Poverty: Its effects on student academic achievement Abstract This study examines the impact concentrating poverty on a school level has on students academic outcomes. Furthermore, it looks at which group of students are affected the most: above average, average or below average students. The methods used are a compilation of bivariate correlation analysis, partial correlation analysis and multivariate linear regression. Statistical significance for all variables is at the .05 level. Three models of poverty are created and thus analyzed for their effect on academic achievement. The conclusions drawn from this study reveal that concentrating poverty does impact student academic outcomes and that below average students are affected worst by concentrating poverty in a school. Literature Review Does having a higher percentage of low-income students in a school reduce learning? If it does, is the reduction greater for Brisson Page 1
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Poverty: Its effects on student academic achievement
Abstract
This study examines the impact concentrating poverty on a school level has on students
academic outcomes. Furthermore, it looks at which group of students are affected the most:
above average, average or below average students. The methods used are a compilation of
bivariate correlation analysis, partial correlation analysis and multivariate linear regression.
Statistical significance for all variables is at the .05 level. Three models of poverty are created
and thus analyzed for their effect on academic achievement. The conclusions drawn from this
study reveal that concentrating poverty does impact student academic outcomes and that
below average students are affected worst by concentrating poverty in a school.
Literature Review
Does having a higher percentage of low-income students in a school reduce learning? If
it does, is the reduction greater for above-average, average, or below-average students? A few
strands of research have been done that shed light on this question. They include peer effects,
school choice, the achievement gap, and to a lesser degree the issue of the segregation of
schools.
Brisson Page 1
The efficiency and production of any industry relies on understanding and analyzing its
production functions. That is, the inputs that go into manufacturing the product and what kind of
product these inputs produce (outputs). The educational field is no different and the term is
known as education production functions. As prominent school policy researcher Eric Hanushek
notes, “Student achievement at a point in time is related to the primary inputs: family influences,
peers, and schools” (Hanushek 1986). Schools want to be as cost effective as possible to
produce the greatest outputs. In the case of the educational industry then, the outputs are student
achievement. It is interesting to note however, that some people, including many in the
educational field, don’t accept this type of research because they don’t believe that “educational
outcomes…cannot be adequately quantified” (Hanushek 1986). Indeed, it seems the jury
remains out on the strength of the relationship between test scores and students’ achievements in
the labor market (Hanushek 1986). The author notes that while most studies use standardized
tests to measure output, some have employed other methods, including “student attitudes, school
attendance rates, and college continuation or dropout rates” (Hanushek 1986).
It is hard to answer questions about student performance because many variables
influence results and many of these variables can be hard to separate from each other and
carefully control. Indeed, Anita Summers and Barbara Wolfe suggest this in their 1977 study
saying, “Casual observation, combined with the education literature, suggests that achievement is
a function of a student’s hard-to-disentangle genetic endowments and socioeconomic status”
(Summers and Wolfe 1977). For the purposes of our question we will examine peer effects and
family influences.
First, do peer effects actually exist? While substantial research exists on peer effects, it is
not conclusive on what the effects actually are. One such study examined how much of a
Brisson Page 2
difference schools make in a student’s education (Summers andWolfe 1977). The conclusion
was a good amount. Using inputs of family, peer effects and school inputs, Summers and Wolf
found that “Black and non-black students benefitted, via most improved achievement, when they
were in schools with a 40-60% black student body rather than in schools that were more racially
segregated”(Summers/Wolfe 1977). The study also concluded that after accounting for
interactions between school inputs, income and race, “no residual impact of income on
Class size Pearson Correlation .291** -.071** .042** 1
Sig. (2-tailed) .000 .000 .001
N 5783 5783 5765 5783
**. Correlation is significant at the 0.01 level (2-tailed).
Brisson Page 11
Table 1 continued:
AvgEOG8th7thm
ath
% Anglo-Asian in
school
AvgEOG8th7thmath Pearson Correlation 1 .484**
Sig. (2-tailed) .000
N 5783 5783
% Anglo-Asian in school Pearson Correlation .484** 1
Sig. (2-tailed) .000
N 5783 5783
**. Correlation is significant at the 0.01 level (2-tailed).
These four variables were the most statistically significant of my control variables found that
were pertinent to academic performance. The variable “Student has an individualized
education plan” was found to be more statistically significant than “Teacher experience,” but
there were so few cases in the sample it did not warrant inclusion.
Results for Hypothesis 1) “In a comparison of students, those who have more days absent will
have lower academic outcomes.”
