Skip to main content

Table 2 Model comparisons for logistic regression models. Alike models were compared by goodness of fit and performance. Note only models using similar response variables were compared and evaluation of two multinomial models can be found in supplementary materials

From: Associations between hair trace mineral concentrations and the occurrence of treponeme-associated hoof disease in elk (Cervus canadensis)

Model

AICca

AUCb

Sensitivity

Specificityc

A priori hypotheses

 A. Hoof health minerals

86.83

0.67

0.93

0.42

 B. Hoof health minerals with sex adjustment

88.06

0.67

0.93

0.42

 C. Hoof health minerals with “Longitude”c adjustment

88.46

0.74

0.76

0.65

 D. Toxicant hypothesis

80.51

0.65

0.76

0.56

 E. Toxicants with “Longitude” adjustment

81.73

0.70

0.90

0.53

Post hoc models

 H. Interaction of “Longitude” and Se

93.55

0.77

0.72

0.72

 I. Se with toxicants and “Longitude” adjustment

85.30

0.74

0.72

0.74

 J. Cr (as possible toxicant) exacerbating Se deficiency

93.39

0.70

0.76

0.58

 K. Co removed and substituted with Mn from Model C

86.49

0.74

0.69

0.70

  1. aAICc Akaike information criterion for small sample sizes
  2. bAUC Area under the receiver operating characteristic curve
  3. cLongitude category = “west” or “east” assigned from available location information (see Methods)