SUPPLEMENTARY MATERIAL:

S1. Average network measures for Higgins Copse (n = 81) and Cammoor/Stimpsons Copse (n = 73). Ki = degree, Si = association strength, Ei = eigenfactor centrality, Ci = clustering coefficient, Bi = betweeness.

I / Ki / Si / Ei / Ci / Bi
Higgins Copse
Cammoor/Stimpsons Copse
/ 36
28
/ 0.97
0.62
/ 0.099
0.098
/ 0.748
0.756
/ 22.09
23

S2. Comparing the observed average of three centrality measures (weighted degree, betweeness centrality and eigenvector centrality) with the average network measures expected by chance for (a) Higgins Copse and (b) Cammoor/Stimpsons Copse. The raw data streams were randomized 1000 times by randomly redistributing individual identities. The filled circle shows the observed average centrality value, the clear circle and box show the average and 25% - 75% quartiles of the randomized data. The dotted lines show the 95% confidence interval of the randomized data.

(a)

b)

S3. Social network for Higgins Copse, contrasting the network positions of individuals with differing patch discovery success. Dark nodes represent individuals who found food patches; numbers inside nodes indicate how many patches it found. Increasing node size indicates increasing betweeness centrality, and edge (line) thickness indicates association strength. Edges have been thresholded to only show those above 0.02 in strength.

S4. Parameter estimates for the top four NBDA models, cumulatively totaling 0.83 in AICc weight for (a) Higgins Copse and 0.98 in cumulative AICc weight for (b) Cammoor/Stimpsons Copse. Unweighted model-averaged parameter estimates for both areas are listed in table (c).

(a)

Model 1
social + asocial
add., n-c / Model 2
social + asocial
multi., n-c / Model 3
asocial / Model 4
asocial
AICcWt / 0.37 / 0.30 / 0.08 / 0.08
Parameters:
Site / - / -1.411 / -0.392 / -0.306
Species / - / 0.012 / 0.292 / -
Sex / - / - / - / -
Age / - / - / - / -
Feeder use / - / - / 0.005 / 0.005
Model 1
social + asocial
multi., n-c / Model 2
social + asocial
multi., n-c / Model 3
social + asocial
multi., n-c / Model 4
social + asocial
multi., n-c
AICcWt / 0.89 / 0.04 / 0.03 / 0.02
Parameters:
Site / -0.725 / -0.810 / -0.610 / -0.812
Species / 0.504 / 0.294 / - / 0.299
Sex / - / - / - / 0.094
Age / 0.821 / 0.637 / 0.831 / 0.629
Feeder use / - / 0.001 / - / 0.001

(b)

(c)

Site / Species / Sex / Age / Feeder Use
Higgins Copse / -0.578 / 0.028 / 0 / 0 / 0.0008
Cammoor/Stimpsons Copse / -0.713 / 0.466 / 0.0019 / 0.794 / 0.0001

S5. Comparing the betweeness of individuals in Higgins Copse who discovered food patches (“yes”) and those that did not (“no”) with two levels of thresholding – “weak links”: where edge-weights above 0.3 in strength were removed from the analysis, and “strong links”: where edge-weights below 0.03 were removed from the analysis. Weak links seem to be more important for patch discovery than strong links, with discoverer’s having a higher betweeness once strong links are removed (*p=0.04), but with no difference in betweeness if weak links are removed (p=0.37).

S6. Summary of model selection statistics for a general linear model of patch discovery at (a) Higgins Copse (n=81, three trials) and (b) Cammoor/Stimpsons Copse (n=73, four trials); (sp) = species, (a) = age, (v) = feeder use, (se) = sex; (ec) = eigenvector centrality, (b) = betweeness

(a)

n / model / Δ AICc / AICcWt / cum.wt
1 / Pb / 0 / 0.13 / 0.13
2 / Pv / 1.01 / 0.08 / 0.21
3 / Pb + Pec / 1.47 / 0.06 / 0.27
4 / Pb + Pse / 1.62 / 0.06 / 0.33
5 / Pa + Pb / 1.81 / 0.05 / 0.39
6 / Pb + Pv / 2.05 / 0.05 / 0.43
7 / Pv + Pse / 2.40 / 0.04 / 0.47
8 / Pa + Pv / 2.70 / 0.04 / 0.51
9 / Pec / 2.94 / 0.03 / 0.54
10 / Pa + Pec + Pb / 3.09 / 0.03 / 0.56
11 / Pec + Pb + Pv / 3.13 / 0.03 / 0.59
12 / Pec + Pb + Pse / 3.35 / 0.03 / 0.62
13 / Pa + Pb + Pse / 3.38 / 0.02 / 0.64
14 / Psp + Pb / 3.48 / 0.02 / 0.66
15 / Pb + Pv + Pse / 3.53 / 0.02 / 0.71
16 / Pse / 3.66 / 0.02 / 0.73
17 / Pa + Pb + Pv / 3.69 / 0.02 / 0.75

(b)

n / model / Δ AICc / AICcWt / cum.wt
1 / Pec / 0 / 0.21 / 0.21
2 / Pa + Pec / 1.35 / 0.11 / 0.32
3 / Pec + Pb / 1.89 / 0.08 / 0.40
4 / Pec + Pv / 2.06 / 0.07 / 0.47
5 / Pec + Pse / 2.17 / 0.07 / 0.54
6 / Pv / 3.05 / 0.05 / 0.59
7 / Pa + Pec + Pb / 3.26 / 0.04 / 0.63
8 / Pa + Pec + Pv / 3.43 / 0.04 / 0.67
9 / Pa + Pec + Psex / 3.51 / 0.04 / 0.70
10 / Pec + Pb + Pv / 4.11 / 0.03 / 0.73
11 / Pec + Pb + Pse / 4.15 / 0.03 / 0.75
12 / Pec + Pv + Pse / 4.30 / 0.02 / 0.78
13 / Pa + Pv / 4.40 / 0.02 / 0.80
14 / Pec + Psp / 4.42 / 0.02 / 0.82
15 / Pb + Pv / 4.72 / 0.02 / 0.84
16 / Pse + Pv / 5.17 / 0.02 / 0.86
17 / Pa + Pb + Pv + Pec / 5.53 / 0.01 / 0.87

S7. Summary of the total Akaike weights in both areas across the NBDA Model: additive vs multiplicative (whether individual level variables affect only asocial learning or whether they affect social transmission also); and a baseline rate of transmission which is either constant or non-constant (an unchanging rate of acquisition or a declining/increasing rate of asocial acquisition). (a) Higgins Copse; (b) Cammoor/Stimpsons Copse.

a)

Social + Asocial: Asocial:
Multiplicative / Additive
Baseline rate:
Constant / 0 / 0 / 0
Non-constant / 0.99 / 0.1 / 0

b)

Social + Asocial: Asocial:
Multiplicative / Additive
Baseline rate:
Constant / 0 / 0.03 / 0.24
Non-constant / 0.33 / 0.37 / 0.03