Evaluating Extension Services and Productive Cash Grants in Rural Brazil

We work with program managers to implement a randomized rollout of the program among the targeted population across three Brazilian states. The targeted population consists of rural households currently enrolled in Bolsa Familia (Brazil’s flagship cash transfer program for the poor). These households are selected at random from Bolsa Familia’s database (Cadastro Único) and screened for program eligibility. A subset is allocated to early treatment, and the control group is simply allocated to future treatment.

We can use randomization due to the staggered rollout nature of the program. “Treatment” families start receiving the Fomento program in 2018 (Minas Gerais) or 2019 (Pará and Rio Grande do Sul). “Control” families do not receive it during the duration of the evaluation, but they are slated to receive Fomento after this study ends, in 2021. Hence, the intervention is totally public, and there is no hidden component.

Families selected to participate in the Fomento program receive non-reimbursable cash grants along with rural extension services. Rural extension is provided by trained technicians employed by public companies and has several components. First, the household receives a diagnosis of what it can do to increase its production. Next, based on this diagnosis, the technician elaborates a tailored business plan in consultation with the household, specifying the steps that the household must follow to achieve the targets. For example, the technician (extension worker) can recommend the adoption of new practices, purchase of equipment, or a change in the crop mix. The household then receives 60 percent of the total cash grant of R$2,400.00 (USD $650). The non-reimbursable funds are transferred by the Ministry of Social Development (MDS) directly to the beneficiary families through the Bolsa Família Program payment structure. This cash must be invested accordingly, adhering to the business plan previously developed. The extension agent then verifies that the household made the necessary investments and the remaining 40 percent of the cash grant is then released. Households are then provided technical assistance, with 5 to 10 visits during the course of one year to help implement the business plan.

2018-10-01

2020-12-31

1. Farm production, income, and profits
2. Total family consumption (durables and non-durables)
3. Total family income (cash and cash plus self-consumption valued at market prices)
4. Food security

1. Farm production is measured by physical quantities produced over a given time period. For example, kilograms of corn over the growing season. Farm income is measured as the value of output. For example, the price of corn times the quantity of corn produced. Profits is farm income minus expenditures on inputs used in the productive process. Expenditures include items such as seeds, fertilizer, and labor. Family labor is valued at market prices.

2. Total family consumption is composed of consumption of durables (existence of a suite of items such as refrigerator, TV, vehicles, air conditioning, etc.). Consumption of non-durables is taken over a certain period and main items include: Expenditures on food, transportation, electricity, fuel and housing.

3. Total family income is measured by summing up cash incomes across sources and family members. We also use an alternative income variable that adds to the cash sources all self-consumption production valued at market prices.

4. Food security is measured using a Brazilian version of the FAO Household Food Insecurity Access Scale (HFIAS).

1. Farm investment
2. Marketable production
3. Labor income
4. Propensity to migrate

1. Farm investment is measured by an index of total assets used in agricultural production and marketing -- e.g., machines, physical structures such as fencing and irrigation equipment, vehicles, greenhouses, etc

2. Marketable production is the difference between total farm production (see primary outcomes) and physical quantities consumed within the households. Alternatively, we can measure marketable production as the production that is actually traded in markets.

3. Labor income: income earned by household members in work outside of the household.

4. Propensity to migrate is measured by the number of household members who emigrate to urban areas, or to other rural areas in search of work within a specified period of time.

This research project employs a randomized controlled trial to investigate the impact of FOMENTO on indicators of interest. The trials will be conducted across the states of Para, Minas Gerais, and Rio Grande do Sul.

In each state, the respective EMATERs (state agencies in charge of managing FOMENTO) selected a number of municipalities for this research project (20 municipalities in Para; 15 in Minas Gerais and; 46 in Rio Grande do Sul). Using Cadastro Unico, MDS then listed all eligible families for the Fomento program in those municipalities (i.e., families living in rural areas with income per capita less than R$85 who have not benefited from the program before). Using this sampling frame, the EMATERs then shortlisted approximately 1,000 families in each of their respective states.

In Minas Gerais, the Fomento project was conducted under the name “Dom Helder” with the Ministry of Agrarian Development (MDA) as the primary resource of funds. Together with the MDA’s Agency for Technical Assistance and Rural Extension (ANATER), MDS also played a key role in the program articulation. The key difference in this arrangement vis-a-vis the arrangements in PA and RS is that in Dom Helder the extension company (ANATER-MG) is paid on a per-visit basis, and more visits per family are scheduled.

Dom Helder’s lottery had to be conducted before the data collection started, because of ANATER’s schedule. Out of a list of families compiled by MDS (based on Cadastro Único), ANATER shortlisted households to participate in the lottery. The research team then conducted the lottery designating families into the control and treatment groups. Families in the treatment group started receiving the Fomento by September of 2018, while those in the control group will receive it in 2020.

Data collection strategy:
This research project plans two waves of primary data collection using household survey instruments developed by the research team. In each wave, 2577 households from treatment and control groups will be interviewed.

