Bricks to Blocks: Information and Coordination Challenges for Transitioning to a Cleaner Building Technology

The study evaluates the transition from traditional fired clay bricks to cleaner concrete/soil blocks in the construction sector through a clustered randomized controlled trial with three arms:

Control Group: Participants in these Upazilas receive no intervention and continue with business-as-usual operations.

Treatment 1 (Information & Facilitation): Contractors and relevant stakeholders receive an information and facilitation package delivered through in-person workshops. This includes a concise booklet summarizing the policy context (government shift towards blocks), comparative benefits and costs, a directory of local block supplier contacts, and basic workmanship guidance.

Treatment 2 (Information & Facilitation, Worker's training): Along with the Treatment 1 package, randomly selected one worker from each contractor in Treatment 2 will get hands on training on the use of blocks.

2025-11-03

2026-01-22

Primary endpoints capture adoption and awareness of eco-friendly blocks, measured separately for each actor type.

Confirmatory primary outcomes:

For contractors (N=792): (1) Ever used eco-blocks - binary indicator for having ever used hollow concrete blocks (HCB) or compressed/stabilized soil blocks (SSB) instead of fired clay bricks (baseline mean = 27.1%, MDE = 11.4 pp or 0.26 SD), and (2) Awareness index - continuous index (0-2) summing policy awareness and information receipt (baseline mean = 0.81, MDE = 0.26 SD).

For construction workers (N=1,584): Ever worked with eco-blocks - binary indicator for having ever worked on a project using HCB/SSB (baseline mean = 31.9%, MDE = 9.0 pp or 0.19 SD).

For private clients (N=528): (1) Awareness index - identical construction to contractors (baseline mean = 0.53, MDE = 0.30 SD); and (2) Used blocks in construction - binary indicator for having ever used eco-blocks in own construction (baseline mean = 3.0%, MDE = 5.3 pp or 0.31 SD). Both are confirmatory primary outcomes. The MDE for client block adoption in SD units (0.31 SD) is statistically equivalent to the awareness index (0.30 SD); the low baseline (3%) is reflected in the power calculation via a conservative rho=0.20 assignment and does not justify downgrading the classification. The preferred estimand restricts to baseline non-adopters (N approx. 512, 97% of clients), where "ever used" at endline equals new adoption since baseline, eliminating the floor-effect concern by construction. Adoption is the terminal behavioral outcome of the theory of change; classifying the mechanism (awareness) as primary while treating the terminal outcome as exploratory would invert the causal logic.

Exploratory outcomes (limited power):

Evaluated tender with blocks (procurement officers, N=152) - binary indicator for having evaluated a government tender that included eco-blocks (baseline mean = 49.3%, MDE = 0.59 SD). Classified as exploratory due to small sample (approximately 2 officers per cluster). Procurement officers attend workshops primarily as a belief signal; any direct behavior change is a positive externality.

Compliance/manipulation check (not a confirmatory outcome):

Received training on blocks (T2 workers, N=528) - binary indicator for having received training on working with eco-blocks (baseline mean = 2.3%, MDE = 3.8 pp or 0.26 SD, using contractor-level clustering with 264 contractors and 2 workers each). Reported prominently but not a confirmatory primary outcome, as it measures intervention delivery rather than substantive adoption behavior.

Block adoption outcomes (ever used eco-blocks, ever worked with eco-blocks, used blocks in construction) are binary indicators equal to 1 if the respondent reports ever using hollow concrete blocks or compressed/stabilized soil blocks instead of fired clay bricks, and 0 otherwise.

The Awareness index is the sum of two binary items: (1) aware of government policy preferring alternatives to fired clay bricks in public procurement, and (2) received information about eco-friendly blocks in the past year. Range: 0-2. One component ("received information") is close to a manipulation check, so a featured sensitivity analysis reports the government policy awareness component alone alongside the composite index in the main results tables. If treatment moves the composite but not the policy awareness component, this suggests the effect is driven by treatment exposure recall rather than substantive information gain.

Evaluated tender with blocks (procurement officers) is a binary indicator equal to 1 if the officer has evaluated a government tender that included HCB/SSB, and 0 otherwise.

Received training on blocks (workers, T2 compliance check) is a binary indicator equal to 1 if the worker has received any training on working with eco-blocks, and 0 otherwise.

