Introduction

Although the number of stunted children under age 5 has fallen in many countries over the past decades, global figures remain alarming, and stunting is declining too slowly to meet the target set by the World Health Assembly for 2025 [1]. In West and Central Africa, the number of stunted children has actually increased by 7.8 million between 2000 and 2019, and little improvement is predicted in the coming years [2]. This puts those children at great risk of impaired physical and cognitive development, poor health, reduced scholarly success, and in the longer term, low economic productivity and poor reproductive outcomes for girls [3,4]. The international nutrition community contends that effective strategies to reduce growth retardation lie in the adoption of multisectoral approaches, which combine nutrition-specific and nutrition-sensitive interventions targeted at the “first 1,000 days” of life (between conception and the child’s second birthday) [5–7].

Promising nutrition-sensitive interventions include cash transfer (CT) programs. Originating in Latin America, these programs proved to be effective in reducing poverty and food insecurity and in increasing school enrollment and use of health services [8]. Despite their great potential, CT programs have demonstrated relatively modest impacts on childhood nutritional outcomes (linear growth, growth retardation, and micronutrient status) [9–11]. “Cash plus” programs, which combine CTs with additional components (e.g., community education and nutritional supplements), tend to have a greater impact than cash alone, especially on longer-term outcomes such as knowledge and behavioral changes, morbidity, and nutrition [12]. Those results have contributed to the popularity and expansion of CT programs worldwide, as reflected by the number of reviews on the topic [8,13–20]. The most recent reviews integrate new and emerging evidence from sub-Saharan Africa, where the number of CT programs has considerably increased since the 2000s [9,21]. Established in 2008, the multicountry “Transfer Project” research initiative provided evidence on the effectiveness of large-scale national CT programs in sub-Saharan Africa and partly rectified the knowledge gap for this part of the world [22]. However, there is still a dearth of evidence in West Africa, where most CT programs were small-scale pilots relying on external funding. The few published impact evaluations from Mali, Burkina Faso, and Ghana did not demonstrate any positive impact of CT programs on nutritional outcomes (stunting or wasting) among young children [23–25].

Optimizing the nutritional impact of future programs will require a better understanding of how CT programs work [26]. Very few evaluations have thoroughly and systematically studied CTs’ mechanisms of action and documented the pathways through which impact on nutritional status was achieved. In that sense, documenting the impact of CT programs on intermediary outcomes along the program’s theoretical impact pathways is as important as documenting their impact on primary nutritional outcomes. In 2014, at the scale of 5 districts, the government of Togo implemented a “cash plus” program combining unconditional CTs (UCTs) with community activities (sensitization meetings and home visits directed at child health, nutrition, and protection, as well as integrated community case management of childhood illnesses and acute malnutrition [ICCM-Nut]) targeted at mother–child pairs during the “first 1,000 days.” The aim of the program was to improve children’s nutrition, health, and protection. We implemented a cluster-randomized controlled trial to assess the impact of this program on children’s linear growth and on multiple intermediary outcomes along the program’s theoretical impact pathways.

We hypothesized that the program would improve children’s growth through a synergy effect between the community activities, which would raise mothers’ awareness and provide them with essential knowledge in health and nutrition, and the cash that would allow them to put into practice what they have learned through those community activities. This would result in improved health and nutrition of mother–child pairs (increased health-seeking behavior, decreased morbidity, higher dietary diversity, increased consumption of nutrient-rich foods), which in turn would help improve children’s growth. Providing women with better knowledge and financial means would also contribute to their empowerment, which has been recognized as an important factor in their own health and that of their children.

