The most convincing proof that your fix worked is often the least trustworthy evidence you have. And the moment you are most likely to be fooled is the moment you feel most sure.

Here is how it happens. Your weakest property has an ugly quarter. Occupancy slides, collections slip, complaints climb, and it lands at the bottom of every list you keep. So you step in. You move a stronger manager onto it, tighten up the process, and spend a few weeks paying close attention to a building that normally gets none. The next quarter, the numbers come back up. Not all the way, but clearly better.

The lesson writes itself. You made a change at the worst-performing property and the property turned around, so the change is what did it. You remember that intervention as something that works and reach for it again the next time another property struggles. There is another explanation, and statistically it is often the more likely one. The improvement may not have been caused by your intervention at all.

Extremes Do Not Stay Extreme

The reason has less to do with property than with how numbers behave. Any measurement that is partly skill and partly luck will, when it lands at an extreme, tend to be less extreme the next time you look. A quarter that bad was not purely the building's true level. It was the true level plus a run of bad luck: a cluster of move-outs that happened to bunch up, a maintenance failure that hit at the wrong moment, a collections month that went sideways. The true level carries into the next quarter. The bad luck does not. So the number rises, on its own, back toward the building's normal range.

This is one of the oldest results in statistics. Francis Galton described it in the 1880s, watching tall parents produce children who were tall but closer to average. He called it regression toward the mean, and the effect now usually goes by regression to the mean. It shows up anywhere a result is not perfectly repeatable, which is to say, everywhere. An extreme reading is a signal wrapped in noise, and the noise, being noise, does not come back for an encore.

The Trap Is That Your Fix Lands Exactly Where Regression Is Strongest

Daniel Kahneman told the sharpest version of this, from his time working with flight instructors. They had noticed that when they praised a cadet for a beautiful maneuver, the next attempt was usually worse, and when they screamed at a cadet for a clumsy one, the next attempt was usually better. Their conclusion was reasonable and completely wrong: praise ruins pilots, criticism sharpens them. In fact both were regression. A superb maneuver is a peak a cadet is unlikely to repeat, and a terrible one is a trough they are unlikely to revisit, regardless of what the instructor yelled in between.

Notice what made the instructors so sure. They only ever praised after the best flights and criticized after the worst ones. Their feedback was applied precisely at the extremes, which is exactly where regression guarantees the next result will move back toward normal. The timing of the intervention made pure statistics look like cause and effect.

Property runs the same trap, and runs it constantly, because of a habit that feels like good management. You intervene when things are at their worst. The building that gets the new manager and the sudden attention is, by definition, the one having its worst stretch. That is the responsible instinct, and it is also the one that all but guarantees the next quarter will look like a win, whether your intervention did anything or not.

The Playbook Fills Up With Fixes That Never Worked

Do this for a few years and something uncomfortable follows. Because you always act at the bottom, and the bottom always tends to rise, almost everything you try will appear to succeed. The good interventions and the useless ones produce the same after-picture: a rebound that was coming anyway. Attributing that rebound to whatever you happened to do is common enough to have its own name, the regression fallacy, and it quietly corrupts what an operation believes it knows about itself.

So the playbook fills with moves that have a perfect track record and no real effect. Worse, the genuinely effective fixes get no more credit than the theatrical ones, because both are graded against the same rebound. You are not learning what works. You are learning what you happened to be doing when the numbers came back up.

You Cannot See It From a Single Before And After

A single property, measured once before and once after, can never tell you the answer, because the rebound is real either way. To know whether your move mattered, you need two things that live outside that one comparison, and most operations track neither.

The first is the building's normal range, so you can see whether the bad quarter was ever outside its ordinary swing. The second is a rough control group: the similar buildings you did not touch over the same stretch. If the ones you left alone rebounded just as much as the one you fixed, your fix did nothing. Seeing one property's history next to the similar properties you did not touch is hard when the data lives in different places. That is the practical value of holding the whole portfolio's history together, which is what a platform like RIOO is for: it gives your intervention the control group it never had. Without both, you are judging the intervention in isolation, and isolation is exactly what makes regression so convincing.

The Discipline Is Being Willing To Say You Do Not Know

None of this means interventions are pointless. Sometimes the new manager really is the reason. It means you should hold your conclusions more loosely than the rebound invites you to, especially when you acted at a low point, because that is exactly the situation designed to fool you.

The habit worth building is small and slightly deflating. When a struggling property recovers after you stepped in, resist the clean story. Ask whether it was already at an extreme, whether the recovery is merely a return to its normal range, and whether the buildings you left alone did the same thing over the same period. Often the honest answer is that you cannot be certain the fix mattered, and that answer, uncomfortable as it is, is worth more than a confident one that happens to be wrong.

An operation that credits every rebound to its own cleverness slowly fills its playbook with superstitions. An operation willing to say I am not sure that worked keeps looking, and eventually finds the things that actually do. The most dangerous fix is not the one that fails. It is the one that never worked and convinced you it did.

FAQ

1. What is regression to the mean?
It is the statistical tendency for an extreme measurement to be followed by one closer to the average. When a result is partly determined by luck, an unusually high or low reading contains luck that will not repeat, so the next reading drifts back toward the normal level. It was first described by Francis Galton in the 1880s and applies to almost any measure that is not perfectly repeatable.

2. What is the regression fallacy?
It is the mistake of attributing that expected drift back to normal to a cause, such as an intervention, when it was going to happen anyway. If you act right after an extreme result and things then improve, the improvement may simply be regression, not the effect of your action. Crediting your action for it is the regression fallacy.

3. Why does intervening on my worst property usually look like it worked?
Because you intervene precisely when the property is at its worst, which is exactly the moment regression to the mean predicts the next result will improve on its own. The building was selected for attention because it hit an extreme, and extremes tend to moderate. So almost any intervention timed to a low point will be followed by a rebound, whether or not it did anything.

4. How can I tell whether an intervention actually worked or was just regression?
Compare against two things a single before-and-after cannot give you. First, the property's normal range of variation, to see whether the swing was ever outside its ordinary noise. Second, a comparison group of similar properties you did not touch over the same period. If the untouched properties improved just as much, the intervention likely did nothing.

5. Does this mean interventions never work?
No. Some interventions genuinely work, and the point is not to stop acting. The point is to be cautious about concluding that a fix worked simply because performance rebounded after it, especially when you acted at a low point. Hold those conclusions loosely, check them against normal variation and untouched comparables, and you will slowly separate the fixes that matter from the ones that only seemed to.