Jmeter extracts sub-samples from the log file and calculate metrics with these extra data points.

[JingshuXia]

Sub-samples or sub-requests should not be involved in stats calculations, it will impact the aggregations, percentiles results.
For the following case, request 'login' has two redirects "https://www.linkedin.com/uas/login-submit" and
"https://www.linkedin.com/nhome/". Naarad capture all three of them as separate requests and aggregates the metrcs, but actually the latter two shouldn't be involved in calculation as separate requests.