Methods

Sampling

Two repeated cross-sectional surveys were conducted, one before the first distribution of cash (baseline survey, May 16 to July 4, 2014) and the other after 24 months of program operation (endline survey, May 26 to June 16, 2016) (Fig 1). Representative samples of mothers and their 6- to 29-month–old children in both arms of the study were surveyed in each village. This age range was the best compromise for reasonable theoretical exposure of children to the program (at least 12 months) and stage of exposure (during pregnancy or not later than 6 months of age during infancy) after 2 years of program operation (S1 Fig). In each village, households were randomly selected from the program census database. The latter was compiled prior to the beginning of the study through an exhaustive household census whose aim was to list the households eligible for CTs in each ICCM-Nut village, i.e., those with at least one pregnant woman or a child under 2 years of age. To avoid the overrepresentation of children aged 6–23 months in our sample, we used these households as “starting points” from which we applied a random-route sampling method. At each “starting point,” a household member rolled a die. When the die indicated 5 or 6, we included the household when eligible for the evaluation study (i.e., when at least one child aged 6–29 months was living in the household) or we included the nearest eligible household otherwise. If the die indicated 1, 2, 3, or 4, we systematically selected the eligible household living next to the “starting point” household. If several households were living next to the “starting point” household (at equal distances, in a same compound), one was randomly selected directly in the field. Within selected households, we surveyed all eligible mother–child pairs. If a mother had more than 1 eligible child, the enumerator randomly selected one. This sampling procedure was repeated identically between baseline and endline surveys.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow chart of the study.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016. CT, cash transfer; ICCM-Nut, integrated community case management of childhood illnesses and acute malnutrition.

https://doi.org/10.1371/journal.pmed.1003388.g001

The sample size was calculated to detect a change of 0.20 z-score in height-for-age (HAZ) among 6- to 29-month–old children while considering the following parameters: 5% significance level, 90% of power, 10% of missing/invalid data, 162 villages to be randomized in 2 arms, and an intraclass correlation coefficient of 0.02. To detect a 0.20 HAZ difference between the intervention and control arms at endline (assuming a variance of 1.3 determined from previous studies), a sample size of 1,020 children per arm and per survey was required. This sample size also allowed detecting a difference in the prevalence of stunting of 9% between the 2 arms. This calculation was conservative because it did not account for the district stratification used for randomization. An average of 13 mother–child pairs were surveyed in each village (n = 128), ranging from 5 pairs in small villages (n = 16) to 17 pairs in large villages (n = 18).

Data collection

Data were collected using the same standardized questionnaire at baseline and endline. The questionnaire was administered face-to-face to women and heads of households during home visits by experienced enumerators speaking the local languages of the study area. These enumerators were identified through a highly selective process and extensively trained over 2 weeks. Their training mainly consisted of the exhaustive review, translation, and pretest of the questionnaire; each question was standardized across the different vernacular languages spoken in the study area. Enumerators also learned how to perform anthropometric measurements and how to collect data through tablets. Built upon the program’s theory of change, the questionnaire covered a wide range of intermediary outcomes meant to document the program’s theoretical impact pathways, which are presented in their simplified version in Fig 2 (for a more comprehensive version, please see [27]). At endline, a complementary module was administered to women in order to document how the program rolled out and how it was used by beneficiaries and to collect information on potential unexpected effects.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Simplified theoretical impact pathways of the program.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016. BCC, behavior change communication; CT, cash transfer; ICCM-Nut, integrated community case management of childhood illnesses and acute malnutrition.

https://doi.org/10.1371/journal.pmed.1003388.g002

Results

Participants’ characteristics at baseline

The number of households and mother–child pairs surveyed at baseline and endline is presented in Fig 1. There were no significant differences between the control and intervention arms at baseline in any of the sociodemographic and economic characteristics documented at the household, mother, and child levels (Table 1). Anthropometric indicators of children were also similar between the 2 arms at baseline. Households had a mean size of 8 persons and were mostly headed by men (>93%) with poor education. The main occupation of households was farming. Around half of households had access to improved sources of drinking water, and less than 5% had access to improved sanitation. Mothers from both arms were 29 years old on average, approximately 74% of them had no education, and almost all were engaged in a marital union, either monogamous (42%) or polygamous (53%). The children in our sample were 17 months old on average, and half of them were boys. The mean HAZ was −1.1 (0.06) in the control arm and −1.2 (0.06) in the intervention arm; the mean BW was slightly above 3,000 g in both arms.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of households and mother–child pairs by treatment arm.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016.

