Thursday, June 09, 2011

GRE and SAT validity

Note Added in response to 2020 Twitter mob attack which attempts to misrepresent my views: Every faculty member involved in admissions has a responsibility to understand the science behind standardized tests. The subject is heavily politicized, but the scientific results are very clear and based on very large datasets -- e.g., decades of student records in large university systems.

I refer you to the detailed 2020 report prepared by a special committee of faculty assigned this task by the University of California system. This committee recommended continued use of the SAT in admissions.

https://infoproc.blogspot.com/2020/02/report-of-university-of-california.html


###################################



If you are a professor at a research university you have probably spent time on graduate admissions. How good is the GRE as an indicator of candidate quality? Is the subject score more useful than the general score? What about relative to undergraduate GPA? Similar questions apply to the SAT and undergraduate admissions.

In both cases the answer is that standardized tests have roughly as much predictive power as GPA (SAT is about as powerful as HS GPA; GRE similar to undergraduate GPA). Not bad for a brief test! When these factors are combined the overall predictive power is increased. My opinion is that standardized tests load more heavily on cognitive ability and less on conscientiousness relative to course grades, hence the non-redundant information in the two measures.



This is a meta-analysis of GRE validity by Kuncel and collaborators. Here are slides from a related talk. See also this blog discussion (social scientists) of the GRE.

I first came across Kuncel's work when I was investigating SAT validity using U Oregon undergraduate data. Unlike many of the people working on this topic, Kuncel is neither an idiot nor an ideologically motivated faker. For example, he understands issues like restriction of range, variation in course and subject difficulty, and self-selection effects. See these slides from his talk on SAT validity. Video and audio here (bonus: Thomas Espenshade; Kuncel starts at 27 minutes in).




Validities for various standardized tests:

29 comments:

  1. So why do most science&Engineering Phd programs rely more on research experience and recommendations,and not GRE  scores?

    ReplyDelete
  2. Anonymous_IV3:28 PM

    Because the candidates are above the range where the GRE-Q (which is the only GRE part relevant to STEM) can distinguish among them.

    ReplyDelete
  3. Anonymous_IV3:32 PM

    Because those programs must distinguish among candidates that are above the level that the GRE-Q (the only GRE section that matters for STEM) can separate.

    ReplyDelete
  4. Anonymous_IV3:34 PM

    Because those programs must distinguish among candidates that are above the level that the GRE-Q (the only GRE section that matters for STEM) can separate.

    ReplyDelete
  5. Anonymous_IV3:39 PM

    [S.Hsu -- please remove the duplicate (and this message).  Sorry.]

    ReplyDelete
  6. MtMoru3:50 PM

    Following that logic you'll have to somehow subtract out the effect of institutional prestige on  career outcomes. Better test scores gets you into a better school.

    ReplyDelete
  7. MtMoru3:54 PM

    Because:
    1) they don't know what they're doing
    2) STEM academics are still academics and far left of center just not as far as non-STEM
    3) they themselves may not have have scored that high.

    ReplyDelete
  8. Grad admission people always tell me that the important criteria are (in order),

    1. Proven research experiences
    2. Undergrad recommendations
    3. Some proof of ability to write competent English
    4. Undergrad GPA

    They claim not to even know why GRE scores are requested since they don't count.

    ReplyDelete
  9. MtMoru3:58 PM

    "works to end the misuses and flaws of standardized testing and to ensure that evaluation of students, teachers and schools is fair, open, valid and educationally beneficial"

    I suppose Steve might (fairly) paraphrase this: "works to assure that everyone gets the same score or at least that there are no big differences between ethicities or the sexes".

    ReplyDelete
  10. Maybe.  In our case, the median score of applicants who make our short list is 800, so the test isn't all that helpful to us.  And if we limited ourselves to 800s, we wouldn't eliminate many applicants.  Maybe it means the test is useful, but it's value in the right tail is pretty limited:  we look at things like what courses applicants have taken, whether they have research experience.  What we really want is passion, but that's hard to measure.  

    ReplyDelete
  11. MtMoru7:24 PM

    You look only at the quantitative section? Did you look at the analytic section (logic problems) in the old days? You don't look at the subject test?

