<p>
</p>
<p>Does that mean you continue to claim that #3 is the wrong probability distribution? If so, what do you claim is the correct distribution to use for calculating the probability (in our E,T,A,Score example) that a given applicant (Score=750, A=True) has attribute T? I listed the two obvious alternatives.</p>
<p>Is it door #1, door #2, or something else? If the last, please describe the distribution.</p>
<p>
</p>
<p>“Selection bias” is a synonym for “using the wrong distribution (or population)”. Your answer is that the conditional distribution is wrong because it’s wrong. But you cannot explain what is the correct distribution, nor what is wrong with using the conditional distribution (except producing results that you don’t like).</p>
<p>Bias, in statistics, is always relative to the correct answer. To show that selection bias takes place, one needs to explain what is the correct probability distribution for making the calculation. Can you explain what the correct distribution is?</p>
<p>The standard answer from statistics is “the joint distribution of the unobserved variables given (i.e., conditional upon) the observed variables”. But that is the conditional distribution that you say is wrong. If in fabrizio-math there is a different answer, what is it?</p>
<p>Does this equation still work if the Asians are spending a lot more time on their school subjects and zero time on SAT I and II prep? I know about a private school in California where several kids have perfect scores or almost perfect, 1/3rd of the school is NMSF, 800s in several subjects. However, some of them have taken subject tests in 9th grade and got their 800s, perfect SATs in 10th etc but don’t do any specific prep for SAT per se. The distribution covers more than just Asians.</p>
<p>I don’t believe that it is possible to have 1/3rd of a school that is solely talented or hardworking. So the only factor I see contributing to this skewed profile is that the affluence drives the performance since the school is forcing them be ahead of the curve on academics which gets them these high test scores.</p>
<p>^That curriculum (at least for the advanced students) is a form of test prep. The students who are way ahead in math don’t need help on the math SAT’s. Math team and other academic clubs are test prep. If you add lots more variables to take into account the quality of the school, details of the curriculum and so on then this will capture the effect of going to a super-competitive high school. Espenshade’s book with Radford has a chapter analyzing patterns of test prep, extracurricular activities, academic summer camps and for all measures Asians are higher in net rate of utilization and also the rate with all other factors being equal (so it appears as a specific “Asian” effect, not from wealth or parents’ education or other things that also contribute).</p>
<p>re: equations, in this type of analysis it’s not necessary that Asians be ahead on every specific factor. In the earlier post I gave a hypothetical example of preparing/studying more in grades 1 to 11, then less in grade 12, which is very similar to your example where SAT prep becomes unnecessary due to earlier work. You just need the overall effect of the extra efforts Asians are making, across a wide range of factors, to be higher.</p>
<p>As far as the performance after college admission is concerned, the talent vs effort (or the analogous thing in a more detailed model) tradeoff would predict that groups relying more on effort through high school, will suffer later in selections that load more heavily on “talent” factors. But there are additional factors that exacerbate this effect for Asians specifically. One is that Asians start earlier, especially in math, and this advantage dissipates the further down the pipeline you go (college, grad school, postdoc, …). The other is “surplus capacity”. Whites’ effort levels start to intensify in college on the individual level in addition to the general population effect of admission selections keeping out the low-effort students. </p>
<p>As was mentioned many times the past two years – and fabrizio just noticed on his own! – this argument applies to hard workers and early starters whether they are Asian or not. With Asians the difference is that it is visible, with a large subpopulation following relatively similar academic strategies that are disparate from those of any other large visible group. Lots of whites might do something similar but they blend into a crowd and you can’t easily see them underperforming en masse in the college years, and there is no demographic check-box to indicate Hard Worker or Lazy Genius.</p>
<p>The example school above is not Asian. It is just a school where the tuition is on par with top 20 colleges and 60 out of 180 students made NMSF. So the main variable seems to be money.</p>
