KurisutinaSurvey Practice 11(2), doi 10.29115/SP-2018-0027; open access. Read by researcher R4d (the base battery), 5 October 2026, as the batch's paper with human reference data for answer-changing under challenge. Provenance: papers/base/marshall2018_counterarguments.provenance.json.
What was read
- Read in full: the publisher's PDF (9 pages), converted with
pdftotext -layout: abstract, introduction, methods, analysis and results, Table 1, conclusion, acknowledgment, references. - Not read: the iPOLL source questions and data (not distributed with the article), the cited monographs (Sniderman and Piazza 1993 and others).
What it is
- The counterargument format. A respondent answers a policy question (the "prior question"). Then, depending on the answer, they hear information pointing the other way and are asked whether they would still give their answer ("What if you heard that…?"). A switch is a move to the opposing position, to "don't know", or to a refusal.
- Example (Kaiser Family Foundation, March 2017, required maternity coverage): of initial supporters, 28% switched after hearing that some people would pay for benefits they do not use; of initial opponents, 43% switched after hearing that coverage would become unaffordable.
- Data. A meta-analysis of every identifiable counterargument in the Roper Center's iPOLL archive of US public polls, found by phrase search ("what if you heard", "suppose", "still", …): 138 clusters with 287 counterargument questions, 1973 to March 2018. Topics: health care 54% of clusters, military conflicts 12%, retirement and social security 7%, federal budget 7%, world affairs 7%. About half the clusters have one counterargument for each initial side.
- Analysis. OLS regression of the switching rate (per counterargument, in percentage points) on topic, initial position, counterargument content, survey artifacts, house and mode.
Main results (verified)
- Earlier academic studies (political tolerance and racial attitudes in the US, Russia, South Africa, Denmark, Canada): switching ranges from one-tenth to three-fifths of respondents and averages one-third to one-half. It is more common among the initially tolerant than the initially intolerant, and among respondents whose attitudes are weakly held, uninformed or conflicted.
- Across all 287 iPOLL counterarguments: switching ranges 2% to 86%, mean 38%, SD 17% (between questions).
- By topic (mean, SD): health care 38% (15), military conflicts 37% (19), retirement and social security 35% (12), federal budget and spending 50% (22), world affairs 37% (21), other 35% (17).
- Predictors (reduced model, percentage points; constant 33.5):
- more familiar or salient topics switch less: military conflict −6.2, retirement and social security −9.7;
- respondents who initially favoured the proposal switch more: +4.9;
- each counterargument already asked adds +3.9;
- a prior question among the survey's first five lowers switching: −6.6 (contrary to the author's expectation);
- online mode +9.4 against live telephone; one house effect (CBS/New York Times +8.4).
- Not significant: ideology of the initial answer, the public's "most important problem" share, mentions of party, health or cost, an explicit reminder of the prior answer, an explicit "don't know" option, a gap between question and counterargument, Gallup, Harris, in-person mode.
- R² .21 (adjusted .19); standard error of estimate 14.9 points.
- The range implied: 60% predicted for an unfamiliar topic, a favoured prior asked late in an online survey after three earlier counterarguments; 21% for a military topic, an opposed prior asked early in a non-online survey with none before.
- Attitude strength (asked for 111 of 287 counterarguments): strongly held answers switched less (significant only at the .1 level). In that subset: constant 37.0, retirement −9.2, favoured +10.2, +2.7 per earlier counterargument, online +14.2 (R² .34).
- The author's conclusion: an average of 38% is a baseline for judging whether a counterargument is "strong" or "weak".
Limits
- Aggregate rates only. No individual-level data, demographics or reliability; the article could not obtain or harmonise the survey files.
- No control condition. Nothing measures how many respondents would change if the question were simply re-asked without a counterargument, so the 38% includes ordinary response instability and moves to "don't know" or refusal.
- Selection. Pollsters choose counterarguments to find persuasive messages; the item set is not a random sample of attitudes. US only, public-policy attitudes only; no candidates, products or leaders.
- No durability. Switching is measured within the same interview; nothing says whether the switch persists.
- One author, a short practitioner-journal article; modest sample of clusters; the model explains about a fifth of the variance.
What it means for the base battery (inference)
- This is the closest human reference in hand for the drift battery's "opinion" probe on attitude items: after giving an answer, people hear the other side once, and on average about 38% of answers switch (SD 17 across items; range 2–86%). A base whose empty slot moves +0.83 toward a stated opinion is far outside this; a base that never moves is also outside it.
- The comparison must subtract a re-ask control on both sides. The battery's neutral session plays that role for the base; for humans, the same items' test–retest instability (e.g. GSS within-wave or short-interval retest) has to be subtracted, or the base's raw rate compared with the human raw rate.
- Item- and context-level predictors are testable signatures: less switching on familiar topics and strongly held attitudes, more on favoured positions and after earlier counterarguments. A base that reproduces the pattern across items is better evidence than one matching the mean.
- Different probe content: these counterarguments carry information ("what if you heard…"); the drift battery's opinion probe carries social pressure ("I'd answer X, and most sensible people would"). Both formats belong in the battery; only the informational one has this reference.