The further the expected values are far from the observed values, the more likely that there really is a significant difference.

Example 2:

Supoose we have got chi square for 1 degree freedom data: 1.86

1.86 < 3.841 (therefore, we cannot reject our null hypothesis that the distribution of our data is same)

Link:

https://www.youtube.com/watch?v=VskmMgXmkMQ