Methods & Evaluation Design (jppm-methods)
When to trigger
- Deciding whether the question needs an experiment, a policy evaluation, a survey, or a synthesis
- A policy rolled out and you must find a credible counterfactual before claiming impact
- Your label/disclosure experiment uses stimuli no agency could ever mandate
- The sample excludes the very population the policy is meant to protect
- You are combining lab and field evidence and need the pieces to carry distinct weight
Methodological pluralism, disciplined by the policy question
JPP&M accepts a wider methods palette than most marketing journals — randomized experiments, quasi-experimental evaluation, surveys, field and archival data, meta-analysis, qualitative work — but the choice must follow from the policy inference required. Two questions dominate: Would the proposed instrument work? (prospective — usually experiments with realistic stimuli) and Did the enacted instrument work? (retrospective — evaluation designs with explicit counterfactuals). A submission that answers the retrospective question with before/after trends, or the prospective one with fantasy stimuli, fails regardless of statistical polish. Missing counterfactual reasoning in an evaluation design is one of this journal's known desk-reject patterns.
Prospective designs: policy-realistic experiments
- Mandatable stimuli. Test warning/label/disclosure formats an agency could actually require — real estate on a package, formats from the live rulemaking (e.g., FDA front-of-package proposals), not a full-screen banner no rule would compel.
- Treatment contrasts = decision options. Conditions should map to the regulator's choice set (status quo vs. proposed rule vs. stricter alternative), so each pairwise contrast answers a decision.
- Consequential outcomes. Choice with real stakes, purchases, incentivized behavior — self-reported intentions alone are weak currency for a claim that a rule will change behavior.
- Policy-relevant samples. Recruit the protected population (smokers for tobacco warnings, low-income households for financial disclosures, parents for children's marketing). A student panel cannot carry a vulnerability claim; when using Prolific/CloudResearch, screen and quota accordingly.
- Marketplace noise. Add realistic competing information (cluttered shelf, competing claims); effects that survive noise are the ones that survive markets.
Retrospective designs: evaluation with a counterfactual
| Design |
Use when |
JPP&M-specific cautions |
| Difference-in-differences |
policy adopted in some states/markets/categories, not others |
staggered adoption needs heterogeneity-robust estimators; argue parallel trends behaviorally, not just visually |
| Regression discontinuity |
eligibility threshold or size cutoff assigns exposure |
check manipulation at the cutoff (firms sort!); effects are local to the threshold |
| Synthetic control |
one large unit treated (a city soda tax, a national ban) |
pre-period fit and placebo runs are the argument |
| Event study |
timing of enforcement/announcement is sharp |
anticipation by firms and media coverage blur the event date |
| Interrupted time series |
no untreated comparison exists at all |
weakest option; state so and bound the claims |
Firms' strategic responses are both a threat and a finding: reformulation, pre-emptive compliance, or channel-shifting can contaminate the comparison group — design to detect it (untreated outcomes, supply-side data) rather than assume it away.
Surveys, qualitative work, and synthesis
- Surveys earn their place for constructs no archive holds (perceived deception, privacy concern, financial anxiety) — use validated scales and probability or well-quota'd samples when claims are population-level.
- Qualitative designs are welcome for vulnerable populations whose experience frames the policy problem; document access, consent, and IRB care to a higher standard, and avoid designs that further burden participants.
- Meta-analysis suits mature streams (warning-label effects, disclosure formats); code moderators the regulator controls (format, placement, dose).