<p>^Discounting scores for (some forms of) wealth and private school status, and bonus points for being poor or first generation, are known practices. The principle is the same. Although everyone understands that “wealth-blind” admission would be affirmative action-style SAT bonus for the rich – not all of their higher SAT scores come from ability – it is not often noticed that exactly the same argument implies that race-blind admission is an SAT bonus for Asians. (Unless one believes that 100 percent of rich folks’ or Asians’ score differentials come purely from the students’ individual characteristics and not their parents, teachers and home culture.)</p>
<p>Frankly, I have no idea what all the fuss and debate here is about, but I have a question that’s been bugging me and this looks like the perfect place to ask it.</p>
<p>As an international student, does stating my ethnicity put me at some sort of a disadvantage, particularly if I’m ‘Asian’?</p>
<p><a href=“Unless%20one%20believes%20that%20100%20percent%20of%20rich%20folks’%20or%20Asians’%20score%20differentials%20come%20purely%20from%20the%20students’%20individual%20characteristics%20and%20not%20their%20parents,%20teachers%20and%20home%20culture.”>quote</a>
[/quote]
</p>
<p>I got into college in spite of, not because of, my Asian-ness. My mother was a single mother; she never interfered in my schoolwork or checked my grades and was generally a hands-off mother. In fact, my home environment was often chaotic and hardly conducive to learning. </p>
<p>The only reason to discriminate based on Asian-ness is based on perceived correlated socioeconomic privilege – so filter by socioeconomic privilege instead and leave the rest of us alone.</p>
<p>sis, you can keep rambling on about your “meritocratic discounting” sham all you like. From my perspective, you’ve already admitted that “If there is a negative relationship between A(sian) and T(alent) it can be marked down to an accident of statistics rather than a “fact about reality” (emphasis mine).”</p>
<p>Yet, despite that, you continue to roll with your hoax and now extend it to wealthy applicants, even though you ignored them when I discussed the case yesterday:</p>
<p>
</p>
<p>The only reason any of sis’s findings (excuse me, “correlations”) materialize is because the score is held constant. If effort goes up, talent has to go down, for any given score.</p>
<p>I think everyone should realize that ultimately, sis is advocating (excuse me, “arguing”) for a Gattaca-esque world, where talent, determined by genetics, reigns supreme. His “pure academic selection” attempts to weed out people whose scores were acquired by effort instead of innate characteristics like talent (under a “meritocratic [sic]” guise!) The only people whose scores will not be discounted are those who achieve them without expending any effort.</p>
<p>So it’s ironic, then, that sis often rails against alleged “genetic supremacists” when really, he is one; he’s simply not a “racial genetic supremacist.”</p>
<p>
</p>
<p>The Espenshade/Chung/Radford study would say that you got into college because the college gave you pro-Asian (or pro-nonwhite) racial preference over socioeconomically equivalent whites. Their regression results were:</p>
<ol>
<li><p>Asian lower-class applicants have 91.5 times higher admissions odds ratio than identical White applicants (610 x 0.15). [odds ratio is chance of admission divided by chance of rejection]</p></li>
<li><p>Asian working-class applicants have 2.94 times higher admissions odds ratio than identical White applicants (19.62 x 0.15).</p></li>
<li><p>Asian lower-class applicants are preferred over black and Hispanic working-class applicants, and thus are preferred over all race and economic categories of applicants other than lower-class URM. (odds-ratios of 8.24 (A,lc) versus 3.16 (H,wc), 5.78 (B,wc), 1.00 (W,wc or higher) )</p></li>
</ol>
<p>That’s in addition to a factor of 1.59 bonus for immigrants. If these studies are right, your application was moved to the head of the line ahead of equivalent and more-qualified whites. Their interpretation of the results is that poor Asians provide a double boost to the college’s diversity numbers, counting in both the socioeconomic and racial categories. They don’t say why immigrants get a boost but it is probably because verbal SAT scores are interpreted more generously. </p>
<p>Oh, yes, one other thing. Are you a female? By far the most solid (but curiously unreported) finding from several different versions of this study, was a factor of 1.5 boost for women. Multiply the different advantage factors to get the net result!</p>
<p>Now, as to “filtering by socioeconomic privilege” , have a look at the following graph. I believe it is from the College Board race/sex data the last time they released it.
