The before/after photo as a survivorship-bias trap

A wall of before-and-after photos is genuinely persuasive, and I am not going to pretend the persuasion is irrational. Those results are usually real. The crowns in the photographs were really placed, the smiles really did transform, and the patients really were pleased on the day the photo was taken. I will concede all of it. If your only question were “can this clinic ever produce a good result,” a gallery answers it, and the answer is yes.

But that is not the question you are actually asking. You are asking how often the result is good, and how badly it goes when it is not. On that question, the gallery is not weak evidence. It is structurally incapable of carrying evidence at all, because of the way it was assembled. A before-and-after gallery is built exclusively from cases that worked. The failures do not post photos. They do not send thank-you notes. They quietly disappear from the dataset you are being shown, and their disappearance is not a flaw in the gallery. It is the gallery’s defining property. This is the literal definition of survivorship bias [1], and once you see it, you cannot unsee it.

The planes that did not return

The cleanest illustration of this error comes from the Second World War, and it is worth telling because the structure maps exactly onto a dental gallery [1].

Military analysts wanted to armor bombers against anti-aircraft fire. They studied the planes that came back, mapped the bullet holes, and proposed reinforcing the areas where the damage clustered. It looked like sound, data-driven reasoning. The statistician Abraham Wald saw the fatal flaw. The analysts were studying only the planes that returned. The planes that took hits in other places, the engines, the cockpit, were the ones that did not come back to be studied. The clustered damage on the survivors marked the places a plane could be hit and still fly home. The armor, Wald argued, belonged where the survivors showed no damage, because damage there was what removed a plane from the sample.

Sit with the shape of that, because it is the whole lesson. The data in front of the analysts was real, accurate and complete for what it contained. The error was not in any single observation. It was in mistaking the visible sample for the whole population, when the sample had been silently filtered by survival. The missing planes carried the most important information, and they were missing precisely because of the thing you most needed to know.

A before-and-after gallery is the returning planes. Every photo is real. The error is identical: treating the curated survivors as if they told you about the full population of cases, when the cases that failed were filtered out by the same mechanism that built the gallery.

Why a gallery cannot contain a failure rate

Here is the falsifiable core of the argument. A failure rate is a fraction: failures divided by total cases. A gallery shows you a numerator made entirely of successes and gives you no denominator at all. You cannot compute a rate from it, not because the photos are fake, but because the quantity you need was removed before you arrived.

Consider two clinics. Clinic A has done a thousand implant cases, with an excellent outcome record. Clinic B has done a thousand implant cases with a poor outcome record and a trail of complications. Now ask what their galleries look like. Both can post hundreds of beautiful before-and-after photos, because both produced hundreds of good results in absolute terms. The galleries are visually indistinguishable. The single most important difference between the two clinics, how often things go wrong, is the one thing the format cannot display. A gallery is a flattering portrait of the numerator with the denominator cropped out of frame.

This is why “look how many happy cases they have” does not work as reasoning. A large gallery tells you a clinic does high volume and curates well. It is consistent with excellent care and consistent with poor care. When a piece of evidence is equally consistent with the conclusion and its opposite, it is not evidence. It is decoration. The technical name for the broader family of this error is selection bias, of which survivorship is a particular form where losses are removed from the sample [2][3].

The base-rate trap riding alongside it

Survivorship bias rarely travels alone. It usually arrives with a base-rate problem [4]. When you are shown a hundred glowing results, your intuition quietly updates toward “this almost always works,” because vivid, available examples crowd out the abstract rate you never see. But the strength of your impression is driven by how many successes were shown to you, which is a marketing decision, not by the underlying frequency of success, which is the base rate.

A hundred photos feels like overwhelming evidence of reliability. It is overwhelming evidence of one thing only: that the clinic chose to show you a hundred photos. The vividness is doing the persuading, and vividness is uncorrelated with the failure rate. This is the same machinery that makes the expected-value cost of a failed implant so counterintuitive: a low but real failure probability, multiplied by a large revision cost, dominates the decision, yet it is invisible in a gallery that contains zero failures by construction.

How to deliberately look for the missing planes

Correcting survivorship bias is not a matter of distrusting every photo. It is a matter of deliberately seeking the data the format excludes. Wald did not throw out the returning planes. He asked where the missing ones were hit. You do the same.

The move is to stop asking for more survivors and start asking for the denominator and the failures. Ask a clinic about its documented complication and failure rates over time. Ask what happens when an implant fails: who pays for the revision, where it is done, and on what timeline. Ask what the warranty actually covers and, more revealingly, what it excludes. Ask how the clinic tracks outcomes for patients who have flown home, since those are exactly the patients most likely to vanish from the visible record. None of these questions can be answered with a photo, which is why they cut through the gallery.

You are also, in effect, asking the clinic to show you its planes that did not return. A clinic with strong outcomes can usually speak about its failures with composure, because a known, managed failure rate is a sign of honesty rather than weakness. Evasion, or a pivot back to more happy photos, is itself information. The structural reasons clinics are incentivized to keep the denominator out of view are the subject of the package-deal overtreatment incentive and the wider dental tourism trust gap.

The checklist

WHEN A GALLERY OR TESTIMONIAL WALL IS DOING THE PERSUADING

Reframe what you are seeing
[ ] Recognize the gallery as the SURVIVORS only (the returning planes)
[ ] Remind yourself: it shows the numerator, never the denominator
[ ] Note that gallery SIZE proves volume + curation, not quality

Ask for the missing planes (the failures / the denominator)
[ ] Documented failure / complication rate over time
[ ] What happens when it fails: who pays, where, how fast
[ ] What the warranty EXCLUDES (more telling than what it covers)
[ ] How outcomes are tracked AFTER patients fly home

Weight the evidence correctly
[ ] Curated photos -> low signal (decoration)
[ ] Independent outcome data + traceability -> high signal
[ ] Calm, specific talk about failures -> a good sign, not a red flag

What a patient should verify

Reduced to its essentials, correcting for survivorship bias is a short discipline you can apply to any clinic’s marketing.

• Treat any before-and-after gallery as a sample filtered by success, not as an outcome rate.
• Refuse to let gallery size stand in for quality; it measures volume and curation only.
• Ask directly for failure and complication rates, and notice whether the answer is specific or deflected.
• Find out, in writing, who bears the cost and effort of a revision if something fails.
• Read what the warranty excludes, not only what it promises.
• Weight independent, longitudinal outcome data far above any quantity of curated images.

The reasoning behind each item is the same and it is falsifiable: a gallery is consistent with both good and bad care, so it cannot move your estimate, while failure rates and warranty exclusions discriminate between them and therefore can.

The honest bottom line

The photos are real. That is the trap, not the exemption. Survivorship bias does not work by showing you fakes. It works by showing you a true but filtered slice of reality and letting you mistake the slice for the whole. The before-and-after wall is the set of planes that made it home, and the cases you most need to learn from are the ones that, by the gallery’s very design, were never going to be on the wall.

So do what Wald did. Do not argue with the survivors. Go looking for the ones that did not return. Ask for the failure rate, the revision terms, the warranty exclusions, the post-departure tracking. If a clinic can meet those questions calmly and specifically, that tells you more than a thousand photographs. If it cannot, the silence where the failures should be is the most informative thing in the room.

For the companion reasoning tools, see the expected-value cost of a failed implant, verifying an implant brand and lot number before surgery, and the records to obtain before you leave a dental clinic abroad. For whether to travel at all, see when to go overseas for dental treatment. Our standards are at methodology and disclosures.