Why need to combine the bins when do Chi-square test

Question

From Chi-Square Goodness-of-Fit test, It says

If some of the counts are less than five, you may need to combine some bins in the tails.

So how to understand this?

Answer 1

First, it is not about the counts but about the expected values for each cell. If many of them are small then the variance of the statistic may not be well-estimated so the results of the test may not be valid. What exactly counts for "many" and "small" is not entirely agreed upon.

Second, combining bins may not be the right solution even if many expected values are small. There are exact versions of the chi-square test that can be used.

Third, there may be some better method than chi-square, but, as you did not give context, it's hard to know.