https://doi.org/10.1371/journal.pmed.1003388.t001

Program implementation and coverage

The program underwent major implementation issues, specifically regarding its CT component. At endline (June 2016), 58% of eligible women from our sample said they never received any cash (n = 635) (Table 2). The main reason for this lies in a major delay in updating the recipient list. Although the plan of implementers was to include eligible women before the program began (September 2014) and then refresh the list every 2 months, in reality, the second round of inclusion did not happen until May 2015. Furthermore, at the time of our endline survey, women identified in 2015 (1 year prior) were still waiting for their beneficiary cards and had not yet received any CTs. Several factors combined to explain this substantial problem with program fidelity. One of the reasons lies in the complex institutional set-up of the program, which involved 4 different ministries—the Ministry of Social Action, the Ministry of Health, the Ministry of Grassroots Development, and the Ministry of the Post—that had difficulty communicating and working together. Interviews conducted with program implementers as part of the process evaluation also indicated a lack of activity planning and an unclear definition of roles between actors, as illustrated by these words from a program stakeholder: “The ball is thrown into my court and I throw it back and that’s where we are […] and the project comes to a standstill.” All these problems resulted in sometimes huge delays in program implementation, as seen in the integration of new beneficiaries. In that specific case, program implementers also condemned a high, suspicious increase in the number of beneficiaries in between the 2 censuses, citing the allure of cash as a catalyst. Because many children either did not have birth certificates attesting to their age or had a certificate with questionable dates procured long after birth, the validation of the beneficiary list (both by the community and by independent consultants) took much more time than expected. In the end, out of the women who did receive the CTs (n = 400), 235 recalled how many times and claimed that they received it 13 times on average (Table 2). The median amount received by beneficiaries over 24 months was 60,000 XOF (min: 10,000–max: 135,000), in a country where in 2016 the average monthly income was around 26,000 XOF, 1 kg of maize cost 100 XOF, and one chicken around 2,000 XOF.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Program coverage and uptake based on retrospective data collected at endline.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016.

https://doi.org/10.1371/journal.pmed.1003388.t002

In comparison to CTs, CCPWs’ BCC activities and ICCM-Nut activities were implemented without major issues. The ICCM-Nut activities were already well orchestrated, and CHWs were able to help less experienced CCPW’s on a daily basis. Moreover, those program components were less multisectoral than the CT component and only relied upon 2 ministries, the Ministry of Health for the ICCM-Nut component and the Ministry of Social Action for the CCPWs’ activities. Although the implementation of those components was smoother and did not lead to substantial issues with program fidelity, they were nonetheless better implemented in the intervention arm than the control arm. Both CCPWs’ and CHWs’ BCC activities were more frequent in the intervention villages than in control villages (Table 2). For instance, women from the intervention arm were more likely to have received at least 1 CCPW visit than those from the control arm (OR: 2.43, 95% CI: 1.49–3.96, p < 0.001). CCPWs from intervention villages were also more likely to organize sensitization meetings than those from control villages (OR: 2.41, 95% CI: 1.51–3.86, p < 0.001). The same was true for CHWs (OR: 1.47, 95% CI: 1.01–2.13, p = 0.044). In turn, women from the intervention villages had higher odds of always attending sensitization meetings organized by CCPWs (OR: 1.81, 95% CI: 1.32–2.49, p < 0.001) and CHWs (OR: 1.43, 95% CI: 1.07–1.91, p = 0.016) than women from the control villages.

Impact on child’s linear growth and stunting

The intervention had a significant impact on the linear growth of children 6-to 29-months old, resulting in a difference of 0.25 HAZ for the benefit of children from the intervention arm (DD = +0.25, 95% CI: 0.01–0.50, p = 0.039) (Fig 3). However, this DD was mainly due to a deterioration of HAZ in the control arm between baseline and endline surveys (−1.10 HAZ at baseline versus −1.34 HAZ at endline, β = −0.24 [−0.41, −0.07], p < 0.01), whereas HAZ remained stable in the control arm (−1.13 HAZ at baseline versus −1.11 HAZ at endline; β = 0.03 [−0.14, 0.21], p = 0.728). The impact found on HAZ did not apply to child stunting (DD = −6.2 pp, ROR: 0.74, 95% CI: 0.51–1.06, p = 0.097; S2 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Mean HAZs ± SE of children aged 6–29 months at baseline (2014) and endline (2016) in the control arm and in the intervention arm, ITT analysis adjusted for child’s age, child’s sex, and districts.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016. CT, cash transfer; DD, difference-in-differences; HAZ, height-for-age z-score; ITT, intention to treat; SE, standard error.