For the adoption outcomes, we pre-specify a tiered adoption analysis: (1) Preferred: baseline non-adopter conversion - restrict the sample to respondents who had not adopted at baseline; for these respondents, "ever used" at endline is equivalent to "new adoption since baseline," providing a clean conversion estimand. This is the preferred adoption analysis. (2) Supplementary: full-sample ANCOVA of "ever used" - reported for completeness but diluted by prior adopters whose outcome is fixed at 1. (3) Supporting robustness: endline-only "new adoption since baseline" - the direct flow measure cannot benefit from ANCOVA adjustment; reported as a robustness check. For private clients, the repeated "ever used" item retains identical wording across waves for ANCOVA comparability; "new adoption since baseline" is measured separately as an endline-only complementary outcome.

Secondary outcomes capture mechanisms consistent with the information and coordination channels in the theory of change.

For contractors (N=792): (1) Received info on eco-blocks - whether the respondent received information about eco-friendly blocks in the past year (baseline = 16.3%, MDE = 0.29 SD), serving as a manipulation check for the information intervention; (2) Aware of government policy - whether the respondent knows government prefers alternatives to fired clay bricks (baseline = 64.4%, MDE = 0.25 SD); (3) Ease of contacting suppliers - measured on a 5-point ordinal scale (Very Difficult to Very Easy); the main analysis uses the full ordinal variable treated as continuous in OLS, with a binary recode (Moderate/Easy/Very Easy = 1) reported as a robustness check (baseline binary = 52.5%, MDE = 0.28 SD); (4) Advised clients to use blocks - whether the contractor has recommended eco-blocks to private clients (baseline = 38.9%, MDE = 0.24 SD).

For construction workers (N=1,562): Peer uses blocks - whether the worker is aware of peer workers who have worked with eco-blocks (baseline = 26.7%, MDE = 0.25 SD).

For private clients (N=528): (1) Knows local block supplier - whether the client knows a local eco-block supplier (baseline = 11.2%, MDE = 0.35 SD); (2) Aware of government policy (baseline = 40.9%, MDE = 0.30 SD).

For the secondary outcome "knows block supplier," the pre-specified main analysis estimates separate effects by respondent type (contractors, private clients). A supplementary pooled interacted model combining contractors and private clients (N=1,320) tests whether effects differ across respondent types. The pooled regression includes respondent-type fixed effects and treatment-by-type interactions; both the pooled model and the separate-sample estimates are reported regardless of interaction significance.

The following outcomes are pre-registered as exploratory due to limited statistical power, small sample sizes, or indirect intervention alignment.

For contractors (government tender participants only, N=182): Proposed blocks in tender - whether the contractor has ever proposed eco-blocks in a government tender (baseline = 13.2%, MDE = 0.48 SD), reclassified from secondary to exploratory because the conditional sample (N=182, approximately 23% of contractors) yields an MDE substantially larger than the primary contractor outcomes (both 0.26 SD) and larger than the client adoption outcome classified as primary (0.31 SD). The study is underpowered for confirmatory inference on this outcome.

For contractors (N=792): Block market access - a continuous index (0-1) measuring supplier accessibility, constructed as the average of perceived block availability and knowing a local supplier (baseline = 25.3%, MDE = 0.37 SD), classified as exploratory as a mechanism variable for the coordination channel.

Barrier perceptions index (contractors) - a z-scored index from 4 binary barrier items with meaningful baseline variation: blocks not easily available (60.5%), not preferred by clients (24.6%), not mentioned in tender (20.6%), and lack of awareness (14.4%). Six additional barrier items excluded due to floor effects (less than 7% prevalence). Captures information and coordination frictions targeted by T1 (N=577 non-adopter contractors, MDE = 0.29 SD).

For procurement officers (N=152): Environment in top 3 bid criteria - whether the officer ranks environmental considerations among top 3 tender evaluation criteria (baseline = 2.6%, MDE = 0.58 SD), classified as exploratory due to small sample and floor effect. Received quality training - whether the officer has received training on assessing eco-block quality in tenders (baseline = 21.1%, MDE = 0.57 SD), reclassified from secondary to exploratory; all procurement officer outcomes are exploratory due to approximately 2 officers per upazila cluster, with MDEs ranging from 0.56 to 0.59 SD across the four PO outcomes.

For private clients (N=528): Received info on eco-blocks - whether the client received information about eco-friendly blocks in the past year (baseline = 12.5%, MDE = 0.30 SD), classified as exploratory because it is a supportive mechanism/manipulation measure rather than a focal confirmatory endpoint.

Endline-only measures: The following are collected only at endline because they capture post-intervention states. ANCOVA adjustment is not available for these outcomes.

Compliance and dosage: (1) Workshop attendance (all treated respondents) - binary indicator equal to 1 if the respondent attended the BRAC/BIGD information workshop. (2) Training dosage (T2 contractors only) - number of workers per contractor who actually received hands-on training (0, 1, or 2), required for the pre-specified dose-response analysis. (3) Used supplier directory (T1/T2 contractors and private clients) - binary indicator equal to 1 if the respondent consulted the supplier directory, separating directory compliance from workshop attendance.

