Chaining models are among the oldest theoretical accounts of serial order memory, with roots in British associationism and the stimulus-response framework of behaviorism. In their simplest form, they propose that items in a sequence are linked by direct inter-item associations: recalling item A activates item B, which activates item C, and so on. Despite their intuitive appeal, chaining models face fundamental challenges that have driven the development of alternative frameworks.
Simple and Compound Chaining
In simple chaining, each item is associated only with its immediate successor. Retrieval proceeds as a Markov chain: the current item serves as the sole cue for the next. The probability of correctly recalling the item at position i+1 depends on the strength of the association from item i to item i+1 relative to all other associations from item i:
Compound chaining extends this by using multiple preceding items as retrieval cues. The cue for position i+1 is some function of items at positions i, i−1, i−2, etc., providing richer contextual information for retrieval.
The Ranschburg Effect and Repeated Items
Chaining models face a critical challenge with repeated items. If item X appears at positions 2 and 5 in a sequence (e.g., A B X C D X F), the chain from X must lead to both C and F. The Ranschburg effect (difficulty recalling the second occurrence of a repeated item) has been both cited in support of and against chaining, depending on the specific model variant.
Empirical Challenges
Several key findings challenge simple chaining accounts. First, transposition errors tend to preserve item position (items move to nearby positions), which is predicted by positional models but not by chaining. Second, the fill-in effect (when item B is omitted from sequence A B C D, item C tends to move forward rather than being lost) is inconsistent with simple chaining because the broken link at B should prevent retrieval of all subsequent items. Third, people can recall sequences in reverse order and from arbitrary starting points, which is difficult to explain with forward-only chains.
Modern Chaining Variants
Despite these challenges, chaining mechanisms persist in modern models as one component of hybrid architectures. The TODAM model uses chaining via convolution-based associations. Lewandowsky and Murdock (1989) implemented a chaining model using TODAM's convolution algebra that could account for many serial recall phenomena. The SOB model (Farrell & Lewandowsky, 2002) incorporated an energy-based mechanism that functions partly through inter-item associations. Most current models of serial order include some form of inter-item association alongside positional coding.
Although pure chaining models have fallen out of favor in memory research, associative chaining remains relevant for highly practiced sequences such as reciting the alphabet, playing a musical piece from memory, or typing familiar words. In these domains, the automatic activation of each element by its predecessor is subjectively compelling and may reflect genuine chaining mechanisms built up through extensive practice.