Robust Wald-type tests for non-homogeneous observations based on the minimum density power divergence estimator (Q723446)

Let \(\{Y_1,Y_2,\ldots, Y_n\}\) be a sample of independent and possibly non identically distributed random variables. Let for each \(i\) absolutely continuous distribution function \(F_{i,\theta}\) of \(Y_i\) depends on parameter \(\theta\in \Theta\subset \mathbb{R}^p\). The minimum density power divergence estimator (MDPDE) \(\widehat{\theta}_\tau\) for non-homogeneous observations is the solution of the system of equations \[ \sum\limits_{i=1}^n \bigg( f_{i,\theta}^\tau(y_i)u_{i,\theta}(y_i)- \int f_{i,\theta}^{\tau+1}(y)u_{i,\theta}(y)\text{d}y \bigg)=0, \] where \(\tau>0\) is the turning parameter, \(\{y_1,y_2,\ldots,y_n\}\) is the realization of random variables \(\{Y_1,Y_2,\ldots, Y_n\}\), \(u_{i,\theta}(y)=\frac{\partial}{\partial y}\log f_{i,\theta}(y)\) and \(f_{i,\theta}\) is the probability density function of the distribution function \(F_{i,\theta}\). The authors of the paper introduce and study the corresponding MDPDE based the Wald-type tests for independent and possibly non-identically distributed data. The Wald-type test statistics are considered for testing simple and composite null hypothesis. The robustness is studied of the proposed Wald-type test statistics. The specific results for the generalized linear models with a fixed design are presented. The normal regression model and the Poisson regression model are considered in details.