This paper presents the analysis of uncertainties on output parameters using a stochastic collocation method based on Lagrange polynomials. The method is applied to the case of a dipole illuminating a specific anthropomorphic mannequin. The uncertainties on the geometry dimensions of the model, using different bases of collocation points, like the sparse grid nodes constructed by the Smolyak algorithm or the tensor product of one-dimensional nodal sets, are studied. The Finite Difference Time Domain (FDTD) method is used to evaluate quantities such as absorbed power and specific absorption rate (SAR). The statistics of theses quantities are then efficiently obtained through the stochastic expansion of the field.