The long-lasting effects of newspaper op-eds on public opinion

Quarterly Journal of Political Science 13(1): 59–87, doi 10.1561/100.00016112. Pre-registered at egap.org "except where indicated". Replication archive doi 10.7910/DVN/3KBCJF (not downloaded). Read by researcher R5b (change with reasons), 5 October 2026, from the first author's website: the published PDF, the online appendix (dated 26 September 2016) and the author's errata note. Provenance: papers/change/coppock2018_oped_persistence.provenance.json.

What was read

• Read in full: the 29-page article (Tables 1–11), the 32-page online appendix (heatmap labels and stars as extracted text, Tables B.1–B.3, D.4–D.5, the full texts of the five op-eds and the outcome items), and the 5-page errata (ten corrections).
• Figure 2 (persistence) was rendered and inspected; values below are read off it (about ±0.03 SD). Figure 1 (effects by party) and the appendix heatmaps were not inspected as images; the appendix's Figures C.6–C.7 were read only as extracted axis text.
• Not read: the publisher's separate "supplementary material" and the replication archive.
• A discrepancy noted: in the extracted text, the appendix's elite heatmaps (Figures A.4–A.5) are identical to the MTurk ones (A.1–A.2), including a climate column the elite study did not have. They may be duplicated figures; this was not resolved.

What it is

• Question: do real op-eds change policy attitudes, among the mass public and among elites, across party, and do the changes last?
• Treatments: five real, unmodified libertarian op-eds, shown with their original mastheads: on Amtrak and infrastructure funding (Newsweek), climate scepticism (Lindzen, Wall Street Journal), a 14.5% flat tax (Rand Paul, WSJ), replacing VA health care with vouchers (New York Times), and Wall Street's benefits (USA Today). Each argues with facts and reasons; none is a two-sided "point–counterpoint".
• Study 1 (MTurk): 3,571 people (the errata correct the text's 3,567), randomized to one op-ed or none; outcomes immediately, at 10 days (84% recontacted) and at 30 days (69%). At wave 2 half saw an unrelated distractor op-ed (on marijuana legalization). Attrition was unrelated to treatment (p = .10 and .61).
• Study 2 (elites): 2,181 respondents (errata) from a list of 32,498 policy professionals (trade associations 26%, academics 20%, finance 14%, government affairs 14%, think tanks 8%, media 8%, Hill staffers 4%); four op-eds (no climate arm); outcomes immediately and at 10 days (1,358 responded). Unpaid; older, better educated, whiter, more male and more partisan than the MTurk sample.
• Outcomes: one pre-registered "main" 7-point item per issue, a factor-score composite of 4–5 items per issue (unit SD in the control group), and an "agreement" dichotomy split at the control group's median (not pre-registered).

Main results (verified)

• Large immediate effects on the target issue, near zero elsewhere.
  • MTurk, main items (1–7 scale): +0.43 (climate) to +0.92 (Wall Street) points, all p < .001; 2 of 20 non-target effects significant.
  • MTurk, composites: +0.28 SD (climate) to +0.70 SD (Wall Street); others 0.49–0.65.
  • Elites, main items: Amtrak +0.44, flat tax +0.41 (p = .0042, corrected by the errata), Wall Street +0.79; veterans +0.05, n.s.
  • Elites, composites: Wall Street +0.57, Amtrak +0.30, veterans +0.15; flat tax +0.10, n.s. One of 12 non-target cells significant (errata).
  • Study means of the target composite effects: 0.52 SD (MTurk), 0.28 SD (elites) (errata; the article's "0.30 to 0.50" was wrong).
• "Agreement": at the median split, the op-eds raised agreement with the author by 20 points on MTurk (from 50% to about 70%) and 12 points among elites; by issue 11–28 points (MTurk) and 2–26 (elites). Splits at the 25th and 75th percentiles gave 13 and 21 points (MTurk), 6 and 11 (elites).
• Elites move less on three of four target scales: the elite-by-treatment interactions are −0.20 (Amtrak), −0.38 (flat tax), −0.49 (veterans), −0.13 (Wall Street, n.s.).
• Party matters for size, not sign. All three party groups moved toward the libertarian author. On MTurk the precision-weighted mean effect on the main items was 0.58 (SE .05) among Democrats and 0.88 (SE .08) among Republicans, a significant 0.3-point gap; joint tests of equal effects across Democrats, independents and Republicans rejected only for Amtrak and flat tax on MTurk. (This partisan analysis replaced the pre-registered BART analysis.)
• Persistence (Figure 2; always-responders): roughly half of the immediate effect remained at 10 days, and almost no further decay by 30 days (a "hockey stick"). Read off the figure, MTurk target-group gaps in SD: Wall Street about 0.63, 0.34, 0.37 at 0, 10 and 30 days; veterans 0.60, 0.31, 0.37; Amtrak 0.49, 0.28, 0.23; flat tax 0.45, 0.21, 0.17; climate 0.25, about 0.10, about 0.07. Elites, 0 and 10 days: Wall Street 0.55, 0.30; Amtrak 0.45, 0.20. Most remained significant.
  • The authors' reading: immediate effects mix priming, which fades, and information, which persists.
  • The wave-2 distractor op-ed did not visibly change persistence (appendix C, axis text only).
• Cost per mind changed: 6 cents to $2.50 under their assumptions (about 400,000 readers per op-ed). Because so few people read any one op-ed, aggregate opinion would move about 0.02 points.

Limits

• Convenience samples (MTurk; a self-selected elite list); persistence is estimated on always-responders, who differ demographically from the full samples.
• Forced exposure, one-sided messages, no competing arguments; the authors call the estimates an upper bound for naturalistic reading.
• All five op-eds are libertarian, from one think tank's authors; reasons and source are not separated (real op-eds, so the causal feature is unknown).
• The agreement measure and the party analysis were not pre-registered; the errata list seven text–table mismatches, none changing a conclusion.
• Effects at one month are measured in the same survey format; whether behaviour changed is not measured.

What it means for Kurisutina (inference)

• The persistence reference the battery lacked. For argument-based persuasion on policy attitudes, about half of the immediate shift survives 10 days and roughly holds at 30. That is a carry-over ratio near 0.5 for one strong, one-sided, reasoned message, below the chat replica's 0.79–0.90. It is a different quantity from Kiley's panel result (most change seen between waves two years apart reverts): one induced shift over a month, not spontaneous change over years.
• Two components of change with reasons: a fast-fading part (consideration made salient) and a lasting part (information absorbed). A rule of change could carry both: an in-session shift that decays within days, and a persistent shift of about half its size when the episode carried new information.
• Spread: opponents move less than allies (0.58 against 0.88 points) and elites less than the public, but nobody moves away. That gives signs and rough sizes for person settings (prior strength, identity), not a person-level distribution.
• For the battery: a matched reference for a "good argument" probe: a real, reasoned, one-sided text; a 7-point item; immediate +0.4 to +0.9 points (0.3–0.7 SD); about half left at 10 days. It fits a persistence test (P7-like) better than Wagner's audience-tuning ratios.