Validation of an automated data collection method for quantifying social networks in collective
behaviours
By Fumiaki Y. Nomano, Lucy E. Browning, Shinichi Nakagawa, Simon C. Griffith, and Andrew F. Russell
Corresponding author: Fumiaki Y. Nomano; email: nomanof@ees.hokudai.ac.jp
Behavioral Ecology and Sociobiology
Table S1 Summary of the parameters for the generalized linear mixed model (GLMM) for stay
duration of individuals across different status (see also Fig. 1a-b).
95% CI
Term
posterior
2.5%
97.5%
5.095
4.902
5.306
Bm-Bf
-2.456
-2.758
-2.172
Hfo-Bf
-2.206
-2.705
-1.675
Hfy-Bf
-2.229
-2.749
-1.731
Hmo-Bf
-2.410
-2.711
-2.151
Hmy-Bf
-2.524
-2.883
-2.139
-0.040
-0.063
-0.012
Bm-Bf
0.082
0.049
0.117
Hfo-Bf
0.015
-0.075
0.111
Hfy-Bf
-0.007
-0.084
0.066
Hmo-Bf
0.101
0.065
0.139
Hmy-Bf
0.115
0.059
0.165
0.187
0.121
0.280
(Intercept)
mean
Status
Brood age
Status × Brood age
SD of individuals
Fixed effects are considered significant if the 95% CI’s of the parameters do not include zero. Brood
ID was removed from the random effects because the estimated variance was close to zero when
included. The effects of individual status were examined by setting the breeding female (B f) as a
reference category. Keys used for other categories are B: breeder, H: helper, f: female, m: male, o:
adult, y: yearling and juvenile. The model was fitted using MCMCglmm (Hadfield 2010) in R
version 2.15.2 (www.r-project.org).
S1 Distributions of stay duration matched well between (c) nest visits in nest video data and (d) nest
visits estimated from PIT-tag data recording visits by the same individuals in the same time frame as
the nest video data. Vertical dashed lines show median values. Data region for long stay duration in
the video is not shown since stay longer than 200 seconds was only 1.7%. Visits of breeding females
are not included.
S2 Relationships between synchrony count per dyad and three possible denominators of synchrony
frequency index for nest-video with breeding females, and PIT-tag. min(Ni, Nj), was best in
predicting the synchrony count.
S3 Examples of synchronous nest visits observed by nest-videos. Horizontal lines correspond to visit
record of individuals. Each horizontal black bar represents time when each individual stayed inside
the nest (from arrival time to exit time). Each shaded area corresponds to a single synchrony bout,
which starts with arrival time of first birds in the synchrony and ends with the arrival time of the last
bird. Labels on the y-axis are h: helper, B: breeder, f: female, m: male. Each panel is from three
different breeding attempts: (a) cek0702, (b) efg0702, (c) efp0801. Occurrence of synchrony was
highly variable across breeding units and time.
S4 An example of sub-grouping of individuals in synchronous visits over time on one whole day of
bfd0801 (PIT-tag data). The synchrony events are presented in the order of occurrence (rows). Filled
cells indicate synchrony bouts which individuals (columns) were observed to take part in. Individuals
have large variation in synchrony rate. Data for the breeding female is not shown due to the problem
in the visit reconstruction by PIT-tag (see texts).
S5 Histograms of synchrony frequency index for (a) nest-video with breeding females, (b) nest-video
without breeding females, and (c) PIT-tag. Full data was used for each.
S6 Comparisons of the total number of synchronous visits per dyad between different monitoring
methods. PIT-tag data tend to overestimate it. Data are shown as either point or line linked to point.
Where more than one data point overlap on the same value, lines representing such additional data
points were added around the point. N = 119 dyads in all cases.
S7 Comparisons of network indices (two descriptive network indices for each individual: node
strength (a-c) and Holme’s weighted clustering coefficient (d-f), and one model-based estimator of
individual connectedness, ‘sociality’ (g-i)) between different monitoring methods. Only nest visits
that occurred in the time frame where both nest-videos and PIT-tag readers were recording were used.
Spearman’s correlation coefficients are shown for each comparison. Note that values for dyad
involving breeding females are excluded.
4
4
3
2
0
1.0
r s 0.79
0.4
0.6
0.2
0.0
1.0
0.8
0.4
0.6
0.2
0.4
0.6
0.8
0.6
1.0
0.8
0.6
0.4
-0.2
0.0
r s 0.944
p 0.001
N 47
-0.4
-0.2
0.0
Sociality i
video with breeding female
0.2
i
Sociality
PIT-tag
0.4
0.0
0.2
i
Sociality
PIT-tag
r s 0.922
p 0.001
N 47
-0.4
-0.2
0.6
(i)
0.6
0.6
0.0
-0.2
-0.4
-0.4
0.4
Clustering coefficient
video without breeding female
(h)
r s 0.962
p 0.001
N 47
0.2
0.0
1.0
0.8
Clustering coefficient
video with breeding female
Clustering coefficient
video with breeding female
(g)
47
0.0
0.0
0.4
0.001
N
0.4
47
p
0.2
0.001
0.6
N
Clustering coefficient
PIT-tag
0.8
p
0.2
Clustering coefficient
PIT-tag
0.8
0.4
0.2
0.0
0.2
0.0
0.4
1
3
0
1.0
1.0
47
0.6
N
0.2
(f)
r s 0.679
0.879
p 0.001
4
3
2
Node strength
video without breeding female
(e)
rs
47
1
0
Node strength
video with breeding female
(d)
0.001
p
N
4
3
2
1
0
4
3
2
1
Node strength
video with breeding female
Clustering coefficient
video without breeding female
47
2
N
1
Node strength
PIT-tag
1
2
47
Node strength
PIT-tag
4
p 0.001
3
rs
0.001
0
Node strength
video without breeding female
0.932
p
N
r s 0.964
0.882
rs
0
Sociality i
video without breeding female
(c)
(b)
(a)
-0.4
-0.2
0.0
0.2
0.4
0.6
Sociality i
video with breeding female
-0.4
-0.2
0.0
0.2
0.4
0.6
Sociality i
video without breeding female
Choice of offset term
For the denominator of synchrony frequency index, we examined (i) total number of nest visits per
dyad: Ni + Nj − Yi,j (Vance et al. 2009), (ii) product of individual visit counts: Ni Nj (Krivitsky et al.
