LESSWRONG
LW

Comment Permalink

You can scale any distribution to have a mean of 100 and a standard deviation of 15, so I'd say it's a type error to ask whether or not the population actually matches that. It could very well be, of course, that the people taking IQ tests are a biased sample. In that case, scaling the test so that the test scores have a mean of 100 and a standard deviation of 15 may correspond to a different distribution among hypothetical IQ scores of people who don't have an actual IQ score.

RobertLumley13y30

You can scale any distribution to have a mean of 100 and a standard deviation of 10, so I'd say it's a type error to ask whether or not the population actually matches that.

The population isn't necessarily normal, and the scaling is not necessarily done (accurately). That was my point.

2Viliam_Bur13y

A serious IQ test should not have this kind of bias -- I think they pay randomly selected people to participate. (Of course a few people will refuse anyway, but you can make an estimate how big error that makes.)

See in context

2 Personality analysis in terms of parameters

by Blackened

20th Jun 2012

1 min read

2

(copied and edited from my post in the Facebook LW group)

I suspect that it would be obvious to most rationalists that the way people judge other people is flawed. Typically for a heuristic approach, it's correct to a degree, but with many faults. And it's wasting a big amount of information and a potential for a more planned approach where you can ask questions that assess certain qualities and exchange information about people's personalities by giving their "parameters".

I needn't think of it in this way, it was natural for me to take this approach as soon as I learnt my first measurable parameter and its implications (it was IQ). Then I explored more of them and researched them some more.

So far, I know about IQ, rationality (Keith Stanovich's), Big Five personality traits, executive functions, intuition for social situations and a few more things. However, I can't seem to find any literature that helps describe them (how do I detect them in people and what are their implications?) and their implications (how *exactly* is someone with a higher IQ different from someone with a lower IQ?). Also, I can't find literature on other traits.

Any literature on any of that would be greatly appreciated. I wonder if there is a book that deals with the whole issue. Also, I need literature about IQ and Big Five, but anything else would still be useful.

Is that sort of thing popular on LessWrong?

Personal Blog

2

New Comment

26 comments, sorted by

top scoring

Click to highlight new comments since: Today at 8:00 AM

[-][anonymous]13y40

how exactly is someone with a higher IQ different from someone with a lower IQ?

Start with the Wikipedia entry on 'Mental Retardation' and move toward the high IQ range. Many links to literature on differences between lower and higher IQ. Small degrees of IQ difference are hard to see, but large differences aren't.

Not from any particular literature: the range of errors that can be made by a high IQ person is greater than the range of errors that can be made by a low IQ person. A super-genius who can speak twelve languages and has degrees in physics and music can crash and burn in ways obvious to a casual observer. But an exceptionally low IQ person will never fail-up to learning twelve languages and having degrees in physics and music.

[-]Blackened13y00

That's again too abstract. I'm looking for a description that would help me make an IQ test. Currently, I'm relatively bad at judging people's intelligence - even though my gut feeling already gives me a relatively good heuristic for that.

J.P. Guilford's Structure of Intellect (1967) model used three dimensions which when combined yielded a total of 120 types of intelligence. It was popular in the 1970s and early 1980s but faded due to both practical problems and theoretical criticisms.

I don't know if this model is working at all, but I'm looking for something like what this sounds - a detailed low-level explanation that will ideally help me see the presence or absence of IQ in every individual thought/idea.

[-]gwern13y60

I don't know if this model is working at all, but I'm looking for something like what this sounds - a detailed low-level explanation that will ideally help me see the presence or absence of IQ in every individual thought/idea.

Why do you expect this to be possible? Even the most heavily g-loaded tasks like matrix tests still have substantial non-g variance to them. Let's remember what IQ is: a statistical tendency found by factor analysis of performance on multiple unrelated tasks. Expecting to find 'the presence or absence of IQ' makes it sound like you're reifying IQ as a specific plausible neurological capability akin to long-term memory or working memory.

