What do you think GM Hikaru IQ is?

Emory university is one of the premier health and humanities universities... I think it's also in the top 25 ranking of Ivy league universities. On their psychology departments website they have a link discussing the interpretation of effect sizes in psychology:

Correlation (emory.edu)

In psychological research, we use Cohen's (1988) conventions to interpret effect size. A correlation coefficient of .10 is thought to represent a weak or small association; a correlation coefficient of .30 is considered a moderate correlation; and a correlation coefficient of .50 or larger is thought to represent a strong or large correlation.

This is the convention they cite... it is directly contrary to your claim that 0.1-0.3 is weak. i.e. 0.2 is in the dead center of the "weak" range. What I said originally was that 0.24 was moderate. Well my claim is much closer to Cohen's convention than yours. Now, as you've pointed out, the correlation is reduced to 0.22... now we probably have to say the effect is mild-to-moderate.

So Cohen wrote this book back in 1988, this is apparently a longstanding convention in psychology, according to the professors at Emory. It's a convention that actually makes sense within the field of psychology... On the other hand, from you I'm just hearing this rote insistence that some other specific math book is the gold standard and defines things differently, according to you... you're making comments about social science conventions you don't even appear qualified to make. I'm assuming you own this math book, but I've seen no actual citation of your math book on this issue from you, so I can't interpret what it's saying contextually... I also see no intelligent reasoning or interpretation of the practice of social psychology provided by you... so no, I don't think I acknowledge this appeal to convention that you're making. Everything I see on the web is saying moderate is about 0.3, this is what makes sense, what I was taught at university when I took statistics in psychology... and what Emory / Cohen go with. Clearly Cohen's work is a longstanding convention, at best the argument is a wash and at worst you're either misreading... or misinterpreting, or stretching the truth, or just full of it, who knows.

Anyway, I find it hilarious that you're doubling down on what the math books say as if this is the most critical point... you're still missing it - I'm not even really hinging the point on this, in other fields certainly 0.2 is a fairly weak correlation. Anyway, what I'm really describing is how social science is practiced. This is what's relevant because you're presuming to do social science. A math book is generally not going to presume to teach very much about the practical aspects of social science to students, it's going to leave that for other courses and mostly just teach students the math itself. Anyone who's majored in or worked in a STEM field will tell you there is a world of difference between the book learning they got in school, and how their field is actually practiced... have you ever worked in a STEM field...? I've worked in two separate STEM fields, engineering and social science. What I'm telling you is social science is a field made up of 0.2 correlations, and so within the field you cannot just dismiss such a correlation or consider it trivial, you would be left unable to do social science. Your study, which yielded an effect of r=0.22, should not be ignored by a serious social scientist. That is the important point, you ignored it but I'm repeating it again for you. And the social scientists who wrote that correction don't disregard the effect, they state explicitly that the conclusion of the study remains unchanged.

In reality an effect in social science is almost always predicted by at least 1-2 dozen core factors. So you end up with a complex equation for predicting your effect, one which has 1-2 dozen inputs and produces the effect as the output. The core factors are usually found through some broad analysis of hundreds of traits, and then through a process of factor analysis. So in our conversation what we'd say is IQ is one of 1-2 dozen factors which predicts chess playing ability, along with many other factors such as hours practiced, studying habits, age at which the person first learned, the persons sex, probably their current age, and a half a dozen other factors you could probably identify. That's how this works in reality.

The other thing I will mention, yet again, is that in statistics the sample correlation r is inversely proportional to confidence interval p. So when you have a confidence interval of p=0.001, an extremely conservative confidence interval, it becomes quite meaningless to squabble over whether a correlation r is moderate or mild-to-moderate. Because I could literally just change p=0.05 or 0.01 and the correlation would shoot right up to moderate immediately, it's quite a meaningless thing you're squabbling over. But you don't really realize how meaningless it is, because again, you're not a social scientist.

The Emory site simply doesn't say that, you did. What it says it 0.1 is a mild correlation, and 0.3 is moderate. It ends there. And I think they're speaking loosely, if you want to know the truth, because they aren't imagining this as some discrete mechanical scale the way that you are. If 0.3 is moderate 0.29 is not going to be mild, it's going to be "basically moderate" - the scale is continuous, these are subjective descriptors, it's not some discrete mechanical scale.

In either case, what we have is a misalignment between your source and mine. So the debate is a wash - I really think Cohen gives a much more useful and practical scale for use within social science, because it corresponds with how psychology is actually practiced in the real world. Your source is just a linear scale, it looks like it was constructed without much thought, or any real consideration for practical interpretation. Obviously the people at Emory agree with me. But you can stick with your little linear scale if you want, it's not too essential.

Btw - you missed a paragraph in the preceding post, you should address that.

I like my little one-liners at the end, they tend to get under the skin of internet egomaniacs such as yourself, something I do enjoy doing.

