Tracking social contact networks with online respondent-driven detection: who recruits whom?

BMC Infectious Diseases

Table 2 Number of reported contact persons per participant per day by different characteristics and relative number of contacts from the Poisson Inverse-Gaussian Regression model

Category	Covariate	Number of participants	Mean (standard deviation) of number of reported contacts	Relative number of reported contacts (95 % CI)^a
Age of participant	0–39	268	20.98 (24.88)	1.00
	40–49	256	25.35 (37.24)	0.97 (0.80–1.17)
	50–64	656	19.94 (35.16)	0.93 (0.79–1.09)
	65+	379	14.19 (39.63)	0.69 (0.58–0.83)
Sex of participant	Female	1010	18.94 (30.78)	1.00
	Male	549	20.83 (42.41)	1.05 (0.94–1.18)
Household size	1	389	17.85 (29.49)	1.00
	2	648	15.73 (23.91)	1.02 (0.89–1.17)
	3	192	26.54 (58.17)	1.44 (1.20–1.73)
	4	218	24.93 (43.10)	1.55 (1.29–1.87)
	≥5	112	25.92 (37.37)	1.81 (1.43–2.29)
ILI	No	1519	19.93 (35.68)	1.00
	Yes	40	7.25 (9.70)	0.37 (0.25–0.53)
Days of the week	Sunday	224	16.68 (51.25)	1.00
	Monday	414	17.94 (32.15)	1.33 (1.12–1.59)
	Tuesday	249	24.27 (36.80)	1.84 (1.52–2.23)
	Wednesday	192	22.41 (31.73)	1.60 (1.30–1.96)
	Thursday	182	21.16 (28.29)	1.61 (1.31–1.99)
	Friday	117	18.76 (28.11)	1.42 (1.12–1.81)
	Saturday	181	16.65 (29.16)	1.27 (1.03–1.57)

^aDispersion parameter λ = 1.7 (95 % CI 1.4–2.1). The Poisson Inverse-Gaussian model is appropriate for modelling correlated counts with long sparse extended tails. The over-dispersion parameter in the model was significantly different from zero, indicating the necessity to use this model instead of a generalised Poisson model. Comparing AIC statistics, the Poisson Inverse-Gaussian model gave a better fit as opposed to a negative binomial model and a generalised Poisson model [22]

ISSN: 1471-2334