International Comparison of Top Students

Dr. Dwight Galster

September 17, 2005

I found an article by Chris Correa called "More on the Elite Myth" after reading a post by Jenny D, "Our Kids Aren't As Good As Their Kids." It's about international comparisons and how our top students compare.

I've given this issue some thought, as well. Beginning with anecdotal evidence, it certainly seems that our foreign college students (primarily Asian but also notably some "Eastern Block" European) in Math and Engineering tend to outperform the best local students mathematically. It is also evident that the winners of competitions in Math and Science are mostly first or second generation Asian immigrants.

This topic really deserves a closer look, beyond looking at the 95%ile. I am concerned that the percentile may not be the best way to determine what we want to know. Ideally, I would like to know the 95%ile score for all countries combined, then the percentage of each country's students above that score. In addition, it seems important to know more precisely the shape of the upper tail of the distribution. Having information on more percentiles would be very useful.

My study of this begins with the TIMMS (2003) rather than the PISA (2003). TIMMS tested 4th and 8th graders (I am only looking at 8th grade results), while PISA tests 15-year olds, whose grade levels ranged from 7 to 11. The TIMMS reports data for 46 countries, while PISA reports data for 39, of which only 17 are common to both studies. Thus the two studies, taken together, include 68 countries. TIMMS includes subscales in five content areas, Number, Algebra, Measurement, Geometry, and Data. PISA sub-divides by Literacy and Problem Solving. The literacy scale is further divided into four subscales for Quantity, Space and Shape, Change and Relationship, and Uncertainty. There does not appear to be a direct mapping between the categories used in the two studies. PISA also provides percentages of students in six achievement levels for each subscale, giving us another way (besides percentiles) to compare our elite students.

In the TIMMS study, 20% of the countries have mean combined scores higher than the US, while 25% of the countries have a higher 95%ile on the combined scale. This suggests we are losing ground as we get into the higher ability levels. The US mean is 504, while Singapore holds the top position with 605. We could say that Singapore's mean is 20% higher than ours, but since I don't know what kind of scale these scores come from, I hesitate to try to interpret that. (Another table shows that on average Singapore's students got 45% more questions right). Looking at the 95%ile, the US has 635, while the top country (Chinese Taipei) is at 733. Again, we could say that is about 15% higher than ours, but interpretation is problematic. It would be much more useful to know what percentage of Chinese students scored above 635, or alternatively, what percentage of US students scored above 733. In the subscale means, the US is 12th in Number, 11th in Algebra, 21st in Measure, 24th in Geometry, and 13th in Data.

In the PISA study, 28 (72%) of the countries have mean literacy scores higher than the US, while 26 (67%) of the countries have a higher 95%ile. The US mean is 483, while Hong Kong holds the top position with 550 (14% higher). Looking at the 95%ile, the US has 638, while the top country (Hong Kong) is at 700 (10% higher). The same problems with interpretation apply as discussed above. However, when we look at the percent of students who achieved a Level 6 proficiency in Literacy, the picture is much clearer and more shocking. In the US, only 2% of the students achieved this level. In Hong Kong, it was 10.5%. Not only that, but there are 16 countries that have 4% or more in this group (double or more the US figure). Canada and Australia have almost three times as many as the US, and Korea, Japan, and Belgium have four times as many. This is strong evidence that we are failing to provide an adequate math education for our best students.

Further research: I would like to try to combine data from the two studies, and also look at comparisons at the lower end of the distribution.

 

International Comparisons II

Dr. Dwight Galster

November 27, 2005


I have obtained the PISA 2003 data. Analysis of PISA data is a daunting task. Before even beginning that analysis, one must study the technical theory and philosphy of the PISA, which I can assure you, is anything but simple or straightforward. I still don't understand it fully, but I have been able to determine how the national scores are derived, and thus can study in more detail the distribution of the upper tail, as I proposed doing in the previous article.

The concern we are addressing here, is that not only is the USA falling behind in the average performance of students in Math, but that our "brightest and best" are falling behind as well. My studies of the upper tail of the distributions support that hypothesis.

I calculated the percentile scores of all countries in the PISA survey for the following percentiles: 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 99.5, 99.75, and 99.9. (PISA reports mainly on 38 countries. There were actually 41 country codes, which include regions such as Hong Kong. In this study, I used all of them, and refer to them as countries for convenience.) Then for each percentile score thus obtained, I found the nearest percentile (to the tenth's place) for that score in each country's data. Thus, I could calculate the estimated percent of students in each country who exceeded those levels. I checked so many levels because I wanted to know if there were any differences in the shape of the upper tail between countries. However, the differences revealed by PISA can be adequately summarized by looking at just the 90th and 95th percentiles.

The 90th percentile score of all surveyed countries is 604 (hereafter "P90"), and the 95th percentile score is 640 (hereafter "P95"). The US students track very close to the international profile all along the scale. Eleven percent of US students exceed P90, and 4.8% exceed P95. By contrast, the top ranked country (Hong Kong) has 31.8 percent above P90, and 18.4% above P95. What this means, is that when we look at the top 10% of math students in the "world" (where "world" means those countries for which the survey is representative), the US is contributing a slightly "above average" share (11%), but Hong Kong contributes three times the "average" share. When we look at P95, the US is a shade "below average," while Hong Kong is producing three and a half times as many at that level. Even Canada produces more than twice as many students at both the P90 and P95 level as the world average.

Country        %0ver P90  %Over P95
HongKong           31.8      18.4
Belgium            27.5      16.0
Netherlands        27.3      14.8
Korea              25.8      14.5
Liechtenstein      25.6      14.9
Japan              25.1      14.2
Finland            24.4      12.9
Switzerland        22.4      12.2
NewZealand         21.9      12.4
Canada             21.3      11.0
Australia          20.6      11.0
Macao              18.7       9.9
CzechRep           18.6      10.1
Germany            17.0       8.2
Sweden             17.0       8.7
Denmark            16.9       8.2
Iceland            16.3       7.3
France             15.7       7.9
Austria            15.3       7.4
UK                 15.2       7.8
Slovokia           12.9       6.4
Ireland            12.0       5.4
Norway             12.0       5.5
Hungary            11.4       5.1
USA                11.0       4.8
Luxembourg         11.0       5.1
Poland             10.5       4.7
Spain               8.9       3.4
Latvia              8.4       3.5
Russia              7.8       3.5
Italy               7.5       3.6
Portugal            5.8       2.2
Turkey              5.8       3.9
Greece              4.4       1.9
Uruguay             3.0       1.2
Yugoslavia          2.5       1.0
Thailand            1.8       0.6
Brazil              1.1       0.6
Mexico              0.4       0.1
Indonesia           0.3        -
Tunisia             0.2        -