I doubt even in chess or any other game that uses elo or an elo-like system (e.g. MMR) is accurate without a ton of games played, and even people with thousands of games and a fairly established ELO rating will occasionally win games against those with higher ELO than them. Over 100 or so matches though, one would likely find that the higher rated person won more than 50% of their games.
Theres probably more variance in individual MMRs a 5v5 game, and the small differences can snowball a somewhat more here which make it sometimes take longer for people to find their 'true' mmr, but it does seem to work, as evidenced by the fact that diamond players will consistently poop on bronze players. Similarly long-term silver players will likely lose against long-time gold players, and so on with various other match ups of differing tier where the players have established themselves.
It can become more apparent that mmr is a good indicator of skill when in premades with people of completely different mmr. The matchmaking will try to find a fair match that'd ideally be people individually the same mmr as each person in your premade, but often ends up settling for a match where the average team mmr is roughly the same. This often results in the people with low MMR ending up getting destroyed, while the high mmr people will do quite well (e.g. winning lane or simply ending up with a much better kda than their lower mmr peers).
People are prone to thinking in relatively small sample sizes (as most people probably won't even play 10 games in a day, much less a statistically significant amount) and the games with feeders/AFKs stand out which makes them feel held back, but if you actually look at the statistics over a longer period in many cases those players actually do have a ~50% win rate or more, and so actually are trending upwards. (And of course in many more cases, players overestimate their own skill level, and so despite having a 50% win rate over a large number of games, they continue to blame their teammates.)
AFKs/trolls are somewhat rarer in higher MMRs (although still unfortunately a thing), but the key thing to remember is that there are 5 members on the opposing team, while only 4 members+you on your team. If you're not an AFKer/troll yourself, then that means the opposing team is more likely to have an AFKer/troll, so in the long run if it was only down to those, you'd consistently rise up in MMR, rather than trend down as many people seem to think. Thankfully there are matches where those sort don't feature at all, and so more interesting skill based competitions happen as well, and the result of those games really help sort out where you 'belong'. AFKs/trolls might make your mmr higher than it 'should' be in the long run, but in the big picture the contribution from afk/troll games compared to skill-testing games is going to be relatively small.