Contamination check: SUTVA is threatened when substantive treatment content reaches the control group, not when respondents merely hear that a workshop occurred. We measure: (1) Received workshop materials (all respondents, especially Control) - binary indicator equal to 1 if the respondent received printed eco-block materials (leaflets, supplier directory, cost comparison sheets) from someone outside their upazila, capturing material leakage. (2) Received detailed block information from external source - binary indicator equal to 1 if someone shared specific information about block costs, supplier contacts, or technical specifications, distinguishing substantive treatment leakage from general awareness.

Adoption dynamics: (1) New adoption since baseline (contractors and private clients) - binary indicator equal to 1 if the respondent used eco-blocks for the first time since baseline. For private clients, this outcome is especially important because baseline adoption is rare (3.0%); it complements the repeated "ever used" measure by isolating first-time adoption after the intervention. (2) Number of block projects (contractors) - count of construction projects using eco-blocks in the past 6 months, capturing the intensive margin; analyzed using Poisson QMLE (negative binomial as robustness check).

Mechanism measures: (1) Cost comparison belief (contractors and private clients) - perceived relative cost of blocks versus bricks on a 5-point scale (1 = much cheaper to 5 = much more expensive, with "don't know" option). (2) Post-use satisfaction (contractors and private clients who used blocks) - overall satisfaction on a 5-point Likert scale.

"Aware of government policy" is a binary indicator equal to 1 if the respondent is aware that government prefers alternatives to fired clay bricks in public procurement, and 0 otherwise.

"Received info on eco-blocks" is a binary indicator equal to 1 if the respondent reports receiving information about eco-friendly blocks in the past year, and 0 otherwise.

"Knows local block supplier" is a binary indicator equal to 1 if the respondent reports knowing a local eco-block supplier and 0 otherwise, directly targeted by the supplier directory covering 122 verified producers across 39 districts (approximately 47 upazilas).

"Ease of contacting suppliers" is measured on a 5-point ordinal scale (Very Difficult, Difficult, Moderate, Easy, Very Easy). The main analysis uses the full ordinal variable treated as continuous in OLS. A binary recode (= 1 if Moderate, Easy, or Very Easy; 0 if Very Difficult or Difficult) is reported as a robustness check.

"Advised clients to use blocks" is a binary indicator equal to 1 if the contractor has recommended eco-blocks to private clients, and 0 otherwise.

"Peer uses blocks" (workers) is a binary indicator equal to 1 if the worker is aware of peer workers who have worked with eco-blocks, and 0 otherwise.

This is a cluster randomized controlled trial with upazila as the unit of randomization. Upazilas are randomized into three arms (Control, T1, T2) stratified by district, with one upazila per arm within each sampled district. Outcomes are measured via surveys of contractors, procurement officers, construction workers, and private clients sampled within each upazila after the endline survey of the same respondents. Main analysis will estimate intention-to-treat effects on respondent-level outcomes, accounting for clustering at the upazila level and stratification by district.

Not available

Randomization was done in office by computer (Stata 17 MP). A random uniform number was generated to order upazilas within each district, and study arms were assigned within district using a modular rule. A reproducible random seed based on a timestamp was used. (Note: Code set seed 20251026 was used in Stata).

Randomization Unit Upazila (Cluster randomization). Randomization was stratified by District, assigning Upazilas within a district to different arms.
Additional randomization (within one arm): Within-contractor randomization of construction workers in T2, selecting exactly one of two sampled workers per contractor for treatment using a uniform random number.

Yes

66 upazilas (clusters), spread across 22 districts, with 22 upazilas per arm.

3,056 respondents to be tracked as a panel from baseline to endline. Baseline composition: 792 contractors, 152 procurement officers, 1,584 construction workers, and 528 private clients.

Control: 22 Upazilas (~1020 respondents)

Treatment 1 (Information): 22 Upazilas (~1,015 respondents)

Treatment 2 (Info + Worker Training): 22 Upazilas (~1,021 respondents)

