The serial position effect is one of the most replicated findings in memory research (Murdock 1962; Atkinson & Shiffrin 1968): free recall of a list traces a U-shape — early items (primacy) and the last few items (recency) are recalled far better than the middle.
This sim runs a simplified multi-store simulation. Each new item enters a rehearsal buffer of fixed capacity K (Miller's "magic number", Cowan's ~4 slots). While an item sits in the buffer it accumulates rehearsal every tick, building a long-term memory (LTM) trace:
buffer.length >= K → evict one random occupant, insert new item
every tick: rehearsal[i] += gain for every i currently in the buffer
LTM(i) = rehearsal[i] / max(rehearsal)
Early items enter before the buffer is full, so they rack up rehearsal for longer before ever risking eviction — that surplus is the mechanistic source of primacy. Faster presentation rates shrink the rehearsal gain per tick (less time to encode each word), flattening the primacy limb — the classic presentation-rate effect (Murdock, 1962).
Items still sitting in the buffer at the end of the list carry a separate, fast-decaying short-term (STM) activation:
STM(i) = inBuffer(i) · e^(−λ · delay) λ ≈ 0.15 s⁻¹
A recall delay filled with a distractor task lets STM decay before test — exactly the manipulation Glanzer & Cunitz (1966) used to show the recency bump selectively vanishes with delay while primacy survives, because LTM traces don't decay on that timescale. Final recall probability blends both stores:
P(recall | i) = clamp( 0.15 + 0.55·LTM(i) + 0.75·STM(i), 0, 1 )
- List length — longer lists overflow the buffer sooner, sharpening the primacy/middle contrast.
- Presentation rate — slower study time per item increases rehearsal gain and strengthens primacy.
- Recall delay — drains STM activation; push it high enough and the recency bump disappears while primacy remains.
- Buffer capacity — how many items can be actively rehearsed at once; a smaller buffer means fewer items enjoy the primacy advantage.