2
$\begingroup$

Do outliers lead chi-squared tests of goodness of fit to fail?

$\endgroup$
2
$\begingroup$

The answer has to depend on what you are testing and how you implement a test as well as the data you have.

  • If one bin containing an outlier has observed frequency 1 and expected frequency almost zero, its contribution to a chi-squared test statistic may be enormous. (The usual advice is naturally: Don't do that! But it's among the possibilities.)

  • If you have bins with equal expected frequency, then sometimes a chi-squared test will not even notice an outlier; it may be one of several values in an open-ended bin for the highest values (or the lowest values, as the case may be).

This arbitrariness is one reason why chi-squared tests of goodness of fit are not often seen outside elementary texts and courses.

I exaggerate a bit, but in most problems better methods are available. For example, chi-square testing of normality is rarely useful compared with customised plots or even customised tests that do not entail arbitrary binning.

| cite | improve this answer | |
$\endgroup$

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.