Tolman's Sunburst Maze 80 Years on: A Meta‐Analysis Reveals Poor Replicability and Little Evidence for Shortcutting

TL;DR

A classic experiment suggesting rats use cognitive maps to navigate has poor replicability, with most studies showing animals rely on simpler strategies instead.

Contribution

A meta-analysis reveals that the original Sunburst maze experiment is not reliably replicated, challenging the evidence for cognitive maps in rodent navigation.

Findings

Only 17% of experiments showed shortcutting, with most animals favoring adjacent or unremarkable paths.

Procedural or associative learning strategies are more common than cognitive map-based navigation in the Sunburst maze.

Neurophysiological evidence undermines the claim that shortcutting depends on cognitive maps.

Abstract

The Sunburst maze, first described 80 years ago by Tolman, Ritchie and Kalish (1946) and popularized by Tolman (1948), is widely regarded as a classic demonstration of cognitive map use in rats. In this task, animals trained on a circuitous path to a reward were presented with new paths, including a shortcut, after the original route was blocked. A substantial proportion of rats selected the shortcut, which Tolman et al. (1946; 1948) interpreted as evidence that animals have an internal spatial representation, or ‘cognitive map’. Despite the influence of this study, attempts to replicate it have been largely unsuccessful. This review critically examines a dozen replications involving rats, squirrel monkeys and humans, highlighting a range of alternative strategies, with only a fraction of experiments demonstrating shortcutting (17%). Instead, most studies found that animals either favoured paths adjacent to the original training route (32%), did not have a preference (26%), chose unremarkable paths (13%) or selected options consistent with previously rewarded responses (6%), suggesting a reliance on procedural or associative learning rather than demonstrating flexible spatial inference. Although the original experiment has been widely criticized for including a visual cue above the reward location, subsequent studies rarely found that this feature guided path choices (6%). Neurophysiological data from hippocampal lesion and head‐direction cell studies further undermine the claim that shortcutting in the Sunburst maze depends on cognitive maps. We argue that this study, though historically significant, is a poor standalone demonstration of map‐based navigation.

In 1946, Tolman et al. reported that rats could take a novel shortcut to a goal after training on an indirect route, supporting the Cognitive Map theory. However, a review of subsequent Sunburst maze studies shows this is an outlier: shortcutting has been observed in only a relatively small fraction of experiments. Instead, animals most often pick adjacent or unremarkable paths, pointing to response‐based strategies or novelty‐related confusion rather than map‐based spatial reasoning.

Figures