Assessing Testing Effects and Retention in a Cluster-Randomized Workshop Intervention on Pollution Taxes

Intervention (Public)
The intervention is a structured educational workshop focused on the principles, design, and economic rationale of pollution taxes. The workshop is designed to improve participants’ understanding of environmental taxation, including its effectiveness, implementation challenges, and impact on behavior and market outcomes.

Workshops are delivered across seven clusters (locations), and each site receives the same curriculum, materials, and facilitator training to ensure consistency.

Clusters are randomly assigned to one of two groups:

Group A receives:

A pretest to assess baseline knowledge of pollution taxes.

The pollution tax workshop intervention.

An immediate posttest after the workshop.

A follow-up test one month later to assess retention.

Group B receives:

The pollution tax workshop intervention.

An immediate posttest after the workshop.

A follow-up test one month later.

The intervention aims to measure not only learning outcomes but also potential testing effects — i.e., whether taking a pretest improves posttest performance or knowledge retention.

Intervention (Hidden)
Participants are not informed of their assignment to different study groups or of the existence of a pretest-only group. The study is presented uniformly as a workshop on pollution taxes followed by a knowledge assessment. This design is intended to minimize expectancy effects and differential engagement.

In Group A, the pretest is administered as a standard entry activity. In Group B, no mention is made of the pretest, and participants only encounter the assessment at the end. The one-month follow-up is framed as a general feedback and learning review to reduce test-related bias.

The covariate-constrained randomization approach was conducted before the workshops using a pre-specified script that minimized imbalance in expected attendance and regional representation. Workshop facilitators are blind to the allocation strategy.

2025-06-13

2026-02-10

Willingness to Pay (WTP) for Pollution Taxes (Immediate Posttest)
Measured via a 4-item behavioural intention scale based on TPB constructs. Score is computed as the average across items.

Change in WTP (Pre–Post, Group A only)
For Group A, willingness to pay is measured both before and immediately after the intervention. Change is calculated as the difference in composite scores.

Posttest WTP Comparison (Group A vs. B)
Between-group comparison of posttest WTP to assess the effect of pretesting on outcome levels.

WTP at 1-Month Follow-Up (All Participants)
Willingness to pay will be reassessed for both groups one month after the workshop. This allows evaluation of retention of intention and delayed effects of pretesting. Group-level comparisons and within-person (for Group A) change from posttest to follow-up will be analyzed.

Moderation by Pro-Environmental Behavior
A 12-item index capturing baseline environmental behavior will be tested as a moderator of both immediate and follow-up WTP outcomes.

Heterogeneous Effects by Demographics
Gender, religion, income, and other background variables will be explored as moderators of WTP outcomes.

Willingness to Pay (WTP) for Pollution Taxes
WTP is measured using a 4-item behavioral intention subscale, consistent with the Theory of Planned Behavior. Items include statements like:

“I intend to pay a pollution tax if it is introduced.”

“I plan to comply fully with a pollution tax policy.”

“I will likely avoid paying a pollution tax when required.” (reverse-coded)
Each item is rated on a 7-point Likert scale, and the composite WTP score is calculated as the mean of the items (after reverse-coding where applicable). This is measured:

Immediately after the workshop (posttest) for all participants, and

One month later to assess retention of willingness to pay.

Change in WTP (Pre-Post, Group A only)
For participants in Group A (who receive a pretest), change in WTP is calculated as the difference between posttest and pretest composite WTP scores.

Long-Term Retention (Follow-Up WTP)
The WTP measure is repeated at a 1-month follow-up for both groups. For Group A, change scores (pre vs. follow-up, and post vs. follow-up) can be calculated. For Group B, follow-up WTP offers a delayed outcome measure not influenced by pretesting. This allows exploration of whether the initial test exposure affects retention.

Moderation and Heterogeneity
Pre-intervention pro-environmental behavior (12-item index) will be used as a moderator of WTP outcomes. Demographics (e.g., gender, income, religion) will also be used in subgroup analyses.

TPB Constructs as Exploratory Mechanisms
Attitudes, subjective norms, perceived behavioral control, and moral norm (each measured by 4 items) will be examined as potential mediators or moderators of WTP outcomes but are not primary endpoints themselves.

Theory of Planned Behavior Constructs (Posttest and Follow-Up)
These include:

Attitudes toward pollution taxes

Subjective norms

Perceived behavioral control

Moral norm

Each construct is measured with 4 Likert-scale items (1–7 scale) and scored as the mean of the relevant items (after reverse-coding where needed).
They will be analyzed as:

Exploratory secondary outcomes (changes from pretest to posttest and follow-up)

Potential mediators of the effect of the intervention on WTP;

Workshop Evaluation (Posttest Only)
At the end of the workshop, participants rate:

Clarity of the workshop content

Perceived usefulness for understanding pollution taxes

Likelihood of recommending the workshop to others
These are measured on Likert scales and will be used for descriptive and internal evaluation purposes only.

