4 ms·

That appears to be a purely statistical argument - some questions appear to be disproportionately hard/easy for certain racial groups. That has some value, but

That appears to be a purely statistical argument - some questions appear to be disproportionately hard/easy for certain racial groups. That has some value, but it would be nice to see some concrete example(s) of how this bias is expressed.

jedberg 5y ago
I posted an example in a sibling comment: Runner is to Marathon as XXXXXXXX is to Regatta If you grew up on the coast in a wealthy family, you have a much higher chance of knowing what a Regatta is and that it involves boat races.
- civilized 5y ago
  Aside from the fact, repeatedly pointed out, that the analogy section has been kaput since 2005... this would be a hard question, the sort that shows the least racial and income gaps.
- allturtles 5y ago
  As I replied there, this example is very old. There hasn't even been an analogy section since 2005.
  - civilized 5y ago
    Definitely my most hated section. Full of rare, obscure, you-know-it-or-don't words for massive bookworms. But if you read the papers cited in the GGGP comment, those hard problems aren't the problems that show the racial gap. The hardest problems show the least gap. Which doesn't fit the pat little sailboat jargon theory at all. However it does fit a common pattern, also seen in longevity statistics, where the most exceptional are more similar to one another than the average are.
    - DenisM 5y ago
      > the most exceptional are more similar to one another than the average are Wait, can you elaborate?
      - civilized 5y ago
        Sure! This idea is interesting and may well be original: As people get older, the mortality rate gaps between different groups get smaller on a relative basis. Like, the ratio between rich and poor 50 year olds' risk of death over the next year might be 3x or more, but as they age, they become more similar, and by 90 the ratio may be more like 1.3x, and after 100 it becomes indistinguishable from 1 (in part because of thin data). The theory is that the genetics and environmental conditions needed to survive to an old age are fairly specific, so once you're down to those people, they have less variation from one another. The health risk factors that distinguish income and racial groups may mostly impact younger age mortality, and the people that get past those younger ages may have avoided or resisted those risk factors, regardless of race or income. Similarly here, students who "survive" to do well on hard SAT questions may be fundamentally very similar to one another (sharing a high g-factor), and this may greatly reduce socioeconomic and racial gaps that are seen on the easier questions. Does this mean the easier questions are biased? Not necessarily, I would argue. The concepts tested in the easy-medium questions are most likely to be intellectually foundational -- you are far, FAR more likely to get obscure rich people jargon in the hard questions than the easy-medium ones. We want people to have the basic intellectual exposure that enables them to do well on those easier questions, and we should reward it. The problem is not the test, the problem is racially biased failure to educate in the lead-up to the test. This explains the pattern perfectly: the strongest students power through regardless of background, while the weaker are more sensitive to resources they were given. So you see the biggest gaps on easy-medium questions where basic educational background matters most to getting them right, rather than raw talent.
        DenisM 5y ago
        I see. Very curious. Shall we call it “adverse convergence” hypothesis? Have you explored broader implications of your idea? For example people who survived tribulations of a political career may end up close to each other despite their initial difference in political opinions, and maybe even across different countries. This doesn’t sound quite as convincing, but I’m kind of grasping in the dark trying to find some other examples.

jedberg 5y ago

I posted an example in a sibling comment: Runner is to Marathon as XXXXXXXX is to Regatta If you grew up on the coast in a wealthy family, you have a much higher chance of knowing what a Regatta is and that it involves boat races.

civilized 5y ago
Aside from the fact, repeatedly pointed out, that the analogy section has been kaput since 2005... this would be a hard question, the sort that shows the least racial and income gaps.
allturtles 5y ago
As I replied there, this example is very old. There hasn't even been an analogy section since 2005.
- civilized 5y ago
  Definitely my most hated section. Full of rare, obscure, you-know-it-or-don't words for massive bookworms. But if you read the papers cited in the GGGP comment, those hard problems aren't the problems that show the racial gap. The hardest problems show the least gap. Which doesn't fit the pat little sailboat jargon theory at all. However it does fit a common pattern, also seen in longevity statistics, where the most exceptional are more similar to one another than the average are.
  - DenisM 5y ago
    > the most exceptional are more similar to one another than the average are Wait, can you elaborate?
    - civilized 5y ago
      Sure! This idea is interesting and may well be original: As people get older, the mortality rate gaps between different groups get smaller on a relative basis. Like, the ratio between rich and poor 50 year olds' risk of death over the next year might be 3x or more, but as they age, they become more similar, and by 90 the ratio may be more like 1.3x, and after 100 it becomes indistinguishable from 1 (in part because of thin data). The theory is that the genetics and environmental conditions needed to survive to an old age are fairly specific, so once you're down to those people, they have less variation from one another. The health risk factors that distinguish income and racial groups may mostly impact younger age mortality, and the people that get past those younger ages may have avoided or resisted those risk factors, regardless of race or income. Similarly here, students who "survive" to do well on hard SAT questions may be fundamentally very similar to one another (sharing a high g-factor), and this may greatly reduce socioeconomic and racial gaps that are seen on the easier questions. Does this mean the easier questions are biased? Not necessarily, I would argue. The concepts tested in the easy-medium questions are most likely to be intellectually foundational -- you are far, FAR more likely to get obscure rich people jargon in the hard questions than the easy-medium ones. We want people to have the basic intellectual exposure that enables them to do well on those easier questions, and we should reward it. The problem is not the test, the problem is racially biased failure to educate in the lead-up to the test. This explains the pattern perfectly: the strongest students power through regardless of background, while the weaker are more sensitive to resources they were given. So you see the biggest gaps on easy-medium questions where basic educational background matters most to getting them right, rather than raw talent.
      - DenisM 5y ago
        I see. Very curious. Shall we call it “adverse convergence” hypothesis? Have you explored broader implications of your idea? For example people who survived tribulations of a political career may end up close to each other despite their initial difference in political opinions, and maybe even across different countries. This doesn’t sound quite as convincing, but I’m kind of grasping in the dark trying to find some other examples.