Table 2: Number of Days Absent (Partial Correlation with controls)
Brisson Page 12
Correlations
Control Variables
AvgEOG8th7thm
ath
Number of
School days
Absent
Years of Teacher Experience
in District & Class size & %
Anglo-Asian in school
AvgEOG8th7thmath Correlation 1.000 -.272
Significance (2-tailed) . .000
df 0 5760
Number of School days
Absent
Correlation -.272 1.000
Significance (2-tailed) .000 .
df 5760 0
After controlling for the control variables, statistical significance still remains. Linear regression
will provide a clearer picture of the strength, direction and variance explained of the
independent variable on the dependent variable.
Table 3: Number Days Absent (Regression)
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .286a .082 .082 8.82338
a. Predictors: (Constant), Number of School days Absent
Brisson Page 13
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 362.530 .152 2378.482 .000
Number of School days
Absent
-.308 .014 -.286 -22.702 .000
a. Dependent Variable: AvgEOG8th7thmath
Our results show that 8.2% of end of grade math test scores are affected by the number of days
a student misses and the relationship is negative as predicted. The variable remains statistically
significant. While 8.2% may not seem a lot, we must remember that many variables will
interact to affect academic outcomes of a student within his/her school environment. The null
hypothesis is rejected.
Results for Hypothesis 2) “In a comparison of students, those who are taught by a more
experienced teacher from their district will perform better academically.”
Table 4: Teacher experience in district (Partial Correlation with controls)
Correlations
Control Variables
AvgEOG8th7thm
ath
Years of Teacher
Experience in
District
Class size & % Anglo-Asian
in school & Number of
School days Absent
AvgEOG8th7thmath Correlation 1.000 .078
Significance (2-tailed) . .000
df 0 5760
Years of Teacher Experience
in District
Correlation .078 1.000
Significance (2-tailed) .000 .
df 5760 0
Brisson Page 14
We can see that after being controlled for, the “teacher experience in district” variable shows a
rather paltry statistical significance. None the less, we will continue to examine its impact on
academic performance.
Table 5: Teacher experience in district (Regression)
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .151a .023 .023 9.08236
a. Predictors: (Constant), Years of Teacher Experience in District
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 358.559 .189 1898.386 .000
Years of Teacher Experience
in District
.158 .014 .151 11.579 .000
a. Dependent Variable: AvgEOG8th7thmath
By only explaining 2.3% of the variance academic outcomes, the impact is negligible, yet
remains statistically significant. The table shows that for every percentage point increase in
years of teacher experience, a students’ test scores will have gone up only 1/10 and a half of a
percent. The relationship was positive as predicted and the null hypothesis is rejected.
Results for Hypothesis 3) “In a comparison of students, those in larger classes will have lower
academic outcomes when compared with students in smaller classes.”
Brisson Page 15
Table 6: Large class size (Partial Correlation)
Correlations
Control Variables
AvgEOG8th7thm
ath Class size
Number of School days
Absent & % Anglo-Asian in
school & Years of Teacher
Experience in District
AvgEOG8th7thmath Correlation 1.000 .165
Significance (2-tailed) . .000
df 0 5760
Class size Correlation .165 1.000
Significance (2-tailed) .000 .
df 5760 0
The variable remains statistically significant after inserting appropriate controls.
Table 7: Large class size (Regression)
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .291a .085 .085 8.80939
a. Predictors: (Constant), Class size
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 354.134 .290 1221.126 .000
Class size .200 .009 .291 23.139 .000
a. Dependent Variable: AvgEOG8th7thmath
Regression analysis reveals that class size explains 8.5% of the variance in EOG test scores.
Interestingly, the relationship is positive, not as predicted, with a percentage increase in class
Brisson Page 16
size being equivalent to a 0.2 percentage point increase in test score. The null hypothesis is
rejected.
Results for Hypothesis 4) “In a comparison of students, having a higher percentage of
Anglo/Asian students in a school will have a positive outcome on student academic
performance.”
Table 8: %Anglo/Asian in school (Partial Correlation)
Correlations
Control Variables
AvgEOG8th7thm
ath
% Anglo-Asian in
school
Number of School days
Absent & Years of Teacher
Experience in District & Class
size
AvgEOG8th7thmath Correlation 1.000 .411
Significance (2-tailed) . .000
df 0 5760
% Anglo-Asian in school Correlation .411 1.000
Significance (2-tailed) .000 .
df 5760 0
This variable proves itself to be the strongest of the control variables.