    ReplyDelete
  12. MtMoru7:27 PM

    Is that the GMAT or the GRE? I thought perfect GMATs were quite rare, like 50 a year.

    ReplyDelete
  13. botti7:42 PM

    Speaking of ideological fakers, SJ Gould has been badly exposed.

    http://johnhawks.net/weblog/topics/meta/gould-morton-lewis-2011.html

    ReplyDelete
  14. Allen Knutson8:58 PM

    It's hard to ensure there being no big differences between ethicities, when the unethical people cheat.

    ReplyDelete
  15. Umm so is 0.37 considered good? Confused there!

    ReplyDelete
  16. See pictures here or in my data mining paper to get an idea of what correlation of .37 looks like. For social science it's considered a relatively strong correlation.

    http://en.wikipedia.org/wiki/Correlation_and_dependence

    ReplyDelete
  17. Why then cant they use the undergraduate equivalent of an academic index,finding a composite of all the three sections with allowance for those who come from non english speaking countries?Wont that be a better predictor than research experince?

    ReplyDelete
  18. I wonder how much the hormone testosterone affects these types of scores: http://www.nytimes.com/2000/04/02/magazine/the-he-hormone.html

    ReplyDelete
  19. ohmyohmy7:41 PM

    There are times when GPA and SAT/GRE conflict; in such cases, which is a better predictor of success?
    ie, what happens when you have low GPA and high SAT/GRE (or vice versa)?

    ReplyDelete
  20. Maria Smith12:23 AM

    Gre exam is very important so students must keep in mind
    that only hardworking can help for passing it

    gre practice

    ReplyDelete
  21. Shan Rose1:00 PM

    Gre exam is very important so
    students must keep in mind that only hardworking can help for passing it

    gre score

    ReplyDelete
  22. Martha Victor2:54 AM

    Another interesting thing about the GRE. Most colleges don't care! When I called one of my prosepective graduate schools they said that the GRE is usually the last thing they look at when deciding whether or not to accept you

    gre practice study

    ReplyDelete
  23. I have been obsessed with IQ analysis for the past 1 year. I think using SAT or GRE score for IQ measurement is total nonsense. The math section is ridiculously easy and measures how well you have been trained in specific techniques you learn in high school. India and China perform exceedingly well since engineering is given extreme focus in both the countries and hence most of students get trained very well in math techniques.

    ReplyDelete
  24. Also Indian and Chinese students put in a lot of hours getting trained in those specific techniques. Americans on the other hand enter all kinds of fields like english, psychology, business, history and so on and also do not tend to put much emphasis on math skills and also do not put much emphasis on education in general. Chinese Students have an average GRE score of 780. Does it mean that all the students coming in have an IQ of 145 (3SD above the mean)? I have seen many of my indian friends (comsci grads) who cannot even write a single program and who fail out of most of their college courses score 740-750 without any prep. The SAT and GRE scores should not be taken seriously as told by prominent thinkers like Taleb.

    ReplyDelete
  25. I think the SAT english section is a far more better indicator of general intelligence as compared to the verbal section (of course only for people having a strong english background).

    ReplyDelete
  26. Both the GRE scores I have mentioned below (780,740) are the math scores and I got the data from the ETS website itself. I have even seen your slides and like the BGI research program but I think taking the GRE or SAT scores is a big mistake. The math score should not be a factor at all for the input set of students and the verbal score cannot be a factor because of the cultural differences among students. I have seen many Indian Phd students struggling to score even 400 in verbal and I have seen all the exceptional Chinese students ( they dominate the grad college in all the best universities) failing their english courses and many Chinese lectures who have published hundreds of papers still have trouble string a sentence together in their college lectures.

    ReplyDelete
  27. 100 Bricks12:20 PM

    Thanks for sharing such a nice information . you can also visit GRE Verbal Course

    ReplyDelete
  28. Thanks for sharing such a nice information . you can also visit. GRE Verbal Course

    ReplyDelete
  29. Thanks for sharing such a nice information .you can also visit GRE Verbal Course

    ReplyDelete