Friday, December 04, 2009

Wordsum serves as valid IQ proxy

Periodically it is asserted (usually by one-time commenters who drop in through search engines here) that using wordsum scores as a proxy for IQ produces no relevant information since intelligence tests are much more than simple vocabulary tests. Looking at wordsum scores by region and comparing them to several other good-faith IQ estimates suggests the test is indeed a useful IQ proxy. In complement to that, the following tables show how wordsum scores interact with other uncontroversial intelligence proxies. For each variable, respondents are divided into five groups; Really Smarts (wordsum score of 9-10, comprising 13% of the population), Pretty Smarts (7-8, 26%), Normals (6, 22%), Pretty Dumbs (4-5, 27%), and Really Dumbs (0-3, 12%).

Since the inception of the GSS, interviewers have recorded their impressions of respondents' apparent ability to comprehend the questions being asked them. The percentages, by wordsum intelligence groupings, who had a "good" grasp comprehending the questions being asked them (n = 24,111):

Comprehended questions
Really Smarts96.9
Pretty Smarts93.5
Pretty Dumbs77.8
Really Dumbs49.8

Language barriers do not present issues, as respondents not fluent in English were excluded from the GSS until 2006, when a Spanish version of the survey was also dispatched. This is evident in the gentle but steady increase over time in the percentage of respondents deemed to have "good" comprehension and a corresponding decrease in the percentage of respondents said to have "poor" comprehension--if a lack of English fluency was obfuscating the relationship, the average level of comprehension would be declining, not rising.

The effective range on the scale used to record respondents' socioeconomic status runs from 17.1 on the low-end to 97.2 on the high-end. Mean values by grouping:

Average SES
Really Smarts61.4
Pretty Smarts52.7
Pretty Dumbs42.6
Really Dumbs37.2

And mean number of years of education:

Educational years
Really Smarts15.3
Pretty Smarts13.8
Pretty Dumbs11.9
Really Dumbs10.5

Scarcely distinguishable from the distributions documented in The Bell Curve. The wordsum test is a relevant and incredibly useful proxy. Without it, GSS yields would be much less copious than they are.



Jokah Macpherson said...

I am always impressed that vocabulary gives such an accurate measure of intelligence. I guess it is because it is something that is picked up in one's everyday life based on curiosity and ability to determine the meaning of words based on context. Still, I wonder, would a hypothetical MATHSUM GSS variable be a better or worse proxy for intelligence?

Your group names reminded me of the scoring system for the Cracker Barrel game for some reason.

Steve Sailer said...

"This is evident in the gentle but steady increase over time in the percentage of respondents deemed to have "good" comprehension and a corresponding decrease in the percentage of respondents said to have "poor" comprehension"

Thanks. Can you put up the numbers for that? It seems like a real world version of the Flynn Effect.

My guess would be that exposure to electronic media, from radio onward, has made people mentally quicker.

Also, it would be interesting to see who is overrated and underrated relative to their test scores. My impression is that people in New York City seem mentally quicker than people elsewhere. Some of that is actually mean IQ difference, but some of it is different affect.

For example, I had lunch with a guy who was a doorman in an Upper East Side apartment building. For the first 15 minutes, I thought, "Wow, everybody must be really smart in New York if the doormen are this smart." By the end of the lunch, I could see, however, that he was really about as smart as you'd expect a doorman to be. But, in New York City, you're expected to be on your toes using whatever brains you've got all the time, whereas in other parts of the country, you aren't expected to use your full mental capacity all the time.

David said...

Interesting. You continue to make me think. Thank you.

FeministX said...

No one doubts that wordsum has some correlation with IQ, the issues are these- how high is the correlation? Is it .4 or .7? The secondary issue is does it matter? If wordsum itself correlates well with SES, that might be enough information to make whatever interpretation about social behavior you are doing. On the other hand, if you are making a statement about the heritablility of some behavior, wordsum might not be equal to IQ in terms of heritability even if it has the same level of correlation with SES.

Anonymous said...

Wordsum is fine as far as it goes, vocabulary is undoubtedly something that is tied closely to IQ.

However you have to be careful about using it in some cases, for example because women tend to do better on it any variable that differs by sex is potentially problematic.

There is another GSS variable that also seems like a reasonable proxy for intelligence- ALIKESUM- which hasn't been done on so many subjects but has a much weaker sex bias. It does seem to produce somewhat different results to WORDSUM on some variables, like POLVIEWS for example.

Audacious Epigone said...


I suspect the two would correlate strongly with one another. VCU's Professor McDaniel used science and math NAEP scores to construct his state IQ estimates. Previously, I'd used math and reading. His results correlate slightly better with previous good-faith attempts at measuring average IQ by state than mine do (though our results correlate at .96, so we're looking at the very margins), and test's like Raven's matrices don't rely on vocabulary at all and yet are solid. So the hypothetical MATHSUM would probably be preferrable, but not in a significant way.


Sure. I'm working on doing so now.

Re: perceptions of intelligence by state, it seems to run as strongly against people with southern drawels. Is it as simple as the rapidity of words flowing out of someone's mouth? New Yorkers talk quickly. The GSS is limiting in that in that it only reports location by nine geographic regions rather than by state, but I'll give it a look as well.


I suspect it's at least .7, and probably somewhere in the .8-.9 range. I agree that for wordsum scores, relative to IQ scores, the remaining slack is probably more 'nurture' than it is 'nature', though.


I'm not finding the variable. Is it slightly different than what you left in your comment? I tried variations LIKESUM, ALIKSUM, ALKESUM but nothing there, either.