Are 1xxxx better than 2xxx or worse? what about 0xxxxx? Are 5xxxxx equivalent to 4xxxxx or will they be more desirable? Is 40xxx less desirable than 45xxxx because it’s closer to 2xxx? Paying attention to these minute differences is very short-sighted if you ask me. Any value in a particular cert is only relevant as long as the perception lasts. When we inevitably get to 7xxxx cert numbers are people going to consistently be able to sort 0xxx to 7xxx in order of quality? Is the first digit really an indication of anything? It’s such an arbitrary cutoff. I can tell you in the past people have even broken down the 2xxxx label into some that are better or worse, where are those people and ideas now? Forgotten.
I’ve looked at the numbers. It’s an undeniable fact that in terms of WOTC cards, PSA has graded less 10s when comparing the end of the 2xxx to the 4xxxx “eras”. Whether this is because they started grading harsher or the average quality of submitted card has fallen is totally up for debate. Keep in mind that 2xxx cards correspond to the time when the most boxes were being broken and graded. Even if we assume that the difference between 2xxx and 4xxx is entirely due to a change in PSA strictness, in the most extreme case you’re looking at a deviation of about 8 in 100 4xxxx cards getting a lower grade than the average baseline expectation. All factors ignored, the vast, vast majority of 2xxx and 4xxx cards are receiving the same grade as expected from the average expectation.
The most relevant thing to consider is even if there was definite proof PSA has suddenly become harsher, that is only established by averages. When you’re buying a single card, you’re not buying an average. From a statistics perspective, it’s like sampling a single data point from two near-completely overlapping Normal distributions. By random chance, (assuming there is a quantifiable difference between 2xxx and 4xxxx) it’s going to be a common outcome to see that the 2xxx is in better than the 4xxx when you’re only sampling from each distribution once and not working with averages (which is captured by @gottaketchumall 's point near the beginning of this thread).
Imagine a modified roulette wheel where 18 spots are red and 19 spots are black. Naturally, all things equal, black is a better bet. If you’re playing over and over and doing long averages, you will benefit from actually paying slightly more fore a black spot than a red spot. But if you’re playing one time and you’re paying 25% more from a black spot… well that’s basically this thread. Of course the caveat here for this analogy is that we are again assuming there is a slight difference between 4xxxx and 2xxx (which is my stance based on the data I’ve seen) and what we are actually caring about is condition (“winning in roulette”) and not what people are perceiving condition to be.
tldr. the real difference between 4xxx and 2xxx is trivial if it even exists. A single random sample doesn’t have to be concordant with the average, ESPECIALLY when the differences are so small. People won’t be talking about 4xxx as anything special as we cycle to the next arbitrary thing, just as 0xxx and 1xxxx and 2xxx have no hierarchy