Penalty and non-penalty estimation strategies for linear and partially linear models