Table 9: %Anglo/Asian in school (Regression)
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .484a .234 .234 8.05954
a. Predictors: (Constant), % Anglo-Asian in school
Brisson Page 17
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 353.846 .186 1898.937 .000
% Anglo-Asian in school .166 .004 .484 42.017 .000
a. Dependent Variable: AvgEOG8th7thmath
The regression model shows that 23.4% of the variance within test scores can be attributed to
the percentage of Anglo and Asians in a school. The relationship is positive. However, it is
interesting to note that while the variance prediction is large, the actual increase in test scores
is only 0.17 of a percentage point per percentage point of Anglo/Asian added to the school.
Class size actually did a better job in increasing test scores while explaining almost 2/3 less of
the variance. The null hypothesis is rejected.
When the statistically significant variables are ran together in a regression we see that
31.6% of the variance is explained. This does not add up to the individual variance percentages
obtained using the one variable regressions but demonstrates a good amount of explanatory
power with %Anglo/Asian in a school doing most of the variance work. It is possible to explain
the loss in variance explanation when considering the unstandardized coefficients. In the
multivariate regression, a students’ class size’s ability to improve scores dropped by over 1/3
and the years a teacher was experienced variables’ ability to improve scores dropped by almost
2/3. This may suggest the overarching importance of the number of days absent and
%Anglo/Asian in a school as bearers on test scores and controls in the study.
Table 10: Control Variables’ Regression
Brisson Page 18
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .562a .316 .315 7.60157
a. Predictors: (Constant), Number of School days Absent, Years of
Teacher Experience in District, Class size, % Anglo-Asian in school
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 352.920 .300 1174.885 .000
% Anglo-Asian in school .138 .004 .403 34.255 .000
Years of Teacher Experience
in District
.069 .012 .066 5.957 .000
Class size .100 .008 .146 12.682 .000
Number of School days
Absent
-.253 .012 -.235 -21.489 .000
a. Dependent Variable: AvgEOG8th7thmath
Having established statistically significant control variables to run with our regressions, I will see
how well African American and Latino students are correlated with poverty in order to further
test whether poverty affects learning in school. It should be noted however that the variable
%Anglo/Asian students in a school cannot be included in the regressions testing different
measures of poverty and their effects on test scores. While it is indeed a very statistically
significant variable, it is too correlated and thus will disrupt the data from giving an accurate
representation of its findings.
Table 11: Correlation between %FRL, %African American and %Latino in school
Brisson Page 19
Correlations
% FRL in School
in year 2008
% Latino in
school in 2008
% African-
American in
school in 2008
% Anglo-Asian in
school
% FRL in School in year
2008
Pearson Correlation 1 .634** .771** -.855**
Sig. (2-tailed) .000 .000 .000
N 5783 5783 5783 5783
% Latino in school in 2008 Pearson Correlation .634** 1 .334** -.611**
Sig. (2-tailed) .000 .000 .000
N 5783 5783 5783 5783
% African-American in school
in 2008
Pearson Correlation .771** .334** 1 -.950**
Sig. (2-tailed) .000 .000 .000
N 5783 5783 5783 5783
% Anglo-Asian in school Pearson Correlation -.855** -.611** -.950** 1
Sig. (2-tailed) .000 .000 .000
N 5783 5783 5783 5783
**. Correlation is significant at the 0.01 level (2-tailed).
Strikingly, the %Anglo-Asian has a more negative correlation with %FRL than %African American
or %Latino has a positive correlation with %FRL. Regardless, all are very highly correlated with
%FRL.
Results for Hypothesis 5) “In a comparison of students, those who attend schools having a
higher percentage of African Americans in the school reduces learning.” Running a linear
regression analysis reveals exactly a 30% explanation of variance of the dependent variable.
We find that %African American does have statistical significance when run with our control
variables. Looking at the Unstandardized B Coefficients it shows that for every percentage
point change in %African American in the school, tests scores will be affected by 0.16%.
Brisson Page 20
Demonstrating this via a mathematical representation we find that the effect equation is as
follows: 362.4 (mean test score of dependent variable)-(.5*.16) = 362.32. The .5 represents a
school that is 50% African American. This regression equation says that with this school
composition a score would drop only 0.08 of a point, which is paltry. The relationship is
negative and the null hypothesis is rejected.
Table 12: %African American in school (Linear Regression)
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .548a .301 .300 7.68537
a. Predictors: (Constant), % African-American in school in 2008,
Number of School days Absent, Years of Teacher Experience in
District, Class size
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 364.575 .421 865.804 .000
Class size .113 .008 .164 14.169 .000
Years of Teacher Experience
in District
.068 .012 .065 5.794 .000
Number of School days
Absent
-.249 .012 -.232 -20.907 .000
% African-American in school
in 2008
-.155 .005 -.378 -31.984 .000
a. Dependent Variable: AvgEOG7th8thmath
Brisson Page 21
Results for Hypothesis 6) “In a comparison of students, those who attend schools having a
higher percentage of Latinos in the school reduces learning.” Running a linear regression to
test the hypothesis it is found that 22.1% of the variance in test scores is explained through this
model. This finding reiterates the findings of table 11 which showed a strong correlation
between %FRL, %African American in school, and %Latino in school, but showed a stronger
correlation between the first two. While the model is less predictive in its variance, it shows
%Latino’s in school to actually have a more negative impact on test scores than did %African
Americans in school (Unstandardized B Coefficient). The relationship is negative and the null
hypothesis is rejected.