Keep trying

I ignored your post mostly, yeah, because I'm losing interest. It was mostly irrelevant, the only interesting part was the part I addressed there. The rest of it... well there was some weird speculation on why I chose to say "meaningful" instead of "significant" somewhere - you're way too suspicious and confused, going down a rabbit hole there for reasons I don't understand. And now it's just becoming: "the scale is discrete! Why would you ever say it's continuous, how could it be?! I win I win you lose you lose!", the conversation is really degrading at this point.

Here's the paragraph you missed though -

The other thing I will mention, yet again, is that in statistics the sample correlation r is inversely proportional to confidence interval p. So when you have a confidence interval of p=0.001, an extremely conservative confidence interval, it becomes quite meaningless to squabble over whether a correlation r is moderate or mild-to-moderate. Because I could literally just change p=0.05 or 0.01 and the correlation would shoot right up to moderate immediately, it's quite a meaningless thing you're squabbling over. But you don't really realize how meaningless it is, because again, you're not a social scientist.

- You added that paragraph afterwards in an edit, but here you go:

You stated that a sample correlation (r) is "inversely proportional" to the confidence interval (p), which is not accurate. Correlation coefficients (r) and p-values (p) are related, but aren't inversely proportional. The correlation coefficient (r) measures the strength and direction of the relationship between two variables, the p-value (p) assesses the likelihood that the observed correlation occurred by chance. A small p-value (such as p < 0.001) means the result is statistically significant, but it doesn't affect the magnitude of r. The confidence interval provides a range within which the true correlation is likely to fall, but the p-value is not equivalent to or directly interchangeable with the confidence interval.

You implied that by changing the p-value threshold (from p=0.001 to p=0.05 or p=0.01), the correlation would "shoot right up to moderate". Incorrect. Changing the p-value threshold does not affect the strength of the correlation. The correlation coefficient remains constant regardless of the p-value threshold. A p-value only indicates how likely the observed correlation is to have occurred by chance. Lowering the p-value threshold (e.g., from 0.001 to 0.05) simply makes it easier to declare the result statistically significant, but it does not change the strength of the correlation. A weak correlation (0.22, in this case) will remain weak even if you change the p-value threshold.

You dismiss whether a correlation is weak or moderate as "meaningless" due to the p-value being small (p=0.001). The distinction between weak and moderate correlations is not meaningless, even if the result is statistically significant. In fact, the strength of the correlation is just as important as statistical significance when interpreting practical importance. A statistically significant weak correlation still implies a weak relationship between variables. Statistical significance tells us whether the relationship exists or not, but the correlation coefficient tells us how strong that relationship is.

You are correct on one point, because I said sample correlation where I should have said population correlation. However, you are wrong on the broader point, your interpretation of this result and of the relationship between r / p:

r=0.22, p<0.001

This is this result of a left-tailed test of correlation. r here is the population correlation, not the sample correlation. You're correct you couldn't change the confidence interval and modify the sample correlation - by changing p what you're modifying is the left-tail of the confidence interval, i.e. the hypothesized population correlation. The correlation result r of the study we're talking about here is a hypothesized population correlation - that's why r is qualified by p, it must be interpreted alongside p. If this study used a normal confidence interval, like p<0.01, the confidence interval will become more narrow and the left tail of the confidence interval, which is 0.22 here - i.e. the hypothesized population correlation - will increase to a moderate correlation. That is the meaningful point about the results of the study I am making. And yes this is an inversely proportional relationship, as the confidence interval widens the left tail of the test, the hypothesized population correlation, approaches zero i.e. reduces in significance... and vice versa.

What this really shows is how difficult it is to interpret statistics correctly, and how precise you really have to be. For example, I've used the term "confidence interval" loosely as well but really p here is the lower confidence limit. The interval is actually the internal space between the upper and lower limit.

So anyway, the point about our conversation remains - if this study were to use a normal confidence interval (or limit), like p<0.01, it would have produced a more moderate correlation (even by your discrete interpretation of that term) between IQ and chess playing ability in the population. How much more moderate? Hard to say. But yes, this debate is meaningless. It has been meaningless the entire time.

It's left tailed because the sample correlation is above 0, so to test the null hypothesis (test whether the confidence interval includes zero) you need the lower limit of the confidence interval.

Not meaning to break up what is ongoing in conversation but can I enquire what the thoughts are on the value of having a known high IQ?

A practical use I've seen of it was taking a closely related 11+ test to get into a selective school. Of course, you could argue that being in Mensa is some kind of perk, that you're in a club with "like-minded" people and that's good for the talk and groups. Some people include Mensa on their CV, but when has that really made a difference?

Googling to check gave results of Kasparov scoring 123 on Raven's and 135 "on another." I was only familiar with the 135 result.

I doubt IQ among GMs averages to 130, but it might, and anyway as you said there's obviously a correlation.

Here's an IQ-like puzzle question. See how you compare against a few top GMs...

