science, artificial intelligence, etc.) are often unable to recognize a decision. This article aims to dispel this darkness.

“When you get to a fork in the road, take it” — Yogi Berra

A decision exists only when you have options to choose from. Without options, there is no decision. Strangely enough, this simple truism is ignored, and analysts present diplomas in data science without being able to list a single decision that they analyzed. What would help them realize that people make decisions quite often? To commute to office by bicycle, car, or bus. To vote for a candidate. To hire more salespeople. It helps to define the decision as a matter of choosing between different options in a specific context. The decision becomes a problem to solve: which option is the rational choice?

Once you have located a decision, how do you know if it’s worth analyzing?

“Would you tell me, please, which way I ought to go from here?”, asked Alice. “That depends a good deal on where you want to get to,” said the Cat. “I don’t much care where…” said Alice. “Then it doesn’t matter which way you go,” said the Cat. — Lewis Carroll, Alice in Wonderland

To analyze a decision, you must evaluate the options against what you care about. Can you write down an assessment of each option? In this, you will encounter the criteria with which to assess each option, so that all the options can be weighted on the same scales. Without a list of criteria, you run the risk of biased evaluation of options because you did not check each option against each criterion.

A simple decision has a few options, such as yes/no or left/right. Some of these decisions can be made rapidly, and are diagrammed in a cascade of decisions with choices in each branch. Such decision cascades are easily outlined in a flowchart. For example, I receive job applications for the Decision Scientist position at CoBot Systems and I decide on what to do with them as a decision cascade composed of a sequence of simple decisions depicted as blue ovals in the diagram, below. I prefer using an oval because the standard diamond-shaped symbol for a decision point in a flowchart looks too spiky to me.

A decision cascade in a flowchart

For multiple criteria decisions, you must assess each option against each criterion. These assessments become the score for that option and criterion. A score can be binary (yes/no), a letter grade (A to F), a number, a sentence, a paragraph … what matters is that you thought about the score. If you score every option for every criterion, you can make a table: a simple multi-criteria decision model with options in rows, criteria in columns, and scores in cells. Now you can select the best option based on the scores, to make the rational choice. The rational choice is to choose the option that scores the best.

The rational choice is easy to make when one option scores the highest in all the criteria: that one option is Pareto Optimal. It doesn’t matter what weights you assign your criteria, the Pareto Optimal choice remains the best one.

The decision model as a table, options as rows, criteria as columns, scores in cells, and the rational choice as the one option that scores the best in all criteria.
The best choice in a multi-criteria decision when one option is Pareto Optimal

In case there is no single Pareto Optimal option, the choice depends on trade-offs. There are a lot of methods for making trade-offs. You can sum up the scores to get a total score. Make weighted sums of scores (multiply each score with the weight for that criterion) for each option. Count “win/loss/draw” in each criterion for each option. But it’s never easy to make trade-offs, and the process of making trade-offs is fraught with subjectivity.

The decision model as a table, options as rows, criteria as columns, scores in cells, and the the need to make trade-offs.
The need for trade-offs in a multi-criteria decision when no option is Pareto Optimal

A man is an angel that has become deranged, Joe Fernwright thought. Once they — all of them — had been genuine angels, and at that time they had had a choice between good and evil, so it was easy, easy being an angel. And then something happened. Something went wrong or broke down or failed. And they had become faced with the necessity of choosing not good or evil but the lesser of two evils, and so that had unhinged them and now each was a man.”
― Philip K. Dick, Galactic Pot-Healer

While modeling decisions, it is important to think of all the criteria you care about. Later you can weed out the criteria that don’t carry weight, or replace highly correlated criteria with one criterion to reduce the complexity and confusion of multi-criterion decision making.

On selecting criteria, here’s a little story about Sidney Morgenbesser. After finishing dinner, Sidney decided to order dessert. The waiter told him he has two choices: apple pie and blueberry pie. Sidney ordered the apple pie. After a few minutes the waiter returned and said that they also have cherry pie. Sidney thought a bit, and then said: “In that case I’ll have the blueberry pie.” Is Sidney irrational?

  • Sidney’s choices are not consistent with maximization of a utility function. He had preferred apple to blueberry. Cherry could either be better than apple, in which case he should switch to cherry, or not, in which case his original preference for apple should prevail (and cherry becomes an “irrelevant alternative”). His choice for blueberry, therefore, is not rational, because adding an irrelevant alternative (cherry) changed Sidney’s decision, even though blueberry was not preferred earlier.
  • Sidney had selected apple pie because it’s sweeter than blueberry pie and Sidney likes sweetness. When the waiter told him about cherry pie, however, he remembered that cherry pie is sweeter than apple pie, but it’s also less healthy based on the diet plan he had committed to follow. Sidney added the “healthy” criterion to his utility function, and rationally changed his decision to blueberry pie based on its combined score for being sweet and healthy.
Adding the cherry pie option changed the rational choice

Adding scenarios to the decision model allows you to explore your options in the context of different scenarios. For example, increasing the price of a product in the context of different scenarios for demand elasticity, or deciding on manufacturing capacity in the context of different scenarios for market growth. The problem of making a rational choice in the face of multiple scenarios adds another dimension to the trade-offs, as the likelihood or weightage of scenarios also has to be incorporated in the choice.

Options, Criteria, Scores, Scenarios, and the Rational Choice

To summarize: the five basic features for recognizing and outlining a decision are Options, Criteria, Scores, Scenarios, and the Rational Choice.

This is a good way to get started in the game of recognizing decisions. For more context, you can read Business Analytics by Rahul Saxena and Anand Srinivasan. Later you can move up to more complexity in modeling decisions, explored in the academic domains of Operations Research, Decision Analysis, Management Science, System Dynamics, etc. and applied in different ways in different industries, organizations, and decision opportunities.

Where does forecasting fit? In generating scores and scenarios. Where does optimization (like the simplex method) fit? It adds the idea of constraints on the decision, and provides a good way to determine the rational choice.

How do you know you are a decision scientist? You have built and used decision models, and can explain how they work. If you’ve not done that in a professional setting, you may be an aspirant who can start as an analyst to learn how decisions can be made more systematic and intelligent.

--

--