Contiguity is limited in free recall

Does the “law of contiguity” apply to free recall? I find that conditional response probabilities, often used as evidence for contiguity in free recall, have been displayed insufficiently, limiting the distance from the last item recalled, and only averaged over all items. When fully expanding the x-axis and examining the conditional response probabilities separately for the beginning, middle and end of the presented item list, I show that the law of contiguity applies only locally, even then only sometimes, and breaks down globally.

💡 Research Summary

The paper revisits the long‑standing assumption that the “law of contiguity” governs free‑recall performance. The law posits that after recalling an item, the next item retrieved is more likely to be one that was presented nearby in the original list. Researchers have traditionally measured this effect using conditional response probabilities (CRPs) that are averaged across all list positions and limited to a narrow lag window (typically –5 to +5). The author argues that such practices obscure the true shape of the contiguity function and may overstate its generality.

To address these concerns, two methodological extensions are introduced. First, the lag axis is expanded to cover the full range of possible distances (from –N to +N, where N is the list length), allowing the detection of long‑range transitions that would be invisible in a truncated plot. Second, the list is divided into three segments—beginning, middle, and end—and CRPs are computed separately for each segment. This segmentation acknowledges that encoding and retrieval strategies can differ across the serial position of items.

Data were drawn from standard free‑recall experiments with word lists of 16–24 items. For each participant, the sequence of recalled items was transformed into a series of lags (the positional distance between successive recalls). CRPs were then calculated for each lag value, both globally and within each serial‑position segment.

The results reveal a nuanced picture. In the early portion of the list, a pronounced peak at lag = ±1 is evident, confirming that local contiguity operates strongly when the most recent recall is among the first few items. In the middle segment, the peak diminishes and the CRP curve flattens, indicating that the influence of immediate neighbors is weaker and that more distant items are retrieved with non‑trivial probability. At the end of the list, the pattern reverses: participants often jump forward to items that were presented later in the list, producing a modest peak at positive lags and a trough at negative lags. When the CRPs are averaged across the entire list, the familiar “U‑shaped” curve emerges, but this shape is driven primarily by the early‑list data; the global curve masks the segment‑specific variations. Moreover, for lags beyond ±10 the CRP approaches chance level, showing that the contiguity effect does not extend across the whole list.

These findings lead to several important implications. First, the law of contiguity appears to be a local, not a global, constraint on memory search. It is most reliable for items that are temporally adjacent early in the study phase, but its influence wanes for items in the middle and can even reverse near the list’s end. Second, the traditional practice of reporting only a narrow lag window and a single averaged CRP may give a misleading impression of a robust, universal contiguity effect. Third, models of free recall that rely heavily on temporal context (e.g., the Temporal Context Model) should incorporate segment‑specific parameters or additional mechanisms—such as semantic clustering, strategic reconstruction, or goal‑directed retrieval—that can account for the observed deviations from pure contiguity.

Methodologically, the paper recommends that future studies present full‑range CRP plots and segment‑wise analyses as standard practice. Doing so would provide a clearer view of how temporal proximity interacts with other cognitive processes across the serial position spectrum.

In sum, the study demonstrates that while contiguity does influence free recall, its reach is limited to local neighborhoods and is highly dependent on where in the list the recall occurs. Global contiguity, as traditionally conceived, does not hold across the entire list, suggesting that free‑recall behavior is shaped by a combination of local temporal associations and broader, non‑temporal strategies. This refined understanding has significant theoretical ramifications for memory research and practical implications for the design and interpretation of free‑recall experiments.