At 2 minutes in the video, pause after you hear the question because due to editing they start talking / answering within a few seconds.

https://www.youtube.com/watch?v=xlUfsrk179I&ab_channel=chess24

Only one of them reasoned out the answer, one of them guessed correctly, and one of them guessed incorrectly.

Emory university is one of the premier health and humanities universities... I think it's also in the top 25 ranking of Ivy league universities. On their psychology departments website they have a link discussing the interpretation of effect sizes in psychology:

Correlation (emory.edu)

In psychological research, we use Cohen's (1988) conventions to interpret effect size. A correlation coefficient of .10 is thought to represent a weak or small association; a correlation coefficient of .30 is considered a moderate correlation; and a correlation coefficient of .50 or larger is thought to represent a strong or large correlation.

This is the convention they cite... it is directly contrary to your claim that 0.1-0.3 is weak. i.e. 0.2 is in the dead center of the "weak" range. What I said originally was that 0.24 was moderate. Well my claim is much closer to Cohen's convention than yours. Now, as you've pointed out, the correlation is reduced to 0.22... now we probably have to say the effect is mild-to-moderate.

So Cohen wrote this book back in 1988, this is apparently a longstanding convention in psychology, according to the professors at Emory. It's a convention that actually makes sense within the field of psychology... On the other hand, from you I'm just hearing this rote insistence that some other specific math book is the gold standard and defines things differently, according to you... you're making comments about social science conventions you don't even appear qualified to make. I'm assuming you own this math book, but I've seen no actual citation of your math book on this issue from you, so I can't interpret what it's saying contextually... I also see no intelligent reasoning or interpretation of the practice of social psychology provided by you... so no, I don't think I acknowledge this appeal to convention that you're making. Everything I see on the web is saying moderate is about 0.3, this is what makes sense, what I was taught at university when I took statistics in psychology... and what Emory / Cohen go with. Clearly Cohen's work is a longstanding convention, at best the argument is a wash and at worst you're either misreading... or misinterpreting, or stretching the truth, or just full of it, who knows.

Anyway, I find it hilarious that you're doubling down on what the math books say as if this is the most critical point... you're still missing it - I'm not even really hinging the point on this, in other fields certainly 0.2 is a fairly weak correlation. Anyway, what I'm really describing is how social science is practiced. This is what's relevant because you're presuming to do social science. A math book is generally not going to presume to teach very much about the practical aspects of social science to students, it's going to leave that for other courses and mostly just teach students the math itself. Anyone who's majored in or worked in a STEM field will tell you there is a world of difference between the book learning they got in school, and how their field is actually practiced... have you ever worked in a STEM field...? I've worked in two separate STEM fields, engineering and social science. What I'm telling you is social science is a field made up of 0.2 correlations, and so within the field you cannot just dismiss such a correlation or consider it trivial, you would be left unable to do social science. Your study, which yielded an effect of r=0.22, should not be ignored by a serious social scientist. That is the important point, you ignored it but I'm repeating it again for you. And the social scientists who wrote that correction don't disregard the effect, they state explicitly that the conclusion of the study remains unchanged.

In reality an effect in social science is almost always predicted by at least 1-2 dozen core factors. So you end up with a complex equation for predicting your effect, one which has 1-2 dozen inputs and produces the effect as the output. The core factors are usually found through some broad analysis of hundreds of traits, and then through a process of factor analysis. So in our conversation what we'd say is IQ is one of 1-2 dozen factors which predicts chess playing ability, along with many other factors such as hours practiced, studying habits, age at which the person first learned, the persons sex, probably their current age, and a half a dozen other factors you could probably identify. That's how this works in reality.

The other thing I will mention, yet again, is that in statistics the sample correlation r is inversely proportional to confidence interval p. So when you have a confidence interval of p=0.001, an extremely conservative confidence interval, it becomes quite meaningless to squabble over whether a correlation r is moderate or mild-to-moderate. Because I could literally just change p=0.05 or 0.01 and the correlation would shoot right up to moderate immediately, it's quite a meaningless thing you're squabbling over. But you don't really realize how meaningless it is, because again, you're not a social scientist.

TL: DR

You lost me with utter nonsense in the first paragraph, anyway. Someone with an IQ of 142 should at least know how to Google Ivy League and learn that none of them are in the South.

It's ok, the conversation is above your head anyway, no loss here. 
But I actually got that from google -

As an independent observer of this discussion, you didn't. You replaced the unfamiliar term "New Ivy" by the widely known term "Ivy League" (which I believe always refers to a specific 8 very high performing Universities, like "Oxbridge" refers to a specific 2 in the UK).

"Emory University is one of 25 top schools in the nation tapped as a "New Ivy" in Kaplan/Newsweek's 2007 "How to Get Into College Guide."

Though this is in danger of becoming a very stupid debate, but not the first.

To back up my claim of independence, I don't view this carelessness as clearly indicating a lack of a high IQ. But checking facts would be definitely a habit worth acquiring.