By the way, I assume your average rating over a large number of games will be the same, but your peak rating will get a little boost by playing people better than you.
Anyway, it really doesn't matter IMO... and I haven't even messed with that setting (I'm at the default whatever that is).
It does help a bit. Natural fluctuations in performance have the biggest impact when the ratings are close.
Oh, and I mean this mathematically
For example start with two equally rated players (they will score 50% and no rating change)
And now increase one player's strength by 100 points, and they will gain 14%
Increase by another 100 and they will gain 12%
Increase by another 100 and they will gain 9%
See how 100 points helps less and less? This is important because the number of points you win (or lose) depends on your expected score (yes, any version of Glicko works the same way).
In simple terms this means crushing a player much lower than you rewards you less in terms of points per amount of skill displayed than when playing someone equally rated... and this is why I say playing people rated higher than you rewards your good form more and punishes your bad form less.
Yes, I know, I use awfully big words for a forum full of 10 year olds, but sometimes someone accidentally posts something worth talking about, and so I talk about it.
If you have three players equally skilled, all rated 1200, and one had only 800 rated opponents, another 1200 rated, the third all 1800 for, say, 100 games, do you think they would have different ratings? I'm wanting to trust that they wouldn't. I also keep wanting to say ELO. 
Maybe that's the wrong question.
If you have three players equally skilled, all rated 1200, and one had only 800 rated opponents, another 1200 rated, the third all 1800 for, say, 100 games, do you think they would have different ratings? I'm wanting to trust that they wouldn't. I also keep wanting to say ELO.
If all are accurately rated and play as well as their rating, then the average rating over 100 games for each player is expected to be 1200.
I know a website that calculates ELO... there's one for Glicko, no doubt. Hmmm...
If you have three players equally skilled, all rated 1200, and one had only 800 rated opponents, another 1200 rated, the third all 1800 for, say, 100 games, do you think they would have different ratings? I'm wanting to trust that they wouldn't. I also keep wanting to say ELO.
If all are accurately rated and play as well as their rating, then the average rating over 100 games for each player is expected to be 1200.
whew... thank you
But a thought experiment like this...
Say 3 players, each 1200. One exclusively plays 1100, one exclusively plays 1200, and last plays only 1300.
Then during each game, we flip a coin for each player (the 1200 and their opponent). Heads their skill level goes up 50 points and tails it goes down 50 points. Over a large number of games the average for each player will be 1200, but the person who only plays 1300s will have the highest peak rating...
... at least this is what I expect. I could write some code to simulate it... this could be a fun blog post I suppose. I'll consider it.
It's an interesting idea.
But a thought experiment like this...
Say 3 players, each 1200. One exclusively plays 1100, one exclusively plays 1200, and last plays only 1300.
Then during each game, we flip a coin for each player (the 1200 and their opponent). Heads their skill level goes up 50 points and tails it goes down 50 points. Over a large number of games the average for each player will be 1200, but the person who only plays 1300s will have the highest peak rating...
... at least this is what I expect. I could write some code to simulate it... this could be a fun blog post I suppose. I'll consider it.
If a 1200 flips heads twice in a row, do they go up 100 points? Or does their rating reset to 1200 after every game and can only go up to 1250 and down to 1150?
But a thought experiment like this...
Say 3 players, each 1200. One exclusively plays 1100, one exclusively plays 1200, and last plays only 1300.
Then during each game, we flip a coin for each player (the 1200 and their opponent). Heads their skill level goes up 50 points and tails it goes down 50 points. Over a large number of games the average for each player will be 1200, but the person who only plays 1300s will have the highest peak rating...
... at least this is what I expect. I could write some code to simulate it... this could be a fun blog post I suppose. I'll consider it.
If a 1200 flips heads twice in a row, do they go up 100 points? Or does their rating reset to 1200 after every game and can only go up to 1250 and down to 1150?
Yeah, the coin flip only lasts for that 1 game.
And I wouldn't adjust their rating, only their skill level... in other words a 1200 who is playing like a 1250 (or 1150). And that'd be for the purpose of simulating someone having a good or bad day.
EDIT
Then if they win, and their new rating is 2010 or whatever, then the next coin flip could put them at 1250 or 1150... I have to remember there are 3 ratings to keep track of which makes this a bit annoying 
There's the intrinsic or base level skill (which will be 1200 and never change)
There's the adjusted skill due to having a good or bad day (which will be 1150 or 1250, and will be the one that influences win %).
And then there's the actual rating which is what will go up or down due to a win or loss.
But a thought experiment like this...
Say 3 players, each 1200. One exclusively plays 1100, one exclusively plays 1200, and last plays only 1300.
Then during each game, we flip a coin for each player (the 1200 and their opponent). Heads their skill level goes up 50 points and tails it goes down 50 points. Over a large number of games the average for each player will be 1200, but the person who only plays 1300s will have the highest peak rating...
... at least this is what I expect. I could write some code to simulate it... this could be a fun blog post I suppose. I'll consider it.
If a 1200 flips heads twice in a row, do they go up 100 points? Or does their rating reset to 1200 after every game and can only go up to 1250 and down to 1150?
Yeah, the coin flip only lasts for that 1 game.
And I wouldn't adjust their rating, only their skill level... in other words a 1200 who is playing like a 1250 (or 1150). And that'd be for the purpose of simulating someone having a good or bad day.
EDIT
Then if they win, and their new rating is 2010 or whatever, then the next coin flip could put them at 1250 or 1150... I have to remember there are 3 ratings to keep track of which makes this a bit annoying
There's the intrinsic or base level skill (which will be 1200 and never change)
There's the adjusted skill due to having a good or bad day (which will be 1150 or 1250, and will be the one that influences win %).