Table 13: %Latino in school (Linear Regression)
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .470a .221 .221 8.11012
a. Predictors: (Constant), % Latino in school in 2008, Number of School
days Absent, Years of Teacher Experience in District, Class size
Brisson Page 22
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 359.508 .399 902.151 .000
Class size .152 .008 .222 18.458 .000
Years of Teacher Experience
in District
.134 .012 .128 10.983 .000
Number of School days
Absent
-.282 .013 -.263 -22.541 .000
% Latino in school in 2008 -.223 .012 -.218 -18.197 .000
a. Dependent Variable: AvgEOG7th8thmath
Results for Hypothesis 7) “In a comparison of students, those who attend schools having a
higher percentage of students on free and reduced lunch in the school reduces academic
performance in the school.” Running a linear regression this model predicts a 27% explanation
of the variance in the dependent variable, while accounting for the least amount of change in
test scores of the poverty models tested. The relationship is negative and the null hypothesis is
rejected.
Table 14: %FRL (Linear Regression)
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .520a .271 .270 7.84828
a. Predictors: (Constant), % FRL in School in year 2008, Number of
School days Absent, Years of Teacher Experience in District, Class
size
Brisson Page 23
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 362.496 .409 887.260 .000
Class size .125 .008 .182 15.424 .000
Years of Teacher Experience
in District
.119 .012 .113 9.994 .000
Number of School days
Absent
-.265 .012 -.246 -21.798 .000
% FRL in School in year 2008 -.114 .004 -.322 -27.283 .000
a. Dependent Variable: AvgEOG7th8thmath
Results for Hypothesis 8) “In a comparison of students, having a higher percentage of low
income students in a school affects lower performing students more than higher performing
students.” To test this hypothesis I first divided the dependent variable into three categories,
as shown in table 15. These formed cutoff points to enable labeling students in below average,
average, or above average test score categories. I then ran three separate regressions with
each category to see which was affected the most by the independent variable (%FRL). For
above average and average students’ regressions, 5.3% and 5% variance explained were
respectively shown. The above average students test scores were affected almost twice as
much as the average students test scores however. Most noteworthy was that the below
average students test scores were affected the most by having a larger %FRL students in class.
12.3% of the variance is explained and the test scores are four times worse than the above
average students scores and just over seven times worse than the average students scores.
Table 15: Frequency Analysis with 3 cut points (EOG 7th/8th math scores average)
Brisson Page 24
Statistics
AvgEOG7th8thmath
N Valid 5783
Missing 0
Percentiles 33.33333333 355.5000
66.66666667 364.5000
Table 16: %FRL Above Average Students change in test scores
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .235a .055 .053 3.46201
a. Predictors: (Constant), Number of School days Absent, % FRL in
School in year 2008, Class size, Years of Teacher Experience in
District
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 352.333 .276 1274.509 .000
% FRL in School in year
2008
-.016 .003 -.110 -5.048 .000
Class size -.006 .007 -.019 -.879 .379
Years of Teacher Experience
in District
-.010 .010 -.022 -1.029 .303
Number of School days
Absent
-.064 .007 -.204 -9.454 .000
a. Dependent Variable: AvgEOG7th8thmath
Table 17: %FLR Average students change in test scores
Brisson Page 25
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .228a .052 .050 2.53555
a. Predictors: (Constant), Number of School days Absent, % FRL in
School in year 2008, Years of Teacher Experience in District, Class
size
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 359.691 .232 1550.339 .000
% FRL in School in year
2008
-.009 .002 -.092 -3.989 .000
Class size .031 .005 .152 6.597 .000
Years of Teacher Experience
in District
.018 .006 .063 2.809 .005
Number of School days
Absent
-.035 .009 -.089 -3.966 .000
a. Dependent Variable: AvgEOG7th8thmath
Table 18: %FRL Below Average students change in test scores
Model Summary
Model R R Square
Adjusted R
Square
Std. Error of the
Estimate
1 .353a .125 .123 4.21883
a. Predictors: (Constant), Number of School days Absent, Years of
Teacher Experience in District, Class size, % FRL in School in year
2008
Brisson Page 26
Coefficientsa
Model
Unstandardized Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant) 374.404 .453 827.122 .000
% FRL in School in year 2008 -.064 .005 -.306 -12.627 .000
Class size -.028 .008 -.086 -3.596 .000
Years of Teacher Experience
in District
.056 .011 .113 5.058 .000
Number of School days
Absent
-.132 .020 -.144 -6.578 .000
a. Dependent Variable: AvgEOG7th8thmath
Conclusion
In conclusion, my study showed a correlation between many variables and their effects on
student end of grade test scores. I attempted to build up three models of poverty. One,
percent of students on free and reduced lunch, is the purest model because there is a perfect
correlation between poverty and this measure. The other two models consisted of ethnicities
(African American / Latino) generally ascribed higher percentages of poverty per their
populations. The three models together described a range of variance of test scores from
22.1% to 30.1%. Interestingly, it was the percentage of African Americans in class that
described the most variance, indicating, as noted in the literature review (Brown-Jeffy, S. 2009),
that racial composition has its own role to play in expressing student academic achievement
beyond just a rank of poverty. The fact that the percentage of free and reduced lunch in school
described the least amount of test score change, while the percent of Latino’s in school
described almost twice as much reiterates this fact.
