Effect of School Poverty on Racial Gaps in Test Scores: The Case of the Minnesota Basic Standards Tests, The

Journal of Negro Education, The, Winter 2004 by Myers, Samuel L Jr, Kim, Hyeoneui, Mandala, Cheryl

1. Individual-specific

a. Race (Black; left-out dummy is White)3

b. Gender (female; left-out dummy is male)

c. Age (in months)

d. Attendance rate (percent of membership days attended)

e. Mobility (number of school changes in past year)4

f. English not spoken at home (dichotomous variable)

g. Individual poverty (receives free or reduced price lunch; not included in 1996 analysis)

2. Program-specific

a. Disability/Special Education (enrolled in Special Education or Disability program)

b. Limited English Proficiency (enrolled in LEP program)

c. Gifted and Talented (enrolled in Gifted and Talented program)

3. School-specific

a. School poverty (percent of students receiving free/reduced price meals)

b. Percent Black in school (percent of all test takers at school who are Black)

c. Percent Asian in school (percent of all test takers at school who are Asian)

d. Percent Hispanic American in school (percent of all test takers at school who are Hispanic American)

e. Percent Native American in school (percent of all test takers at school who are Native American)

f. Top quintile school-math (a dummy variable indicating whether a school ranked among the top 20% of math scores)

g. Top quintile school-reading (a dummy variable indicating whether a school ranked among the top 20% of reading scores)

h. Bottom quintile school-math (a dummy variable indicating whether a school ranked among the bottom 20% of math scores)

i. Bottom quintile school-reading (a dummy variable indicating whether a school ranked among the bottom 20% of reading scores)

RESULTS

Table 1 reports the estimated coefficients on the school poverty variable on mathematics and reading test scores. The estimated coefficients are obtained from data for 1996, 1998, and 1999. Estimates are included for models that take account of both school poverty and individual poverty (for 1998 and 1999) as well as for school poverty alone; this is to gauge the impacts of possible confounding influences of school and individual poverty measures. Estimates are obtained using Ordinary Least Squares (OLS) as well as using Hierarchical Linear Modeling (HLM).5 To create greater comparability between the OLS and HLM results, we also estimated the OLS model with interaction terms between school poverty and all individual level variables in the model. To expand the logic of using HLM, we also included interaction terms between all school level variables and individual level variables. Thus, Table 1 reports the results of estimation of 80 different regression equations, 40 each for Black and White students, 20 for each race for mathematics, and 20 for reading. Of the 20 regressions for mathematics or reading, 5 have no interaction terms, 10 have interaction terms with school poverty, and 5 have interaction terms with all school level variables. Half of the estimations use OLS and half use HLM methods. In short, Table 1 represents a comprehensive summary of impacts of school poverty on eighth-grade MBST scores across recent years.

Each entry in the table represents the effect of school poverty on test scores. The first two columns of the table report the estimated coefficients on the school poverty variable obtained from the linear regression equations using the OLS method. The entry is interpreted as the change in the test score that accompanies a one-percentage point increase in school poverty. Thus, the first cell for Black mathematics examination in 1996 has the value of .1401, which is statistically significant at the 5% level. T-statistics are reported in parentheses below the cell entry, and whenever an entry is statistically significant at the 5% level, we have presented the entry in bold type.