Can Household Engagement Shift Care Burdens? A Randomized Experiment with Unpaid Caregivers in Antioquia, Colombia

The experiment is embedded in the apuesta formativa (formative program) of the Sistema Departamental del Cuidado de Antioquia (SCA),
an initiative of the Governorate of Antioquia designed to strengthen the autonomy, leadership, and well-being of women caregivers. The
formative program started in 2025 and has an ongoing 2026 cohort. It is organized into three thematic schools: Del pasado, presente y
futuro (identity, life planning, and productive projects), Del amor propio (self-care and economic autonomy), and De las autonomías
(gender rights, labor autonomy, and civic participation). Each school comprises several modules delivered through in-person group sessions
at the municipal level. The 2026 stage, on which this study focuses, covers nine modules across several sessions and emphasizes deepening
the reflections from 2025 into concrete tools for leadership, economic autonomy, and rights-based participation. The program is directed at
women identified by municipal administrations and Governorate databases as unpaid caregivers of children, the elderly, or persons
with disabilities or illness. Participation is voluntary, and the program does not involve the women's household members in any of its
standard activities.

The two interventions evaluated in this study are designed to extend the program's reach into the household by involving a co-resident
adult member alongside the participating woman. Both are additions to the standard program and do not modify its content or delivery.

Component B — WhatsApp-based information campaign

Component B consists of a WhatsApp-based information campaign that delivers content for each thematic component of the formative program. The intervention has three arms, with women assigned in a 2:1:1 ratio to control, individual-delivery treatment, and household-delivery treatment respectively.

For each thematic component, treated women receive two pieces of content: an infographic and an audio message. The infographic is a
didactic image that summarizes the main learning of that component and includes conversation-opener questions intended to prompt
discussion between the woman and her household member about the topics covered in the session, including care responsibilities,
gender roles, and time use. The audio message covers the same content in spoken form. The two formats carry identical information and are
always delivered together, so that all participants have an accessible way to engage with the material regardless of reading proficiency or
visual limitations.

The three arms differ only in whom receives the content:

Arm 0 (Control, N ≈ 703): Women receive no infographic or audio message. The research team does not contact them or any household
member through this campaign.
Arm 1 (Individual delivery, N ≈ 351): The infographic and audio message are delivered only to the woman herself.
Arm 2 (Household delivery, N ≈ 352): The infographic and audio message are delivered to both the woman and a co-resident adult
household member she nominated at baseline. Both recipients receive identical content simultaneously.

This three-arm design allows us to separately identify: (i) the effect of the information campaign itself, comparing individual
delivery to control; (ii) the incremental effect of involving a household member, comparing household delivery to individual delivery;
and (iii) the total effect of the household-inclusive intervention, comparing household delivery to control.

The study universe for Component B is 1,406 women, corresponding to those who completed the baseline survey (LB26) among the 2,063
attendees of the 2026 formative program cohort who were reachable by mobile phone. Women who did not complete the baseline survey are
not included in Component B but remain eligible for Component C.

Component C — Family closing session

Component C consists of a family closing session held at the end of the formative program. Women attending the program in treatment
municipalities are invited to bring a co-resident adult household member or close family contact to a final program session. The
session is facilitated by program staff and focuses on the themes covered throughout the formative program, with particular emphasis
on care co-responsibility, time use, and household task distribution. Women in control municipalities receive the standard program closing
without this session. To avoid anticipatory effects or changes in behavior prior to the session, women are not informed about the
existence of Component C until the invitation is issued at program closure.

2026-05-27

2026-11-01

The study has six primary outcome variables, all measured at endline (October–December 2026). Effects on these outcomes will be estimated separately for Component B (information campaign) and Component C (family session), with multiple hypothesis correction across the two co-primary components using the Benjamini-Hochberg FDR procedure (q=0.05).

