Introducing errors into the test data set