A Simulated Ecology of Communication: Geography, Workgroups, and the Structure of Scientist Networks Christopher Liu Rotman/University of Toronto Organization.

A Simulated Ecology of Communication: Geography, Workgroups, and the Structure of Scientist NetworksChristopher LiuRotman/University of Toronto

Organization Science Winter ConferenceFebruary 7, 2013

2

Research Themes• Unravel the antecedents to relationships and

network positions.

• Ecology + Networks: – “Individuals may be universally instrumental, but their opportunities

to engage in strategic relationships are unevenly distributed across ecological and socio-demographic space.”

• Across multiple settings and levels. – Biotech; Senators; Regional Clusters

• Causal Inference: – Quirky settings and natural experiments…– Simulation Methods

3

Motivation & Research Question• Examine the effects of micro-geography on relationships in

the workplace.– Separate geographic and (confounding) organizational factors.– Examine geography in a knowledge production setting.(Festinger, et al., 1950;

Marmaros & Sacerdote, 2006)

• Examine alternative (i.e., mosaic vs. uniform seating) organizational policies.

– Most often, it is not possible to do an “experiment”.

• What might we learn about individual- vs. organizational- network topographies?

Research Question: What are the consequences to changes in laboratory size and

geographic distribution on relationships and networks?

4

Setting and Data (AMINO)• Setting:

– Explored in one biotechnology firm. – Network data.

• Complete email logs • => xcutting ties & count of open triads.

• Spatial Ecology:– Geographic distance

(floorplan/blueprint data). – Social (expertise) overlap.

• Scientists are organized into laboratories (i.e., workgroups)

5

Spatial Ecology at AMINO

5

N

Org Policy: “Like with like”. -Senior mgmt wanted “short arrows” -Individuals with longer arrows are Brokers..Implication: Spatially diverse positions allow broader networks…

Geography

Homophily

(choices) Xcutting TiesBrokerage

6

Counterfactual Policy (Mosaic)

N

Desired Experiment: with a Mosaic, would you see more cross-cutting ties and brokerage?

7

Network Simulation (I)I. Collect very rich data! a. Environmental Factorsb. Socio-demographicsc. (email) networks

d. Identify correlates of relationshipsa. Dyad-level regression

II. Model Network(1) People meet (2) People choose

whether (or not) to interact

(3) Measure network variables.

2) Preferencesa. P(Same sex)b. P(Same ethnicity)c. P(Same cohort)d. P(Same education)e. P(Same discipline)f. P(Same building-floor)g. P(Same laboratory)

1) Opportunity Structuresa. P(Randomly)b. P(Same building-floor)c. P(Same laboratory)d. P(Friends of Friends)

3) Network

Repeat 4X

(Jackson and Rogers, 2007)(Mayer and Puller, 2008)

8

Network Simulation (II)

III. Tune Parameters-Choose parameters & run model (II)-Δ Obs(I) – Exp (II.3)

-Revise parameters to identify plausible (Px)s.

IV. Simulation “Experiment”-hold parameters constant- generate “Mosaic”- Change lab composition- Measure changes in the simulated network.

(Jackson and Rogers, 2007)(Mayer and Puller, 2008)

3) Network Characteristics1. All Ties-#2. All Ties-SDLocal Ties3. Collocated/Same-Lab Ties-#4. Collocated/Same-Lab Ties-SDCrosscutting Ties5. Collocated/Diff-Lab Ties-#6. Collocated/Diff-Lab Ties-SD7. Non-Collocated/Same-Lab Ties-#8. Non-Collocated/Same-Lab Ties-SDDistant Ties9. Non-Collocated/Diff-Lab Ties-#10. Non-Collocated/Diff-Lab Ties-SD

4) Counterfactual Experiment

9

Dyad-Level Regression

10

Model Calibration

• Calibration is not (yet) perfect, but pretty darn close. • But the relative magnitudes of parameters seem to make sense.

Individuals meet with a 9% probability each cycle.

11

Preferences, Homophily, Friends, Org Structure, Geography.

All Ties

Collocat

ed/Sa

me-Lab

Ties-

#

Collocat

ed/D

iff-Lab Ti

es-#

Non-Collocat

ed/Sa

me-Lab

Ties-

#

Non-Collocat

ed/D

iff-Lab Ti

es-#

0

0.2

0.4

0.6

0.8

1

1.2

No HomophilyNo FoFsNo LabNo Bldg

12

Mosaic & Lab-size Counterfactual

To Be Determined…

(next week)

13

Concluding Thoughts• Networks and relationships are constructed on a

spatial ecological scaffold, even for strategic actors.

• Simulation methods help us to do “experiments” that organizations won’t, or can’t do. – Can explore gender, ethnicity, cohort, etc.

• This is a general methodology that may be applicable across multiple levels.

14

Setting and Data• Setting:

– Explored in the US Senate.

• Network data (DV). – Complete bill

cosponsorship patterns (1979-2001).

• Geography:– Spatial distance (chamber

seating data). • Rearrangements over time.

• Controls:– FE; Ideology; tenure;

0.0

05.0

1D

ens

ity

0 100 200 300 400Number of bills co-sponsored by senators i and j (within a congress)

Number of bills co-sponsored by Senators i-j