I think a good horror game should include physical (in the virtual sense) and psychological horror, but again it's really hard to pull off.
I feel like a big problem with horror games stems from the fact that they're games; not because they aren't real, but because they can be MASTERED. Once someone essentially masters a game and/or its mechanics, its not scary anymore.
A game with all psychological horror and no physical will be very scary on its first run, but on a second playthrough the player knows that it's all air and no substance.
A game with specifically programmed physical horror can also be scary, but loses that horror when the player figures out how to deal with the threat in a non-risky way. Imagine a pack of lumbering shadowy forms vaguely similar to wolves; you have weapons but none seem to do much at all. They swarm and chase and the entire experience can be pretty frightening given the right atmosphere and effects. Now say you later tell the player "The weakpoint is the front foot". Next time you see them you put a bullet in one of the wolf's front paws and it dies instantly. Sure they can still attack but it becomes less scary and more of an action scene.
These wolves can also become less scary when instead of giving the player an easy way to dispatch of them, the developer uses them too much. At this point, the player becomes familiar with the enemy, its attack patterns, how to avoid it, etc. The threat is no longer "scary and mysterious" but "Known and some-what redundant". The solution to this also isn't just "Use the enemy much less" either, as it goes back to games losing their horror on multiple playthroughs. You know what is coming, you know what it does, go through the movements.
For a truly unique horror to stand out, aside from the obvious such as proper settings, atmosphere, characters and story, etc. It needs to have an element of uncertainty at several points in the game. For example, lets discuss Amnesia: Dark Descent when it was released. There were a few areas in Amnesia where monsters had a chance to spawn or not spawn in a certain area and even if the player died, they would see that the area the monster came from is open now. Monsters also traveled seemingly at random, meaning that players never knew which room and hallway were safe, especially when noises they made would change the monster's location.
But the most IMPORTANT thing horror games need to stand out is simpler than one would think: It's unique gameplay. As i said earlier, players will master games, but that also extends to those games controls and mechanics. Amnesia: Dark Descent was praised so highly because it's use of no-weapons, running, hiding, opening/closing mechanics, and don't look at the monster in a First-Person environment was pretty unique. If Outlast came out before Amnesia, it probably would have gotten the same amount of praise Amnesia did at its release. While I don't want to use it as a point, FNAF 1 still terrified people even without the jumpscares (the problem with FNAF is that it's easy to master).
TLDR:
If you want to make a good horror game, make something new and unique. Amnesia: Dark Descent style horror games aren't as scary as they normally would be when it's the 4th game you've played with the same mechanics.
Also give your game randomness and uncertainty if you want it to stay scary through multiple encounters and playthroughs.
