Jamie R. Crait and Merav Ben-David. 2007. Effects of river otter activity on terrestrial plants in trophically altered Yellowstone Lake. Ecology 88:1040–1052.

Appendix A. A description of model selection techniques used to predict occurrence of river otter latrine sites on Yellowstone Lake and tributary streams, and (1) a table of habitat characteristics measured for latrine and non-latrine sites; (2) a table of candidate logistic regression models used to predict occurrence of otter latrine sites; (3) a table of comparisons of logistic regression model selection for predicting occurrence of river otter latrine sites; (4) frequency distribution showing distances from all latrines to the nearest non-latrine site; (5) frequency distribution showing distances from latrines to the nearest non-latrine site where stable isotope samples were collected, and (6) literature cited.

We recorded habitat characteristics of 66 river otter latrine sites and 102 non-latrine sites on the shoreline of Yellowstone Lake and tributary streams. Each site was characterized with respect to topography, composition of terrestrial vegetation, and composition of river or lake substrates (Appendix A, Table A1). These variables were then introduced to a logistic regression equation (Hosmer and Lemeshow 2000) to develop a model that could be used to describe habitat selection by otters (Bowyer et al. 1995, Ben-David et al. 1996, Manly et al. 2002). Variables chosen for the models were based on previous studies of latrine site selection by river otters (Melquist and Hornocker 1983, Dubuc et al. 1990, Newman and Griffin 1994, Reid et al. 1994, Bowyer et al. 1995, 2003, Ben-David et al. 1996, Swimley et al. 1998). Selection of a "best" model for the stream and lake shorelines was based on percent correct classification (prediction success) of the site type and significant contribution (P < 0.05) of the variables (Appendix A, Table A2; Fielding and Bell 1997).

Variable selection for regression models based solely on statistical hypothesis testing has come under some criticism (Anderson et al. 2000, Burnham and Anderson 2002, Johnson and Omland 2004, Guthery et al. 2005, Stephens et al. 2005). To further evaluate the predictive strength of our selected models, AIC_c values were calculated, and models were ranked by ∆_i AIC_c (Appendix A, Table A3; Anderson et al. 2000, Burnham and Anderson 2002). Akaike weights (w_i) and evidence ratios were calculated to determine a relative strength for each model over other candidate models (Anderson et al. 2000, Burnham and Anderson 2002). Model rankings based on percent overall correct classification and AIC_c were then compared to evaluate the degree of agreement between the two model selection techniques (Appendix A, Table A3). Where these methods identified different "best" models (Appendix A, Table A3), we tested the performance of the top AIC_c and overall correct classification models using a two-fold cross validation (Hosmer and Lemeshow 2000) that we repeated five times with randomly selected data (Hosmer and Lemeshow 2000, Harrell 2001; Appendix A, Tables A2 and A3). The model with the highest prediction success in all iterations was chosen as the final model. Although correct classification (prediction success) is not generally considered a robust measure for model comparison, it is appropriate in studies where site classification or prediction of site types is the primary objective (Hosmer and Lemeshow 2000). As we were principally concerned with using these models to evaluate the effects of site characteristics and otter N deposition on the terrestrial plant community (described in text), we based our final model choices on prediction success and variable significance.

Classification results from the two logistic regression models (86.9% overall correct for the lake and 77.4% correct for tributaries) suggest that we may not have measured all appropriate variables for predicting latrine site selection in this system. For example, other studies documented the importance of elevation of sites above the water-line, water depth, and shoreline diversity (Dubuc et al. 1990, Bowyer et al. 1995, Swimley et al. 1998). Nevertheless, correct classification percentage in this study was comparable to other studies of otter latrine site selection (Dubuc et al. 1990, Bowyer et al. 1995, Ben-David et al. 1996, Swimley et al. 1998).

TABLE A1. Habitat characteristics measured for latrine and non-latrine sites on Yellowstone Lake and tributary streams, summer 2002 and 2003.

TABLE A2. Logistic regression models used to predict occurrence of river otter latrine sites on Yellowstone Lake and tributary streams, summer 2002 and 2003. Models are ranked in order of percent overall correct classification.

TABLE A3. Comparisons of logistic regression model selection for predicting occurrence of river otter latrine sites on Yellowstone Lake and tributary streams, summer 2002 and 2003. Comparisons are based on Akaike’s Information Criteria (AIC_c) and percent correct classification.

