On my camera, a Canon PowerShot G1X, I can select picture sizes from S to L.
Now I wondered, if I select a smaller picture size, that would theoretically lead to more sensor surface per pixel, possibly making the camera perform better in low-light or other 'harsh' conditions.
However, when I try this, I can't notice any difference in behaviour. The end result just looks like a resized version of the larger picture, and there are no differences in shutter speed and ISO to get to that result.
So, it looks like the camera just takes a large picture and then downscales it, which was against my expectations (or at least against my hopes).
It certainly has advantages over downscaling by software, because it makes more efficient use of the hardware (faster reading, better noise reduction).
I think the answers below don't really answer my question. Neither do the answers of the supposedly 'duplicate' of this question, although the answer by mattdm came closest by at least suggesting that a camera might support this 'hardware-level binning', which allowed me to continue my search.
So far I couldn't find evidence that the Canon G1X supports binning, so I'm afraid it does not. But if someone know if it does (or maybe knows a hack or trick to enable it), please let me know. Previously I had a Canon PowerShot SII, and there was a way to install 3rd party firmware to unlock extra features. Maybe something like that is possible for the G1X as well?