Confusing test items bias respondents toward agreement, not disagreement

Evidence

Yet you often find more agreement than disagreement overall on balanced, valid psychological tests. On the RWA Scale, for example, with its long, complicated items, the breakdown is usually about 60-40. Moreover, if you give subjects a chance to correct earlier mistakes by having them reanswer the RWA Scale, compare their two sets of answers, and then decide what their final answer will be, they move about 10% of their responses from one side of the neutral point to the other—mostly from agreement to disagreement. These changes tend to occur on early items (especially the first contrait) and on longer items where confusion may have led to a yea-say (EOF, p. 45). — Bob Altemeyer, The Authoritarian Specter, 809

Categorization

• XREF: Connects to the acquiescence/yessing response set literature in survey methodology — Sidanius and Pratto’s work on response sets, and contemporary work on survey design and item wording bias. — connects to: Pre-1962 survey phrasing ambiguously mapped opposition to integration. — Ilyana Kuziemko and Ebonya Washington, Why Did the Democrats Lose the South?…, 1330
• QUESTION: Wonder whether this yea-say bias on long/complicated items is specific to authoritarian-content scales or generalizes across survey instruments — is it really about item length and confusion, or content?