https://doi.org/10.1371/journal.pmed.1003388.g003

Impact on intermediary outcomes

The food and nutrient pathway.

See Table 3. Globally, we observed no significant impact of the program on child feeding practices. However, when disaggregating the analyses by age groups, we found that older children (18–29 months) experienced positive impacts of the program on DDS7 (DD = +0.29, 95% CI: 0.03–0.54, p = 0.031). Children aged 18–23 months, who were still benefiting from the program at endline, also had higher odds of eating at least 2 types of ASF (DD = +9.1 pp, ROR: 2.65, 95% CI: 1.01–6.98; p = 0.048). The program also had a positive impact on the proportion of women who had at least 3 meals on the day prior to the survey (DD = +6.6 pp, ROR: 1.61, 95% CI: 1.07–2.40, p = 0.022) and on the proportion of women who consumed at least 2 types of ASF (DD = +4.5 pp, ROR: 2.24, 95% CI: 1.09–4.61, p = 0.029). No significant impacts were found on WDDS-10 or MDD-W. At the household level, the program had a positive impact on food security. Households from the intervention arm had lower odds of experiencing severe food insecurity than those from the control arm (DD = −10.7 pp, ROR: 0.63, 95% CI: 0.43–0.91; p = 0.016).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Program’s impact on intermediary outcomes along the food and nutrient pathway.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016.

https://doi.org/10.1371/journal.pmed.1003388.t003

The health and hygiene pathway.

See Table 4. Overall the program had no significant impact on children’s medical follow-up, except on the proportion of children who received vitamin A (DD = +6.7 pp, ROR: 1.65, 95% CI: 1.07–2.56, p = 0.024). The program had no impact on child morbidity (DD = −3.5 pp, ROR: 0.80, 95% CI: 0.56–1.14, p = 0.214). However, sick children from the intervention arm were less likely to forgo visits to the health center because of financial constraints than their control counterparts (DD = −26.4 pp, ROR: 0.23, 95% CI: 0.08–0.66, p = 0.006). The program had a positive impact on several antenatal care variables, including the proportion of women who received iron (DD = +4.5 pp, ROR: 2.08, 95% CI: 1.13–3.84, p = 0.019) and intermittent preventive treatment (IPT) for malaria (DD = +4.7 pp, ROR: 2.08, 95% CI: 1.04–4.16, p = 0.039). Cash-recipient women also had higher odds of delivering in a health facility as compared to nonrecipients (DD = +10.6 pp, ROR: 1.53, 95% CI: 1.10–2.13, p = 0.012). The program had a protective effect on LBW (DD = −11.8 pp, ROR: 0.29, 95% CI: 0.10–0.82, p = 0.020). Regarding postnatal care, beneficiary women were more likely to attend the postdelivery visit (DD = +12.9 pp, ROR: 3.28, 95% CI: 1.85–5.79, p < 0.001) and to receive postnatal iron supplementation (DD = +10.0 pp, ROR: 1.76, 95% CI: 1.20–2.59, p = 0.004) than nonbeneficiary women. The program had a positive impact on overall hygiene (DD = +9.0 pp, ROR: 1.44, 95% CI: 1.01–2.06, p = 0.046), which was mainly driven by the positive impact observed on women’s hygiene (DD = +9.0 pp, ROR: 1.54, 95% CI: 1.03–2.29, p = 0.035).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Program’s impact on intermediary outcomes along the health and hygiene pathway.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016.

https://doi.org/10.1371/journal.pmed.1003388.t004

Enabling environment.