PRIMARY OUTCOMES:
1. Hours of unpaid care work per day (continuous)
- Total daily hours dedicated to direct care (children, elderly, ill household members) plus indirect care (cooking, cleaning, household errands)
- Reference period: typical weekday in the past four weeks
- Source: endline survey, time use module

2. Hours of paid work per day (continuous)
- Total daily hours dedicated to all forms of paid work (formal employment, informal work, self-employment, occasional jobs)
- Reference period: typical weekday in the past four weeks
- Source: endline survey, labor module

3. Index of gender norms (Anderson standardization, continuous)
- Standardized index combining items measuring beliefs about gender roles in care work, women's economic participation, and household decision-making
- Includes 1st-order norms (respondent's own beliefs) and 2nd-order norms (perceived community beliefs)
- Standardization following Anderson (2008), with sign-adjusted direction (positive values indicate more egalitarian norms)
- Source: endline survey, norms module, compatible with baseline norms module

4. Disposition to engage in paid work (Likert composite, continuous)
- Composite measure combining willingness to seek paid work, perceived feasibility given household constraints, and concrete intentions in the next 12 months
- Standardized to mean zero, unit variance
- Source: endline survey, disposition module

5. Disposition to engage in recreational activities (Likert composite, continuous)
- Composite measure combining self-reported frequency of leisure and rest activities relative to other household members, perceived legitimacy of dedicating time to leisure, and concrete intentions to engage in recreational activities in the next 12 months
- Standardized to mean zero, unit variance
- Source: endline survey, disposition module

6. Index of internalized gender norms around care (Anderson standardization, continuous)
- Standardized index combining items measuring the degree to which the respondent feels her personal value in the household is tied to time dedicated to domestic tasks, guilt experienced when not fulfilling care responsibilities, and perceived social and family pressure to conform to caregiver roles
- Standardization following Anderson (2008), with sign-adjusted direction (positive values indicate lower internalization of norms that tie women's worth to care work)
- Source: endline survey, norms module

1. Intra-household distribution of care tasks (index, continuous)
-Standardized index combining the share of domestic and care tasks performed by the respondent versus other household members, and whether tasks are performed alone or shared with a co-resident adult
-Sign-adjusted direction: positive values indicate a more equal distribution of tasks within the household
-Standardization following Anderson (2008)
-Source: endline survey, household tasks module

2. Perceived barriers to care redistribution (index, continuous)
-Standardized index combining items measuring the degree to which the respondent perceives lack of household support, lack of communication, absence of decision-making power over task division, and belief that no one else will perform the tasks if she does not
-Sign-adjusted direction: positive values indicate fewer perceived barriers to redistribution
-Standardization following Anderson (2008)
-Source: endline survey, barriers module

3. Emotional well-being (index, continuous)
-Standardized index combining items measuring the frequency with which the respondent feels guilt when unable to dedicate time to household tasks, the emotional response when household tasks are not completed, and self-reported life satisfaction
-Sign-adjusted direction: positive values indicate higher emotional well-being
-Standardization following Anderson (2008)
-Source: endline survey, well-being module

4. Household decision-making autonomy (index, continuous)
-Standardized index combining items measuring the respondent's perceived role in household decisions about finances, time allocation, and care arrangements
-Sign-adjusted direction: positive values indicate greater autonomy in household decision-making
-Standardization following Anderson (2008)
-Source: endline survey, autonomy module

Experimental Design

The experiment evaluates two household-inclusive components of a
public caregiver training program (Sistema Departamental del Cuidado
de Antioquia, SCA) in Colombia: (1) a WhatsApp-based information
campaign delivering content to the woman and a co-resident adult
household member throughout the program (Component B), and (2) a
family closing session attended jointly by the woman and a household
member at the end of the program (Component C). These two components
are co-primary and form a 3x2 factorial design.

Study Population

The experimental universe consists of women attending the 2026
cohort of the SCA's formative program (apuesta formativa) across
46 prioritized municipalities in Antioquia (following exclusion of
five municipalities with persistent operational issues: Santo Domingo,
Vigía del Fuerte, San Juan de Urabá, Jericó, and Murindó). All
women who have attended at least one program session in 2026 in
these municipalities are eligible for Component C. Women who also
completed the baseline survey (LB26) are eligible for Component B.

Design Overview

The experiment combines two independent randomizations:

Component B — WhatsApp-based information campaign (co-primary):

At the individual level, 1,406 women who completed the baseline
survey are randomized to one of three arms in a 2:1:1 ratio:
- Arm 0 (Control, N ≈ 703): no information campaign
- Arm 1 (Individual delivery, N ≈ 351): infographic and audio
message delivered only to the woman
- Arm 2 (Household delivery, N ≈ 352): infographic and audio
message delivered to both the woman and a co-resident adult
household member she nominated at baseline

Randomization is stratified by municipality × sub-region × number
of prior attendance sessions (categorized as 0, 1–2, 3+), using
blocks of 4.

