I just said so. Like in that very post?
The point-buy method is, in part, designed in order to approximately match the spread of using the advertised rolling method
as written. If you chop off the bottom third or quarter of it, you are very directly making PB-based characters weaker--or, if you prefer, making
all characters stronger (on average) since a rational player will choose to roll under such conditions.
E.g. let's say we switch from regular 4d6k3 to the same, but you reroll all 1s, meaning it's functionally 4d5k3+3. Not only does this raise the average by a meaningful degree, it also narrows the spread, making nearly all values (over three quarters!) 12 or higher. This is functionally equivalent to throwing out any array which generates a value lower than 6, and makes 6 extremely rare (0.16% of single rolls of 4d5k3+3.) Most stats will be 12, 13, or 14--but you have a very good chance of getting at least one 18, which, at least in 5e, isn't possible via point buy, because the highest you can buy is 15. In fact, you have more than a 1-in-5 chance of rolling at least one stat higher than you could possibly buy.
So...you'll rarely, if ever, have any downside from rolling--worst you're even
remotely likely to get is a single 7--but you get a whole bunch of upsides. That distorts the power curve of the game in many ways. Sure, the GM can try to compensate by (for example) just pretending that your level is 1 higher than it really is, or whatever, but that's a kludge, not a solution. With 5e already being...tenuously balanced at best, this could very easily result in tearing the whole thing wide open. One of the bigger problems of an already unbalanced game; there are degrees of imbalance just like balance, but it's far easier to accidentally make things much worse than it is to make things much better. Patterns are easy to break and hard to build from scratch, that sort of thing.