See Table 5. We observed a positive impact of the program on the global knowledge of women (DD = +14.8, ROR: 1.86, 95% CI: 1.32–2.62; p < 0.001) but a rather negative impact on their decision-making power (DD = −0.53, 95% CI: −1.01 to −0.05, p = 0.030). This result is mainly driven by a negative impact on their decision-making pertaining to children (DD = −0.31, 95% CI: −0.54 to −0.08, p = 0.008). Regarding IPV, women receiving the CTs had lower odds of having experienced physical violence in the last 12 months than nonbeneficiaries (DD = −7.9 pp, ROR: 0.60, 95% CI: 0.36–0.99, p = 0.048). We did not find any impact on controlling behavior or emotional violence. In line with the consumption of ASFs among women and children, the program had a positive impact on the amount that households spent at market to buy ASFs (DD = +619, 95% CI: 268–971, p = 0.001).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Program’s impact on enabling factors.

CT cluster-randomized controlled trial, Northern Togo, 2014–2016.

https://doi.org/10.1371/journal.pmed.1003388.t005

Discussion

Our study showed UCTs associated with community activities (including ICCM of childhood illnesses and BCC activities related to children’s health, nutrition, and protection) and implemented over the “first 1,000 days” in Togo had a protective effect on linear growth among 6- to 29-month–old children when compared with community activities only. The program also had positive impacts on several intermediary outcomes along the “food and nutrient” pathways, those being household food security and consumption of ASFs by women and older children (18–23 months); along the “health and hygiene” pathway, positive impacts were seen on prenatal and postnatal iron supplementation, the proportion of women receiving IPT for malaria and sleeping under an impregnated mosquito net during pregnancy, the percentage of deliveries at health services, the proportion of LBW (among 6- to 19-month–old children), the proportion of women attending at least 1 postnatal consultation, and the hygiene of the mother. In the enabling environment, the program had a negative impact on women’s decision-making power but positive impacts on women’s knowledge, physical IPV experienced by women over the past 12 months, and household purchases of ASFs at markets. All of these impacts were found despite a suboptimal coverage of the CT component of the program.