[-]Blackened13y00

That doesn't change anything - then I want to know what constitutes pattern recognition, working memory (I know that some of it comes from executive functions - so it's not as simple as it looks like), spatial. Then I want to look at how each of them looks like in someone's reasoning. Plus maybe some of the not so highly g-loaded tasks.

[-][anonymous]13y00

I don't know if this model is working at all, but I'm looking for something like what this sounds - a detailed low-level explanation that will ideally help me see the presence or absence of IQ in every individual thought/idea.

That sounds like looking at a single coin and knowing it's good for buying grain but not cloth if it is combined with many other single coins. Many coins can purchase many things, but that information isn't encoded in a single coin.

[-]Blackened13y00

I'm not sure I got what you're saying. That low-level psychological mechanisms always participate in more than a single process? I don't see what's wrong with that. Or that there are several low-level mechanisms for that process, I still don't see anything wrong with that. Also, by "low-level", I mean "as low-level as I can get while still efficiently understanding and applying everything", so probably no neurology.

[-][anonymous]13y00

Here's an exact quote from a specific book. I'd like you to tell me something meaningful about the book it's from. The quote is: "r."

What I'm saying is one letter can't tell you anything about a specific book, one coin can't tell you about the purchasing power of many coins, and one thought / idea can't tell you anything about an individual's IQ. We seem to agree, reading that you write there's a low level under which understanding and application don't happen.

[-]brekken13y20

Judgement of other people can be pretty accurate actually. Observer judgement of the big five personality traits have more predictive power than self judgement.

See http://psycnet.apa.org/journals/apl/79/2/272/

[-]gwern13y20

Thin-slicing works well for judging Big Five. A paper just came out on detecting Big Five by correlation to choice of shoes, for example.

[-]fubarobfusco13y00

Unfortunately, the article in question (by Gillath et al) is expensively paywalled.

From this lack of Openness (ha ha) I predict that the article won't begin with the opening line of R.E.M.'s "Good Advices".

[-]OphilaDros13y00

You can get the 'author's version' of the paper 'for personal use, not redistribution' by going to this website, and providing your email ID:

http://web.ku.edu/~gillab/pubs.html

[-]OphilaDros13y00

Hmm.. Not sure about the etiquette of posting the link in a public forum since it's not meant for redistribution, but will keep it for now.

[-]gwern13y00

No, it's not. I know I read it without going through my academic proxy; following a link on Marginal Revolution, IIRC.

[-]Viliam_Bur13y00

how *exactly* is someone with a higher IQ different from someone with a lower IQ?

I guess a higher IQ predicts a higher chance of solving a problem successfully; also solving the same problem in a shorter average time. The prediction works best for problems that are general enough (don't require too specific background knowledge), otherwise the domain knowledge becomes important too.

Is this the kind of answer you want?

It would be more exact with a specific way of measuring complexity of the problem (if everything else fails, we could just measure them by "how many people can solve this problem in one minute"), and then a two-dimensional table with IQ in rows, problem complexity in columns, and probability of solving the problems and average time in the cells. I never heard about something like this, so perhaps it would make a good thesis for a psychology student -- we just need a volunteer.

It would be interesting to know e.g. how much intelligence can be substituted by speed. For example if you give a person with IQ 150 one minute to solve a problem, and a person with IQ 100 gets unlimited time for the same problem, which kind of problem will give advantage to which one?

[-]Blackened13y00

Is this the kind of answer you want?

Not really, I was looking for an answer that explains the exact way IQ works. Some problems might be less g-loaded. The explanation I'm looking for is something like that:

In a hypothetical situation, a problem is given to two (or more) people with very different IQs. Then there is a detailed explanation of how did they reach to the answer they gave - describing their thoughts and how did IQ lead them to think that. Also, something that accounts for the speed.

In other words, I want to know the patterns through which IQ works. Something that if I learn perfectly, I'll be able to predict the minimal IQ for the understanding of any phrase/idea (under given conditions, of course).

