Track 9 · Behaviour · lesson 5
Researching until you agree with yourself
10 min
Track 9 · Behaviour · lesson 5
10 min
Four cards on a table. Each has a letter on one side and a number on the other. You can see E, K, 4 and 7.
The rule to test: if a card has a vowel on one side, it has an even number on the other.
Which cards must you turn over? Decide before reading on.
Predict
Confirmation bias is not usually a refusal to look at contrary evidence. It is something quieter: the searches you run, the sources you open and the questions you ask are all shaped by what you already think, so the contrary evidence never arrives to be refused.
The output feels like research, because it is research. Hours went in. Sources were read. The problem is that the sample was selected by the conclusion, and a selected sample cannot test the thing that selected it.
This is the part that makes the bias dangerous rather than merely common.
Someone who spends ten minutes forming a view holds it loosely. Someone who spends forty hours holds it firmly, and is right to — forty hours ought to count for something. But if all forty hours were spent gathering supporting material, the confidence went up and the evidence did not, in exactly the way a crowd's confidence goes up without its evidence growing.
So the effort is real, the feeling of rigour is real, and the belief ends up more certain and no better tested. The people most vulnerable to this are therefore not the lazy ones. They are the ones who do the reading.
The bias operates through three distinct steps, and they can be interrupted separately.
Search. The question you type decides most of the answer. "Is this a good investment" and "why do people lose money on this" return different internets.
Interpretation. Ambiguous evidence gets read as supporting whichever side you were already on. Both sides of a genuine disagreement can read the same report and each come away more confident.
Recall. Later, the supporting items come to mind easily and the awkward ones do not, so even a balanced sample becomes lopsided in memory.
Check
Telling yourself to be open-minded does approximately nothing, for the same reason telling yourself to be well-calibrated does approximately nothing. What works is changing the question, because the question is the part that selects the sample.
Ask what would change your mind, and write it down before looking. If nothing would, you are not holding a belief about the world, you are holding a preference, and it should be labelled as one.
Look for the strongest version of the other side, not the weakest. The weakest version is always available and its only function is to make you feel finished.
Find someone whose job or money depends on the opposite view. They have already done the work of assembling the contrary case, and they had an incentive to do it properly.
Two readings
Gathering material about a position until enough has accumulated to feel settled.
Naming in advance what evidence would overturn the position, then going to look for it specifically.
Both sides accept
Both accept that effort matters and that more information is generally better than less. The disagreement is about whether the direction of the search is part of the method or a detail.
Which is which? Put each one on a side.
Opening the reviews and sorting by lowest rating first
Asking a friend who agrees with you whether it sounds sensible
Writing down, before you start, the number that would make you walk away
There is a failure mode on the other side, and it is worth naming because this course could easily produce it.
Someone who has learned about confirmation bias can use it as a weapon: any argument they dislike gets dismissed as motivated reasoning, and any evidence that agrees with them is exempt because they thought about it carefully. The bias becomes something other people have.
The only defence is that the test is procedural, not psychological. It does not ask whether someone is being fair-minded, which is unknowable. It asks what evidence was searched for, whether the falsifier was named before the answer was known, and whether anything was ever found that moved the position. Those are questions you can answer about your own last three decisions, and the answers are usually informative.