Obfuscatory language, information acquisition, and opinions: a 2×2 online experiment

We run an online, individual-level randomized experiment with a 2×2 design. Factor 1 is Language: participants read a brief news-style description written in either Active voice (clear agent; control) or Intransitive syntax (more obfuscatory; treatment). Factor 2 is a salience cue (Z) applied to the “See facts” option that allows respondents to view a short facts summary before answering opinion questions. In Z=1, the label reads “See facts (~10–15 seconds)”; in Z=0, it reads “See facts.” Z changes text only (not option order/size/placement or the facts content). We measure (i) whether the respondent clicks to see facts (information acquisition) and (ii) opinions in a Likart scale about the presented news.

Language arm (50/50): After consent and demographics, respondents view a news-style vignette about a fatal traffic stop. In the Active arm, the opening lines and facts panel use active voice (e.g., “A police officer shot and killed Jordan Reyes, 29…”). In the Intransitive arm, the text uses intransitive/nominalized phrasing (e.g., “Jordan Reyes, 29, died in an officer-involved shooting…”). Within each language arm, the facts panel content is identical across Z; only syntax differs by arm.

Salience Z (50/50, independent): Immediately after the vignette, respondents choose between two options presented with identical order/size/placement: (i) “See facts” and (ii) “Continue to next section.” In Z=1, the first option includes the parenthetical “(~10–15 seconds)” to describe expected reading time; in Z=0, it does not. Clicking “See facts” opens the panel (brief bullet points; typical read time ~10–15s); otherwise the respondent proceeds directly to opinion questions.

Randomization unit: individual; 2×2 design with independent randomization.

Note: Z is content-neutral (text-only); it does not alter content, order, size, placement, or page flow aside from labeling the “See facts” option.

Ethics: Exempt determination pending from an independent U.S. social/behavioral IRB. Data collection will begin only after approval/exemption.

2025-12-08

2025-12-12

Primary outcomes: (a) Acquire = 1 if “See facts” was clicked, 0 otherwise; (b) Opinion Index = mean of z-scores of moral-responsibility and legal-penalty items (higher = more punitive), along with average response for each question.

The opinion questions are the following:

Q1. Moral responsibility of the shooter (7-point Likert): 1 = Not at all responsible … 7 = Completely responsible.
Q2. Should the shooter face legal penalties (e.g., criminal charges)? (7-point Likert): 1 = Strongly oppose … 7 = Strongly support.

Secondary/mechanism outcomes: Jail recommendation (binary), recall items, perceived agency clarity, polarization.

Questions are the following (excluding verification questions):

Q3. “How clear was it who caused the death in the description you read? 1=Not at all clear … 7=Completely clear.”
Q4. Jail recommendation (binary): 0 = Do not recommend jail; 1 = Recommend jail.
Q6. According to the description, what happened during the traffic stop? (multiple choice; one correct).
A. During a late-evening traffic stop in Midtown, a bystander was killed while Jordan Reyes was uninjured.
B. During a late-evening traffic stop in Midtown, Jordan Reyes, 29, was taken to a hospital and later died.
C. During a late-evening traffic stop in Midtown, Jordan Reyes was treated at the scene and released.
D. During an afternoon traffic stop in Midtown, Jordan Reyes suffered minor injuries but survived.
E. During a late-evening traffic stop in Midtown, property damage occurred but there were no injuries.
F. Not sure.
Q7. Was any body-camera footage mentioned?
A. Yes
B. No
C. Not sure

Polarization for a given participant, for a given response to Q1 or Q2, is measured as the distance (absolute value) of their response to the average. Variations include distance to the median of responses; and to the median of the Likert scale.

Individual-level randomized online experiment with a 2×2 design. Factor 1 (Language): Active voice (control) vs Intransitive (treatment). Factor 2 (Salience Z): text-only cue on the “See facts” option (Z=1 “See facts (~10–15 seconds)” vs Z=0 “See facts”). After reading the vignette, respondents choose whether to view a brief facts summary before answering opinion questions.

Flow. Consent → demographics → vignette (Language assigned) → choice page (Z assigned) → optional facts panel (if clicked) → outcomes.
Language (50/50). Opening lines and facts panel use Active vs Intransitive syntax; content is otherwise parallel across arms.
Salience Z (50/50, independent). Choice page shows two options with identical order/size/placement: “See facts” and “Continue to next section.” In Z=1, the “See facts” label includes “(~10–15 seconds)”; in Z=0, it does not. Z changes text only; it does not alter content, order, size, placement, or page flow.

Randomization is implemented at the individual level using SurveyMonkey’s A/B testing / random assignment features. Language (Active vs Intransitive) is randomized 50/50, and Salience Z (0 vs 1) is randomized 50/50 independently, producing four cells with equal expected proportions.

Individual respondent. No clustering; no stratification.

No

None — individual-level randomization (no clustering)

Target sample: N = 2500 participants (online sample)

2×2 design, equal allocation (expected):
• Active × Z=0: n ≈ 625
• Active × Z=1: n ≈ 625
• Intransitive × Z=0: n ≈ 625
• Intransitive × Z=1: n ≈ 625