Data S1

Question 1: Is the boldness of the single boldest member of the group (the keystone) associated with # of attackers that respond to simulated prey? Model Structure: GLMM, normal error distribution, identity link function. Response variable: Average number of attackers across three staged trials. Predictor Variables: keystone boldness, arid/wet site distinction, and their interaction term. Random effects: site ID

Effect Summary

Source / LogWorth / PValue /
Keystone Boldness / 9.407 / / 0.00000
Keystone Boldness *Arid or Wet site / 9.385 / / 0.00000
Arid or Wet site / 1.116 / / 0.07654

Summary of Fit

RSquare / 0.465393
RSquare Adj / 0.458626
Root Mean Square Error / 1.285881
Mean of Response / 3.08574
Observations (or Sum Wgts) / 241

Parameter Estimates

Term / Estimate / Std Error / DFDen / t Ratio / Prob>|t| /
Intercept / 1.7436143 / 0.355166 / 13.32 / 4.91 / 0.0003*
Arid_Wet[Arid] / 0.6201628 / 0.290176 / 5.988 / 2.14 / 0.0765
Keystone Boldness / 0.0038837 / 0.000594 / 231.9 / 6.54 / <.0001*
(Keystone Boldness -341.593)*Arid_Wet[Arid] / 0.0038783 / 0.000594 / 231.9 / 6.53 / <.0001*
Random Effect / Var Ratio / Var Component / Std Error / 95% Lower / 95% Upper / Wald p-Value / Pct of Total /
Site ID / 0.3738864 / 0.6182176 / 0.3893398 / -0.144874 / 1.3813096 / 0.1123 / 27.214
Residual / 1.6534906 / 0.1538581 / 1.3889824 / 2.0019055 / 72.786
Total / 2.2717081 / 0.4166959 / 1.6340297 / 3.3738367 / 100.000

-2 LogLikelihood = 848.54112299

Note: Total is the sum of the positive variance components.

Total including negative estimates = 2.2717081

Fixed Effect Tests

Source / Nparm / DF / DFDen / F Ratio / Prob > F /
Arid or Wet site / 1 / 1 / 5.988 / 4.5676 / 0.0765
Keystone Boldness / 1 / 1 / 231.9 / 42.7528 / <.0001*
Keystone Boldness *Arid or Wet site / 1 / 1 / 231.9 / 42.6326 / <.0001*

Q1: Arid sites only

Figure: Regression plots depicting the relationship between the boldness of the single boldest member of the group (x-axis) and the average number of attackers that respond to simulated prey (y-axis) for arid sites in aggregate (top panel) and for each site individually (bottom panel).

Summary of Fit

RSquare / 0.394294
RSquare Adj / 0.388981
Root Mean Square Error / 1.509835
Mean of Response / 3.775833
Observations (or Sum Wgts) / 116

Parameter Estimates

Term / Estimate / Std Error / DFDen / t Ratio / Prob>|t| /
Intercept / 1.0424515 / 0.56745 / 10.18 / 1.84 / 0.0955
Keystone Boldness / 0.0077515 / 0.001094 / 111.4 / 7.08 / <.0001*

REML Variance Component Estimates

Random Effect / Var Ratio / Var Component / Std Error / 95% Lower / 95% Upper / Wald p-Value / Pct of Total /
Site ID / 0.2673344 / 0.6094163 / 0.5638608 / -0.495731 / 1.7145632 / 0.2798 / 21.094
Residual / 2.2796029 / 0.3060172 / 1.7813014 / 3.0219344 / 78.906
Total / 2.8890192 / 0.6362135 / 1.9578994 / 4.6901996 / 100.000

-2 LogLikelihood = 443.17547084

Note: Total is the sum of the positive variance components.

Total including negative estimates = 2.8890192

Fixed Effect Tests

Source / Nparm / DF / DFDen / F Ratio / Prob > F /
Keystone Boldness / 1 / 1 / 111.4 / 50.1754 / <.0001*

Q1: Wet sites only

Leverage Plot

Figure: Regression plots depicting the relationship between the boldness of the single boldest member of the group (x-axis) and the average number of attackers that respond to simulated prey (y-axis) for wet sites in aggregate (top panel) and for each site individually (bottom panel).

Summary of Fit

RSquare / 0.322064
RSquare Adj / 0.316553
Root Mean Square Error / 1.03654
Mean of Response / 2.445333
Observations (or Sum Wgts) / 125

Parameter Estimates

Term / Estimate / Std Error / DFDen / t Ratio / Prob>|t| /
Intercept / 2.4513428 / 0.450501 / 4.556 / 5.44 / 0.0038*
Keystone Boldness / -3.539e-6 / 0.000595 / 120.5 / -0.01 / 0.9953

REML Variance Component Estimates

Random Effect / Var Ratio / Var Component / Std Error / 95% Lower / 95% Upper / Wald p-Value / Pct of Total /
Site ID / 0.5798506 / 0.6230007 / 0.5369725 / -0.429446 / 1.6754475 / 0.2460 / 36.703
Residual / 1.0744159 / 0.1387054 / 0.8470466 / 1.4079489 / 63.297
Total / 1.6974166 / 0.5534854 / 0.9793695 / 3.6375289 / 100.000

-2 LogLikelihood = 386.49688222

Note: Total is the sum of the positive variance components.