Changes in TPB Constructs Over Time
For participants with pretest and follow-up data (Group A), changes in attitudes, norms, control, and moral obligation will be tracked to explore how stable these constructs are post-intervention and how they relate to sustained WTP.

TPB Construct Scores
Each Theory of Planned Behavior construct (attitudes, subjective norms, perceived behavioral control, moral norm) is measured using 4 Likert-scale items (7-point scale). Scores are computed as the mean of the items within each construct, after reverse-coding negatively worded items. These scores will be analyzed:

As outcomes in their own right (pre-post and post-follow-up)

As potential mediators in models predicting willingness to pay (WTP)

Workshop Evaluation Metrics
Three post-intervention Likert items assess perceived clarity, usefulness, and likelihood of recommendation. Each item is treated as an individual ordinal outcome. These will be used descriptively to inform future improvements to workshop delivery but are not inferential endpoints.

Longitudinal Change in TPB Constructs
For participants with complete pretest, posttest, and follow-up data (Group A), within-subject change scores will be calculated for each TPB construct. These will be used to assess persistence or decay of psychological constructs over time.

Let me know when you’re ready to move on to the Analysis Plan or revisit any earlier registry section!

This is a cluster-randomized field experiment conducted across seven university workshops. Clusters (workshops) are randomly assigned to one of two groups:

Group A: receives a pretest, the pollution tax workshop, a posttest, and a 1-month follow-up.

Group B: receives the same workshop, a posttest, and a 1-month follow-up, but no pretest.

The design allows for comparison of outcomes across and within groups, including follow-up measurement of retention

This is a two-arm cluster-randomized trial involving seven university-based workshop sites. The clusters (workshops) are the unit of randomization.

Clusters were randomly assigned to either:

Group A (Pretest–Posttest–Follow-Up)
Participants complete a pre-intervention survey (measuring willingness to pay for pollution taxes and Theory of Planned Behavior constructs), attend a standardized workshop, complete an immediate posttest, and complete a follow-up survey one month later.

Group B (Posttest–Follow-Up Only)
Participants attend the same standardized workshop but do not complete a pretest. They complete the immediate posttest and the 1-month follow-up survey.

🎲 Randomization Procedure
Randomization was done at the cluster level, using covariate-constrained randomization:

All possible 3:4 splits were generated.

For each split, a balance score was computed based on:

Total expected attendance

Regional distribution (Southern vs. Northern workshops)

The allocation with the lowest imbalance score was selected. This approach ensures better covariate balance given the small number of clusters (n = 7).

🧪 Purpose of Design
This modified Solomon-like design allows us to:

Estimate the effect of pretesting on posttest and follow-up outcomes.

Measure within-group change for Group A (pre to post and follow-up).

Assess retention via 1-month follow-up in both groups.

Test moderators and mediators using Theory of Planned Behavior constructs and pro-environmental behavior scales.

Randomization was done in office by computer using a covariate-constrained algorithm written in Python. All possible 3:4 allocations of clusters were evaluated, and the assignment that minimized imbalance on expected attendance and regional representation was selected

Randomization was conducted at the cluster level, where each cluster corresponds to a university workshop session. All participants within a given workshop were assigned to the same experimental condition. No individual-level randomization was used.

Yes

7 clusters (workshops), with approximately 3 assigned to the pretest group (Group A) and 4 to the posttest-only group (Group B).

Approximately 1,200 university students, based on expected attendance figures across the 7 workshop clusters.

Group A (Pretest + Posttest + Follow-Up): 3 clusters, approximately 600 students

Group B (Posttest + Follow-Up only): 4 clusters, approximately 600 students

Power Calculation: Minimum Detectable Effect Size for Main Outcomes Assuming: 7 clusters (3 in Group A, 4 in Group B) 600 students per group (≈85–215 per cluster) Intra-cluster correlation (ICC): 0.01–0.05 (typical for attitudinal survey outcomes) Two-sided α = 0.05, power = 0.80 Standard deviation (SD) of WTP scale: assumed to be 1.0 (standardized Likert-based composite) The minimum detectable effect size (MDES) is approximately: Standardized effect size (Cohen’s d): 0.30–0.35, depending on ICC This translates to about 30–35% of one standard deviation in WTP or TPB-based scale scores. Power was calculated using formulas for two-group comparisons in clustered designs, accounting for unequal cluster sizes and between-group allocation (3 vs. 4 clusters).