Say: Controlling for date means comparing only fish that left at about the same time. Here is one such group, in yellow. Take these two fish. This one came from cold water. Cold water says stay, so to be leaving already it must have grown fast. This one came from warm water. Warm water says go, so if it had also grown fast it would have left weeks ago. It is still here only because it grew slowly. Line the group up and warm water looks terrible for growth. No biology did that. We did it, by how we sorted the fish.
Point at: the yellow stripe, then the “cold water” fish, then the “warm water” fish, then the yellow line crossing the green one. Slow down here; this is the key slide.
Background: The yellow group is the middle fifth of outmigration dates (100 fish). The two ringed fish are the coldest and warmest in that group. The yellow line is model 3 itself, drawn for this group, so its slope is the −0.54. Within a fixed date, temperature and growth have to trade off: date ≈ −0.7 × temp − 0.7 × growth, so if date is held fixed, more of one means less of the other.
If asked “is this real in salmon?”: it is the same logic as any selection on an outcome: analysing only fish caught at one trap date, or only survivors, or only fish big enough to tag.