Teacher Education and Development Study in Mathematics - IEA

PART 1: TEACHER RELATIVE PAY AND STUDENT PERFORMANCE
39
Table 5: PISA 2003, Lowest and Highest Test Scores as Function of Income Distribution and GDP per Capita
Dependent Variable
Variable Score Diff. of Mean Score of Mean Score of Mean Score of
Highest 10% to Lowest 10% Lowest 25% Highest 10%
Lowest 10% Relative of Test Scorers of Test Scorers of Test Scorers
to Mean Score
Gini Coefficient 0.002** -2.917*** -3.148*** -3.904***
(x 100) (2.85) (-4.03) (-3.96) (-3.92)
GDP per Capita -0.0000024*** 0.0017*** 0.0018*** 0.0017**
(-3.40) (3.06) (2.90) (2.19)
Intercept 0.52 416.4 485.3 714.9
Adjusted R 2 0.41 0.53 0.52 0.46
N 36 36 36 36
Note: **Statistically significant at 5% level of significance; ***statistically significant at 1% level.
Source: OECD PISA 2003.
Student performance and income distribution within countries
We now turn to similar estimates across states within the same country. We use two
examples—Mexico and the United States.
Mexico
During its participation in PISA 2003, Mexico drew random samples of 15-year-olds
in each of 31 (of the 32) states that took part in the study. (The data from one state,
Michoacan, could not be collected in that year.) As was the case in the international
survey across countries, PISA test results for public school students across states
correlated negatively and significantly with the Gini coefficient of income distribution
in each state, even when GDP per capita in the state was controlled for (Table 6). When
we added other controls for the percentage of the age cohort in each state attending
higher secondary school and the percentage of those in the PISA sample in each state
attending higher secondary school, the coefficient of the Gini was smaller but still
statistically significant at the 5% level.
As was the case with the cross-national estimates, the independent variables explained a
relatively high fraction of the total variance in test scores across states. The coefficients
for the percentage of the age cohort in upper-secondary school and for the percentage
of 15-year-olds taking the PISA tests and attending upper-secondary school (as opposed
to still being in middle school) were what we had predicted (negative for the percentage
attending higher secondary in each state and positive for the percentage of the PISA testtakers
in each state attending higher secondary), and both were statistically significant.
Thus, the states in which a relatively high percentage of 15-year-olds were attending
school were likely to be the states where students had the lower PISA scores, other
factors being equal. Also, the states where a higher percentage of 15-year-olds who took
the PISA test were in Grade 10 rather than in Grade 9 were likely to be the states with
students who gained the higher scores on the PISA tests, all other factors being equal.