0

Mitigating problem when cell count $<5$ or almost zero
a. 2x2 table with cell count zero
b. When event rate is small in logistic regression: such that when seen further by an independent categorical variable it leads cell counts less than 5 or almost zero.

Are these options?

One is to take this variable out of the regression model. It might not be a good option, but it could help in verifying the problem. The other option is to collapse across some of the categories to increase the cell size.

kjetil b halvorsen
  • 77,844
SR1
  • 31
  • Cell counts per se are not a problem at all. Are you perhaps trying to ask about conducting chi-squared tests when expected counts are small? Please clarify. – whuber Jul 27 '22 at 15:11
  • Thanks @ whuber for your reply. I understand when cell count is less than one should go fisher excat test instead of chi square. What if a) cell count is close to zero. Another scenario while doing regression or finding odds ratio between categorical variables, there are chances that cell count in some of the cells can be close to zero. In that case it’s not possible depict odds ratio. How to mitigate this problem? Should one collapse the categories - that is merging subcategories ? – SR1 Jul 27 '22 at 17:49
  • That understanding is incorrect. Even zero counts are not necessarily a problem. Once again: you need to take special care in count models, if at all, only when the expected counts are small. Merging the categories is problematic, because it can lead to incorrect p-values. – whuber Jul 27 '22 at 17:55
  • Thanks. I observed if cell counts are near to zero, stata does not consider subcategory and eliminates it from analysis. What sometimes you need that variable? If we can’t collapse the subcategories under that variable- what’s the best approach ? – SR1 Jul 27 '22 at 18:04
  • I'm sure Stata does not behave like that. What the best approach might be depends on the specific analysis you are performing and on why you're doing it. – whuber Jul 27 '22 at 18:05
  • Thanks for your reply. I am working on logistic regression analysis. – SR1 Jul 27 '22 at 18:07
  • The details matter. Please edit your post to describe your analysis. – whuber Jul 27 '22 at 18:11
  • Let us continue this discussion in chat. – SR1 Jul 27 '22 at 23:34

0 Answers0