Your numbers are confusing to me. You retain 98% of the read pairs (both surviving) when filtering with either >q20 and >q30? When exactly do you keep only 30%? When you filter at q20?

From the title I guess it's

100% - filter at q20 - retain 30% pairs - subsequent filter at q30 - retain 98% pairs

100% - filter at q30 - retain 98% pairs

Have you checked the raw read quality statistics with FastQC or similar tools?

I tried to re-analysis with condition below

1. raw reads ---filter at q20 ---> -----filter at q30 ---> filtered reads
2. raw reads ---filter at q30 ---> filtered reads

Then, I got exactly same rate of both surviving.

In fact, I found my code was bad.

Sorry, Carambakaracho, arup and everyone.