[File:1995-SAT-Income2.png</a> - Wikimedia Commons](<a href=“http://commons.wikimedia.org/wiki/File:1995-SAT-Income2.png]File:1995-SAT-Income2.png”>File:1995-SAT-Income2.png - Wikimedia Commons)</p>
<p>You can see that differences in average SAT between racial groups are relatively independent of income across a wide range. (In the lowest income levels, there is an exception to this where whites are advantaged, maybe because fewer whites are immigrants with low verbal SAT). The implication is that race-blind “economic affirmative action” would still be statistically equivalent to an SAT boost for Asians and a penalty for URM. Unless you believe that after accounting for income the racial difference comes entirely from characteristics of the individual students and not the race/culture/parent dependent environment.</p>
<p>Asians may find some discrimination in college admissions but I think it is mostly in the schools with the most selective admissions. For those schools, every little detail counts. They want to diversify their pool - they don’t only want engineers, etc. NYU told us some years ago that they didn’t screen for this - but then again they basically want candidates that are willing to come up with the money.</p>
<p>
</p>
<p>When Asians are ORM? UVA didn’t need to boost its racial categories for Asians, Siserune. McDonnell’s Confederate-friendly administration is already suspicious of them. (At 55% of the student population I hardly think females get a leg up – unless you’re looking at MIT.)</p>
<p>And how do you judge “identical” white applicants?</p>
<p>
</li>
</ol>
<p>Averages can be deceiving. My poor Asian classmates didn’t go anywhere out of state. Heck, most of the migrants from China didn’t even go to a four-year-school.</p>
<p>^ adding to above - how old is this so called eppenshade study and does it have any bearing for 2012?</p>
<p>All studies that use statistics serve a few useless purposes for the rest of us - to get the academics published and others can use those to propagate their own shallow point of view. For the rest of us in real world, life goes on.</p>
<p>
</p>
<p>Say it ain’t so, TPG!</p>
<p>xiggi - I did write a master’s thesis in college and read several papers and dissertations during that process. So I have some idea how these things go. </p>
<p>I have seen how statistics can be interpreted to benefit a study and know a few doctoral students over the last few years that ran into an academic juggernaut when they found their professor’s entire existence was called into question based what they found in their own research that contradicted several years of research from the same lab in the past. So they either had to bend their data or start over in some other area. There are always studies that contradict each other completely and then there are studies about cold fusion. Then there are a whole bunch of drug trials that get totally cooked up which only come to light when a few young people die and it becomes Perry’s fault.</p>
<p>It is easy to contradict studies in sciences because someone has to replicate them and find out quickly it is not so. Who is going to contradict studies about education? its not like any of us really care right?</p>
<p>It is amazing how few studies there are about these about college selection…
Why are there not more, given that it is a pressing issue?</p>
<p>Maybe academia does not want to look to deeply into the activities of the institutions that employ them, lest hey bite the hand that feeds them…?</p>
<p>I would think that professors would be adamant about the qualities of the students and the overall student body they teach, even though I doubt highly that they would not want to spend their precious time in the selection process itself. I guess they are happy enough with the students, and might react at the “tipping point”… both by voicing their concerns to their own college’s admissions administration, and by doing research studies!!</p>
<p>
</p>
<p>For UVA, you’re in luck, statistical studies were done on their admissions. Asians were found to have a small admissions advantage over whites. </p>
<p>The Center for Equal Opportunity, a right wing think tank devoted to dismantling affirmative action, performed the same type of admissions study as Espenshade for the UVA admissions process in 1996 and replicated the findings in 1999. They fitted a statistical regression (logistic) model to predict probability of admission from race, gender, SAT math, SAT verbal, class rank, legacy status, and in-state residency. The sample was the entire set of applications each year.</p>
<p>They found, in addition to the expected affirmative action effects for black and hispanic applicants, a positive effect (odds ratios of 1.19 and 1.22 compared to whites, for the two years considered) of being Asian. This was consistent with the slightly higher acceptance rates for Asians overall, but is a sharper result because it tries to “control” for the effect of the other variables. The Asian coefficients were not statistically significant in either year. In short, either Asian was generally a positive factor, or Asians and whites were treated the same. </p>
<p>For the law schools at UVA and College of William and Mary the same type of analysis found 2-to-1 and 3-to-1 advantages for Asians in 1999, controlling for LSAT and other variables. Pro-Asian affirmative action, in other words. </p>
<p>It doesn’t make for splashy journalism to report non-discrimination. Claims of perennial racial victimization are much more publishable. The CEO website appears to be down but the report can be found on other sites by its title “Preferences at the University of Virginia”.</p>
<p>
</p>
<p>Females do not need a leg up to reach or surpass parity, except at the schools with very high SAT requirements, or a science/engineering focus. At UVA 1999 the study saw a very small negative gender effect (1.03 times lower odds ratio of admission); men and women were treated essentially the same. The supply of females starts to deplete at the high SATs compared to men and the consequences of this are probably what the Espenshade & Chung study detected.</p>
