Summary

  • A World Bank–funded cash-transfer experiment in Liberia, run with Liberia’s government and the nonprofit GiveDirectly, identified a positive total effect of joint planning on household income and a conditional effect on intimate partner violence that was confined to couples with a documented history of violence.
  • Random assignment licenses intervention-level causal claims for the planning component; the design holds the $750 cash amount constant across arms and therefore cannot identify any separate effect of cash magnitude on violence.
  • Two mediator candidates — deviation detection through a shared wall poster and a redistribution of decision-making power from men to women — are consistent with the released findings; the design does not adjudicate between them.
  • An asymmetric-detection explanation, in which the planning arm’s structured follow-up increased reporting rather than incidence, is at most a residual possibility in the released reporting and is not supported by the article’s description of identical baseline and one-year interviews across arms.

A 2022 World Bank–funded experiment in Maryland and Bomi counties randomized more than 2,300 low-income Liberian couples between cash alone and cash plus a 45-minute joint planning session, with each household receiving $750 in four installments — more than double the $250 average annual cash aid the Liberian government typically provides, according to Aurelius Butler, Liberia’s national coordinator for social protection. The experiment, carried out with the nonprofit GiveDirectly, produced two reported effects at one-year follow-up: Sarika Gupta, senior economist at the World Bank and a lead researcher on the study, said that “households that received the planning actually were even better off economically than households who received only the cash transfer,” and that among couples classified as high-risk based on pre-treatment interviews, those who received the planning reported “an increase in intimate partner violence” that did not appear in the cash-only arm.

What the randomization establishes

The study’s design — random assignment of planning versus no planning, with cash amount held constant — supports a direct causal claim about the planning component’s effect on intimate partner violence. Because treatment was randomized, all back-door paths from the planning treatment to intimate partner violence are blocked by design, and the reported high-risk subgroup finding is treatment-effect heterogeneity rather than a confounded association. The Liberia cash-transfer program operates against a baseline of approximately 35% self-reported intimate partner violence in the sampled counties, against a global average of 27% the article cites. Within that context, the conditional finding is precise: planning had no detectable effect on reported intimate partner violence in the broader sample, and a positive effect only in couples with a documented prior history of violence.

The two candidate mechanisms

The study team’s own framing, as conveyed by Gupta and by Amber Peterman, an impact evaluation specialist at UNICEF who was not involved in the research, identifies two pathways.

The first is deviation detection. Couples who went through the planning session jointly listed spending priorities on a poster that, Gupta said, “stays with the couple, and they can put it on their wall so they can see it” and serves as “like a constant reminder of what they’ve kind of committed to.” Couples who could not write used stickers — “a stack of books for education, a stethoscope and syringe for healthcare” — to mark their priorities. Gupta described the conflict pathway directly: “Anytime one partner or the other maybe deviates from that plan, the other person kind of can call them out. And so that there could be that kind of downside to planning.”

The second is power redistribution. The article reports that the standard government program, per Butler, operates on the principle “we target the female in the house,” and that many men had reported feeling “excluded” by that approach. The experiment’s design — equal cash and an equal say in spending — was constructed to address that exclusion. In the article’s framing of the study team’s published findings, the rise in the high-risk subgroup is presented as flowing from “power shifts or deviations from agreed plans.” The published framing characterizes the planning as having “shifted the power dynamic from men to women, possibly spurring partner violence.” Peterman summarized the determination: “It is quite clear that it’s something in the shift about couples deciding together that led to the increase in violence.”

Christian, a participant who provided only her first name, works on a farm while her husband gathers latex for rubber making. In their planning session they agreed on education, food, and home repairs; she used her cash to pay school fees for their five children, buy rice, and pay off debt, while her husband bought solar light bulbs and cement to patch their leaky roof. The article does not specify whether later arguments arose from the allocation.

What the design cannot separately identify

Because the cash transfer was $750 across both arms, any direct effect of cash magnitude on violence is unidentified by the study. The standard government program provides $250 annually, and whether a different cash amount paired with no planning would have shifted violence cannot be answered from this design. The natural direct and indirect effects of planning on violence cannot be non-parametrically identified from the data released alongside the article. The published reporting also does not separately measure whether the wall-poster artifact produces conflict independent of deviation — for example, as a status symbol or a source of embarrassment — versus through the deviation pathway alone. M1 and M2 may not exhaust the mechanism space; correlates of the planning session itself, such as communication frequency or time spent in structured conversation, are not explicitly eliminated by the data.

