1 of 24

Network experiments in the lab

Ashley Harrell

Assistant Professor of Sociology

Duke University

2 of 24

Manipulate one (or more) variables (the IV) to observe effects in another variable (the DV), while holding constant all other variables

experiments in

3 of 24

Can allow researchers to make causal inferences

experiments in

4 of 24

Can allow researchers to make causal inferences

Classic example in networks: homophily vs. influence

experiments in

5 of 24

Can allow researchers to make causal inferences

Classic example in networks: homophily vs. influence

experiments in

6 of 24

Can allow researchers to make causal inferences

Classic example in networks: homophily vs. influence – observation doesn’t tell causal story

experiments in

7 of 24

Can allow researchers to make causal inferences

Classic example in networks: homophily vs. influence – observation doesn’t tell causal story

experiments in

8 of 24

experiments in

Have two hallmarks:

1) Random assignment

9 of 24

Participants assigned to experimental conditions randomly. Idiosyncrasies among people, groups are distributed equally across conditions, making the groups “statistically identical”

10 of 24

Participants assigned to experimental conditions randomly. Idiosyncrasies among people, groups are distributed equally across conditions, making the groups “statistically identical”

11 of 24

experiments in

Have two hallmarks:

  1. Random assignment
  2. Procedural control

12 of 24

All other variables except IV held constant. Diffs between conditions thus must be caused by IV, not extraneous variables

13 of 24

Random assignment + Procedural control allow for the establishment of causal relationships

14 of 24

lab experiments

Allow for a highly controlled test of theory

15 of 24

lab experiments

Allow for a highly controlled test of theory

Can complement observational data, uncover mechanisms

16 of 24

17 of 24

Cook and Emerson 1978; Yamagishi and Cook 1993

18 of 24

Harrell and Quinn 2023; Harrell and Wolff 2023

19 of 24

lab experiments in

Allow for a highly controlled test of theory

Can complement observational data, uncover mechanisms

Often focus on things (behaviors) that are difficult to observe or measure outside the lab

20 of 24

21 of 24

22 of 24

lab experiments have strengths and weaknesses…

High degree of control = high internal validity

Strongest possible evidence for causality, insight into mechanisms

Precise measurement, relatively simple models

23 of 24

lab experiments have strengths and weaknesses

External validity of lab experiments often criticized as settings don’t always resemble “the real world”

Sample, participants aren’t representative of the population (WEIRD, ”the college sophomore”, small)

24 of 24

Thank you!