1. Diffusion & Interventions
Diffusion of innovations
Behavior change
Behavior change is short term whereas diffusion looks at the long view of how new behaviors spread 1 1

2. Diffusion of Innovations
New ideas and practices originate enter communities from some external source. These external sources can be mass media, labor exchanges, cosmopolitan contact, technical shifts and so on. Adoption of the new idea or practice then flows through interpersonal contact networks. 2 2

3. Diffusion of Innovations
Rogers wrote consecutive texts on this topic:
1962 1st Edition
1971 2nd Edition (with Shoemaker)
1983 3rd Edition
1995 4th Edition
2003 5th Edition
Synthesized, elaborated, codified, explained diffusion of innovations 3 3

4. ELEMENTS OF THE DIFFUSION OF INNOVATIONS
1) The rate of diffusion is influenced by the perceived characteristics of the innovation such as relative advantage, compatibility, observability, trialability and complexity, radicalness, and cost. 2) Diffusion occurs over time such that the rate of adoption often yields a cumulative adoption S-shaped pattern. 3) Individuals can be classified as early or late adopters. 4) Individuals pass through stages during the adoption process typically classified as (1) knowledge, (2) persuasion, (3) decision, (4) implementation or trial, and (5) confirmation. 4 4

5. Characteristics of an Innovation
Relative advantage
Compatibility
Complexity
Trialability
Observability
Cost
Radicalness 5 5

6. 4 Elements according to Rogers
Innovation: An idea or practice perceived as new
Perceived attributes: relative advantage, compatibility, complexity, trialability, observability
Communication channels
Homophily vs. heterophily
Time: at the individual & macro levels
Social system 6 6

7. Hypothetical Cumulative and Incidence Adoption Curves for Diffusion Homogenous Mixing7 7

8. 8 8

9. Behavior Change Stages in Four Models9 9

10. Diffusion
Takes time
Is difficult even when something is seemingly worthwhile
Is guided and influenced by many factors, some obvious, some not so obvious
Provides a macro – micro perspective on behavior change 10 10

11. Diffusion
Process by which an innovation is communicated through certain channels over time among the members of a social system
Communication is special in that it attempts to reduce uncertainty about the innovation
Diffusion vs. Dissemination vs. Technology Transfer 11 11

12. 12 12

13. Hypothetical Diffusion When Adopters Persuade Non-adopters at a Rate of One Percent (Homogenous Mixing)
Time
Cum.
Ado.
Non
New 1 2 3 4 5 6 7 8 9 10
9.75
18.55
33.66
55.99
80.63
96.25
99.86
100 95
90.25
81.45
66.34
44.01
19.37
3.75
0.14
4.75
8.8
15.11
22.33
24.64
15.62
3.61 13 13

14. Hypothetical Cumulative and Incidence Adoption Curves for Diffusion Homogenous Mixing14 14

15. The Diffusion of Knowledge, Attitudes and Practices (KAP)15 15

16. Example of Diffusion 16 16

17. The Two-Step Flow Hypothesis of Mass Media Influence
Friends
Family
Mass
Media
Opinion
Leaders
Coworkers
Others 17 17

18. Mathematical Models Used to Derive Diffusion Rate Parameters
Valente, T. W. (1993). Diffusion of innovations and policy decision-making. Journal of Communication, 43(1), 18 18

19. History
Early pre-paradigmatic research by Anthropologists, Economists, & Sociologists interested in cultural change ( )
In 1943, Ryan & Gross published a study farmers’ adoption of hybrid seed creating the paradigm
By 1962 Rogers published “Diffusion of Innovations” which solidified the paradigm
Coleman, Katz & Menzel’s (1966) study of Medical Innovation solidified the theory on diffusion networks 19 19

20. Ryan & Gross
Studied the diffusion of hybrid seed corn, retrospectively
2 communities in Iowa, 255 of 257 farmers adopted
Contrasted economic and social variables
Established diffusion paradigm 20 20

21. Number of Diffusion Publications Over Time21 21

22. Diffusion Publications and Research Innovations: Ratio of Innovations to Publications Remained Constant 22 22

23. Reasons for Decline
It was perceived as fallow intellectually (15 of 18 variables used by Ryan & Gross)
Political climate was against cultural imperialism. It was politically incorrect – associated with technological hegemony
Environment suffered from the spread of technological innovations (pesticides, herbicides)
Social scientists not trained in matrix methods to investigate network reasons for diffusion 23 23

24. Research on Innovation Diffusion in Many Fields
In Demography and fertility transition studies
In Sociology by re-newed attention on diffusion networks
In Communication as a tool to evaluate communication campaigns
In Organizations as a means to understand and plan change 24 24

