Generalized Additive Mixed Models Initial data-exploratory analysis using scatter plots indicated a non linear dependence of the response on predictor variables. To overcome these difficulties, Hastie and Tibshirani (1990) proposed generalized additive models (GAMs). GAMs are extensions of generalized linear models (GLMs) in which a link function describing the total explained variance is modeled as a sum of the covariates. The terms of the model can in this case be local smoothers or simple transformations with fixed degrees of freedom (e.g. Maunder and Punt 2004). In general the model has a structure of: Where and has an exponential family distribution. is a response variable, is a row for the model matrix for any strictly parametric model component, is the corresponding parameter vector, and the are smooth functions of the covariates, . In regression studies, the coefficients tend to be considered fixed. However, there are cases in which it makes sense to assume some random coefficients. These cases typically occur in situations where the main interest is to make inferences on the entire population, from which some levels are randomly sampled. Consequently, a model with both fixed and random effects (so called mixed effects models) would be more appropriate. In the present study, observations were collected from the same individuals over time. It is reasonable to assume that correlations exist among the observations from the same individual, so we utilized generalized additive mixed models (GAMM) to investigate the effects of covariates on movement probabilities. All the models had the probability of inter-island movement obtained from the BBMM as the dependent term, various covariates (SST, Month, Chlorophyll concentration, maturity stage, and wave energy) as fixed effects, and individual tagged sharks as the random effect. The GAMM used in this study had Gaussian error, identity link function and is given as: Where k = 1, …q is an unknown centered smooth function of the kth covariate and is a vector of random effects following All models were implemented using the mgcv (GAM) and the nlme (GAMM) packages in R (Wood 2006, R Development Core Team 2011). Spatially dependent or environmental data may be auto-correlated and using models that ignore this dependence can lead to inaccurate parameter estimates and inadequate quantification of uncertainty (Latimer et al., 2006). In the present GAMM models, we examined spatial autocorrelation among the chosen predictors by regressing the consecutive residuals against each other and testing for a significant slope. If there was auto-correlation, then there should be a linear relationship between consecutive residuals. The results of these regressions showed no auto-correlation among the predictors.