Alcohol and Agriculture: Results from a Field Experiment in Kenya

David Murphy is an assistant professor of economics at Colgate University.

Why look at alcohol in Africa?

While alcohol abuse is a global problem, it may be most acute in Sub-Saharan Africa (SSA). The World Health Organization (WHO) reports that SSA as a region has the highest age-standardized burden of disease and injury of alcohol as well as the highest percentage of drinkers who engage in heavy episodic drinking (binge drinking). Moreover, as African countries continue to develop, greater disposable income will likely lead to greater consumption of alcoholic products.

Unlike the United States and other developed countries, alcohol consumed in SSA is usually from informal sources. Although technically illegal in some countries (such as Kenya), most drinkers get their alcohol from households in their neighborhood or village that specialize in producing homebrew. This creates a challenge for policymakers attempting to stem the tide of alcohol abuse, as supply side levers (such as taxation) are impractical. In addition, these illicit brews can be dangerous, containing hazardous byproducts (e.g. methanol) or additives (e.g. potassium), which often generate fatalities among drinkers.

The program

In a recent working paper, I analyze the results of a randomized control trial with an intervention that utilizes peer effects and intrahousehold relationship ties to decrease alcohol consumption and generate positive economic outcomes. Peer effects, such as those found in groups like Alcoholics Anonymous, can potentially lead individuals to learn from others who face, or have faced, similar challenges. They also provide a source of pride or shame to the participant given the expectation that they will share their successes and failures with others in the group. Strengthening intrahousehold relationships can also potentially decrease alcohol abuse, as family-system approaches towards therapy have been shown to be most effective in alcohol remission (Kaufman and Pattison, 1981).

A local partner organization implemented an alcohol remission program in early 2021 with 200 randomly selected treatment households in 20 villages (randomly selected from a list of 30 random villages) in western Kenya. Each of the households comprised two cohabiting spouses, in which at least one was a regular consumer of alcohol. The organization invited both of the spouses to participate in the treatment program, which comprised two major parts:

1. Group therapy and discussion: This was a facilitator-led discussion group that met over three non-consecutive full days. During these discussions, participants shared challenges, frustrations, and successes with the group regarding alcohol and self-control. Social workers also introduced and led discussions on common topics each day, such as dealing with peer pressure, coping strategies, etc.
2. Spousal counseling: after the first phase, the counselors shifted to personal visits to the treatment households, guiding the participants towards individual goals and empowering spouses to check each other’s alcohol consumption. Additional personalized training on coping strategies and methods of self-control were introduced during these sessions. In this phase, counselors visited the households eight times over a two-week period.

Because withdrawal symptoms from alcohol remission can be dangerous, we had a full-time nurse on staff. This nurse made visits to all treatment households, and prescribed medication to aid withdrawal symptoms where needed.

Data and results

Enumerators first conducted a short baseline survey in late 2020. I then randomly placed any household with at least one spouse who was found to be a consumer of alcohol in the baseline study into treatment or control groups for further participation in the study (motivation to decrease alcohol consumption was not a criterion for eligibility and was not measured at baseline). Following the intervention described above, enumerators conducted detailed follow-up surveys with treatment households, other households in the same village (a spillover control group), and households in the ten remaining, randomly selected non-treatment villages (a pure control group). The first survey, which measured short-run results, was given to households an average of 6-weeks post intervention, with a subsequent medium-run survey given an average of 16-months post intervention. Enumerators collected data on alcohol consumption by both directly asking the respondent how much of various types of alcohol they had consumed in the past 24 hours, 7 days, and 30 days, as well as privately and concurrently asking their spouse how much they consumed. The spouse-reported data thus serves as a crucial check to ensure the results are not biased by embarrassment or experimenter demand effects (EDE), where a respondent may not answer truthfully – instead telling the enumerator what they think the enumerator wants to hear (I include additional checks for EDE in the paper). I then converted the consumption data to US standard drink equivalents for ease of interpretation. As the reader can see in Table 1, average levels of alcohol consumption are very high among men in the sample. One reason for this is the average local serving size: from the data collected, the average purchased drink was 300 milliliters, was 34% alcohol, and cost less than 1 USD! (For reference, a 12oz can of beer is 355 milliliters, and contains around 5% alcohol.)

In analyzing the effects of the program, I find large and significant decreases in alcohol consumption. Sixteen months post-intervention, intention-to-treat effects show the program decreases likelihood of drinking in the pooled gender sample by 12 to 17 percentage points (Figure 1) and decreases overall alcohol consumption by 4.0 to 4.4 US standard drinks (Figure 2) in the week prior to the survey relative to those in non-treatment villages (representing a 37 to 45% decrease from the control mean).

Some of the most interesting effects of the program that I observe are on agricultural outcomes (all households in the sample farm at least some land). ITT results show that random assignment to the treatment increases per-acre fertilizer usage and harvest values by approximately 25% compared to the pure control mean, controlling for pre-intervention values (Figure 3). Why is this? While there are many possible mechanisms, along with finding a decrease in spending on alcohol, I find an increase in agricultural expenditures. The evidence suggests a reallocation of spending from alcohol to agriculture, increasing per-acre fertilizer use and thus harvest values.

Why this research matters

As described above, alcohol abuse is a serious and increasing problem in SSA and has many negative impacts, ranging from lower work productivity to violence in the household. This research shows that a relatively inexpensive self-help alcohol remission program leads to significant decreases in alcohol consumption, both in the short- and medium-run, and increases agricultural productivity. These results suggest that peer effects and spousal therapies can be effective in increasing self-control in a developing country context, making this intervention a potential model for larger-scale programs to address substance abuse in the region.

______________________________________________________________

Interview with the Author

1. This is a very innovative research idea. Could you speak a little bit as to how you came up with it?

The idea came from observations I made while I was doing my dissertation fieldwork in this area of western Kenya. While I was visiting these villages, we noticed that many people, especially men, were either too inebriated to participate in the research or were functional alcoholics. It wasn’t too far of a leap to guess that this was having negative effects on agricultural productivity, in an area where agriculture is the most important economic activity. While it is difficult to study the negative effects of alcohol on agricultural productivity (we cannot randomize inebriation), I thought I could look at the opposite and implement a project to invite a randomized sample to participate in an alcohol remission program. That way, I could see the effect on alcohol consumption and determine if there was an effect on agricultural productivity. So, to answer the question, it was just casual observations that led me to think of this research idea.

2. What was the biggest lesson you learned about the research process from this project?

I learned a few things in this project, but one was how to work with a non-research organization as a key implementing partner. Incentives are very different between those in academia and NGOs: in development economics, we are interested in clean research designs, while NGOs are oftentimes motivated to help as many people as possible. While this is an admirable goal, we obviously cannot identify a treatment effect without clearly distinguishable treatment and control groups. Therefore, I soon realized the need to ensure that the partner organization from the director down to individual counselors understood that we were only treating those who were on the “treatment list,” which was counter-intuitive to many of those on-the-ground NGO workers.

3. Your results show some important short and medium-run effects of the intervention. Do you expect long-run effects? What could these look like?

This is a good question because based on the data I expect that positive effects with respect to alcohol consumption and agricultural outcomes will persist. This question is whether in a few more years we will be able to identify them given the diffusion I already see happening in, and between, villages. Given the intensity of the program, the results appear to have changed the culture with regard to alcohol in these villages and the social networks themselves (see paper here). On the other hand, it is completely possible that there will be backsliding and a return to higher alcohol consumption. Either way, I hope to return to do a five-year follow-up unless attrition is unreasonably high, which due to internal migration from this area, could, unfortunately, be the case.