The impacts of CT programs on nutritional outcomes have been reported in meta-analyses and reviews [9–11,19], but none of them found conclusive evidence regardless of the indicators considered. Most of the data came from conditional CT studies conducted in Latin America, even though evidence recently emerged from sub-Saharan Africa, including a small handful of West African countries, where no impact was found on nutritional outcomes. In Togo, the program had a positive impact on HAZ and LBW (in 6- to 19-month–old children). Although the positive impact of the program on LBW was clearly due to an improvement in the intervention arm, its impact on mean HAZ was mainly driven by a deterioration in the control arm, in which children surveyed at endline had lower HAZ than those surveyed at baseline. We assumed that this deterioration was mainly due to a bad conjuncture between 2014 and 2016, and that without the CTs’ distribution, the same thing would have happened in the intervention arm. A few elements in our data bolster that assumption. Looking at a series of shocks (disease, job loss, crop failure, etc.) households might have experienced over the past 12 months, we found that households had higher odds of having experienced losses of livestock at endline than at baseline in both the intervention (OR: 1.56, 95% CI: 1.26–1.93, p < 0.001) and control arms (OR: 1.40, 95% CI: 1.05–1.85, p < 0.05). In the intervention arm, however, households were less likely to report a high impact of those losses on their economic situation at endline than at baseline (OR: 0.51, 95% CI: 0.37–0.70, p < 0.001), pointing to a protective effect of the CTs. Moreover, given the low coverage of the program, the effect of the CTs in the intervention arm may also have been mitigated by children who never received it. PP analysis, even if exploratory, shows that CTs not only prevented a deterioration in children’s HAZ but actually improved it (DD = 0.62, 95% CI: 0.36–0.88, p < 0.001, S3 Fig). Those results are also corroborated by the PP results found on stunting (DD = −12.0 pp, ROR: 0.52, 95% CI: 0.35–0.77, p = 0.001; S2 Fig). The positive results found for both HAZ and LBW may be explained by several factors associated with the program. By targeting the 5 districts of the country with the highest rates of acute and chronic malnutrition in children under 5 and then in those districts selecting the villages with the least access to health services, the program had greater chances of success. Some studies have indeed shown that it is easier to provide meaningful changes in at-risk communities than in the overall population, in which the potential for improvements is lower [10–11]. On the other hand, adressing stunting through a single intervention in such at-risk communities, in which the issue is due to multiple and entangled factors (poverty, food insecurity, infectious diseases, limited access to basic services) may be more challenging than in other contexts, in which there are fewer contributory factors. Positive results seen on HAZ and LBW could also be explained by the fact that the program specifically targeted mother–child pairs during the “1,000 days” window of opportunity for addressing growth retardation and other forms of malnutrition, which probably maximized its impact [10, 26]. The transfer size of the program was also significant for this area. The literature suggests that transfers ranging from 15% to 25% of total monthly household expenditure are more likely to have impacts [10, 40]. In this study, we estimate that the transfers represented a minimum of 25%–30% of the total monthly expenditure of woman–infant pairs, even though it is very likely that part of the funds also benefited the entire household. The fact that CTs were combined with intensive BCC activities relating to health, hygiene, nutrition, and child protection certainly contributed to the success of the program by increasing participants’ awareness and knowledge of key childcare practices. During home visits, frontline workers provided individual psychosocial support and counseling on how to use the cash, which may have increased beneficiaries’ self-confidence and optimized the use of cash towards the adoption of such practices. Moreover, our results on program coverage suggest that the distribution of UCTs affected the service provision, as BCC activities seemed to be better implemented in the intervention arm than in the control arm. This certainly explains part of the impacts on HAZ and intermediary outcomes (consumption of ASF, maternal hygiene, maternal knowledge). This differential implementation may be due to higher motivation of frontline workers in the intervention villages. Because no blinding of the intervention was possible here, frontline workers may have been influenced by the knowledge of the allocation. Knowing that they were part of the intervention arm, they may have been more prone to be involved in their job because they would assume that families might be able to make better use of their recommendations thanks to the CT and also because they would feel watched by the program’s implementers and funders, who were awaiting results.

Regarding intermediary outcomes, we observed that the “cash plus” program increased households’ disposable income and that they used the cash to purchase food at markets, particularly ASFs. This resulted in a decrease in households’ experienced food insecurity, echoing the findings from 8 CT programs in sub-Saharan Africa (Ethiopia, Kenya, Lesotho, Malawi, Mozambique, South Africa, Uganda, and Zambia), as reviewed in De Groot and colleagues [6], and 1 food voucher program in Senegal [41]. This also led to higher consumptions of ASFs in older children and mothers, but not in dietary diversity indicators at the individual level or in other standard IYCF indicators. Evidence of CT programs’ impact on individual dietary indicators is limited, probably because intrahousehold allocation of foods may not be favorable to all members in several contexts. Furthermore, in our study we used qualitative indicators to reflect individuals’ diets, which may not be sensitive enough to detect significant changes [42,43].

Within the “health and hygiene” pathway, the program had no impact on children’s health or on children’s medical follow-ups, which were already quite good in both groups at baseline. In the literature, the effects of CTs on children’s health status are mixed, with some studies showing reduction in the prevalence of diarrhea or reduction in episodes of illness and other studies showing no impact at all [18]. Nevertheless, we showed that the program helped overcome the financial barrier for seeking healthcare for children in case of illness. Likewise, in Malawi, beneficiaries of the Mchinji Social CT program were more likely to receive care when sick compared with nonbeneficiaries [44]. The intervention also positively affected maternal antenatal and postnatal care, as well as the proportion of deliveries at health facilities and the proportion of LBW babies, which confirms the benefit of targeting women in early pregnancy, as shown in other studies from Latin America, Nepal, and India (reviewed in [45]).