25. Diffusion Networks A specific branch and approach to diffusion study
Some might argue that diffusion is only diffusion when one looks at networks and that other “diffusion” studies are behavior or social change
Diffusion networks has been historically the branch of networks focused on behavior change 25 25

26. Lineage of Diffusion Network Models
From Valente (2006)
Type (1) Social integration
Social Factors are important - Ryan & Gross 1943
Social Integration - Coleman Katz & Menzel 1966
Opinion Leaders - Rogers 1964
Norms - Becker 1970
Rogers & Kincaid 1981
Type (2) Bridges & Structure
Weak Ties - Granovetter 1973
Burt
Watts (2002) 26 26

27. Lineage (cont.) Type (3) Critical levels Schelling 1972
Thresholds - Granovetter 1978
Critical Mass - Marwell, Oliver et al. 1988; Markus 1988
Network Thresholds - Valente 1995/1996
Type (4) Dynamics
Marsden & Polodny 1990
Spatial & Temporal Heterogeneity – Strang & Tuma, 1995
Valente 27 27

28. (1) Social Integration/ Opinion Leaders
Integration can be measured many ways
Behavior is a function of being embedded within a/the community
Usually operationalized as receiving ties 28 28

29. Coleman Katz & Menzel 1966 Actually 1957 was first paper
Data collected
Interviewed all MDs in 4 Illinois cities: Peoria, Bloomington, Galesburg, & Quincy
Sampled prescription records first 3 days of each month to measure Time of Tetracycline Adoption 29 29

30. Diffusion of Tetracycline for Marginal versus Integrated Doctors30 30

31. Diffusion Network Simulation w/ 3 Initial Adopter Conditions (Valente & Davis, 1999)31 31

32. Diffusion Network Game
Distribute red, white & blue chips
Give:
Red to OLs
Blue to Randoms and
White
Allow them to give chips to those people who nominated them 32 32

33. 33 33

34. Diffusion Network Game
Distribute Red, White & Blue Chips to different initial starts
Red = awareness
White = attitude
Blue = behavior
Can only receive a white chip if have red one; only receive a blue one if have red & white 34 34

35. 35 35

36. (2) Structural Structural models require data from the entire network
Can use sociometric data to identify bridges
Can also use to measure structural equivalence and constraint 36 36

37. Granovetter, Strength of Weak Ties (1973), AJS
Seminal article
Cited thousands of times
Granovetter was White’s student
First faculty appointment at JHU
Left JHU for Stonybrook, now at Stanford 37 37

38. Granovetter, Strength of Weak Ties (cont.)
Cognitive balance inclines friends of friends to know friends - transitivity. Granovetter shows Figure 1 which is the forbidden triad, i.e., this type of network configuration rarely occurs. C
The Forbidden Triad A B C
If A & B are linked and
A & C are linked then it
implies that C & B are linked A B 38 38

39. SWT: Bridges created shorter paths
Bridges - individuals who link otherwise disconnected sub-groups. Individuals who act as bridges have weak ties. So a bridge is composed of weak ties, but not all weak ties are bridges. H I G D J F C A B K E L
Weak Tie 39 39

40. (3) Critical Levels Tipping points
Macro vs. micro tipping points, critical mass vs. thresholds
Most CM/threshold models were not explicitly social network explanations 40 40

41. (4) Dynamics Can model how ideas/behaviors spread through a network
Simplest model assumes static (fixed) network and the idea spreads on that network
Start with initial adopters and let the behavior percolate through the network 41 41

42. Network Exposure = Non User = User Exposure=33% Exposure=66%42 42

43. Exposure Equation
where E is the exposure matrix, S is the social network, A is the adoption matrix, n is the number of respondents, n+ indicates the sum of each row, and t is the time period. The exposure equation is a very general model in which the social network can be direct relations, positional relations, narrowly focused, or broadly focused. 43 43

44. Computing Network Weighted Scores Such as Network Exposure
Nx1
Vector
of Row Totals
Nx1
Vector
of Scores
Nx1 Vector of
Network Weighted Scores
…………...……….N
………….….N
N x N Adjacency
Matrix (or weight
matrix) X – = : 44 44

45. Computing Network Weighted Scores Such as Network Exposure
Nx1
Vector
of Row Totals
Nx1
Vector
of Scores
Nx1 Vector of
Network Weighted Scores
…………...……….N
………….….N …. …. …. …. …. . 1 . 2 3 . .5
1.0
.33 . X : – = 45 45

46. NxT Matrix of Exposure Scores …. …. …. . 46 46

47. 4. Personal network exposure
Personal network exposure is the degree an individual is exposed to an innovation through his/her personal network. Network exposure provides: 1. awareness information 2. influence/persuasion 3. detailed information on how to get the innovation, possible problems, updates, refills, enhancements, novel uses 4. something to talk about 47 47

