What is the correlation with accuracy and ELO?

It is partially, but rating comes into play as well 

https://support.chess.com/article/1135-what-is-accuracy-in-analysis-how-is-it-measured

So to summarize this, I would say that accuracy and ELO are totally unrelated, accuracy depends almost always on whether your opponent is aggressive or defensive.  Chess.com has an ELO explanation: Elo Rating System - Chess Terms - Chess.com.  Does this help??

Interesting to note that in IM Daniel Rensch's video 0 of his opponent's moves were shown as inaccurate or worse, but his opponent's CAPS score was only 98.3, not 100. So if CAPS is a percentage we still don't know what it's a percentage of.

The CAPS2 documentation makes it clear that the CAPS2 score is definitely not a percentage of anything, so it would probably be a good idea if people stopped putting "%" signs every time they refer to it.

There is a reasonable objective meaning to percentage accuracy, namely 100 x the number of accurate moves / the total number of moves,  where an accurate move in a theoretically winning position is one that reduces the number of ply (half moves) to mate by 1 (whichever side is considered) and an accurate move in a theoretically drawn position is one that doesn't result in a theoretically winning position (necessarily for the opponent).

The problem is that nobody including the engines can reliably tell what those moves are unless they're covered by a tablebase and they've only recently started making the relevant tablebases with the 50 move rule taken into account. In any case they may not be reliable either after you've made a few moves they don't approve of because they still won't take the triple repetition rule into account.

So exactly how your CAPS2 scores relate to actual accuracy is anybody's guess. If you give chess.com analysis a sequence that is known to be 100% accurate taken from a Nalimov tablebase, Coach may very well have lots of nasty comments to make about it.

And if you're losing it's theoretically impossible to blunder. But accuracy is also not necessarily desirable in practice if you're losing, You would rather swindle a draw (or win) than achieve a perfectly accurate loss. 

Agreed, In my opinion accuracy is overated, what matters are the wins.

I am in the mid-high 400s, and I play around 60-70 accuracy each game. Is that normal for my rating?

This is Rybka/Nalimov tablebase playing itself. Every move by both sides is not just perfect but also perfectly accurate.

So if you get 67 as Black that must be pretty good.

Since most of the replies were just excuses for why they repliers cannot or don't want to answer the OP's question i found a blog that actually adresses this topic very well statistically:
https://www.chess.com/blog/hissha/accuracy-and-ratings-on-chess-com

is there a calculator to calculate estimated elo compared to the accuracy in a single match?

Even if there is, it must be really bad. That correlation is much weaker than you imagine, mostly because your accuracy greatly depends on your opponent's moves. Do they make obvious mistakes? Or do they put pressure on you?

If a guy rated 1500 plays against a beginner, he takes all the pieces and looks like Paul Morphy.

If the same 1500 guy plays against a master, he looks like he can barely play chess.

Chess.com developed something similar (estimating the players' rating based on the game), and it's ridiculous.

Hello everyone,

The relationship between Elo rating and move accuracy has been a topic of much discussion. To shed light on this, I analyzed 35,000 blitz games (10+0) from Lichess, evenly distributed across ratings from 400 to 2600. Using Stockfish 17.1, I measured the accuracy of each move.​

Key findings:

• There's a clear, continuous correlation between Elo rating and move accuracy.

• Even when players of similar ratings (within 50 Elo points) face off, higher-rated players consistently make more accurate moves.​

These results suggest that higher-rated players not only understand the game better but also maintain higher precision, even in evenly matched games.​

If you're interested in the detailed analysis and methodology, feel free to read the full article:

👉 Do Stronger Players Play More Accurately? I Analyzed 35,000 Games to Find Out

Looking forward to your thoughts and discussions!

Did you use lichess's definition of "accuracy", or chess.com's, or your own?

Do you even realize that these are not necessarily the same?

The accuracy was calculated the same way as on Lichess and Chess.com — by measuring the difference between each move and the best move suggested by Stockfish.

However, to ensure a fair comparison and eliminate any discrepancies, I recalculated the accuracy locally on my PC under identical conditions for all the games. That way, we’re comparing apples to apples.

This is not true on chess.com. It used to be, but they changed the formula some years ago to give higher accuracy to low rated players.

Ok Good to know !