Total including negative estimates = 1.6974166

Fixed Effect Tests

Source / Nparm / DF / DFDen / F Ratio / Prob > F /
Keystone Boldness / 1 / 1 / 120.5 / 0.0000 / 0.9953

Question 2A: Is colony foraging aggressiveness (average # attackers) associated with persistence in situ?

Model Structure: GLMM, binomial error distribution, logit link function. Response variable: colony persistence (1/0). Predictor Variables: Average number of attackers across three staged trials, arid/wet site distinction, and their interaction term.

Whole Model Test

Model / -LogLikelihood / L-R ChiSquare / DF / ProbChiSq /
Difference / 32.8466439 / 65.6933 / 3 / <.0001*
Full / 134.149956
Reduced / 166.9966
Goodness Of Fit Statistic / ChiSquare / DF / ProbChiSq /
Pearson / 247.8279 / 237 / 0.3014
Deviance / 268.2999 / 237 / 0.0794
AICc /
276.4694

Effect Summary

Source / LogWorth / PValue /
Average # Attackers / 8.327 / / 0.00000
Average # Attackers*Arid_Wet / 4.846 / / 0.00001
Arid_Wet / 0.016 / / 0.96451 / ^

Effect Tests

Source / DF / L-R ChiSquare / ProbChiSq /
Arid_Wet / 1 / 0.0019802 / 0.9645
Average # Attackers / 1 / 34.30709 / <.0001*
Average # Attackers*Arid_Wet / 1 / 18.836747 / <.0001*

Parameter Estimates

Term / Estimate / Std Error / L-R ChiSquare / ProbChiSq / Lower CL / Upper CL /
Intercept / -2.2953226 / 0.4252576 / 39.270925 / <.0001* / -1.508585 / -3.1874977
Arid_Wet[Arid] / -0.0071864 / 0.161516 / 0.0019802 / 0.9645 / -0.308751 / 0.3269206
Average # Attackers / 0.684522 / 0.1337731 / 34.30709 / <.0001* / 0.965555 / 0.437222
(Average # Attackers-3.08574)* Arid_Wet[Arid] / 0.528898 / 0.1337731 / 18.836747 / <.0001* / 0.808943 / 0.279979

Q2A: Arid sites only

Model Structure: GLM, binomial error distribution, logit link function. Response variable: colony persistence (1/0). Predictor Variables: Average number of attackers across three staged trials, Site ID, # attackers * site ID. Site ID was designated as a random effect in the combined model because we were merely interested in contrasting wet vs. dry sites and were disinterested in the particulars of any given site. For the separate wet vs. dry models this designation was changed to a fixed effect because we were specifically interested in testing for differences in the relationship between # attackers and colony survival across sites.

Logistic Fit for Survival Arid Sites

Effect Summary

Source / LogWorth / PValue /
Average # Attackers / 17.424 / / 0.00000
Site ID / 1.832 / / 0.01471
Average # Attackers*Site ID / 1.286 / / 0.05177

Whole Model Test

Model / -LogLikelihood / DF / ChiSquare / ProbChiSq /
Difference / 38.446849 / 7 / 76.8937 / <.0001*
Full / 41.336423
Reduced / 79.783271
RSquare (U) / 0.4819
AICc / 100.019
BIC / 120.702
Observations (or Sum Wgts) / 116

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / -6.1798731 / 1.3648482 / 20.50 / <.0001*
Average # Attackers / 2.03512189 / 0.4566824 / 19.86 / <.0001*
Site ID[Boegoeberg] / -1.7362606 / 0.6295414 / 7.61 / 0.0058*
Site ID[Kalkrand] / -0.4402858 / 0.7341707 / 0.36 / 0.5487
Site ID[Rehoboth] / 1.35395271 / 1.0960651 / 1.53 / 0.2167
(Average # Attackers-3.77583)*Site ID[Boegoeberg] / -1.2632103 / 0.5068731 / 6.21 / 0.0127*
(Average # Attackers-3.77583)*Site ID[Kalkrand] / 0.03547491 / 0.6834384 / 0.00 / 0.9586
(Average # Attackers-3.77583)*Site ID[Rehoboth] / 1.07289067 / 1.0993798 / 0.95 / 0.3291

For log odds of 1/0

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
Average # Attackers / 1 / 1 / 75.4411321 / <.0001*
Site ID / 3 / 3 / 10.5071617 / 0.0147*
Average # Attackers*Site ID / 3 / 3 / 7.73718063 / 0.0518

Q2A: Wet sites only

Model Structure: GLM, binomial error distribution, logit link function. Response variable: colony persistence (1/0). Predictor Variables: Average number of attackers across three staged trials, Site ID, # attackers * site ID. Site ID was designated as a random effect in the combined model because we were merely interested in contrasting wet vs. dry sites and were disinterested in the particulars of any given site. For the separate wet vs. dry models this designation was changed to a fixed effect because we were specifically interested in testing for differences in the relationship between # attackers and colony survival across sites.