And then there's the actual rating which is what will go up or down due to a win or loss.
import java.util.Random;
public class Main
{
public static void main(String[] args) {
Random r = new Random();
int x = 1200; //player who plays 1100
int y = 1200; //player who plays 1200
int z = 1200; //player who plays 1300
int games = 50;
for (int i = 1; i<= games; i++) {
int chance1 = r.nextInt(2);
int chance2 = r.nextInt(100);
//if they play like 1250
if(chance1 == 0) {
if(chance2 <= 57) {
x -= 12;
}
else
x += 20;
}
//if they play like 1150
else {
if(chance2 <= 70) {
x -= 12;
}
else {
x += 20;
}
}
}
//player who plays 1200
for (int i = 1; i<= games; i++) {
int chance1 = r.nextInt(2);
int chance2 = r.nextInt(100);
//if they play like 1250
if(chance1 == 0) {
if(chance2 <= 57) {
y+= 16;
}
else
y -= 16;
}
//if they play like 1150
else {
if(chance2 <= 43) {
y += 16;
}
else {
y -= 16;
}
}
}
//player who plays 1100
for (int i = 1; i<= games; i++) {
int chance1 = r.nextInt(2);
int chance2 = r.nextInt(100);
//if they play like 1250
if(chance1 == 0) {
if(chance2 <= 70) {
z += 12;
}
else
y -= 20;
}
//if they play like 1150
else {
if(chance2 <= 57) {
z += 12;
}
else {
z -= 20;
}
}
}
System.out.println("New rating of player who played 1300: " + x);
System.out.println("New rating of player who played 1200: " + y);
System.out.println("New rating of player who played 1100: " + z);
}
}
This might not be 100% correct since I spent 5 minutes doing this but after 50 games, the player who played the 1100 the most gains the most points.
https://www.mark-weeks.com/aboutcom/aa04l04.htm
Whoops, line 74 is wrong. Should be z -= 20 instead. I also should've initialized i as 0 instead of 1 but that shouldn't really matter. Comments are also wrong. Should be:
int x = 1200; //player who plays 1300
int y = 1200; //player who plays 1200
int z = 1200; //player who plays 1100
Its also interesting you say John Bartholomew started the idea. Because I have never seen him smurf in rated matches. What he does is choose his rating range, as described in this thread. Bu8t he only does it in casual matches. At least in all of his streams that I watched. And he uses his real account. So in his case, there is nothing wrong with that and its respectable. And his streams truly are educational. They are not teaching kids how to manipulate ratings like Naroditsky and HIkaru. The guy usually has actual lesson plans and plays as best as he can accordingly on the board. His 5 part beginner serious is pretty legendary.
And the majority of people would be upset at having to play against a speedrunning GM. I don't know in what world you think most unsuspecting players would not be. Its simply not sporting and it encourages rating manipulation. There is a reason they are playing anonymously.
In the climbing the rating ladder series he would choose a rating range. The opponent could have chosen rated or unrated. But he was playing rated games if they wanted to.
If you look at his regular chess series, those were rated games. 15 minute rated. He even began the series by saying and I quote, I'm putting my mighty 23xx rating on the line. There was virtually no one with a rating as high so he played players from 1100 to 18 or 1900 on average. Occasionally hitting some people around 2000 to 2100 but there just weren't that many in the pool.
Nowadays he does challenges rated or unrated but it is all 3 minute Blitz and it's got nothing to do with any educational value. You just get to play a high rated player just like all other streams.
And the majority of people would be upset at having to play against a speedrunning GM. I don't know in what world you think most unsuspecting players would not be. Its simply not sporting and it encourages rating manipulation. There is a reason they are playing anonymously.
If you play one of those accounts the points you lose if you lose, are refunded. How does that have anything to do with ratings? The only thing you can do is win the game, typically by cheating, and then you get to keep the points provided you don't get caught cheating. They don't take away points you in, they just give back any points you might lose. So the only terrible wrong being committed is that the person winning against the speedrunning streamer gains points that they are not supposed to have. At least not by your definition. I guess Daniel's supposed to be upset if he gets beat and loses rating. But he's not.
And the majority of people would be upset at having to play against a speedrunning GM. I don't know in what world you think most unsuspecting players would not be. Its simply not sporting and it encourages rating manipulation. There is a reason they are playing anonymously.
If you play one of those accounts the points you lose if you lose, are refunded. How does that have anything to do with ratings? The only thing you can do is win the game, typically by cheating, and then you get to keep the points provided you don't get caught cheating. They don't take away points you in, they just give back any points you might lose. So the only terrible wrong being committed is that the person winning against the speedrunning streamer gains points that they are not supposed to have. At least not by your definition. I guess Daniel's supposed to be upset if he gets beat and loses rating. But he's not.
me and probably everyone else would love a chance to play a GM
It does help a bit. Natural fluctuations in performance have the biggest impact when the ratings are close.
Oh, and I mean this mathematically
For example start with two equally rated players (they will score 50% and no rating change)
And now increase one player's strength by 100 points, and they will gain 14%
Increase by another 100 and they will gain 12%
Increase by another 100 and they will gain 9%
See how 100 points helps less and less? This is important because the number of points you win (or lose) depends on your expected score (yes, any version of Glicko works the same way).
In simple terms this means crushing a player much lower than you rewards you less in terms of points per amount of skill displayed than when playing someone equally rated... and this is why I say playing people rated higher than you rewards your good form more and punishes your bad form less.