Fuck stats,use your eyes.
Stats are a numerical representation of "eye-test". They aren't competing evaluation mechanisms.
For example, if you watched Brighton last season, you would think "jesus they create so many chances that they should win more matches, but the finishing is non existent". They scored 40 goals, whereas they had an xG of 51. xG is the measure of expected goals scored based on the type/location/quality of shots taken. In this case, xG suggests that they should have scored 51 goals with average finishing, but their forwards were not clinical and so they under-performed by 10 goals. That passes the eye test doesn't it? It's completely grounded in reality.
But why look at Brighton, let's bring the discussion closer home. You know based on eye-test, there's a common belief that we don't concede a lot of shots, but the ones we do end up straight in the net. It's extremely frustrating and people blame Ederson a little, but often we look at those shots and think "well what could he even do?!". Let's see what the stats have to say about that.
City had a (PSxG - GA)/90 of 0.03. That's quite a mouthful. PSxG is expected goals based on high likely the goalkeeper is to save the shot. GA is goals allowed. So (PSxG - GA) is measuring "what's the difference between how many goals a goalkeeper
should have conceded and how many he
actually did". Dividing by 90 just gives us per match number. A value of 0.03 suggests that actually Ederson didn't under or over-perform. The kind of shots that he faced were of such high quality that on average a GK wouldn't be expected to save them anyway. Top of the (PSxG - GA)/90 table is Aston Villa with a value of 0.2 and I think anyone who watched them last year would agree that their GK (Martinez) had a monster season, producing some unreal saves. And the stat completely aligns with reality!
If you've the time to pour over stats like these, you will find that they almost always agree with the eye-test. So what is the use of stats then? Well if based on a real-world comparison you find that xG is pretty reliable, then instead of watching every match, you can just look at a number and deduce if the team had good/average/bad finishers.
Now say I'm looking for a player who is a great finisher. As a scout, I don't have to physically watch 10,000 matches a year anymore to find that hidden gem. I can use stats to find the profile of players who typically overperform xG. Say that is a list of 5 players. Now with a smaller set of players, I can go to the actual matches and look at these 5 players specifically to confirm if they do indeed pass the eye test.
So stats are useful. The underlying framework which enables us to generate stats in football is the same framework that captures the text on a physical paper accurately when you take a photo using your phone. It's all algorithms and models designed for a specific purpose, with a limited set of variables/features and trained using real world data.