Regarding the enabling variables, our findings showed, contrary to what could have been expected, that the program rather had a negative impact on women’s decision-making, at least when considering a score based on sole decisions by women. Taking a closer look at data, we showed that this result was mainly due to a decrease in beneficiaries’ decision-making power regarding decisions relative to children (such as health, nutrition, and schooling). Interviews with beneficiary women and their husbands, conducted as part of the process evaluation, indeed showed that fathers interfered more in decisions on children since the start of the program because they wanted to ensure "proper" use of the CTs. However, this does not mean that women had no say. Besides, the validity of decision-making measures have been criticized [46]; for example, the concept of “sole” versus “joint” decisions—potentially with or without disagreement—and how individuals perceive the notion of “ultimate decision maker” is difficult to capture quantitatively. Regarding IPV, women from the intervention arm had lower odds of being physically assaulted by their partner than women in the control arm. Very few studies looked at the impact of CTs on IPV, and to our knowledge, none of these were conducted in West Africa. These studies usually showed that CTs reduced physical and/or sexual IPV, mostly through an economic security and emotional well-being pathway [47,48]. Although the intervention had no overall impact on psychological/emotional violence, it had a positive impact on the proportion of women who were humiliated by their partner (DD = −6.4 pp, ROR: 0.61, 95% CI: 0.39–0.96, p = 0.031). Finally, the impact found on women’s knowledge could be explained not only by a better implementation of the BCC activities in the intervention arm, as already mentioned, but also by the strong motivations of beneficiaries themselves. They were incentivized both by the bonus and by the untrue belief that they needed to attend sensitization meetings to continue receiving the CTs.

Nevertheless, improving women’s awareness and lifting financial constraints may not always be sufficient. Demand-side interventions such as CTs should also invest in the supply side to be fully efficient. If the cash incentive allows women access to health services and food markets, then both quality healthcare and nutrient-rich foods should also be available to enhance their health and diet. To improve their impact, CT programs should also tackle issues that may optimize their implementation and coverage, such as communication and work sharing between stakeholders, staff organization and management, and planning and delivery of activities. Our study indeed showed that in remote and resource-poor settings, implementing such complex multisectoral programs according to protocol and ensuring their full coverage was challenging. This was especially true here given that the program was a pilot and that most of its operators were inexperienced. The program’s results could have been maximized through further experience, but although it was temporarily extended to other districts based on its positive results, the program did not continue in its current configuration. Nevertheless, it opened the road for CTs in Togo, as evidenced by the recent creation of a governmental CT program to support those who have lost their income because of the adoption of response measures against COVID-19. Moving beyond pilot and emergency programs to implement long-lasting and large-scale social safety nets will require more time and experience, but Togo now seems to be on the right track.

One major strength of our study lies in the use of a well-designed randomized controlled trial, in which we documented the program impact on a wide range of intermediary outcomes along the hypothesized program impact pathways, in addition to the impact on mean HAZ. Providing a global picture of the program impact rather than fragmented results helped in understanding the results found on linear growth. As foreseen in the theory of change of the program, our analyses confirmed that the intervention had an impact on multiple factors acting at different times over the “1,000 days” course to influence children’s growth. Examining the relative effect of each factor over the entire course of the intervention would have required performing a mediation analysis on the subsample of children aged 18–19 months, i.e., children who took full advantage of the intervention from early pregnancy to age 2. Unfortunately, our study was not powered for this. Besides, using a longitudinal design rather than repeated cross-sectional surveys would have allowed examining the change in children’s growth charts between groups to document the incidence of stunting rather than its prevalence. Because stunting results from a cumulative process, a longitudinal design would have provided more insight on children who were in the process of becoming stunted. Longitudinal designs may also offer opportunities for dose–response analyses. On the other hand, the usual delays in program roll-out and full implementation can have negative effects on an evaluation study cohort (e.g., children aging out of the “window of opportunity” for maximum nutritional impact of the intervention), making it unusable for assessing impact on age-sensitive outcomes such as stunting. Additional constraints of longitudinal designs include the logistical complexity and related additional cost of tracking individual children over time, with high risk of loss to follow-up. Finally, an information bias may also occur because individuals from a cohort are followed more closely than the whole population, with possible consequences on representativeness and external validity, whereas cross-sectional design better preserves real-life conditions.

