Shrinkage and selection estimators are widely used to increase the accuracy of an estimator and to increase the interpretability of the estimated model by deleting irrelevant variables. However, both shrinkage and variable selection will decrease the accuracy when the shrinkage factor is too large (overshrinkage) or when relevant variables are deleted from a model (overselection). We propose a shrinkage and selection estimator that controls the amount of overshrinkage uniformly over the unknown parameter of interest. As a byproduct the estimator controls overselection at the same level. It is shown that this estimator has attractive properties both in terms of risk as in terms of interpretability. The estimator dominates the unbiased estimator in terms of risk when the number of parameters exceeds three. Furthermore, it is highly practical since it allows a researcher to choose the maximum rate of overselection.