It was suggested there that the data may be subject to recall or measurement errors. Assuming that these recall errors are potentially concentrated among high counts, reestimate the Poisson and NB2 models after sequentially deleting counts greater than
(a) 15,
(b) 20, and
(c) 25.
Which parameters from the estimated model would you expect to be most sensitive to such deletion? Does your expectation match the observed outcome?