LEARNING AND STATUS IN SOCIAL NETWORKS

The patterns in which individuals interact with each other have important con-
sequences for determining outcomes across a wide variety of contexts. One notable
phenomenon that relies on these interactions is social learning. Social learning oc-
curs when asymmetrically informed individuals have the opportunity to observe the
choices of others and to incorporate this information when making their own choices.
Under certain assumptions this process leads to information cascades in which the
ability to learn from others ceases before any sound conclusion is reached. To the
extent this theory is valid, the implications for information aggregation are quite
negative. However, casual empiricism suggests that such an ine±ciency is unlikely:
many people making similar decisions over time are very unlikely to be continually
wrong. The ¯rst two chapters address this discrepancy.
Experiments that implement a standard social learning paradigm are reported.
A novel feature of the data is that we examine long sequences of decisions (up to
forty) and study the e®ects of di®erent signal qualities. In contrast to standard
equilibrium predictions, a pattern of cascade formation, collapse, and re-formation is
routinely observed. The main implication of these dynamics is that learning continues
throughout the sequence of decisions, so that the truth is nearly revealed. Quantal
Response Equilibrium, augmented to allow for non-rational belief updating in the
form of base-rate neglect, explains nearly all the features of the data.
One assumption underlying most studies of observational learning is that the order
in which individuals choose is exogenously ¯xed. In many applications, however,
strategic considerations play a major role in determining the timing of decisions. To
understand how timing issues impact the ability to learn from others' decisions, I
study a model in which decision times are strategic variables and individuals have
heterogeneous signal qualities. The main ¯nding is that with two players, the player
with better information announces ¯rst in (the unique) equilibrium. Consequently,both players make the same decision, as in a herd, but because of the sorting e®ect, the
outcome is informationally e±cient. It is also shown that the game ends immediately
as the time interval vanishes. In comparison to the standard exogenous sequence
assumption, welfare is always higher under strategic timing. When there are many
players, a herd forms immediately, and it is always on the correct action because early
decision makers have the best information, and their choices reveal the true state.
We next study a model that addresses how social networks form in strategic
settings. Individuals allocate a budget of resources across others, creating directed,
continuously valued links. In many applications, a (directed) link confers bene¯t to
both individuals involved in the link. By separating this bene¯t °ow into \giving"
and \taking" components, we are able to study the implications for e±ciency. The
main ¯nding is that ine±ciencies at equilibrium are due only to the giving of bene¯t.
We also relate network structures to underlying heterogeneity of players.
The ¯nal chapter analyzes large-scale social networks. The main question concerns
how correlation patterns in links across individuals a®ect the di®usion of a virus or
rumor. A surprise is that in all of the simulations considered, the Susceptible-Infected-
Susceptible (SIS) model behaves identically on networks with varying correlation
patterns. These are the ¯rst results in this context that isolate the role of correlation
structures.

Download PDF now.