<p>
</p>
<p>It means “same values of the control variables”: gender, SAT scores, class rank, etc. The regressions purport to detect the result of being Asian “all else being equal”. In reality the interpretation is more complicated but in ideal conditions where the model is accurate, that is what the factor 1.19 or 1.22 would mean.</p>
<p>
</p>
<p>Your outraged posting that “everyone knows” Asians are discriminated was based on a news report whose sole factual underpinning, if you trace it back to sources, was the second Espenshade study on 1997 data. But that study found that you are in a strongly favored admissions category. Similar analyses from 1996 and 1999 at your school, indicate a lack of discrimination, unless it was in favor of Asians. When the studies don’t support the preconceived notion that discrimination against Asians took place, then “averages are deceiving”.</p>
<p>
</p>
<p>The “race” and “ethnicity” categories are asked about for American students (citizens and permanent residents). For international students, the relevant category is being an international student, and that is how you will be reported in the federal statistics. Here are some examples: </p>
<p>[U-CAN:</a> Harvard University](<a href=“ucan-network.org”>ucan-network.org) </p>
<p>[U-CAN:</a> Brown University](<a href=“http://members.ucan-network.org/brown]U-CAN:”>http://members.ucan-network.org/brown) </p>
<p>[U-CAN:</a> Claremont McKenna College](<a href=“ucan-network.org”>ucan-network.org) </p>
<p>[U-CAN:</a> Reed College](<a href=“ucan-network.org”>ucan-network.org)</p>
<p>
</p>
<p>When reasonable conversation is impossible, sometimes you have to post a calculation.</p>
<p>A verbal SAT score of 750 is at about the 98th percentile; around 2 percent of the general population of test takers get 750 V or higher. With that in mind:</p>
<p>Assume, as in our stylized running example, that there are rare binary attributes “Talent” and “Effort” (representing unusually high levels of those traits, such as being in the top percentile of talents or hard workers) and that SAT 750+ is attained exactly by those who possess Talent, or Effort, or both. Binary means that for each attribute, one either has it, or not. (The names of these attributes are arbitrary, and they do not have to correspond in any way to the ordinary meanings of talent and effort.)</p>
<p>Talent is distributed at random to one percent of the population. At every birth, the Talent Fairy randomly chooses a number from 1 to 100, and if the number is 100, the lucky child is given the gift of “Talent”.</p>
<p>Effort is distributed differently for Asians and non-Asians. Within the non-Asian population it occurs at random in 1% percent of people, and within the Asian population it is present at random in 3%. </p>
<p>From a set of hundreds of both Asian and non-Asian “applicants” about whom you have limited information – specifically their name, race and the fact that they scored 750+ (the precise score is not provided) – you want to select 100.</p>
<p>Your goal in the selection, for whatever reason, is to maximize the number of chosen applicants who have Talent. </p>
<p>Consider the effect of two different choices.</p>
<ol>
<li><p>Select 100 non-Asians. Then the expected number of those chosen who have “Talent” is approximately 50, because in the non-Asian population the chances to get 750+ through T or E are the same, plus a much smaller chance that one attained it through both channels.</p></li>
<li><p>Select 100 Asians. Then the expected number who have the attribute “Talent” is half as many, about 25, since of those who make the cut to be in the selection pool, 3 out of 4 made it through Effort, again with a small correction to account for those who got in through both paths.</p></li>
</ol>
<p>For the numbers used in this example, a score of 750 would be close to the 96th percentile for Asians and the 98th percentile for whites, which is not that different from the College Board data (percentile ranks by ethnic group) considering the rounding errors in the tables. Also, the same conclusion would be true if the Talent and Effort rates were doubled within the Asian population; the talent distribution would then be more favorable for Asians but the relative chance of reaching the cutoff through T would still be about 1 in 4. </p>
<p>The words “conditional distribution”, “correlation”, and “paradox” do not appear in the text above. If you are the admissions officer in the story above, it does not matter at all whether you join Fabrizio in pondering the presence of correlations that might be present, and if they are, debate the angels-on-pinheads question of whether they should be called “real” or “spurious” or “aardvark” or “zebra”. What matters is the consequences of actions and decisions, which are immune to empty philosophizing and announcements of personal opinion.</p>
<p>At what point do Asians disappear into the main stream and become jews of 2012 where they are not labelled separately? I raise this issue because I have a younger kid in middle school and several of the Asian kids in the middle school seem to be 2nd generation, in some cases of mixed heritage. So although I moved to this country, some of my kid’s Asian/mixed heritage classmates have parents who were born and brought up in US.</p>
<p>I find the above definition of 2% of non-asians achieving 750 odd. I have to wonder if they are all jewish who are also considered to work hard. Afterall, 25% of Harvard is considered Jewish for a population percentage that is no different than Asians. I think this fact alone breaks the T and E correlation since all jews are lumped into non-asians and they do extremely well in their standardized tests. </p>
<p>So I fail to see that the top 2% of non-asians got 750 with talent alone if most of them turn out to be jewish. That breaks your whole premise.</p>
![http://commons.wikimedia.org/wiki/File:1995-SAT-Income2.png]File:1995-SAT-Income2.png](http://commons.wikimedia.org/wiki/File:1995-SAT-Income2.png%5DFile:1995-SAT-Income2.png){kind=link}
