Are the Reading Rich Getting Richer? Testing for the Presence of the Matthew Effect

Abstract

The Matthew effect, where good readers get increasingly better over time compared to relatively lower-ability readers, is an often cited phenomenon in reading research. However, researchers have not always found empirical evidence supporting a Matthew effect. We used hierarchical growth curve modeling to test for the presence of the Matthew effect using a longitudinal sample of 1,573 children in high-poverty, low-performing schools. Our results failed to support the presence of a Matthew effect, but instead we found a compensatory growth trajectory whereby reading achievement gaps closed over time. We discuss several possible explanations for the growth patterns.

Notes

¹Title I is a federal program that is designed to assist specific schools with high percentages of economically disadvantaged students and provides additional funding to assist students in meeting academic benchmarks.

Note. FRPL = free or reduced price lunch.

Note. FRPL = free or reduced price lunch.

²The Level 1 model was Y _tij  = π_0ij + π_1ij (GRADE −2) +e _tij , where Y _tij was the reading scaled score of student i in school j at time t; π_0ij was the ending status of student i in school j; π_1ij was the slope of student i in school j; and e _tij was the time-specific error term of student i in school j at time t. GRADE was recentered (where kindergarten was originally GRADE = 0) so that the intercept represented scores at the end of the second grade, and the slope still indicated the growth rate over time (Singer & Willett, 2003).

The slope (i.e., π _1ij , or the growth estimate) was of particular interest in the study. If the high-performing group had a positive association with the slope and the low-performing group had a negative association with the slope, then a widening of the gap was evident and empirically supported the presence of a Matthew effect. However, if the association with the slope estimates of the high-performing group was negative and the growth for the low-performing group was positive, a compensatory trajectory was present, as gaps were narrowing over time.

For the intercept term and the reading slope, the Level 2 models were

where r _sij was the corresponding Level 2 random effect and MALE (1 = yes), WHITE (1 = yes), and ECONDIS (1 = economically disadvantaged) were dummy-coded student covariates. Finally, at Level 3, the model was β _sqj  = γ _sq0 + γ _sq1(%FRPL) + γ _sq2(%MINORITY) +u _{sq j}, where s = 0 was the ending status, s = 1 was the growth slope, q = 0 − 6 were the beta coefficients from the Level 2 model, %FRPL was a function of the percentage of students eligible for FRPL, and %MINORITY was the percentage of non-White students at the school. For s = 0, u _{sq j} was the Level 3 random effect. For q > 0, all other Level 2 coefficients were fixed: β _s1j  = γ _s1j … β _s6j = γ _s6j.

Note. FRPL = free or reduced price lunch.

*p < .05. **p < .01. ***p < .001.

Note. FRPL = free or reduced price lunch.

*p < .05. **p < .01. ***p < .001.