2009), (iii) smaller individual visit count in dyad i, j: min(Ni, Nj), where Ni, is visit count of
individual i and Yi,j is the number of synchrony for the dyad i, j. min(Ni, Nj) corresponds to the largest
possible number of synchronous visits per dyad given the individual visit counts Ni and Nj. Although
all three variables were expected to be correlated with the synchrony count per dyad, the degree of
correlation may be different depending on the structure of the data (Whitehead 2008). These offset
variables, Ni + Nj − Yi,j and Ni Nj have been used in the same class of models in other studies (Vance
et al. 2009; Krivitsky et al. 2009), while association index similar to Yi,j /min(Ni, Nj) is recommended
when there is large variation in individual detection in more general analyses (Whitehead 2008).
Relationships between these three types of offset variables and the dyadic synchrony count were
examined. We chose one of the three variables that was best correlated with the synchrony count as a
denominator term of the synchrony frequency index.
As expected, the synchrony count per dyad and the three measures of individual visit counts
per dyad (Ni + Nj − Yi,j, Ni Nj, and min(Ni, Nj)) were correlated, and dyads with individuals showing
high nest visit rate had high synchrony count. Yet, the strength of correlations was different across
these measures. The relationship of synchrony count and the total number of visits per dyad, Ni + Nj
− Yi,j, was proportional for some dyads, but the synchrony count was very low for some dyads even
though they had high total visit counts, Ni + Nj − Yi,j. This resulted in two discrete regions of data
mass in the correlation plot especially for PIT-tag data where the visit counts were more variable and
asymmetric within dyad (Fig. S2, nest-video, spearman correlation coefficient: rs = 0.545, N = 221, p
< 0.001; PIT-tag, rs = 0.673, N = 634, p < 0.001). The product of visit counts Ni Nj was also
correlated with the synchrony count. But the correlation became weak for data with a larger number
of nest visits (Fig. S2, nest video, rs = 0.734, p < 0.001, N = 221; PIT-tag, rs = 0.942, N = 634, p <
0.001). Finally, min(Ni, Nj), the visit counts for individuals that had smaller counts compared to the
other in the dyad, was linearly correlated to the synchrony counts in both monitoring methods (nestvideo: spearman correlation coefficient = 0.745, p < 0.001, N = 221; PIT-tag: spearman correlation
coefficient = 0.958, N = 634, p < 0.001, S1). This result is consistent with the fact that the upper limit
of synchronous visit count per dyad is determined by min(Ni, Nj). Thus, we chose min(Ni, Nj) for the
denominator and defined the ratio Yi,j /min(Ni, Nj) as the synchrony frequency index in this paper.
This index dictates the proportion of synchronous visits that actually occurred to the number of
possible synchronous visit given the nest visits of individuals in the dyad. The synchrony frequency
based on other two indices examined here should be appropriate when the nest visits of individuals
are more symmetric as in species of biparental care (Mariette & Griffith 2012).
BUGS code for the GLMM with ‘sociality’ random effect
###########
model {
for (i in 1:N.d){
sync[i] ~ dbin(p[i],count[i])
# count[i] is denominator of synchrony frequency
log(p[i]) <- b.0
+ r.ring[ring1[i]] + r.ring[ring2[i]] #sociality
+ r.nest[nest[i]]
+ r.dyad[dyad[i]]
}
#random effect prior
##sociality (individual ID)
for(k in 1:N.ring){
r.ring[k] ~ dnorm(0, tau.r)
}
tau.r <- pow(sigma.r, -2)
sigma.r ~ dunif(0, 100)
##brood ID
for(k in 1:N.nest){
r.nest[k] ~ dnorm(0, tau.n)
}
tau.n <- pow(sigma.n, -2)
sigma.n ~ dunif(0, 100)
##dyad ID
for(k in 1:N.dyad){
r.dyad[k] ~ dnorm(0, tau.d)
}
tau.d <- pow(sigma.d, -2)
sigma.d ~ dunif(0, 100)
b.0 ~ dnorm(0, 0.0001)
}
############
References
Hadfield JD 2010 MCMC methods for multi-response generalized linear mixed models: the
MCMCglmm R Package. J Stat Softw 33:1–22.
Krivitsky PN, Handcock MS, Raftery AE, Hoff PD 2009 Representing degree distributions,
clustering, and homophily in social networks with latent cluster random effects models. Soc
Networks 31:204–213
Mariette MM, Griffith SC 2012 Nest visit synchrony is high and correlates with reproductive success
in the wild zebra finch Taeniopygia guttata. J Avian Biol 43:131–140
Vance EA, Archie EA, Moss CJ 2009 Social networks in African elephants. Comput Math Organ Th
15: 273–293
Whitehead H 2008 Analyzing Animal Societies: Quantitative Methods for Vertebrate Social Analysis.
University of Chicago Press, Chicago