*Ecological Archives* E094-035-A3

Audrey Barker Plotkin, David Foster, Joel Carlson, Alison Magill. 2013. Survivors, not invaders, control forest development following simulated hurricane. *Ecology* 94:414–423. http://dx.doi.org/10.1890/12-0487.1

Appendix C. Additional statistical output for ANCOVA and breakpoint analyses.

Table C1. ANCOVA analysis summary table. The linear models were fit using the lm function in R 2.9.2, and are of the form response variable = β_{0} + β_{1} × year + β_{2} × treatment + β_{3} × year × treatment. Polynomial models were of the form response variable = β_{0} + β_{1} × year + β_{2} × treatment + β_{3} × year × treatment + β_{4} × year^{2} + β_{5} × treatment × year^{2}. For the polynomial functions, the 'year' variable was centered prior to analysis. Parameter estimates that are significantly different from 0 are shown in italics.

Basal area | Litterfall, 1991–1996 |
Litterfall, 1997–2010 |
Trees with basal sprouts |
Herb cover | Shrub cover* | |

Source of variation | ||||||

Year | F_{1,6} = 67.87P = 0.0002 |
F_{1,8} = 8.35P = 0.0202 |
F_{1,24} = 0.32P = 0.5754 |
F_{1,10} = 3.27P = 0.1007 |
F_{1,8} = 4.93P = 0.0571 |
F_{1,6} = 18.51P = 0.0051 |

Treatment | F_{1,6} = 462.58P < 0.0001 |
F_{1,8} = 75.13P < 0.0001 |
F_{1,24} = 0.94P = 0.3418 |
F_{1,10} = 17.40P = 0.0019 |
F_{1,8} = 0.36P = 0.5633 |
F_{1,6} = 6.22P = 0.0468 |

Year × Treatment | F_{1,6} = 6.70P = 0.0414 |
F_{1,8} = 16.48P = 0.0036 |
F_{1,24} = 0.38P = 0.5452 |
F_{1,10} = 12.40P = 0.0055 |
F_{1,8} = 2.82P = 0.1316 |
F_{1,6} = 10.47P = 0.0178 |

Year^{2} |
NA | NA | NA | F_{1,10} = 23.24P = 0.0007 |
NA | F_{1,6} = 10.34P = 0.0182 |

Year^{2} × Treatment |
NA | NA | NA | F_{1,10} = 2.53P = 0.1429 |
NA | F_{1,6} = 5.10P = 0.0647 |

Parameter estimates (SE) | ||||||

Intercept (β_{0}) |
0.976(0.036) |
1.099(0.050) |
1.064(0.145) |
268.142(22.262) |
0.858(0.169) |
2.338(0.288) |

Year (β_{1}) |
0.012(0.003) |
-0.011 (0.013) |
0.000 (0.010) |
13.218(3.139) |
-0.006 (0.015) |
0.140(0.028) |

Treatment (pulldown) (β _{2}) |
-0.655(0.051) |
-0.531(0.071) |
-0.064 (0.205) |
38.019 (31.484) |
0.406 (0.239) |
0.141 (0.407) |

Year × Treatment (pulldown) (β _{3}) |
0.011(0.004) |
0.074(0.018) |
0.009 (0.015) |
-15.597(4.439) |
-0.036 (0.021) |
-0.071 (0.039) |

Year^{2} (β_{4}) |
NA | NA | NA | -1.965(0.434) |
NA | -0.003 (0.005) |

Year^{2} × Treatment(pulldown) (β _{5}) |
NA | NA | NA | 0.975 (0.613) |
NA | -0.015 (0.007) |

*The polynomial shrub model was significantly better (*F*_{2,6} = 7.72, *P* = 0.0219) than a linear model.

Table C2. Breakpoint analysis for Litterfall. Breakpoint analysis was done using the breakpoint function in strucchange package (Zeileis et al. 2002). The same analysis for the control data showed no breakpoints. Optimal (m+1)-segment partition: Call: breakpoints.formula(formula = littere ~ 1), where 'littere' included all litter data from the experimental (pulldown) plot only.

Breakpoints at observation number: | Corresponding to breakdates: | ||||||||

m = 1: 6 | m = 1: 1996 | ||||||||

m = 2: 3, 9 | m = 2: 1993, 1999 | ||||||||

m = 3: 3, 9, 12 | m = 3: 1993, 1999, 2002 | ||||||||

m = 4: 3, 9, 12, 15 | m = 4: 1993, 1999, 2002 | ||||||||

RSS | 0.8620209 | 0.4165424 | 0.3098453 | 0.2720658 | 0.2458000 | 0.2313259 | |||

BIC | -0.1351537 | -8.6895142 | -8.6163513 | -5.2254757 | -1.2645268 | 3.5131322 | |||

Confidence intervals for breakpoints of optimal 2-segment partition: 2.5% breakpoints 97.5% | |||||||||

Breakpoints at observation number: 3, 6, 8 | Corresponding to breakdates: 1993, 1996, 1998 |

Literature Cited

Zeileis, A., F. Leish, K. Hornik and C. Kleiber. 2002. strucchange: an R package for testing for structural change in linear regression models. Journal of Statistical Software 7:1–38.