Why reduce input records is greater than map output records?

I ran some Job on Hadoop and puzzled by this question. I am attaching a screenshot from jobtracker. In addition, the difference varies from run to run and is sometimes absent.

Combiner is not used in this case