Disclaimer: I work in academia
- Brian Nosek's team examined 100 studies from high-ranking psychology journals in 2015, and could only reproduce 1/3 of them.
- Tim Errington did the same for cancer papers, and could not reproduce most of them either (he spent 8 years for this efforts btw)
- When you aggregate the reported p-value in scientific publications, it often reveals a "funny" distribution (Leggett 2013, Ookubo 2016)
They are not picking up rare misconducts by low-profile researchers. Fraudant research (from p-hacking to data rigging) is very common and a very serious issue.
The biggest thing though is just this idea that a non-reproducing paper is a failure of science. Journal articles are the beginnings of conversation in a discipline, not the last word on it.
You can see what I mean, though: people who probably couldn't name 3 important researchers in a field see people working on replications in those fields (Nosek, Errington) as celebrities. Because reported failures to replicate are newsworthy, and the day-by-day grind of incremental findings and negative results aren't.