A second opinion is worth exactly what its independence is worth

Why asking twice helps at all

The reason two assessments beat one is not that a second person knows more. It is that each assessment contains signal plus error, and if the errors are unrelated they partly cancel when the answers are combined while the signal does not. That is the whole mechanism, and everything about the value of a second opinion follows from how well the assumption of unrelated errors holds.

When it holds, the gain is substantial and cheap. When it fails completely — when the second answer would have made exactly the same mistakes for exactly the same reasons — the second opinion adds nothing except confidence, which is the worst possible thing to add to an unreliable estimate.

The ways independence quietly disappears

The most common is simply telling the second person what the first concluded. Once an answer is in the room it operates as a starting point, and the second judgement becomes an adjustment from it rather than a fresh estimate. This happens by default in almost every real process, because handing over the file means handing over the conclusion written at the top of it.

The others are structural. Both people may be working from the same underlying material, so any error in the material is inherited twice. They may share a training and a professional consensus, which means they share whatever that consensus gets wrong. And the second may have been chosen by the first, which selects for agreement in a way nobody intends and nobody notices.

Agreement between dependent sources is not confirmation

Two people reaching the same conclusion feels like strong evidence, and how strong it actually is depends entirely on how much they share. Two readings of the same ambiguous image by people with the same training will often agree, and their agreement is largely a fact about the training.

This is the practical reason to ask about provenance before treating concurrence as confirmation. The useful question is not whether they agree but whether they could have disagreed — what would have had to be different for the second opinion to come out otherwise. If nothing available would have changed it, the second look was ceremonial.

What a genuinely independent second look requires

Three things, roughly. The second assessor forms and records a judgement before being told the first. They work from source material rather than from a summary written by the first, since a summary encodes the first person’s interpretation. And they are not selected by, employed by or dependent on the party whose conclusion is being checked.

Meeting all three is often impossible, and the point of listing them is not to insist on perfection. It is that each one you cannot meet reduces the value of the exercise by a known amount, so you can at least tell whether you have bought a real check or a reassurance.

Where a dependent second opinion is still worth having

Plenty of checking has nothing to do with independent judgement. Verifying an arithmetic result, confirming that a step was performed, catching a transcription error — these are cases where the second person is looking for a slip rather than forming an opinion, and sharing all the same background is irrelevant to whether they can spot it.

The distinction is between errors of execution and errors of judgement. A shared framework does not stop anyone catching a mistake within that framework; it stops them noticing that the framework was the wrong one to apply. Knowing which kind of error you are worried about tells you what kind of second look to arrange.

When you cannot get independence

Two partial substitutes are worth using. The first is to seek a source with a different professional background, accepting that they will know less about the specifics, because their errors are at least unlikely to be the same errors. Overlapping partial views are often more informative than a second identical one.

The second is to stop asking for a conclusion and start asking for the reasoning behind it. Two people who agree may have very different grounds, and comparing the grounds reveals whether the agreement is convergence or inheritance. Ask what would change their mind. If both give the same answer to that, you have one opinion held by two people.

Neither substitute makes the check as good as a properly independent one, and saying so plainly is part of using them well. What they buy is a better sense of how much weight the concurrence deserves, which is usually less than it felt like before the question was asked.