Social Networks and Health
Spreading of depressive symptoms over adolescent friendship networks.
Robert Eyre, Centre for Complexity Science Frances Griffiths, Warwick Medical School Thomas House, Centre for Complexity Science
University of Warwick, Coventry, United Kingdom, CV4 7AL
Background
Parametric Inference
Depression affects 350 million people worldwide1.
It is characterised by a set of symptoms which could be affected by social
networks.
We explore this possibility using a range of statistical methods.
Considered total CES-D score and 7 component symptoms - anhedonia, poor appetite,
poor concentration, dysphoria, helplessness, tiredness, and worthlessness.
Worsening:
(a)
(b)
Data
or
National Longitudinal Study of Adolescent Health (Add Health) - health
behaviours of United States adolescents in 1994-95 and 1996 2.
Includes friendship data - friendship network.
Includes Centre for Epidemiological Studies Depression (CES-D) scale
questions - quantised depression score3.
Improving:
(c)
(d)
Empirical Data Analysis
53
0
1
2
3
4
5
6
7
8
9
10
37
29
21
Parametric Inference - Example Results
(i)
37
45
53
0
6
Wave II
13
21
29
37
45
53
Wave II
0.08
0.06
0.02
0.0
0.04
Frequency
0
20
(iv)
(iii)
40
0.8
Difference between scores at wave II and wave I
Figure 1: (i) and (ii) - grid plots of CES-D scores at the first time point
(wave I) against the second time point (wave II). (i) - coloured by worse
Data
off friends. (ii) - coloured by better off friends. (iii) - Empirical distribution
of CES-D score change.
Empirical data shows an effect from higher and lower scoring
friends on the change in CES-D score.
The score changes appear to follow an exponential distribution.
Both transmit
Neither transmit
0.0
Parametric Inference - Poor Appetite
1.0
0
2
4
6
8
10
Number of lower scoring friends, k
Observed frequency
Neither transmit
Worsening transmits only
Improving transmits only
0
2
4
6
8
10
Number of higher scoring friends, k
Worsening transmits only
Improving transmits only
Observed frequency
Both transmit
Neither transmit
Worsening transmits only
Improving transmits only
0
2
4
6
8
10
Number of lower scoring friends, k
Figure 2: Results for Helplessness. Red - model 1. Blue - model 2. Green model 3. Purple - model 4. Black - observed frequencies. (i) and (ii) depend
on worse off friends, (iii) and (iv) on better off friends. Model 1 is preferred
for all symptoms except poor appetite.
Conclusions
0.0
0.0
Probability of worsening, pk
0.2
0.4
0.6
0.8
Probability of improving, qk
0.1 0.2 0.3 0.4 0.5 0.6
Both transmit
Observed frequency
0.0
−20
Probability of improving, qk
0.2
0.4
0.6
0.8
−40
0
2
4
6
8
10
Number of higher scoring friends, k
1.0
0
2
4
6
8
10
Number of higher scoring friends, k
Probability of worsening, pk
0.2
0.4
0.6
0.00
0.8
29
0.0
21
Probability of improving, qk
0.2
0.4
0.6
13
Probability of worsening, pk
0.2
0.4
0.6
0.8
6
(ii)
1.0
13
0
(iii)
Model 1 - (a) and (c). Model 2 - (b) and (d). Model 3 - (a) and (d).
Model 4 - (b) and (c).
Model parameters - inferred from data using maximum likelihood estimation.
Models were compared using their Akaike Information Criterion.
0
6
Wave I
53
29
21
13
6
0
Wave I
37
45
0
1
2
3
4
5
6
7
8
9
10
45
(ii)
(i)
or
0
2
4
6
8
10
Number of higher scoring friends, k
Figure 3: Results for Poor Appetite dependent on worse off friends. Model to
line correspondence is the same as figure 2. Unlike with all other symptoms,
model 4 is preferred.
Supported by the Engineering and Physical Sciences Research Council.
go.warwick.ac.uk/complexity/people/students/dtc/students2013/eyre/
email: r.w.eyre@warwick.ac.uk
For most symptoms, the emotional state of friends can have both a positive or
negative effect on the state of an individual.
This effect only occurs for smaller numbers of friends.
These symptoms reflect the total CES-D score.
Appetite is the only symptom to not follow this pattern.
Empirical findings suggest that the change in symptoms occurs progressively
between time points - more testing is needed.
[1] World Health Organisation. Depression - Fact Sheet No. 369. 2014. URL: http://www.who.int/mediacentre/factsheets/fs369/en/.
[2] E. Whitsel, J. Hussey, J. Tabor, P. Entzel, K. Harris, C. Halpern, and J. Udry. The National Longitudinal Study of Adolescent Health: Research Design.
2009. URL: http://www.cpc.unc.edu/projects/addhealth/design.
[3] L. S. Radloff. "The CES-D scale a self-report depression scale for research in the general population". Applied psychological measurement 1.3
(1977), pp. 385-401.