When thinking of how to design information interventions, I would often think about targeting the people who are better connected. After all, they can spread the word to more folks. And since they are connected, maybe they have more credibility.
Recently I read a paper by Lori Beaman and Andrew Dillon that gave me pause.
The setting: Mali.
The program: agricultural extension. Teaching farmers how to compost.
Beaman and Dillon are looking to see how social networks transmit information…or not. So they set up an experiment where they train some farmers, give them some calendars with the agriculture information to distribute and then measure who knows what (as well as who gets the calendar).
Now before proceeding, it’s important to get some network concepts out there. One idea is degree, which is the total number of links in your network. The second concept is betweenness centrality, which Beaman and Dillon define as “the share of shortest paths from all pairs of nodes in the network that connect to the node.” Its basically about how influential you are and the more of it you have, the more likely you are to connect unconnected cliques of people (you are a bridge).
Into 52 Malian villages, Beaman and Dillon bring their intervention. And to look at network structures, they (randomly) target 3 different types of people across villages. Some villages (23 to be precise) have 4 random folks (2 men and 2 women) targeted to be the information starting point. Others have the men and women whose households have the highest degree targeted (so the popular folks). And the third group is villages where household with the highest betweenness measure were targeted (and then randomizing over whether the man or woman was targeted).
Then it’s off to the races. Training is followed a month later by a knowledge quiz for the whole sample. And the enumerators are on the lookout for calendars they distributed (with codes so they could be traced back to the original farmer).
First result: the training worked. Farmers who were trained scored 3 points (53% higher) on the knowledge quiz. And they gave out their calendars: if you were directly connected to the treated farmer, you had a 30 percent chance of getting a calendar. This drops by 9.4 percentage points if you have an indirect connection and 31.9 percentage points if you aren’t connected at all. But wait…if you are a woman, your chances are much lower – for men the direct link translates into a 40 percent chance of getting a calendar but for women its only 13 percent.
So that’s the swag. What about the knowledge itself? Knowledge degrades quickly: direct connections show a 31 percent decline in knowledge, and it gets worse as we move to indirect links and those with no connections. Again, men do significantly better, with the direct links scoring 26 percent lower, and indirect links scoring 31 percent lower. Women, on the other hand, lose 35 percent with the direct links, and this drops to 40 percent lower for indirect links. So all networks are showing information frictions (the old game of telephone is alive and well), but women’s networks are showing a more rapid decline.
Does who you target matter? On average not. So it doesn’t matter if you target the more popular, more connected or some random farmer. But, there are differences by gender. If the targeting was based on betweenness (that influential connectedness), women in those villages show significantly lower knowledge than those in degree or randomly selected farmer villages. So choosing a certain kind of focal farmer can reinforce exclusion of more marginalize populations (women have smaller networks on average than men in Beaman and Dillon’s data).
So, while we can expect some information diffusion from our interventions, we have to be careful about who we choose. Choosing those influential folks may exacerbate existing social exclusion and worsen outcomes for those who are marginalized. On the other hand, choosing a random person isn’t a bad choice. For efficiency, the options deliver the same result. Not for equity.
Join the Conversation