Source
Harris, Blower, Rodgers, Lagator, Page, Burton, Urlichich & Speekenbrink (2019), LIT-1067, read in NOTE-876 from the authors’ submitted manuscript (the version of record was not reachable). Strack & Mussweiler (1997), LIT-1064, read in NOTE-877.
What was actually shown
The preregistered replications. Each experiment asked a comparative question (is the quantity higher or lower than the anchor), then a word-recognition task, then an absolute estimate. Anchors were set at the 5th and 95th percentiles of a calibration sample, except in Experiments 3 and 4, where they were chosen to match the critical words. Low against high anchor means (Table 2): Experiment 1, 12.68 against 14.70°C; Experiment 2, £10,754 against £19,742; Experiment 3, 10.71 against 15.88°C; Experiment 4, £13,817 against £61,452; Experiment 5, 11.03 against 16.16°C; all ps < .001, computed d ≈ 0.87, 1.06, 1.39, 0.85 and 1.34. These held although 70 lexical-decision trials (40 identification trials in Experiment 5) came between the comparison and the estimate. The authors: “the robustness of anchoring effects is remarkable”.
The original studies. In Strack and Mussweiler’s three studies (N = 32, 32 and 67), participants told the anchors “were randomly selected” still assimilated to them when the estimate was on the dimension compared (Study 1: a difference of 1.75 z between high and low anchors), and implausible anchors, such as a 45,000 km Elbe, “were at least as effective as plausible ones” (Study 3). When the estimate was on another dimension the difference fell to 0.24 z (Study 1), and when it was about a very different object the effect reversed to contrast (Study 2).
What could have come out the other way. Five preregistered experiments at about four times the originals’ N could have found a small or absent effect, as they did for the accessibility measure run alongside it; they found a large one each time.
What this does not say
- Why the estimate moves. Neither paper measures the carrier. Harris et al. found no change in word recognition alongside the shift, with a task that also missed direct priming (THEORY-335). Their leading account of the durability is a provisional answer formed at the comparison, which “may result from (e.g.) selective accessibility, anchoring-and-adjustment, scale distortion…”.
- That anchoring holds outside this paradigm. Both papers use a comparative question about a general-knowledge quantity, with samples in Germany (students) and the UK. Nothing here is about anchors met in conversation, negotiation or expert judgement, or about anchors not explicitly compared.
- That any anchor works the same. The moderations by dimension and object (Strack and Mussweiler, Studies 1 and 2) rest on n = 8 per cell and have not been replicated in the record; Study 2’s anchor main effect was F < 1, with the effect reversed for the other object.
- That anchoring is a cost. No study here scores accuracy against a true value in a way that separates harm from help. Whether cue-led judgement is a bias needs a standard outside the search (CLAIM-292).
- How long it lasts. Seventy trials is a lower bound on the delay the effect survives in these designs, not a measured decay.
Connections
- THEORY-335 is the mechanism proposed for this effect and its status. This account holds whichever mechanism is right.
- QUESTION-109 asks when cue-dependent retrieval produces anchoring. This establishes the anchoring in one paradigm; it does not show that retrieval produces it.
- CLAIM-184 says anchors entrench in a record; that is an analogy to this effect, not supported by it.