In those statistical roundups homicide is treated as a proxy for crime in general, so the best we can rigorously say is that homicide rates have decreased - which is, obviously, great. Researchers treat homicide as a proxy because they know not all crimes are reported.
Anecdotally, living in [big city] between 2014 and 2021 my street-parked car was broken into ~10 times, and stolen once (though I got it back). I never reported the break-ins, because [city PD] doesn't care. In [current suburb] a drive by shooting at the other end of our block received no police response at all, and won't be in the crime stats.
Are those types of crimes increasing? I don't know! I'd had my car broken into before 2014, and I witnessed (fortunately only aurally - I was just around the corner) a drive-by in the nineties. But... That's the point: no one knows! These incidents aren't captured in the statistics.
Personally, I think the proxies are broadly accurate, and crime in general is lower, and I shouldn't trust my anecdotal experiences. However, I think the general lack of trust in the quality of American police-work (much of it for good reason, sadly) biases most people towards trusting anecdotal experience and media-driven narratives.
I am more skeptical of homicide rate stats than you are, given the garbage data I see for crime in general, but even I am willing to admit they're much more robust than the rest.
I work with stats. I think even very honest people with high incentive to tell an accurate story and good data have trouble with stats. Now add politicians and police and bad data into that mix with winner-takes-all politics at stake and the stats get gamed.
Also I believe my eyes and when I see crimes happening in my neighborhood I don't rush to "the stats" to ask them what I saw.
But "what you saw" isn't necessarily representative of the state of things, either. Arlington, VA is (was?) one of the nicer places in VA; generally expensive, etc. When I drove through there, the van in front of me at a light was car-jacked, and the person in it chased down. I'm uncomfortable driving through Arlington because of that; even though it's not representative of the area. Admittedly, this was years ago... but the point stands. My experience is not representative of the actual facts.
Stats are also "not necessarily representative of the state of things". At the very best they are a single factoid about a very complex human existence.
Stats only get worse from there: at neutral they contain no information, at worst they are dis-info.
So we have stats, that's the closest we have to objective, but I guess we can't trust those. You say your anecdote contradicts "the stats", and I genuinely believe you. Sincerely, what's the alternative? Vibes? We gotta steer this ship (society) based on something.
How else do you condense down myriad and often conflicting datapoints of this complex human existence in order to get trends you can make decisions on?
Longer answer: this is a fundamental problem across many domains. I don't think anyone has solved it.
I think of a story of Bezos being told by his Amazon execs that customer support wait times were meeting X service levels. In the meeting room with his execs, Bezos dials up customer service, gets some wait time of >>>X and makes the point that service levels are not up to his expectations.
I don't think that story is a great analogy for running society but is interesting nonetheless.