The information campaign is delivered via WhatsApp throughout the
treatment period (May to October 2026), with one delivery per
thematic component of the formative program. Each delivery consists
of an infographic and an audio message. The infographic is a
didactic image summarizing the main learning from that component
and including conversation-opener questions designed to prompt
discussion between household members. The audio message covers the
same content in spoken form. The two formats carry identical
information and are delivered together to ensure all participants
have an accessible way to engage with the material, regardless of
reading proficiency or visual limitations.

The three-arm design allows separate identification of: (i) the
effect of the information campaign delivered to the woman alone
(Arm 1 vs Arm 0); (ii) the incremental effect of extending the
delivery to a household member (Arm 2 vs Arm 1); and (iii) the
total effect of the household-inclusive intervention (Arm 2 vs
Arm 0).

Component C — Family closing session (co-primary):

At the cluster level, all 46 prioritized municipalities are
randomized 1:1 to treatment (approximately 23 municipalities) or
control (approximately 23 municipalities), stratified by sub-region
and municipality operational status. Women attending the program
in treatment municipalities are invited to a final program session
together with a co-resident adult household member or close family
contact. Women in control municipalities receive the standard
program closing without this session. Women are not informed about
the existence of Component C until the session takes place at
program closure.

Factorial Structure

Components B and C are crossed, generating six cells based on the
three-arm Component B design and the two-arm Component C design:

| | C=0 (no family session) | C=1 (family session) |
| B=0 (Control) | ~352 | ~351 |
| B=1 (Individual) | ~176 | ~175 |
| B=2 (Household) | ~176 | ~176 |

The main effects of each component and pairwise contrasts within
Component B (Arm 1 vs Arm 0, Arm 2 vs Arm 0, Arm 2 vs Arm 1) will
be estimated as pre-specified analyses. Interaction effects between
Components B and C are treated as exploratory. Cell sizes assume
proportional allocation to Component C treatment and control
municipalities.

Not available

Randomization will be conducted in office using a reproducible computer algorithm (Python), with documented seeds for full replicability. The experiment includes two independent randomizations:

(1) Component B (WhatsApp-based information campaign, all 2026 program attendees): individual-level block randomization within strata defined by municipality × sub-region × number of prior attendance sessions (categorized as 0, 1–2, 3+). Block size of 4, with 1:1 allocation to treatment (cartillas delivered to the woman and her nominated household member) versus control (no information campaign).

(2) Component C (family closing session): cluster-level randomization at the municipality level. All 51 prioritized municipalities are randomized 1:1 (approximately 25 treatment vs. 26 control), stratified by sub-region and municipality operational status. Women in treatment municipalities will be invited to attend a final program session with a co-resident adult household member or close family contact; women will not be informed of this component until the invitation is issued at program closure.

All randomizations will use the Python random library with documented seeds. Scripts and assignment files will be archived prior to implementation to ensure reproducibility. Balance checks will be conducted post-randomization on observable characteristics immediately, with re-randomization triggered if more than 10% of variables show significant imbalance (p<0.05).

The experiment uses two levels of randomization:

(1) Component B: individual-level randomization. Each attending woman is independently assigned to receive the information campaign (T=1) or control (T=0).

(2) Component C: cluster-level randomization at the municipality level. All 51 prioritized municipalities are randomized as clusters; all attending women within a treated municipality are eligible for the family session, while women in control municipalities are not.

Components B and C form a 2x2 factorial design and are the co-primary components of the study.

Yes

46 municipalities

1,406 women

The following figures are based on program attendance and baseline
survey records available as of the randomization date. The
Governorate of Antioquia has not yet confirmed the final 2026
cohort size; as additional women enroll, attend program sessions,
and complete the baseline survey, the experimental universe may
expand. The randomization will be updated accordingly as new
attendees are incorporated.

