Why Better Connections Can Mean Less Experimentation

Producing knowledge versus sharing it

Better communication makes discoveries easier to share. It can also make discovering something yourself less attractive: why accept the risk of trying an unfamiliar option when a neighbor might supply the answer?

In The Impact of Connectivity on the Production and Diffusion of Knowledge, Gustavo Manso and Farzad Pourbabaee study this trade-off mathematically. Their results distinguish knowledge production from collective welfare. Connectivity discourages experimentation in the settings they identify, but its welfare consequences depend on population size and how agents choose their actions.

A two-period experiment

The authors model people choosing between a safe action and a risky alternative over two periods. The safe action pays zero. The risky action produces either a positive payoff or a loss, depending partly on an unknown state shared by everyone.

In the favorable state, an experiment can succeed but need not. In the unfavorable state, success is impossible. Consequently, a success conclusively reveals that the state is favorable, while a failure leaves uncertainty.

Everyone starts with the same belief about the state. After the first period, agents observe their own outcomes and outcomes from a randomly selected set of others. They update their beliefs and choose again. Because they value future payoffs, they may initially experiment even when the risky action has a negative immediate expected return: information can improve their second decision.

These are theoretical results about strategic choices, not measurements of experimentation in actual organizations.

Why connections discourage exploration

The central incentive is free-riding on information. When others’ experiments become easier to observe, an individual can choose the safe action first and still learn before making the second choice.

At sufficiently pessimistic initial beliefs, everyone chooses the safe action. At sufficiently optimistic beliefs, everyone experiments. Between these regions, the model supports pure-strategy equilibria in which some agents experiment and others do not. It also identifies a symmetric mixed equilibrium, where everyone independently experiments with the same probability.

The leading network specification is an Erdős–Rényi random graph: each pair has the same connection probability. In the local version, agents observe only immediate neighbors. Holding the initial belief fixed within the interior mixed region, greater connectivity lowers each agent’s exploration probability.

The authors also distinguish more contacts from more uncertainty about contacts. In their general contact model, a mean-preserving spread in the number of contacts lowers the belief threshold for full exploration. Thus, increasing average access and changing its distribution are not interchangeable.

Less experimentation need not mean less welfare

The paper’s welfare conclusions require separating three cases. Here, social surplus means the sum of agents’ expected, discounted payoffs; per-capita surplus divides that total by the population.

Finite populations with symmetric mixing. Within the interior mixing region, expected per-capita equilibrium surplus is constant as local connectivity increases. People experiment less, but information travels more easily. The exploration probability adjusts by exactly enough to keep the probability of receiving no successful signal from others unchanged. Since agents are indifferent between exploring and taking the safe action, their equilibrium expected payoff stays unchanged too.

Finite populations with pure strategies. The number of explorers must be an integer. While that number stays fixed, better connections improve information sharing and raise surplus. But when connectivity induces one fewer person to experiment, surplus drops discontinuously. The authors show how such downward jumps can interrupt the gains from diffusion.

The large-population local-network limit. These per-capita jumps disappear. Pure-strategy and symmetric mixed-strategy surplus converge to the same limit, which is weakly increasing in connectivity. For beliefs between the no-exploration cutoff and the threshold for a positive immediate expected risky payoff, surplus eventually becomes flat. This is not a claim that connectivity always strictly improves welfare.

Wider diffusion and the social optimum

The authors separately study global observation, where information can travel throughout an agent’s connected component. In the large-population random graph, a giant component emerges above a connectivity parameter of one. They show a rapid tightening of the full-exploration region near this transition when success probabilities are small, with a precisely characterized scaling limit. This analysis concerns exploration incentives; the local-network welfare conclusions should not simply be transferred to it.

Finally, the paper compares private choices with a planner maximizing collective surplus. Individuals ignore the information their experiments provide to others. The resulting equilibrium therefore exhibits under-exploration relative to the social optimum, including a weakly lower exploration probability under symmetric mixing in the stated belief range.

The distinction is essential: connectivity can spread existing knowledge more effectively without strengthening incentives to create it. Even when equilibrium welfare stays constant, experimentation can remain below the socially desirable level.