Behavioral Science. Cognitive biases in decision making: my journal experiment

The Repeater Problem

Routinely. Even when the information in front of them points to an equal or better alternative, the default pull is to do what they did last time.

That is the entire problem with cognitive biases in decision making in one sentence. Not exotic brain glitches. Not rare pathologies. A heuristic — action repetition — that worked well enough in a stable environment and now quietly runs the show in a world where the environment keeps changing underneath it.

The most expensive bias in any decision isn't the wrong answer. It's the answer you copied from your last decision without checking.

The standard framing in behavioral science is the Dual Process account: a fast, intuitive system, often called Type 1, and a slower, more analytical one, called Type 2. Type 1 is the repeater. It is economical, fast, and usually good enough. It is also the system that anchors on the first number it sees, locks in expectations, and walks you back into the same restaurant, the same vendor, or the same career move because doing the familiar thing is frictionless and doing the new thing is work.

Type 1 does most of the routine work in ordinary decisions. Type 2 is more likely to appear when something feels wrong, when the stakes are obvious, or when somebody deliberately creates a reason to slow down. By then, Type 1 has often already produced a preference and supplied a tidy explanation for it.

Most articles about cognitive biases in daily life stop here, in the catalogue of errors: confirmation bias, anchoring, availability, overconfidence. The list is long, and the usual advice is to be aware of it. As if awareness were a tool. It isn't. Awareness is a flag. Without a system attached to it, the mind drifts back to the same heuristic the moment you get busy, tired, or emotionally invested in the outcome.

The more useful question is not which bias you have. Everybody has several. The useful question is where your own decisions become repetitive, overconfident, defensive, or strangely resistant to new information. That is where a tracking system can do more than a general warning ever will.

The Catalogue Nobody Reads Twice

Here is the short version of four biases that show up reliably in both strategic and everyday decision contexts. They are not exotic. They are ordinary defaults that become expensive when nobody interrupts them.

Note what is not on the list: lack of willpower, lack of discipline, or lack of motivation. Those may affect whether you follow a process, but they are not the basic cause of the bias. The faster system is handling most routine decisions, and it does not pause simply because you know the vocabulary of behavioral science.

Confirmation bias is particularly easy to confuse with careful analysis. You collect several pieces of evidence, but the evidence is not being evaluated symmetrically. The information that supports your first impression feels relevant and nuanced. Contradictory information feels incomplete, unusual, or somehow less trustworthy. The conclusion can still look reasoned from the outside because the reasoning has been organized around a preferred answer.

Anchoring works earlier, often before you realize a decision has begun. The first price, deadline, estimate, or option changes the reference point for everything that follows. You may later adjust away from it, but the initial figure remains in the room. This is why a number placed at the beginning of a negotiation or a product comparison can influence the whole conversation even when nobody treats it as authoritative.

Overconfidence is less about arrogance than calibration. A person can be cautious, self-critical, and still systematically assign too much confidence to a forecast. The problem becomes visible when confidence and accuracy are logged separately. A prediction can be wrong without being unreasonable; the more revealing case is a prediction made with strong confidence that repeatedly misses in the same direction.

Escalation of commitment adds a time dimension. Once money, effort, reputation, or emotion has been invested, stopping feels like admitting that the earlier decision was wrong. The past investment cannot be recovered, but it can still exert pressure on the next choice. A failing project receives another month. An unsuitable subscription remains active because cancelling would make the wasted payments feel more final.

These biases often combine. An initial anchor shapes the forecast, confirmation bias selects the supporting evidence, overconfidence raises the certainty level, and escalation of commitment keeps the plan alive after the original assumptions have changed. The result is not one dramatic error. It is a chain of small decisions that all protect the first decision.

A subtle point from the descriptive model developed by Neal and colleagues is that the depth of cognitive processing and susceptibility to bias are independent dimensions. You can think longer and harder and still end up more committed to a wrong answer, especially if you happen to be an expert in the domain. Reflection is not a debiasing tool on its own. It is a tool that can sharpen whatever direction the fast system already selected.

Thinking harder doesn't disinfect the decision. It just polishes the bias you arrived with.

This is the part motivational advice tends to skip. Working more on the problem is not automatically the fix. Working differently on the problem is.

That difference matters in ordinary situations where there is no dramatic warning signal. Choosing a familiar supplier, extending a deadline, renewing a service, or accepting the first plausible explanation does not feel like a cognitive failure. It feels like efficiency. Sometimes it is efficiency. The point of tracking is not to treat every quick decision as suspicious. It is to identify the repeated decisions where speed has stopped being useful and has become a substitute for checking the environment.

