There's a 2002 experiment by Patrick Haggard that I keep turning over. The setup is simple: a clock face with a hand sweeping one revolution every 2.56 seconds. You press a key whenever you want. Sometimes a tone sounds 250 milliseconds after the keypress, sometimes not. You report where the clock hand was at the moment of either the keypress or the tone.
When there's no tone, people place the keypress where it actually happened. When there's a tone afterward, something different occurs: the keypress is judged as about 15 milliseconds later than it actually was, and the tone is judged as about 46 milliseconds earlier than it actually was. Both events are pulled toward each other. The 250-millisecond gap between action and effect, as perceived, shrinks to roughly 190.
That's not the interesting part yet.
The interesting part is what happens when the experiment is run with involuntary movements instead. A TMS device induces a muscle twitch — the finger moves, but you didn't initiate it. The twitch is followed by the same tone 250 milliseconds later. Now the errors reverse: the twitch is judged as earlier than it actually was, the tone as later. The two events are pushed apart.
So the direction of the error depends entirely on whether you meant to move. Voluntary action: events compress together. Involuntary action: events spread apart. The actual elapsed time is the same in both cases. What changes is who authored the movement.
This is usually described as "intentional binding," and one interpretation is that it reflects a predictive mechanism. When you intend an action, the motor system generates a prediction of what sensory consequence will follow. The binding might be what happens when that prediction and the actual result are integrated — the two events get treated as belonging to the same thing, so their felt timing converges. The tone wasn't an independent event that happened to follow your action; it was the effect your action produced, and the perceptual system registers that as closeness in time.
The inverse makes the logic cleaner. When the movement was involuntary — when you didn't produce the prediction — the events remain separate. They're registered as unrelated, which shows up as increased perceived distance.
What this means is that the felt timing of an event isn't a neutral record of when it occurred. It's already been edited. The gap between your action and its effect is not perceived as-is; it's been compressed or expanded based on a prior judgment about authorship. By the time the event reaches conscious report, the causal story has been baked in.
There's a finding with schizophrenia that extends this. Patients with schizophrenia show an amplified binding effect — the compression is stronger than in controls. The causal attribution mechanism appears to be over-binding, pulling together events that don't belong, marking connections that aren't actually there. Some researchers think this might relate to experiences like thought insertion, where external events feel self-caused. The mechanism that normally correctly stamps "I caused that" starts stamping incorrect connections with the same stamp.
And there's a dissociation: binding can occur without an explicit sense of agency. You can show the temporal compression while someone reports not feeling particularly in control. The implicit causal marking and the explicit narrative of authorship can come apart. This suggests that what shows up in the perceived timing is not identical to what someone would tell you if you asked them whether they felt responsible. Two things that are supposed to be the same thing — "I caused that" — seem to be computed by at least partially separate systems.
I don't know how to think about this completely. The obvious conclusion is something like: the question of when something happened and the question of whether you caused it are entangled at the level of perception. They're not answered separately and then combined into experience. The causal attribution happens first, or at least simultaneously, and shapes the temporal record before it arrives at awareness.
But I'm not sure what that leaves us with. If you ask someone when they pressed the key, they give you a number. That number is not what the clock would have said. It's the number after the edit. Whether they know an edit happened — whether there's any felt sense of the gap having been adjusted — seems to be no. The compression is seamless. The perceived timing feels like timing, not like a verdict about agency dressed up as timing.
The version that troubles me is: if the felt moment of "when" already contains a judgment about "whether I caused it," and that judgment is sometimes wrong, then there's a way in which the very form of temporal experience — the structure of before and after, cause and effect — is not a neutral scaffolding. It has a stake in the answer.