With 22 clusters (upazilas) per arm, MDEs assume 80% power, alpha = 0.05 (two-sided), ANCOVA adjustment (baseline-endline correlation rho = 0.10-0.50 assigned per outcome based on baseline prevalence), and 5-10% attrition. The design has df = 66 - 22 - 2 = 42 (clusters minus strata minus treatment arms). ICCs for primary outcomes are low (rho_ICC = 0.010-0.018 for contractors, 0.012 for workers), yielding design effects of 1.1-1.5. Confirmatory primary outcome MDEs: Contractor block adoption (ever used eco-blocks; baseline = 27.1%): MDE = 11.4 pp (0.26 SD) at 5% attrition, 11.7 pp at 10% attrition - detects a 42% relative increase. Contractor awareness index: MDE = 0.26 SD. Worker block usage (ever worked with eco-blocks; baseline = 31.9%): MDE = 9.0 pp (0.19 SD) - best-powered primary outcome. Private-client awareness index: MDE = 0.30 SD. Private-client block adoption (used blocks in construction; baseline = 3.0%, preferred estimand: non-adopter subsample N approx. 512): MDE = 5.3 pp (0.31 SD) - statistically equivalent to client awareness index. Procurement tender evaluation (exploratory; N=152): MDE = 29.3 pp (0.59 SD). Proposed blocks in tender (exploratory; contractors, N=182): MDE = 16.4 pp (0.48 SD). Both classified as exploratory due to limited power from small conditional samples. Worker training receipt (compliance check; within-contractor clustering, 264 contractors): MDE = 3.8 pp (0.26 SD). Estimation strategy: The primary estimand is ITT. The pre-specified confirmatory contrasts are: (1) T1 vs. Control (beta_1 - information intervention effect) and (2) T2 vs. Control (beta_2 - combined information + training effect). These two contrasts carry confirmatory status. The T2 vs. T1 contrast (beta_2 minus beta_1 - marginal training effect beyond information) is pre-specified and reported for all primary outcomes but is treated as secondary because it tests the incremental value of training beyond information, a more demanding and less-powered comparison. Binary outcomes are estimated using a linear probability model (LPM). Count outcomes (number of block projects) are estimated using Poisson QMLE; negative binomial regression is reported as a robustness check only. Multiple comparisons: The study has a limited number of confirmatory primary outcomes across substantively distinct respondent domains (contractors, construction workers, private clients), tested at two arm-vs-control contrasts. Because confirmatory inference is restricted to five pre-specified primary outcomes and two contrasts, and because all other analyses are clearly classified as secondary or exploratory, we do not apply a formal multiple-testing correction. Instead, we rely on ex ante outcome prioritization, transparent reporting of all estimates, and design-based inference using randomization inference and wild cluster bootstrap p-values for all primary outcomes. The confirmatory primary outcomes by respondent type are: contractors (2): non-adopter conversion and awareness index; workers (1): ever worked with eco-blocks; private clients (2): awareness index and used blocks in construction (preferred estimand: non-adopter subsample, N approx. 512). Worker training receipt is a compliance check, not a confirmatory outcome. Procurement outcomes are exploratory due to small sample size (approximately 2 per upazila); proposed-in-tender is exploratory due to conditional sample (N=182). Inference robustness (reported in main tables): Randomization inference (Fisher exact p-values, 10,000 permutations within district strata) and wild cluster bootstrap p-values (Cameron, Gelbach, and Miller 2008) are reported in the main results tables for all primary outcomes, given the modest number of clusters (22 per arm). Additional robustness checks: results with and without baseline covariates; pooled T1+T2 vs. Control; SDS covariate sensitivity. We collected the 13-item Marlowe-Crowne short-form Social Desirability Scale (SDS) at baseline to capture individual differences in socially desirable responding. The total SDS score (0-13) is the sum of binary-coded socially desirable responses. We construct two subscales: an Attribution subscale (0-5, claims of positive traits) and a Denial subscale (0-8, denial of negative traits). The Denial subscale has better reliability (alpha = 0.59) than the full scale (alpha = 0.48) or the Attribution subscale (alpha = 0.28). The mean total SDS score is 8.79 (SD = 1.81). Procurement officers show the highest social desirability (mean = 9.57), followed by contractors (9.03), private clients (8.83), and workers (8.59). SDS shows very weak correlations with primary behavioral outcomes (all |r| less than 0.05), suggesting self-reported block adoption is relatively unaffected by social desirability bias. Given this weak reliability and near-zero outcome correlations, SDS is excluded from the primary covariate set and addressed as a single robustness exercise: re-estimate primary specifications adding the standardized SDS total score (or the Denial subscale) as a covariate; if treatment effects are substantively unchanged, results are considered robust to response bias. We do not pre-specify SDS heterogeneity analyses (e.g., by SDS tertile) because the scale's weak reliability and near-zero outcome correlations make such tests uninformative and increase fishing risk. 53 respondents (1.7%) were flagged for data quality review: 32 with perfect SDS scores (13/13), 8 with very low scores (3 or below), and 13 with incomplete scales. These are retained in main analyses but examined in sensitivity checks.