Levels of analysis
The same behavior can be explained at several layers—from neurons to evolution—each a valid lens.
What it means
Levels of analysis is the principle that any behavior or mental phenomenon can be described and explained at multiple, complementary levels, and that a full account requires more than one of them rather than reducing everything to a single layer. Two influential schemes formalize this. Tinbergen's four questions distinguish explanations by mechanism (the immediate causes), development (how the behavior arises over a lifetime), function (its adaptive value), and evolution (its phylogenetic history), separating 'how' questions from 'why' questions in biology. Marr's three levels of analysis for information-processing systems distinguish the computational level (what problem is being solved and why), the algorithmic level (what representations and procedures solve it), and the implementational level (how it is physically realized in the brain). The key insight is that explanations at different levels do not compete—a behavior has a neural mechanism and an evolutionary function at once—so confusing or collapsing levels produces spurious debates. It matters because it disciplines theorizing in behavioral science, clarifying what kind of question is being asked and what kind of evidence answers it.
Examples
Asking why birds migrate, one can answer with hormonal triggers (mechanism), the survival advantage of following food (function), or ancestral history (evolution)—all true at different levels.
Ask why someone flinches at a spider: startle circuits firing (mechanism), a childhood fright (development) and ancestors who avoided venom (function) are all correct at once.
For vision, Marr's point is that 'recognise a face' is the problem being solved, the matching procedure is the algorithm, and firing neurons are the implementation — three answers, one system.
First described in Niko Tinbergen (1963); David Marr (1982).