The first wave (baseline) took place between July and September of 2018 before the treatment was delivered to the selected households. The second wave (end line) will re-interview the same households 18 months after treatment is delivered.

Data collection was performed by a specialized firm hired by MDS through an open public tender. MDS also hired an independent consultant to assess the quality of data collection activities by performing field spot checks and random phone calls to interviewed families will help to ensure data quality. The survey instrument was administered using tablets, with automatic field validation, consistency checks, and voice recording.

There is one survey instrument. It will be answered by the person who spends more time in rural activities within the household and contains questions on land use and crop choices; water usage; income; labor, farm production; technology use; household consumption; food security; women empowerment and subjective well-being.

Finally, we may use automated response calls to mobile phones to collect high-frequency data on consumption patterns and food security. Automated calls are a cost-effective way to collect data on issues such as food consumption and reactions to weather shocks, which require high-frequency monitoring. Despite being economically vulnerable, up to 85% of the target households use mobile phones, and coverage in the region is good.

For Rio Grande do Sul and Pará, the research team handed the shortlists to the data collection firm that was hired by MDS. The firm was supposed to interview 900 families in each state (approximately 100 families were left in a substitution list, to be used in case the firm could not interview the 900 families) between September and December of 2018. In Rio Grande do Sul, the firm successfully interviewed 870 families; in Pará, the firm successfully interviewed 866 families. These families form the sample for those two states. Once baseline data collection was finished, the Ematers agreed to a roughly 50-50 split of families within each municipality into treatment and control groups. Families in the treatment group will receive the Fomento program in 2019, and families in the control group will receive the program in 2021. These families are entered into a lottery, conducted by the research team. To conduct the lottery, the research team first stratified the sample by municipality and gender of the respondent of the baseline questionnaire. Within each stratum, families are sorted by the respondent’s age, and grouped in tuples of four; if there is more than one tuple, any malformed tuple (i.e., tuples comprising one, two, or three families) is grouped with the previous four-tuple. A simple lottery is then done within each tuple, starting with the first (youngest) tuple. This yields 444 families in the treatment group in Rio Grande do Sul, and 439 families in the treatment group in Pará.

For Minas Gerais, the lottery procedure is slightly different. To conduct the lottery, the research team first stratified the sample by municipality and gender of the household head (as per Cadastro Único). Within each stratum, families were sorted by the respondent’s age, and grouped in tuples of four; when the number of same-sex heads of households in a municipality is not a multiple of four (which can result in one tuple of fewer than four households), the remaining individuals from both sexes are regrouped together, forming new tuples ordered by age. The research team then conducted a simple lottery within the 4-tuples – with a 50-50 split of families within each tuple into treatment and control groups. This slightly different procedure means there is at most one tuple of mixed head of household genders in a municipality.

The research team handed the shortlists to the data collection firm that was hired by MDS, containing 1001 families, of which 900 families should be interviewed with priority, while 101 formed a replacement list. Baseline data collection took place between September and December of 2018.

In total, 1093 families participated in the lottery, and 418 will be in the treatment group. The research team handed 1093 families to the survey firm, with the goal of interviewing 900 families. Of these 1093 families, 841 were successfully interviewed.

Simple household level randomization (no clusters).

No

There are no clusters.

2577 households - 841 (418 in treatment, 423 in control) in Minas Gerais, 866 (439 in treatment, 427 in control) in Para, and 870 (444 in treatment, 426 in control) in Rio Grande do Sul.

1301 households receive a cash grant and technical assistance (FOMENTO), 1276 households are controls.

We estimate the MDEs using a log transformation and a transformation in the form of log(x+1) for variables 1 to 3. The following minimum detectable sizes are calculated using a power of 0.8 and alpha of 0.05. For the log(x) transformations: 1) Log of Total farm income [income from land rents + total sales revenue]: MDE (in logarithmic points): 0.3234 | Percentage: 4.29% of the untreated mean | Mean: 7.539 | SD: 2.236 2) Log of Total Family Expenses: MDE (in logarithmic points): 0.07332| Percentage: 1.14% of the untreated mean | Mean: 6.417 | SD: 0.647 3) Log of Total Family Income: MDE (in logarithmic points): 0.3687 | Percentage: 4.77% of the untreated mean | Mean: 7.722 | SD: 2.205 For log(x+1) transformations: 1) Log(Total farm income + 1): MDE (in logarithmic points): 0.4494 | Percentage: 10.9% of the untreated mean | Mean: 4.102 | SD: 4.095 2) Log(Total Family Expenses+1): MDE (in logarithmic points): 0.088 | Percentage: 1.38% of the untreated mean | Mean: 6.384 | SD: 0.799 3) Log(Total Family Income+1): MDE (in logarithmic points): 0.5645 | Percentage: 9.78% of the untreated mean | Mean: 5.769 | SD: 3.86 Variable 4 has not been transformed in any way: 4) Food Security Index - standardized by the control: MDE: 0.1104 | Mean: 0 | SD: 1 The stratification by municipality and gender and the within-tuple randomization implies that these are lower bound estimates.