Brisson Page 27
My study also revealed that below average students are hurt the worst when placed in
classrooms with increasing percentages of low income students. In fact, the variance described
between the effect of higher percentages of poverty concentrated in a classroom and below
average students was over two and a half times greater than either above average or average
students. It is beyond the scope of this paper to describe why this situation is occurring.
However, explanations such as lower income students on average do not receive much home
environment educational support and thus when interacting with a large group also not seeing
education as being valued will only reinforce this attitude. Also, teachers are just as susceptible
as anyone else in society to the negative connotations associated with poverty and academic
abilities. It is plausible teachers simply do not expect as much from lower income students and
those students will quickly pick up on that attitude and internalize it.
Bibliography
Brown-Jeffy, S. (2009). School Effects: Examining the Race Gap in Mathematics Achievement. Journal of African American Studies. Vol. 13 (4). Pp.388-405
Coleman, J.S., & others. (1968). Equality of Educational Opportunity. Equity and Excellence in Education. Vol.6 (5). Pp.19-
Godwin, K., & Leland, S.M., & Baxter, A.D., & Southworth, S. (2006). Sinking Swann: Public School Choice and the Resegregation of Charlotte’s Public Schools. Review of Policy Research. Vol.23 (5). Pp. 983-997.
Hanushek, E.A. (1986). The Economics of Schooling: Production and Efficiency in Public Schools. Journal of Economic Literature. Vol.14 Pp.1141-1177.
Heyneman, S.P., & Loxley, W.A. (1982). Influences on Academic achievement across high and low income countries: A re-analysis of IEA data. Sociology of Education, Vol.55 (1). Pp.13-21.
Robertson, D., & Symons, J. (2003). Do Peer Groups Matter? Peer Group versus Schooling Effects on Academic Attainment. Economica, New Series. Vol.70 (277). Pp.31-53.
Brisson Page 28
Rumberger, R.W., & Willms, J.D. (1992). Impact Of Racial and Ethnic Segregation on the Achievement Gap in California High Schools. Education and Policy Analysis. Vol.14 (4). Pp.377-396.
Schutz, G., & Ursprung, H.W., & Wobmann, L. (2005). Education Policy and Equality of Opportunity. CESIFO WORKING PAPER NO. 1518 CATEGORY 3: SOCIAL PROTECTION. www.CESifo-group.de.
Summers, A., & Wolfe, B. (1977). Do Schools Make A Difference? The American Economic Review. Vol. 67 (4). Pp. 639-652.
Zimmer, R., & Toma E.F. (2000). Peer Effects in Private and Public Schools Across Countries. Journal of Policy Analysis and Management. Vol.19 (1). Pp.75-92.
Appendix of Variables
-Dependent Variables-
AvgEOG7th8thmath (this variable was created by computing a new variable in SPSS. EOG 8th grade math and EOG 7th grade math were added together and divided by two)
-Independent Variables-
AFAMSCH [%African American in school in 2008] (used as the main variable in poverty model #1)
LATINOSCH [%Latino in school in 2008] (used as the main variable in poverty model #2)
FRLSCH2008 [%FRL in school in 2008] (used as the main variable in poverty model #3)
-Control Variables-
S_DAYSAB [Number of School days Absent]
CLSSIZE [Class Size]
T_DISTEXP [Years of Teacher Experience in District]