Discriminating observations the substrate supports

The economic and intimate partner violence outcomes are not coextensive: the planning effect is broad for income but narrow for violence. Any pathway that affected only the high-risk subgroup cannot be the same pathway that produced the broad economic gains. All four candidate pathways satisfy this constraint — the deviation and power-shift pathways because they specify a high-risk-moderated effect, and the asymmetric-detection and prior-history-confounder pathways because their effects would plausibly concentrate where prior incidents exist to report or to repeat.

Within the substrate, the literacy/sticker treatment is a candidate discriminator. The deviation pathway depends on a record the couple can be reminded of; calling out a deviation presupposes an artifact to consult. The power-shift pathway does not require the artifact to be load-bearing. The article establishes that “literacy rates are low in rural areas” and that the design provided stickers for couples who could not write. Couples who relied on stickers would have the deviation-conflict pathway muted, while the power-shift pathway and the prior-history-confounder pathway would still operate. If, among high-risk couples, the reported violence effect tracks whether the couple used the artifact as intended, the deviation pathway gains weight.

A factorial variant of the planning component — holding cash constant and varying the format between individual-only sessions, in which each partner meets the researcher separately and records a private list; joint-without-artifact, in which a couple meets together and agrees verbally; and the current joint-with-poster arm — would strip the joint decision and the wall artifact independently. If the high-risk effect tracks the joint artifact rather than the joint decision, the deviation pathway gains support; if it tracks the joint decision regardless of artifact, the power-shift pathway gains support.

Residual hypotheses

Two explanations sit in the substrate but are not supported by the released study description.

The first is asymmetric detection. If the planning arm had more structured follow-up contact with the research team than the cash-only arm, the high-risk subgroup differential could reflect asymmetric reporting rather than asymmetric incidence. The released reporting describes only a baseline interview and a one-year follow-up for both arms, with no indication of differential mid-year contact; the asymmetric-detection pathway is at most a residual possibility. A variant internal to the planning component — the wall-poster artifact normalizing household-decision reflection and lowering the threshold for intimate partner violence disclosure at follow-up — is consistent with the source without requiring asymmetric contact.

The second is the prior-history-as-confounder pathway. Prior intimate partner violence history may act as a proxy for a latent relationship-stability variable that independently predicts which couples are most likely to deviate from any joint plan and which are most likely to experience escalation when they do. The high-risk subgroup increase would then reflect the joint distribution of two consequences of an underlying factor that prior violence partially marks, rather than the marginal effect of planning conditional on prior history. If additional baseline markers — prior household financial conflict, prior household dissolution events, a physiological marker of stress — predicted the planning response at least as well as prior intimate partner violence does, this pathway would gain weight; if prior intimate partner violence remains the dominant predictor, the moderator interpretation is reinforced.

Policy implications and sequel

Peterman, the UNICEF specialist, characterized the planning component as net beneficial at the household level: “We know that this add-on actually did a lot of good at the household level,” and asked, “how could it be tweaked?” The researchers, per the article, said they hope to repeat the experiment with modifications to reduce conflict, provided they can secure funding.

The mechanism-level question — whether the harm in the high-risk subgroup flows from the deviation pathway, the power-redistribution pathway, or a combination — is the question on which any modification turns. If deviation detection is the active ingredient, a planning variant that uses individual commitment rather than a shared poster, or one that frames deviations as forgivable, should attenuate the intimate partner violence effect. If power redistribution is the active ingredient, those design tweaks should not move the effect, because the harm would flow from the redistribution itself rather than from the confrontation it occasions. A complementary counterfactual — whether the same high-risk couple’s violence would be eliminated by re-targeting cash at women only, removing the equal-say design that the standard government program already implements — is also open under either pathway.

Verdict

The randomized design establishes two things: that adding a joint planning session to a cash transfer improves household economic outcomes on average, and that it increases reported intimate partner violence in the specific subgroup of couples with a documented prior history of violence. The design does not, on the released reporting, adjudicate between the candidate mechanisms, and the difference between what the study establishes — that planning increased violence in the high-risk subgroup — and what it does not yet establish — why — is the one on which any scaling decision turns.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Causal DAG
Maps cause and effect as an explicit directed graph, exposing confounders and mediators (Pearl).