Component B (individual-level randomization, three arms, 2:1:1
allocation):
~703 women: Arm 0 — Control (no information campaign)
~351 women: Arm 1 — Individual delivery (infographic and audio
message delivered only to the woman)
~352 women: Arm 2 — Household delivery (infographic and audio
message delivered to both the woman and a co-resident adult
household member she nominated at baseline)
Total: 1,406 women who completed the baseline survey, stratified
by municipality × sub-region × number of 2025 attendance sessions
(categorized as 0, 1–2, 3+)

Component C (cluster-level randomization at municipality level):
~23 municipalities: control (no family session)
~23 municipalities: treatment (family session offered)
Total: 46 prioritized municipalities (following exclusion of five
municipalities with persistent operational issues: Santo Domingo,
Vigía del Fuerte, San Juan de Urabá, Jericó, and Murindó),
stratified by sub-region and municipality operational status,
1:1 allocation

Power calculations are reported for each co-primary component separately. Because baseline variance estimates and intraclass correlation coefficients are not yet available for this population, MDEs are discussed across a range of plausible assumptions rather than fixed to a single point estimate. All calculations assume 80% power and a two-sided significance level of 0.05. Component B (individual-level randomization, three arms) The three-arm design (Control, Individual delivery, Household delivery) allocated in a 2:1:1 ratio yields three pre-specified comparisons, each with different statistical power due to differences in cell sizes. Comparison 1: Individual delivery vs Control (N ≈ 351 vs 703) Under a simple difference-in-means specification, the MDE is 0.146 SD. Where baseline data are available and explain 20–40% of endline variance, the MDE falls to 0.113–0.131 SD. Using time use estimates for rural women in Colombia from the Encuesta Nacional de Uso del Tiempo as a reference (SD ≈ 3–4 hours per day), this corresponds to a detectable effect of roughly 20–35 minutes per day in unpaid care work. Comparison 2: Household delivery vs Control (N ≈ 352 vs 703) Statistical properties are essentially identical to Comparison 1 due to similar cell sizes. MDE is 0.146 SD without baseline controls, or 0.113–0.131 SD with baseline controls explaining 20–40% of variance. Comparison 3: Household delivery vs Individual delivery (N ≈ 352 vs 351) This comparison identifies the incremental effect of extending the information campaign to a household member beyond delivering it to the woman alone. Under a simple difference-in-means specification, the MDE is 0.211 SD. With baseline controls explaining 20–40% of endline variance, the MDE falls to 0.163–0.189 SD. This comparison is the least well-powered of the three; it will detect large incremental effects of household involvement but may miss smaller ones. Under a worst-case scenario in which fewer women complete the baseline survey than expected (e.g., 70% of the current 1,406, or N ≈ 984), the MDEs for comparisons 1 and 2 would rise to approximately 0.175 SD without baseline controls, and to 0.135–0.157 SD with baseline controls. The MDE for comparison 3 would rise to approximately 0.252 SD. The design retains the ability to detect meaningful effects on comparisons against control under this scenario, though with reduced precision for the pairwise treatment comparison. Component C (cluster-level randomization) With 46 municipalities (approximately 23 treatment and 23 control) and an average cluster size of approximately 47 women (2,162 eligible women distributed across 46 municipalities), the MDE depends critically on the intraclass correlation (ICC) at the municipality level. Applying a t-distribution correction with 44 degrees of freedom, at low ICC values (0.01–0.02), typical of community-level interventions in similar settings, the design can detect effects of approximately 0.147–0.170 SD. At higher ICC values (0.10–0.20), the MDE grows to 0.290–0.393 SD, reflecting the limited number of available clusters. The cluster structure is fixed at 46 municipalities regardless of individual-level attendance, so variation in enrollment would affect the average cluster size but leave the number of clusters unchanged. At ICC = 0.05 and a cluster size of 25 (reflecting a substantial enrollment shortfall), the MDE rises from approximately 0.223 SD to 0.276 SD, which remains within a plausible range for norm and time use outcomes in this context. Multiple comparisons within Component B Because Component B involves three pre-specified pairwise comparisons, we will apply the Benjamini-Hochberg false discovery rate procedure (q = 0.05) across the three tests to control the family-wise error rate. Power calculations will be updated in an amendment to this pre-registration once empirical variance estimates from the baseline data are available.