48. Network Exposure (cont.)
5. social support needed to face opposition
6. reinforcement and a sense of belonging
7. relay experiences
Exposure computed on direct ties; and on ties of ties by using the geodesic and weighing the ties by its inverse.
Every network has a different maximum geodesic measure so we need to approximate the influence of any one point on any other point. Luckily the flow matrix has been created which does precisely that. 48 48

49. Three Studies with Data on Time-of-adoption & Social Networks
Medical Innovation
Brazilian Farmers
Korean
Family Planning
Country
USA
Brazil
# Respondents
125 Doctors
692 Farmers
1,047 Women
# Communities 4 11 25
Innovation
Tetracycline
Hybrid Corn Seed
Time for Diffusion
18 Months
20 Years
11 Years
Year Data Collected
1955
1966
1973
Ave. Time to 50% 6 16 7
Highest Saturation
89 %
98 %
83 %
Lowest Saturation
81 %
29 %
44 %
Citation
Coleman et al (1966)
Rogers et al (1970)
Rogers & Kincaid (1981)
Valente, T.W. (1995). Network models of the diffusion of innovations. Cresskill, NJ: Hampton Press 49 49

50. Datasets Provide static view of network
1 based on observational data on adoption (but it is sampled)
2 based on recall- though recall is probably pretty good
They are varied and the network data are pretty good 50 50

51. Two of these Datasets Have Received the Most Attention
Medical innovation by Coleman, Katz & Menzel (1966):
Burt, 1987; Marsden & Podolny 1990; Strang and Tuma, 1993; Valente, 1995; 1996; Van den Bulte & Lilien, 2001
Korean family planning by Rogers & Kincaid (1981):
Dozier, 1977; Montgomery, 1994; Valente, 1995; 1996. 51 51

52. Regression on Time to Adoption by Network Exposure & External Contacts
Medical Innovation
N=125
Brazilian Farmers
N=792
Korean Fam. Plan.
N=1,025
Exposure Direct Contacts
0.54
1.31*
1.09
Exposure via SE
0.88
2.85**
1.02
Attitude toward Science
0.61*
Journals
1.16*
Income
1.01*
Visits to City
1.00
# of children
1.10**
Campaign Exposure
1.04* 52 52

53. Maximum Likelihood Logistic Regression on Adoption by Time, Ties Sent/Received & Network Exposure.
Medical Innovation
N=947
Brazilian Farmers
N=10,092
Korean Fam. Plan.
N=7,103
Time (as %)
0.21
0.72
0.31
Time – Log
0.68
1.94
0.67
Sent
0.91
0.90
0.96
Received
1.06
1.02
1.06**
Exposure via Direct Contacts
0.64
1.07
1.19
SE Exposure
0.93
2.47*
1.12 53 53

54. Exposure Adoption?
Represents a challenge to the diffusion and other behavior change models
Could be a function of location on the diffusion curve – more likely after critical mass
Very disappointing from a replication perspective
What model can explain this? 54 54

55. Network Threshold = Non User = User PN Threshold=33% PN Threshold=66%
Valente, T.W. (1996). Social network thresholds in the diffusion of innovations. Social Networks, 18,
PN Threshold=33%
PN Threshold=66%
PN Threshold=100% 55 55

56. Graph of KFP Communication Network
Rogers & Kincaid, 1981 56 56

57. Graph of Time of Adoption by Network Threshold for One Korean Family Planning Community
100%
Threshold 0% 57
Time
1973
1963 57

58. Table: Adjusted Odds Ratios for the Likelihood of Low and High-threshold Adoption.
Cross-Sectional Data
(N=611)
Panel Data
(N=141)
Low
Threshold
High
Campaign Exposure
2.36**
1.92
1.71*
1.26
*p<.05; **p<.01
Controls for education, age, income, and number of children
Valente, T.W., & Saba, W. (1998). Mass media and interpersonal influence in a reproductive health communication campaign in Bolivia. Communication Research, 25, 58 58

59. Network Structure Network structure is partly defined by centrality.
Central members, popular students for example, both influence and are influenced by group norms
Central members can also contribute disproportionately to peer influence at micro level 59 59

60. Agent Based Models
Advent of computing has enabled scientists to generate hypothetical scenarios and model how people interact
Fundamental issue is:
Do assumptions match reality
Are the processes reasonable 60

61. First Contact Diffusion (Rumor)/Random Seeds61 61

62. Rate of Diffusion
Network Structure Real Rnd Cent Clustered Seeds Leaders Random Between Marginals 62

63. Simulated Network Structural Properties
Size
Density
Centralization
Clustering
Recip. %
Real
150
1.46%
3.28
15.4
43.0
Random
3.25
1.85
1.05
Centralized
7.67
1.63
1.01
Clustered
2.81
15.5
10.3 63