=+73. A kernel function is fit to the data in a sample of size n. Later, a researcher realizes that the largest observation in the sample was actually a typographical error and, because the original lab data no longer exists, this data point is removed from the sample, leaving a sample of n21 measurements.

The researcher wants to fit a new kernel function to the reduced sample of n21 data points. To produce a graph that has about the same smoothness as the original kernel function, will the value of 

have to be raised or lowered?