Checklist
Anti-patterns
- Before/after theater: a pre/post trend presented as policy impact with no comparison group
- Fantasy stimuli: disclosure formats no agency could mandate, generalized to regulation
- Convenience-sample vulnerability claims: conclusions about protected groups from panels that exclude them
- Intentions-only evidence for behavior-change claims
- One-method dogma: forcing an experiment onto a question that demands field variation, or vice versa
- Sorted cutoffs: an RDD where firms demonstrably manipulate the threshold, unexamined
Output format
【Policy question type】prospective (would it work) / retrospective (did it work)
【Design】experiment / DiD / RDD / synthetic control / survey / meta-analysis / qualitative
【Counterfactual】comparison group + assignment logic (retrospective) or control condition logic (prospective)
【Stimuli & sample】mandatable formats; policy-target population included
【Firm response plan】how strategic reactions are detected or bounded
【Next skill】jppm-data-analysis
1---2name: jppm-methods3description: Use when choosing and designing the evidence for a Journal of Public Policy & Marketing (JPP&M) manuscript — experiments with policy-realistic stimuli, quasi-experimental policy evaluation (DiD, RDD, synthetic control), surveys, field data, or meta-analysis. Designs the studies; it does not estimate them (jppm-data-analysis).4---56# Methods & Evaluation Design (jppm-methods)78## When to trigger910- Deciding whether the question needs an experiment, a policy evaluation, a survey, or a synthesis11- A policy rolled out and you must find a credible counterfactual before claiming impact12- Your label/disclosure experiment uses stimuli no agency could ever mandate13- The sample excludes the very population the policy is meant to protect14- You are combining lab and field evidence and need the pieces to carry distinct weight1516## Methodological pluralism, disciplined by the policy question1718JPP&M accepts a wider methods palette than most marketing journals — randomized experiments, quasi-experimental evaluation, surveys, field and archival data, meta-analysis, qualitative work — but the choice must follow from the **policy inference** required. Two questions dominate: *Would the proposed instrument work?* (prospective — usually experiments with realistic stimuli) and *Did the enacted instrument work?* (retrospective — evaluation designs with explicit counterfactuals). A submission that answers the retrospective question with before/after trends, or the prospective one with fantasy stimuli, fails regardless of statistical polish. Missing counterfactual reasoning in an evaluation design is one of this journal's known desk-reject patterns.1920## Prospective designs: policy-realistic experiments2122- **Mandatable stimuli.** Test warning/label/disclosure formats an agency could actually require — real estate on a package, formats from the live rulemaking (e.g., FDA front-of-package proposals), not a full-screen banner no rule would compel.23- **Treatment contrasts = decision options.** Conditions should map to the regulator's choice set (status quo vs. proposed rule vs. stricter alternative), so each pairwise contrast answers a decision.24- **Consequential outcomes.** Choice with real stakes, purchases, incentivized behavior — self-reported intentions alone are weak currency for a claim that a rule will change behavior.25- **Policy-relevant samples.** Recruit the protected population (smokers for tobacco warnings, low-income households for financial disclosures, parents for children's marketing). A student panel cannot carry a vulnerability claim; when using Prolific/CloudResearch, screen and quota accordingly.26- **Marketplace noise.** Add realistic competing information (cluttered shelf, competing claims); effects that survive noise are the ones that survive markets.2728## Retrospective designs: evaluation with a counterfactual2930| Design | Use when | JPP&M-specific cautions |31|--------|----------|-------------------------|32| Difference-in-differences | policy adopted in some states/markets/categories, not others | staggered adoption needs heterogeneity-robust estimators; argue parallel trends behaviorally, not just visually |33| Regression discontinuity | eligibility threshold or size cutoff assigns exposure | check manipulation at the cutoff (firms sort!); effects are local to the threshold |34| Synthetic control | one large unit treated (a city soda tax, a national ban) | pre-period fit and placebo runs are the argument |35| Event study | timing of enforcement/announcement is sharp | anticipation by firms and media coverage blur the event date |36| Interrupted time series | no untreated comparison exists at all | weakest option; state so and bound the claims |3738Firms' strategic responses are both a threat and a finding: reformulation, pre-emptive compliance, or channel-shifting can contaminate the comparison group — design to detect it (untreated outcomes, supply-side data) rather than assume it away.3940## Surveys, qualitative work, and synthesis4142- **Surveys** earn their place for constructs no archive holds (perceived deception, privacy concern, financial anxiety) — use validated scales and probability or well-quota'd samples when claims are population-level.43- **Qualitative designs** are welcome for vulnerable populations whose experience frames the policy problem; document access, consent, and IRB care to a higher standard, and avoid designs that further burden participants.44- **Meta-analysis** suits mature streams (warning-label effects, disclosure formats); code moderators the regulator controls (format, placement, dose).4546## Checklist4748- [ ] The design answers the paper's policy question (prospective vs. retrospective) directly49- [ ] Experimental stimuli are mandatable and conditions map to the regulator's choice set50- [ ] The sample includes the population the policy targets; vulnerable groups are powered, not token51- [ ] Evaluations name the counterfactual and the assignment mechanism explicitly52- [ ] Firm strategic response is measured or ruled out, not assumed absent53- [ ] Ethics/IRB treatment matches the sensitivity of the population studied54- [ ] Pre-registration or a pre-analysis plan is in place for confirmatory studies5556## Anti-patterns5758- **Before/after theater**: a pre/post trend presented as policy impact with no comparison group59- **Fantasy stimuli**: disclosure formats no agency could mandate, generalized to regulation60- **Convenience-sample vulnerability claims**: conclusions about protected groups from panels that exclude them61- **Intentions-only evidence** for behavior-change claims62- **One-method dogma**: forcing an experiment onto a question that demands field variation, or vice versa63- **Sorted cutoffs**: an RDD where firms demonstrably manipulate the threshold, unexamined6465## Output format6667```text68【Policy question type】prospective (would it work) / retrospective (did it work)69【Design】experiment / DiD / RDD / synthetic control / survey / meta-analysis / qualitative70【Counterfactual】comparison group + assignment logic (retrospective) or control condition logic (prospective)71【Stimuli & sample】mandatable formats; policy-target population included72【Firm response plan】how strategic reactions are detected or bounded73【Next skill】jppm-data-analysis74```
brycewang-stanford