[-]patrickscottshields13y00

There are many different ways we could represent a personality (to varying degrees of accuracy.) I have not found a widely-accepted format, but I think we can each make our own for now. Whenever you wonder why someone acted a certain way, think about what the relevant parameters might have been and write them down. If several people work on this and share their results, perhaps one or more standardized personality representation formats will emerge.

The parameters collected by online user profiles such as those maintained by Facebook, Google Plus, or OkCupid might provide some inspiration.

If we had a good dataset of people and their personality attributes along with some performance measures, we could use machine learning to do neat things like predict relationship compatibility between two people. Imagine a rationalist dating service that used personality data to suggest matches!

Schema.org defines a "Person" model but it focuses primarily on circumstantial attributes rather than mental state.

[-]roll13y00

There's probably a lot of low hanging fruit, for example use of correct priors, e.g. given Gaussian prior distribution, a quite strong proof should be needed before you should believe someone (including yourself) has very high intelligence or expertise on a task.

Furthermore, many ways of evaluating people are to some extent self reinforcing as the people being evaluated are aware of evaluation. A smart person or expert can relatively cheaply demonstrate intelligence and/or expertise, in some way that provides very strong evidence, and will do so even for relatively moderate payoffs. Likewise, very selfless people can cheaply (in terms of their utilons) help strangers, and will do so for lower payoffs than people with more selfish utilities.

Other issue is gradual updates on huge amounts of weak, possibly non-independent evidence.

[-]RobertLumley13y00

I don't know about the others, but IQ is designed to have a mean of 100 and a standard deviation of 15. Whether or not the population actually matches that, I don't know for certain, but suspect strongly that it does not.

[-]Kindly13y60

[-]RobertLumley13y30

You can scale any distribution to have a mean of 100 and a standard deviation of 10, so I'd say it's a type error to ask whether or not the population actually matches that.

The population isn't necessarily normal, and the scaling is not necessarily done (accurately). That was my point.

[-]Kindly13y-10

Of course, distributions don't have to be normal to have a standard deviation.

[-]RobertLumley13y30

Sure, but it's a relatively meaningless description of a distribution without the caveat that it's normal.

[-]Kindly13y20

Sure, but it's a relatively meaningless description of a distribution without the caveat that it's normal.

To the extent that your statement is true, it is only true by convention.

In full generality, a standard deviation is a less useful description of a distribution without the caveat that it belongs to a specific family. Knowing the mean and standard deviation tells us the location and the scale of a distribution. Normal distributions are an example of a family that is parametrized by location and scale, but this is not an exclusive club in the least. Uniform distributions, for instance, would also work.

However, statisticians model things as normal distributions by default, and they are often right to do so. This means that there's a bunch of handy tables for tail probabilities and such for normal distributions, and these rely on knowing mean and standard deviation.

On the other hand, uniform distributions (for example) are not a very good description of most actual data. But if they were, then everyone would know that mean and standard deviation uniquely pick out a uniform distribution. And one might then object that scaling IQ to have a mean of 100 and a standard deviation of 15 is relatively meaningless: what if it turns out that IQ is not in fact the uniform distribution on the range [74, 126]?

But of course whatever the distribution of IQ is, it makes sense to fix the mean and standard deviation to our favorite numbers (I would have preferred 0 and 1 to 100 and 15, but I am in the minority), because the location and scale of the distributions are just an artifact of the way we quantify IQ. You could consider using another measure of scale instead of standard deviation, but what would you prefer? Interquartile range?

And finally, even if you don't know anything about your distribution, standard deviation can in fact be put to good use, by means of Chebyshev's inequality.

[-]Viliam_Bur13y20

It could very well be, of course, that the people taking IQ tests are a biased sample.

A serious IQ test should not have this kind of bias -- I think they pay randomly selected people to participate. (Of course a few people will refuse anyway, but you can make an estimate how big error that makes.)

[-]arundelo13y40

Nitpick: IQ's standard deviation is 15 or sometimes 16.

[-]RobertLumley13y10

Fixed. Thanks. I'd always thought it was 10. 15 makes more sense given IQ scores I've heard.

Moderation Log

Curated and popular this week

26Comments