Why the Journal Experiment Is Worth Running

A decision journal is the working-differently tool. Not because it makes you smarter — it doesn't — but because it inserts a small amount of friction between the impulse and the action. That friction gives the analytical system something concrete to work with.

The format is unglamorous. Before a non-trivial decision, write down:

• what you are choosing between;
• what you expect to happen;
• why you believe that outcome is likely;
• how confident you are;
• what would have to be true for the alternative to be better.

After the outcome becomes observable, record what actually happened and compare it with the forecast.

The reason this matters is structural. The fast system builds its case inside your head, where you cannot inspect it easily. The journal externalizes the case. Once the prediction is on paper, you can return to it later and check whether the reasoning was doing useful work or merely decorating a gut call.

This is the point at which cognitive bias tracking becomes practical. You are no longer asking, in the abstract, whether you are prone to confirmation bias. You are looking at a series of decisions and asking more specific questions:

• Did I look for disconfirming information before committing?
• Did the first number determine the range of options I considered?
• Was my confidence higher than my past accuracy justified?
• Did I continue because the plan was still sound, or because I had already invested in it?
• Did I record the alternative seriously, or only as a formality?

Over time, the journal becomes a private dataset of your calibration errors. It shows where confidence and accuracy separate, which decisions are most vulnerable to repetition, and which explanations only appeared after the result was known. That record can begin to retrain the heuristic because repeated, specific feedback is more useful than an abstract warning to be less biased.

The journal is not a motivational log, a gratitude diary, or a place to record what you learned in a flattering form. Those formats can easily confirm what you already wanted to believe. A decision journal serves the opposite purpose: it preserves the original forecast so that hindsight cannot quietly rewrite it.

That preservation is essential. After an outcome is known, the past often looks more predictable than it was. You remember the signs you noticed and forget the uncertainty that surrounded them. If the decision was successful, the original reasoning can appear stronger than it was. If it failed, the mistake can look obvious in retrospect. Neither interpretation is reliable without the earlier record.

There is one important caveat. Research on reflective debiasing is not uniformly optimistic. Neal and colleagues note that reflective processes can occasionally intensify certain biases under specific conditions, particularly when domain experts bring a well-developed but distorted framework to the problem. A journal will not eliminate confirmation bias in daily decisions. What it can do is shrink the blind spot, gradually, and make the pattern visible enough to build a countermeasure for the bias that affects you most often.

The journal is therefore not a verdict on your character. It is an observation instrument. Its value lies in the repeated comparison between prediction and outcome, not in whether each entry sounds intelligent.

What the Experiment Actually Looks Like

If you wanted to run this on yourself — and the entire point of a journal experiment is that it should be small enough to continue — the structure is straightforward.

1. Choose one repeated decision domain.

Hiring, purchases above a chosen threshold, schedule choices, vendor selection, project estimates, or recurring health and work routines can all work. Repetition gives the journal traction. If every entry concerns a completely different kind of decision, patterns are harder to see and explanations become too general.

2. Record the decision before it is settled.

Write down the options you are genuinely considering, not the options you want to appear reasonable afterward. Include the preferred choice, the alternative, the predicted outcome, and the main assumption carrying the decision. If the assumption turns out to be wrong, the alternative should have a better chance.

3. Separate confidence from preference.

You may prefer an option while remaining uncertain that it will work. Those are different signals. A simple confidence scale is enough, provided you use it consistently. The point is not to produce a scientific measurement of your personality. The point is to make strong certainty visible before the result arrives.

4. Define what would count as a meaningful outcome.

A forecast that cannot be checked later cannot teach you much. “This will go well” is too vague. A useful entry identifies what you expect to see, by when, and what result would make you reconsider the choice. The level of detail should match the decision; the journal is not supposed to turn every coffee purchase into a research protocol.

5. Log the result without editing the forecast.

When the outcome is observable, describe what happened and how it differed from the prediction. Avoid converting the result into a lesson immediately. First preserve the comparison. Interpretation is more reliable after the basic record is intact.

6. Review at a fixed interval.

A monthly review is realistic for many people, though the exact rhythm matters less than having one. Look for repeated gaps between confidence and accuracy, recurring assumptions, and domains where you keep choosing the familiar option without comparing it with current conditions.

One useful addition is a surprise rating. The surprise score is not a measure of how dramatic the event felt. It records how far the outcome moved from your prior expectation. Research from Brown University indicates that genuine, prediction-violating surprise is associated with pupil dilation and a norepinephrine response involved in updating prior expectations. In practical terms, an unexpected result can create an opening for the brain to revise its model.

