Tid: 22 november 2017, kl 13-14
Plats: B705


Hierarchical generalized linear models (HGLM), with random effects, are widely used for modelling correlated response. Lee and Nelder (1996) proposed an adjusted profile likelihood estimation algorithm for HGLM, which works through iteratively fitting inter-connected generalized linear models, and is known to be computationally faster than other existing algorithms for this model class. Because spatial HGLMs require spatially correlated random effects, these models cannot be directly fitted with Lee and Nelder’s (1996) algorithm, which requires independent random effects. In this talk, I will show that with minor programming efforts, spatial HGLM can be fitted with existing algorithm for HGLM (Alam et al., 2015). I will discuss computational speed, and finite sample properties of these estimates using simulation studies. I will also present a real data example using reindeer pellet group counts data. The data were collected from a mountain area in the north of Sweden, where two wind farms were built during 2010-2011. Comparing the abundance of reindeer faecal droppings, between the periods of preconstruction, construction, and operation of the wind farms we find that wind farm (and possible human disturbances associated with them) significantly affect the reindeer habitat preference (Skarin and Alam, 2017).


Alam, M., Rönnegård, R. and Shen, X. (2015), Fitting conditional and simultaneous autoregressive spatial model in hglm, The R Journal 7, 5-18.

Lee, Y., and Nelder, J. A., (1996), Hierarchical generalized linear models, Journal of the Royal Statistical Society (B) 58, 619-678.

Skarin, A., and Alam, M. (2017), Reindeer habitat use in relation to two small farms, during preconstruction, construction, and operation, Ecology and Evolution 7, 3870-3882.