This impact evaluation had some limitations. Although the program was running for 30 months, the impact study had to be conducted over a relatively short period of time—24 months—to avoid any seasonal effect between baseline and endline surveys. Theoretically, this time lapse was largely manageable in terms of the program’s correct functioning; nevertheless, there were still major implementation issues at the time of the endline survey, which probably weakened the impact of the program. This encouraged us to use PP analyses to discuss our results on children’s linear growth and stunting. Although this helped to confirm some ITT trends, we are aware that PP analysis is not without bias, especially because our PP analyses only concerned the CT component of the program and did not deal with participation in BCC activities. Because monitoring data were not made available to us, our analysis of program coverage is only based on retrospective declarative data collected from interviewees at endline. Such data are subject to recall bias. Beneficiary women may have been more likely than nonbeneficiaries to remember the number of times frontline workers organized sensitization meetings in their village or visited them. Moreover, we did not collect data on the functioning of health facilities, which could have helped explain certain results. Another limitation lies in the fact that the program itself may have influenced some interviewee’s responses, either because they wanted to stay in or because they wanted to make a good impression as beneficiaries. We attempted to mitigate such social desirability bias through the extensive training of enumerators. We are confident in the fact that our enumerators were trained enough to minimize this bias, however. Lastly, the external validity of our study is limited to rural at-risk areas of Western African countries.

To conclude, we demonstrated that CTs implemented during the “first 1,000 days,” in combination with community activities (including BCC sessions, home visits, and ICCM-Nut), had a protective effect on child’s growth among young children living in a vulnerable area of northern Togo. Our study contributes to the existing literature on CTs by providing new evidence from a West African country where data are missing and fills a gap by examining the effect of CTs on multiple intermediary outcomes along the theoretical impact pathways. Besides, the impacts found on various factors along those pathways confirm that to be efficient in the fight against stunting, interventions should address several determinants at a time and give particular attention to the conception and preconception periods, as evidenced by the positive effects observed on pregnancy- and birth-related outcomes. Our study also suggests that to impact growth, CT program should meet certain conditions including articulating clear nutrition objectives, providing sufficient amounts of cash, targeting the most at-risk populations over the first 1,000 days, and combining CTs with supply-side investments and educational components. It indeed supports the view that moving toward “cash plus” programs—i.e., programs that bundle CTs with additional components in an effort to improve the effect of the money—is key to making social protection truly transformative [49]. To be fully conclusive on this matter, studies comparing CT only with CT+ additional components such as BCC would be necessary. Researchers should also develop theory-driven evaluations that precisely estimate the contribution of a wide range of mediating factors to the final impact on linear growth (e.g., diet quality and quantity, health and medical follow-up, hygiene and sanitation, women’s empowerment) using, for instance, mediation analysis or structural equation modeling. This would help design future CT programs by identifying the most effective levers of actions to improve child’s growth. Finally, rigorous process evaluations addressing issues relating to CT programs’ implementation, coverage, and uptake are also needed to help lift the remaining barriers that prevent CTs from realizing their full potential.

Supporting information

Acknowledgments

Our acknowledgments go to Marie Ruel and Rahul Rawat from the International Food Policy Research Institute (IFPRI), Viviane Van Steirteghem and Christine Jaulmes from UNICEF, and Patrick Eozenou from the World Bank for their valuable inputs on the study protocol. We also thank Albertine Azambo Aquitème, Bénédicte Gnansa from the Ministry of Social Action and Emmanuel Ayayivi, Jean-Pierre Somana, and Frédéric Akakpo from the Ministry of Community Development, who facilitated our relationships with the local authorities in the field. We also acknowledge Joseph Meunier and Pascaline Rollet for their precious help on data management. We express our sincere gratitude to the field enumerators and supervisors for their dedicated work on data collection; a special thank goes to Sara Tandandja, one of our field supervisors, who also helped in multiple tasks throughout the study. Finally, we would like to extend a heartfelt thanks to study participants for their time and warm welcome.