A multiplier bootstrap procedure for construction of likelihood-based condence sets is considered for nite samples and a possible model misspecication. Theoretical results justify the bootstrap consistency for a small or moderate sample size and allow to control the impact of the parameter...