Title: Willard Van Quine, professor of philosophy and mathematics emeritus from Harvard University who is r
1Modeling Peer Influence
Willard Van Quine, professor of philosophy and
mathematics emeritus from Harvard University who
is regarded as one of the four most famous living
philosophers in the world, wrote his doctoral
thesis on a 1927 Remington typewriter, which he
still uses. However, he "had an operation on it"
to change a few keys to accommodate special
symbols. "I found I could do without the second
period, the second comma -- and the question
mark. "You don't miss the question
mark? "Well, you see, I deal in certainties."
2Modeling Peer Influence Issues of Selection
Endogeneity
- Selection
- That some unobserved factor, z, creates both
friendships and the outcome of interest. - Endogeneity
- That the causal order of peer relations and
outcomes is reversed. Peers do not cause Y, but
Y causes friendship relations
3Selection
- What do we know about how friendships form?
- Opportunity / focal factors
- - Being members of the same group
- - In the same class
- - On the same team
- - Members of the same church
- Structural Relationship factors
- - Reciprocity
- - Social Balance
- Behavior Homophily
- - Smoking
- - Drinking
4Selection
Network Model Coefficients, In school Networks
5Selection
How to correct this problem?
- Essentially, this is an omitted variable problem,
and my solution has been to identify as many
potentially relevant alternative variables as I
can find.
6Endogeneity
Estimated Y b0 b1(P) e where P some
peer function. But the actual model may really
be P b0 b1(Y) e
7Endogeneity
Does it matter?
Algebraically the relation between y and p should
be direct translation of the coefficients
since
The statistical problem of endogeneity is that
when you estimate b1, it does not equal 1/b1,
because of our assumptions about x, and hence e.
(see Joel H. Levine, Exceptions are the Rule, for
a full discussion of this)
8Endogeneity
Fully specified peer influence models
Where W is a matrix of interpersonal weights,
calculated from the friendship adjacency matrix
(the mixed-regressive autoregressive peer
influence model, see Friedkin 1999, chapter 2,
Doreian, 1982, SMR)
These models can be estimated directly with Add
Health data, but again the problem is that W may
be determined by Y.
9Endogeneity
Possible solutions
- Theory Given what we know about how friendships
form, is it reasonable to assume a bi-directional
cause? That is, work through the meeting,
socializing, etc. process and ask whether it
makes sense that Y is a cause of W. - Models
- - Time Order. We are on somewhat firmer ground
if W precedes Y in time. Thus, using the
in-school friendship structure to predict wave 1
outcomes is useful. - - Simultaneous Models. Model both the friendship
pattern and the outcome of interest
simultaneously.
10Endogeneity
- Simultaneous models
- One way to do this is with SEMs, but for
identification, you must find some variables that
predict friendship that do not also predict Y,
which given the very nature of the endogeneity
problem, is hard to do. - Could also model a mixed network of relations
and behaviors using dyads and a p style model to
predict ties within the joint network
11Endogeneity
A mixed selection and influence model
Simultaneous balance on friendship and
behavior. Two linked models a) actors seek
interpersonal balance among friends b) actors
change their opinions / behaviors as a weighted
function of the people they are tied to, with W
weighted by number of transitive ties
12Endogeneity
Mij b0ij bx(Xij)bq(qi) bp(pj) eij
13What are the policy implications?
If we find a relation between peer and behavior,
do these types of problems make a difference for
what we should recommend to policy makers, such
as school officials?