I've been thinking a lot about cache difficulty lately (not terrain which is a different attribute) because the official guidance is too qualitative in my opinion – e.g. 'easy', 'relatively easy', 'mild', etc. – and therefore based on individual experiences. Since I’m an analytical data nerd at heart, I think a better way would be to quantify cache difficultly as a quotient of finds to total play logs (excluding reviewer activity, notes, enables/disables, maintenance, etc.) where total play logs equals Finds plus DNFs. In this way the difficulty rating would be a ratio representing how often the cache is (or isn't) found on average.
For example, there's a D-5 close to our house hidden back in 2015. There are 85 total play logs (39 Finds, 46 DNFs) as of today including our own DNF we logged a few days ago. Some quick division tells us this cache (on average) is found only 46% of the time and DNF'ed the other 54%. If a cache is DNF'ed more often than it's found, I think a D-5 rating is appropriate.
By contrast, we found a D-1.5 the other day in less than 5 minutes. To date this cache has a 96% found rate (331 total play logs) which makes it a better candidate for D-1 difficulty in my opinion.
To exclude error from new hides, caches would need to have a significant number of play logs (let's say 10, for example) before a difficulty rating would generate, but this might be viewed as unpopular by some cachers. My most recent hide (9 play logs) has a 56% found rate so far which is surprising to me (I only gave it a D-2), but the log sample is likely too small yet to get a reliable rating.
Some basic statistics would be necessary to plot the distribution of difficulty ratios across all caches and normalize the universal ratings scale. One might assume caches would have a normal distribution clustered around D-2.5, but the data may not actually support this, so an alternate model might be more appropriate to determine the most representative difficulty scale.
For the record, I am not a statistician by trade, but what I’m proposing here any Statistics 101 undergrad would make quick work of. By no means is this a perfect system because logs can be deleted, not recorded, etc., but the point is to make difficulty ratings data-based (within a margin of error of course) and objective. This would certainly be more reliable than asking CO’s to simply guess their hide difficulty which is entirely subjective as is the consensus here.