	Streams								Correct Classification (%)
Model	Model Parameters		log(L)	K	AICc	iAICc	wi		Non-latrine	Latrine	Overall	Rank
1	Brush, Fir, Forbs, Large Rock Substrate, Other Understory, Sand Substrate, Slope		-39.667	8	97.254	0.000	0.419		85.11	62.86	75.61	2
2	Brush, Fir, Forbs, Large Rock Substrate, Other Understory, Sand Substrate, Slope, Spruce > 60%		-38.674	9	97.780	0.526	0.322		82.98	62.86	74.39	3
3	Brush, Fir, Forbs, Large Rock Substrate, Other Understory, Sand Substrate, Shading		-40.515	8	98.950	1.696	0.179		79.59	65.71	73.81	4
4	Brush, Forbs, Large Rock Substrate, Other Understory, Sand Substrate, Shading		-42.836	7	101.145	3.891	0.060		83.67	68.57	77.38	1
5	Forbs, Large Rock Substrate, Other Understory, Sand Substrate, Shading		-45.133	6	103.356	6.102	0.020		83.67	60.00	73.81	4
	Lake								Correct Classification (%)
Model	Model Parameters		log(L)	K	AICc	iAICc	wi		Non-latrine	Latrine	Overall	Rank
1	Alder, Grasses, Lagoon, Rock Formation, Shading, Tributary Stream		-22.665	7	60.804	0.000	0.490		90.57	80.65	86.90	1
2	Alder, Grasses, Gravel Substrate, Lagoon, Rock Formation, Shading, Tributary Stream		-21.874	8	61.668	0.864	0.318		86.79	80.65	84.52	5
3	Grasses, Lagoon, Rock Formation, Shading, Tributary Stream		-24.795	6	62.681	1.876	0.192		88.68	80.65	85.71	3
4	Alder, Grasses, Gravel Substrate, Lagoon, Rock Formation, Spruce Overstory, Tributary Stream		-28.023	8	73.966	13.162	0.001		92.31	77.42	86.75	2
5	Brush, Grasses, Lagoon, Rock Formation, Spruce > 60%, Tributary Stream		-31.114	7	77.701	16.896	0.000		90.38	77.42	85.54	4

FIG. A1. Frequency distribution showing distances from latrines to the nearest non-latrine site. Distances shown are for all sites, summer 2002 and 2003. See Fig. 1 for distribution of latrine sites around Yellowstone Lake and tributary streams.

FIG. A2. Frequency distribution showing distances from latrines to the nearest non-latrine site.  Distances shown are between latrine and non-latrine sites where stable isotope samples were collected, summer 2002.

LITERATURE CITED

Anderson, D. R., K. P. Burnham, and W. L. Thompson. 2000. Null hypothesis testing: problems, prevalence, and an alternative. Journal of Wildlife Management 64:912–923.

Ben-David, M., T. R. Bowyer, and J. B. Faro. 1996. Niche separation by mink and river otters:  coexistence in a marine environment. Oikos 75:41–48.

Bowyer, R. T., G. M. Blundell, M. Ben-David, S. C. Jewett, T. A. Dean, and L. K. Duffy. 2003. Effects of the Exxon Valdez oil spill on river otters: Injury and recovery of a sentinel species. Wildlife Monographs:1–53.

Bowyer, R. T., J. W. Testa, and J. B. Faro. 1995. Habitat selection and home ranges of river otters in a marine environment:  effects of the Exxon Valdez oil spill. Journal of Mammalogy 76:1–11.

Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel inference:  a practical information-theoretic approach, Second edition. Springer, New York.

Dubuc, L. J., W. B. Krohn, and J. Owen, Ray B. 1990. Predicting occurrence of river otters by habitat on Mount Desert Island, Maine. Journal of Wildlife Management 54:594–599.

Fielding, A. H., and J. F. Bell. 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24:38–49.

Guthery, F. S., L. A. Brennan, M. J. Peterson, and J. J. Lusk. 2005. Information theory in wildlife science: Critique and viewpoint. Journal of Wildlife Management 69:457–465. 

Harrell, F. E. 2001. Regression modeling strategies: with applications to linear models, logistic regressions, and survival analysis. Springer-Verlag, New York.

Hosmer, D. W., and S. Lemeshow. 2000. Applied logistic regression, 2nd edition. John Wiley and Sons, Inc., New York.

Johnson, J. B., and K. S. Omland. 2004. Model selection in ecology and evolution. Trends in Ecology & Evolution 19:101–108.

Manly, B. F. J., L. L. McDonald, D. L. Thomas, T. L. McDonald, and W. P. Erickson. 2002. Resource selection by animals: Statistical design and analysis for field studies, 2nd edition. Kluwer Academic Publishers, Dordrecht, The Netherlands.

Melquist, W. E., and M. G. Hornocker. 1983. Ecology of river otters in west central Idaho. Wildlife Monographs 83:1–60.

Newman, D. G., and C. R. Griffin. 1994. Wetland use by river otters in Massachusetts. Journal of Wildlife Management 58:18–23.

Reid, D. G., T. E. Code, A. C. H. Reid, and S. M. Herrero. 1994. Spacing, movements, and habitat selection of the river otter in boreal Alberta. Canadian Journal of Zoology 72:1314–1324.

Stephens, P. A., S. W. Buskirk, G. D. Hayward, and C. Martínez del Rio. 2005. Information theory and hypothesis testing: a call for pluralism. Journal of Applied Ecology 42:4–12.

Swimley, T. J., T. L. Serfass, R. P. Brooks, and W. M. Tzilkowski. 1998. Predicting river otter latrine sites in Pennsylvania. Wildlife Society Bulletin 26:836–845.