Source
Strack & Mussweiler (1997), LIT-1064, read in NOTE-877. Harris et al. (2019), LIT-1067, read in NOTE-876. Mussweiler & Strack (1999), LIT-1078, is Unavailable. Mussweiler & Strack 2000a, the studies Harris et al. replicate, is not held; what is said of it is Harris et al.’s account.
What was actually shown
The account. The comparison is solved by testing whether the target has the anchor value, which builds a model of the target from anchor-consistent knowledge; that knowledge stays accessible for the absolute estimate. In the abstract’s words, “anchoring is a special case of semantic priming”. The paper proposes it for plausible anchors; for implausible ones it allows an adjustment-like route.
The moderations (Strack and Mussweiler, N = 32, 32, 67). With the anchor number fixed, the effect was large when the estimate was on the dimension compared and small when it was on another (Study 1, interaction F(1, 28) = 10.50, n = 8 per cell), and reversed to contrast when the estimate was about a very different object (Study 2, interaction F(1, 28) = 6.30). With plausible anchors the comparison was slower and the estimate faster than with implausible ones (Study 3, interaction F(1, 66) = 72.81). A numeric-priming account predicts none of these, and the paper argues against it on that ground. No study measures what was retrieved.
The direct test, repeated. Harris et al. ran Mussweiler and Strack 2000a’s test (does a high or low anchor speed recognition of anchor-consistent words?) five times, preregistered, N = 128 each, four by lexical decision and one by continuous identification. The anchor × word-type interaction was null every time (η²G ≤ .004), and an exploratory meta-analytic Bayes factor found the data “21.10 times more likely under the null hypothesis than the alternative”. The originals had interactions at p = .043 and .016, with none of four simple effects significant. Then the same tasks failed to detect direct picture priming whose manipulation check was strong. The authors’ conclusions: “these are not good signature tests”, the results do not “necessarily provide additional evidence for or against that theory, which we still consider a good explanation”, and past support from the method “is not a strong part of the evidence base”.
What this does not say
- That selective accessibility is refuted. The failed test cannot detect priming at all, so its null says nothing about retrieval either way. What is rejected is the earlier lexical-decision result as evidence, not the account.
- That accessibility has been shown. Every remaining line of support infers it: moderation by applicability and representativeness (small samples), latencies (which time the estimate, not what is retrieved), coded free reports (LIT-1078 and 2000a, unread), and an overlap result whose explanations, Harris et al. note, rest on “presumably”. Harris et al.’s continued preference for the theory rests on its fit with hypothesis testing and priming, a theoretical virtue, not evidence.
- That accessibility fixes the judgement. The account has two stages. Accessibility is “not sufficient to guarantee its use and determine the direction of its influence”, and Study 2 turned the same activated knowledge into contrast. A frame acts on retrieval and on evaluation, and these designs do not separate the two.
- That it covers all anchoring. Implausible anchors anchored at least as strongly, and the paper assigns them to adjustment. Insufficient adjustment is not ruled out. Harris et al.’s leading account of durable anchoring, a provisional answer formed at the comparison and recalled, is neutral between mechanisms.
- That the retrieval is selection from a fixed store. The paper’s mechanism is construction: judges “retrieve or generate” and “construct a mental model”; the biasing information “is generated by judges themselves”.
- That hypothesis-consistent retrieval trades accuracy now for bias later. The paper says the positive test strategy “seems to generate information that improves the validity of comparative judgments but distorts subsequent judgments for which no specific hypothesis has been provided”. That is the authors’ reading of their model, not tested.
Connections
- THEORY-337 is the effect this account explains, and stands whatever happens to it.
- QUESTION-109 asks when cue-led retrieval helps and when it produces anchoring. This gives a conditional, untested answer for one case: the cost arises when the anchor is plausible (knowledge is consulted), the retrieved knowledge applies to the later judgement, and it is representative of the target. Harris et al.’s durability result fits the question’s path-dependence worry: the carrier outlasts brief lexical activation.
- CLAIM-298 (whose psychological form is now CLAIM-315) cites this account as the psychological form of its thesis and restates its defeat condition on the same terms as this account’s former promote_when, now QUESTION-323: a measure first shown sensitive to direct priming.
Note of 2026-10-30: the dropped clause
The ADR-047 kind-1 sweep sorted what the twelfth status review dropped from this entry’s title. The review changed no scope. It attributed the moderations to Strack and Mussweiler and the repeated test to Harris et al., and it corrected one clause: version 1 called the test “the lexical-decision test” and its measure “a task that also missed direct priming”. That was imprecise. Harris et al. used lexical decision in four of the five experiments and continuous identification in the fifth, and it was direct picture priming that both tasks missed (NOTE-876). It is a mis-statement, not a wider claim, so it gets this note and no Proposed entry. Whether the accessibility step is real is QUESTION-323.