Compensated data show negative events smeared below zero. Does that mean compensation is wrong?

Why

Compensation subtracts a fixed fraction of the signal in one detector from another. The subtraction is exact on average, but the measured light is noisy. Photon counting error scales with the square root of the signal, so a cell bright in PE has a large error in the detector PE spills into.

After subtraction that error remains as spread, centered on zero, including negative values. Roederer showed this spreading is unavoidable and gets worse with red dyes and large spillover. He concluded that compensation cannot be judged by eye on dot plots, and that fluorescence-minus-one controls are needed to place gates.

A log display hides values at or below zero and piles them on the axis, which looks like a problem. A biexponential display shows the spread symmetrically. Real errors look different.

Undercompensation tilts a positive population upward into the other channel. Overcompensation tilts it downward, with the median of the positives below the median of the negatives. Those are judged by comparing medians, not by the shape of the smear.

For example, CD8 BV510 bright cells spreading in BV605 can be correctly compensated while their spread still hides a dim BV605 marker.