I'm currently working on an Econometrics project and I've come to a point where I've dropped a high leverage point as identified by cook's distance and a leverage plot (had observations that were twelve standard deviations away from mean and a nonsensical entry). My regression model before I dropped the high leverage outlier (Sample Size=408) had every regressor be statistically significant at the 95% confidence level. However, after I drop this one observation (New Sample Size=407), one of my variables becomes slightly statistically insignificant (New P=0.051 where Old P=0.045). What is the explanation for this?

  • 2
    $\begingroup$ I suggest you to read this paper $\endgroup$ – user603 Mar 4 '13 at 4:49

First, a change in p from 0.045 to 0.051 is essentially meaningless. It simply points out, yet again, how using a simple cutoff of p < .05 leads to silliness.

Second, a high leverage point forces the regression line to be near the point, even if it is a data entry error or an extremely odd case. Thus, it can radically change the regression line and, therefore, change all the associated statistics.

Third, since in your case the high leverage point was "nonsensical" your first regression is not to be taken seriously and should be ignored. If that high leverage point was "real" - that is, simply an unusual case - then there might be more to say.

  • $\begingroup$ A high leverage point is a point with high influence on the solution. Another way to look at it: It is a very informative data point (if correct ...), and so, dropping a highly informative point should be expected to change conclusions. $\endgroup$ – kjetil b halvorsen Apr 24 '16 at 16:28
  • 1
    $\begingroup$ Unless it is caused by a data error you should wonder why you are getting such an extreme point. Sometimes it reflects that a variable that should be in the model is not. Or something occurred at that point in your data that is unusual. Which might be worth considering (although it can be a pain to do so of course). :) $\endgroup$ – user54285 Apr 19 '19 at 23:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.