Educational Testing Service: Stupid or Smart?

Since the Educational Testing Service is responsible for measuring intelligence, it is disturbing when they appear . . . not intelligent. A Chinese student of mine sent me the following question, which appears in a set of study questions. You are supposed to identify the “flaw” in the argument.

The article entitled ‘Eating Iron’ in last month’s issue of Eating for Health reported that a recent study found a correlation between high levels of iron in the diet and an increased risk of heart disease. Further, it is well established that there is a link between large amounts of red meat in the diet and heart disease, and red meat is high in iron. On the basis of the study and the well-established link between red meat and heart disease, we can conclude that the correlation between high iron levels and heart disease, then, is most probably a function of the correlation between red meat and heart disease.

By “is a function of” I suppose means “is due to”. Sure, there are several imperfections, unstated assumptions, in every argument, including this one, just as every piece of research has several imperfections. But are there obvious important flaws in this argument? I think it is reasonable to assume that red meat is the main source of iron.

Since many scientists have trouble interpreting correlations (they think “correlation does not equal causation” is not misleading) presumably an ETS question writer has even more trouble. And this question reflects that. But maybe not.

10 thoughts on “Educational Testing Service: Stupid or Smart?

  1. The flaw they are looking for is that dietary iron could be a proxy for red meat intake and that something else in red meat could be the source of the association with heart disease, not dietary iron per se. Come on, Seth!

  2. Vic, my take on it is different. I think the flaw in the argument is that something else in the diet (besides red meat) may be the source of the high iron that’s referenced in the first sentence. In any case, the passage is difficult to analyze because it’s ambiguous and poorly worded.

  3. Vic, the conclusion isn’t that the reason for X (correlation between heart disease and red meat intake) is Y (correlation between heart disease and iron intake). It is that the reason for Y is X. I take “Y is a function of X” to mean “Y is caused by X.”

  4. To write a so-called intelligence test question where you have to guess the meaning of an ill-worded passage of English seems pretty stupid. So there’s the answer to your headline question: “stupid”. mIn fact, there’s probably a general point here: I suspect that IQ tests aren’t very reliable when used on subjects who are markedly more intelligent than the setters of the tests. Has this been investigated?

  5. FWIW, the SAT and other ETS exams are intended to measure “scholastic aptitude”, not intelligence. Sadly, they are not the same thing. 8-(

  6. I gave up university when I found out that not even my economics professors understood the maths behind their own articles, but sent them to a math professor so that he could do it for them. This math professor OTOH did not know how to program a computer, to say nothing about stochastic simulations.

    If you tried to ask them for instance if the model was stable, you just got an empty stare, and some non-answer that that it was statistically significant or that it was a well known economic model.

    My conclusion was that nobody in economics (academia?) really understood what they were doing, since the models and the statistics they were using were so complicated, that they used their whole brainpower just to learn to use them, leaving little time to understand what they really was doing.

    This is probably even worse in other faculties, such as psychology or medicine, where I suppose that statistical models are something one just use, without really understanding them at all.

    If we take correlation as an example, few understand that given enough variables, and a limited sample space, you would get significant correlations just from the randomness-variable alone, without there really being any underlying correlation representing the real world.

    The implication of this, is that significant correlation alone is basically worthless. The correlation needs to be BIG, if one is to conclude that something is good or bad for everybody.

    If one goes back to the example of high iron, undiagnosed Hemochromatosis alone would give you a significant higher mortality for the high iron eaters, rendering the correlation worthless for us from say Scandinavia, who have been eating fish, meat and dairy for the last 4000? years, and before that mostly read meat. (A lot of us can’t eat gluten though)

    “Some researchers believe that hemochromatosis originated more than 40,000 years ago in the area we now know as Ireland with a single person whose genes mutated so that he or she could over-absorb iron to compensate for an iron-poor diet.

    Today, with iron-enriched foods, iron supplements and plenty of red meat, there’s no need to pull in extra iron, yet many still carry the ancient mutated genes that cause their bodies to do so, at toxic levels.

    Left untreated, hemochromatosis can lead to everything from early menopause and infertility to diabetes, heart failure, cirrhosis, primary liver cancer and even death. But if caught before damage is done, hemochromatosis patients can be saved, and their health restored through a process called bloodletting, or phlebotomies.”

    https://www.americanhs.org/Irish%20in%20the%20blood.htm

  7. From the hemoc
    “Some researchers believe that hemochromatosis originated more than 40,000 years ago in the area we now know as Ireland with a single person whose genes mutated so that he or she could over-absorb iron to compensate for an iron-poor diet.”

    dearieme ““more than 40,000 years ago in the area we now know as Irelandâ€: golly, are you sure?”

    Not at all, and it was not really important here.

    The reason I mentioned Hemochromatosis is that you would get exactly this result if your sample space concisted of people with partly Irish decent, like in the UK, or the USA, since it is known that undiagnosed Hemochromatosis is a riskfactor for heart disease.

    To make my point clearer, if I have subpopulations with different levels for what amounts to dangerous or beneficial, then it would be wrong to assume a model where you treat the subpopulations as one population, IF you want to say what is dangerous to the average person.

    To go back to the Hemochromatosis example, the level they would find Iron was beginning to become dangerous in their model, would not be relevant for the majority of the population, since what would be to high would be the triggering of the ones with undiagnosed Hemochromatosis.

    From the same article:

    “In 1997, a year after researchers isolated the mutated genes that cause hemochromatosis, a genetic mail-order test was developed that allows people to be screened for the disease. Those who test positive for the gene can keep their iron levels in check, and prevent iron from ever invading their organs and destroying their lives.

    Thomas is on a mission to make people aware of the $125 test, as well as simple blood tests (serum iron, total iron binding capacity and serum ferritin) that doctors seem so reluctant to give.

    It’s not that there’s a conspiracy against patients with iron overload. It’s just that many doctors haven’t been educated about the latest facts regarding the disease.

    And since people aren’t routinely tested for iron overload, doctors must be able to recognize warning signs. Unfortunately, the symptoms of hemochromatosis are often serious diseases themselves, and doctors wind up treating the heart trouble, the diabetes or the liver cancer without looking for an underlying cause.”

    Last about the 40.000 years among the Irish, I think it is correct that they present it so, based on their own models, even if geology say that it imposible. If scientists started tweaking their models to make them fit the wanted result, it would be a lot harder to recalibrate the results afterwards when we find out a better model.

Leave a Reply

Your email address will not be published. Required fields are marked *