Logging the prediction makes that opening easier to notice. You cannot be surprised in the same way by an outcome you explicitly wrote down as possible. When reality contradicts the forecast, the discrepancy is no longer a vague feeling. It is a visible piece of information.

The journal doesn't make your next decision better. It makes the decision after that better, and the one after that, because your heuristic is finally receiving feedback from your own record.

Two things the journal is not are especially important. It is not a place to record what you should have done in hindsight. That is narrative editing, and it teaches the brain to rewrite the past instead of updating the model. It is also not a place for vague feelings. An entry such as “I had a good feeling about this” may be honest, but it gives you little to examine later unless you also record what the feeling predicted.

The process should remain lightweight enough to survive an ordinary week. If writing the entry takes longer than the decision deserves, the system will be abandoned or used only when you are already unusually motivated. The most useful journal is the one that captures routine decisions before the mind has had a chance to make them look inevitable.

The Surprise Mechanism

The Brown University finding is worth lingering on because it points to one of the few debiasing levers that does not depend entirely on willpower. Genuine surprise — the kind that violates a specific prediction — produces a measurable physiological response and can help reset prior expectations. The brain has a mechanism for breaking out of expectation lock-in. You do not have to summon it through a speech about discipline. You have to create conditions in which a prediction can be compared with reality.

A decision journal creates those conditions by fixing the forecast in advance. Without the earlier record, almost any outcome can be absorbed into a flexible story. If the choice worked, you were right for the reasons you remember. If it failed, the warning signs were supposedly obvious. The written forecast makes both stories harder to construct.

This is also why environmental design usually beats motivational pep talks. You can arrange a kitchen so that vegetables are visible and cookies are not, reducing the number of choices that depend on discipline. In the same way, you can arrange a decision process so that the prediction is logged before the choice is locked in. The system then creates a pause at the point where the bias normally operates.

The analogy should not be pushed too far. A decision journal cannot control the environment in the way a rearranged kitchen can. It also cannot guarantee that a person will interpret feedback correctly. The value is narrower and more concrete: it makes the starting assumption harder to lose and the eventual surprise harder to explain away.

That distinction protects the experiment from becoming another self-improvement ritual. The purpose is not to feel more analytical. It is to discover whether your predictions are calibrated and whether the same error pattern keeps returning. If the journal does not change what you notice, it is not yet doing useful work.

The Fail-Safe

Here is the pragmatic close, because the question was never how to think harder. It was how to make sure the next decision is not simply a copy of the last one.

Three fail-safes offer the most leverage.

• Change the default. Most repeated decisions are not really decisions; they are paths of least resistance. Change the default and the bias has to rebuild its case. Try a different vendor, meeting time, commute, or review process when the existing choice is being repeated without fresh evaluation. The point is not novelty for its own sake. It is to interrupt the action-repetition heuristic when circumstances may have changed.
• Log before committing. Five lines in a notebook or note are enough for a minimum viable journal: choice, alternative, prediction, confidence, and the assumption that would have to fail. The format is deliberately plain. A complicated system gives you another reason to postpone the entry until after the decision.
• Schedule surprise reviews. At a regular interval, look through the journal with one question in mind: where was I most confident and most wrong? That gap is a useful part of your bias profile. It may point to a domain where you anchor too quickly, a kind of information you routinely dismiss, or a tendency to continue with a plan because stopping would make the previous investment feel wasted.

A fourth safeguard is to invite a deliberately opposing view when the stakes justify it. This does not mean asking someone to disagree for sport. It means identifying the strongest evidence against the preferred option before the decision is final. For confirmation bias, the quality of the opposing evidence matters more than the mere existence of a dissenting voice.

Another is to separate the decision from the identity attached to it. If changing course feels like admitting that you are careless, inconsistent, or incompetent, escalation of commitment becomes much more likely. A plan can stop making sense without the original choice becoming a moral failure. That is not a comforting slogan; it is a practical condition for updating.

The system does the work, not you. That is the entire point.

Cognitive bias tracking is not a personality project. It is an environmental one. The goal is not to become a person who never relies on intuition, repeats a choice, or makes a confident mistake. That person does not exist. The goal is to put a small amount of friction where the bias lives, preserve the forecast before hindsight gets involved, and make feedback specific enough to change the next round.

The journal does not turn decision making into mathematics. It does something more modest and more useful: it makes your own patterns difficult to ignore. Once you can see that the same first number keeps shaping your estimates, that the same kind of evidence keeps winning arguments, or that confidence keeps outrunning accuracy, the problem is no longer an abstract list of cognitive biases in daily life. It is a process you can redesign.

That is the experiment. Not eliminating bias, but catching the moment when a familiar answer starts pretending to be a considered one.