Nominal Logistic Fit for Survival Wet Sites

Source / LogWorth / PValue /
Average # Attackers*Site ID / 0.381 / / 0.41634
Site ID / 0.212 / / 0.61399 / ^
Average # Attackers / 0.067 / / 0.85681 / ^

Whole Model Test

Model / -LogLikelihood / DF / ChiSquare / ProbChiSq /
Difference / 2.424603 / 7 / 4.849207 / 0.6784
Full / 83.059204
Reduced / 85.483807
RSquare (U) / 0.0284
AICc / 183.36
BIC / 204.745
Observations (or Sum Wgts) / 125

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / -0.5419647 / 0.5202265 / 1.09 / 0.2975
Average # Attackers / 0.03956414 / 0.2191554 / 0.03 / 0.8567
Site ID[Ladysmith] / -0.0976133 / 0.4206221 / 0.05 / 0.8165
Site ID[Outjo] / 0.46738333 / 0.3583988 / 1.70 / 0.1922
Site ID[Rundu] / -0.2957814 / 0.4220266 / 0.49 / 0.4834
(Average # Attackers-2.44533)*Site ID[Ladysmith] / -0.2934648 / 0.436802 / 0.45 / 0.5017
(Average # Attackers-2.44533)*Site ID[Outjo] / 0.21042826 / 0.3904579 / 0.29 / 0.5899
(Average # Attackers-2.44533)*Site ID[Rundu] / 0.40479425 / 0.2873783 / 1.98 / 0.1590

For log odds of 1/0

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
Average # Attackers / 1 / 1 / 0.03255682 / 0.8568
Site ID / 3 / 3 / 1.80434864 / 0.6140
Average # Attackers*Site ID / 3 / 3 / 2.84378567 / 0.4163

Question 2B: Is colony foraging aggressiveness (average # attackers) associated with persistence in situ?

Model Structure: Logistic regression model for each site. Response variable: colony persistence (1/0). Predictor Variables: Average number of attackers across three staged trials.

Arid Sites Only — The association is always significant

Kalkrand:

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / 6.7541062 / 2.3837654 / 8.03 / 0.0046*
ColonyAgg / 2.0705968 / 0.7190677 / 8.29 / 0.0040*

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 23.5110378 / <.0001*

Rehoboth:

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / 8.87697676 / 4.1647738 / 4.54 / 0.0331*
ColonyAgg / 3.1080126 / 1.4142681 / 4.83 / 0.0280*

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 22.3907114 / <.0001*

Upington:

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / 5.94194716 / 2.1514318 / 7.63 / 0.0057*
ColonyAgg / 2.1899666 / 0.8502974 / 6.63 / 0.0100*

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 21.3227386 / <.0001*

Boegoeberg:

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / 3.14646227 / 1.4658218 / 4.61 / 0.0318*
ColonyAgg / 0.7719116 / 0.3110034 / 6.16 / 0.0131*

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 8.93916964 / 0.0028*

Wet Sites Only — The association is never significant

Ladysmith:


Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / -0.0780413 / 1.0234081 / 0.01 / 0.9392
ColonyAgg / 0.25390066 / 0.5343537 / 0.23 / 0.6347

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 0.22710561 / 0.6337

Outjo:

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / 0.58914861 / 1.0251724 / 0.33 / 0.5655
ColonyAgg / -0.2499924 / 0.4570083 / 0.30 / 0.5844

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 0.30263975 / 0.5822

Rundu:

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / 1.82760376 / 1.0462322 / 3.05 / 0.0807
ColonyAgg / -0.4443586 / 0.2628961 / 2.86 / 0.0910

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 3.19033264 / 0.0741

Weenen:

Parameter Estimates

Term / Estimate / Std Error / ChiSquare / ProbChiSq /
Intercept / -0.1708516 / 1.0664076 / 0.03 / 0.8727
ColonyAgg / 0.28219357 / 0.4527259 / 0.39 / 0.5331

For log odds of 0/1

Effect Likelihood Ratio Tests

Source / Nparm / DF / L-R ChiSquare / ProbChiSq /
ColonyAgg / 1 / 1 / 0.39922245 / 0.5275

Q2B: Are the colony aggressiveness vs. persistence relationships different between wet and arid sites?

Model Structure: Simple T-test

Data Matrix:

Site ID / wet_arid / AggVSurvival / AggVEgg
Kalkrand / arid / 2.07 / 0.37
Rehoboth / arid / 3.11 / 0.74
Upington / arid / 2.19 / 0.97
Boegoeberg / arid / 0.77 / 0.94
Ladysmith / wet / 0.25 / -0.51
Outjo / wet / -0.25 / 0.17
Rundu / wet / -0.44 / -0.07
Weenen / wet / 0.28 / 0.48

Colony Aggressiveness vs. Colony Survival