Discovery of The Double Helix

Central figures: Watson, Crick, Wilkins, and Franklin

In 1951, Watson saw Wilkins present at a conference in Naples

This encounter prompted Watson to move to Cambridge and change the direction of his research

Research Question

Vast literature on spillovers, long-term interactions

"Cricket spills over into business" - Enrico Moretti's The New Geography of Jobs

Research Question

Vast literature on spillovers, long-term interactions

"Cricket spills over into business" - Enrico Moretti's The New Geography of Jobs

What is the effect of short-term interactions on collaboration and knowledge diffusion?

Does networking have scientific value (collaborations, knowledge diffusion)? Or just "private" value (finding a job)?

Empirical challenges

Data on short-term interactions is hard to find

Endogeneity of colocation - people aren't randomly in the same place

Measurement

My approach

Quasi-experimental

Repeated interactions

Long-term outcomes

NIH study sections

NIH panels that evaluate grant applications

Convened by a Scientific Review Officer from the NIH

Organized around topics e.g. "Acute Neural Injury and Epilepsy", "Cellular Signaling and Regulatory Systems"

Criteria for reviewer selection

1. Expertise

2. Study section must have enough breadth to cover its scientific area

Permanent vs. Temporary Members

Permanent members serve 4-year terms, 3 meetings a year

Temporary members are recruited on an ad hoc basis

Temporary memberships are sometimes used as tryouts for permanent membership

What happens in a study section?

Each application is assigned 3 reviewers

Reviewers present their assessment of the application and discussion opens up to rest of panel

What happens in a study section?

Each application is assigned 3 reviewers

Reviewers present their assessment of the application and discussion opens up to rest of panel

What potential is there for knowledge diffusion? What do reviewers actually talk about?

Are reviewers different enough?

"I was always a minority, a clinical cardiologist, not an engineer. So it was my job to defend cardiology science and explain it" - Dr. David Sahn, Oregon Health Sciences University

Do they learn from each other?

"I've come away grateful for the opportunity and consciously aware that I learned some new science and developed skills to interact with people with different attitudes and opinions" - Dr. Alice Clark, University of Mississippi

Intellectual similarity

If scientists are too similar, may already be familiar with each other's work

If scientists are too distant, may be harder to use each other's knowledge

Intellectual similarity

If scientists are too similar, may already be familiar with each other's work

If scientists are too distant, may be harder to use each other's knowledge

Effects may vary by intellectual similarity and could be non-monotonic

Study section rosters

List of reviewers who attended a particular meeting for a particular study section

Author-ity

Dataset of disambiguated author names ("John Smith" problem)

Derived from MEDLINE, repository for biomedical literature

Citations

Clarivate Analytics Web of Science

Setup

Scientist pair (not ordered): $(i,j)\equiv (j,i)$

Pair-year panel

Sample

Over 14000 individuals matched out of 19000

235 study sections

Each panel has 29 members on average

Time period: 1992 to 2010, but most meetings observed are in 2006 or earlier

Identification

Limit to permanent members of study sections

Make use of fixed 4-year terms

Identification

Treatment intensity

Identifying assumption

Use reviewer-pairs who served within 3 years of each other

If scientists $i$ and $j$ served on a study section, when they were on the study section is exogenous

Model

$$y_{ij,t}=\sum _{k=-10}^{10}{\beta }_{k}1(K_{ij,t}=k)1(N_{ij}\in \{1,2,3\left\}\right)+{\gamma }_t+{\delta }_{ij}+{ϵ}_{ij,t}$$

$k\equiv$ number of years since first colocation

$N_{ij}\equiv$ number of years of overlap

$y_{ij,t}\equiv$ number of times $i$ and $j$ have cited each other

${\delta }_{ij}\equiv$ pair fixed effects

${\gamma }_t\equiv$ year fixed effects

Pre-trends

Pre-trends

Post-treatment

Post-treatment

Post-treatment

95% confidence intervals

NULL

Discussion

Repeated interactions matter - familiarity matters, not just awareness

Effects may be persistent

Future work

Other outcome variables of interest: collaborations, text measures of intellectual distance

Compare results for temporary members

Estimate specification with more structure

Conclusion

Repeated short-term interactions matter

Some evidence (not shown) that heterogeneity matters