Вы находитесь на странице: 1из 9

School Psychology Review,

2006, Volume 35, No. 1, pp. 134-141

RESEARCH BRIEE

Does the Timing of Grade Retention Make a Difference?


Examining the Effects of Early Versus Later Retention

Benjamin Silberglitt
St. Croix River Education District

Shane R. Jimerson
University of California Santa Barbara

Matthew K. Burns and James J. Appleton


University of Minnesota

Abstract. Research examining the effectiveness of grade retention has provided


overwhelming and seemingly irrefutahle evidence that grade retention is an
ineffective and potentially harmful practice. However, proponents of grade re-
tention often advocate that retention in the early elementary grades (e.g., kinder-
garten, first and second grade) is the justified exception. This longitudinal study
examined the reading growth trajectories of students (n = 49) from first through
eighth grade. Hierarchical linear modeling analytic procedures provided novel
insights regarding the relative reading growth trajectories among retained stu-
dents, comparing those students retained in kindergarten through second grade
with those students retained in Grades 3-6. The results revealed that the growth
trajectories of students retained early (Grades K-2) were comparable to those
retained later (Grades 3-5). These findings failed to support the efficacy of
retention at an earlier grade in elementary school.

Grade retention, defined as requiring a quently used and controversial intervention


student to remain at his or her current grade (Jimerson. Pletcher, Graydon, Schnurr, Nick-
level the following school year despite spend- erson, & Kindert, 2006). Approximately 2.4
ing a full school year at that given grade million children, or 5-10% of the school-aged
(Jackson, 1975), remains a relatively fre- population, have been retained each year

Correspondence regarding this article should be addressed to Benjamin Silberglitt. St. Croix River
Education District, 425 South Dana Avenue, PO Box 637, Rush City, MN 55069; E-mail:
bsilberglitt@scred.kl2.mu.us
Copyright 2(X)6 by the National Association of School Psychologists, ISSN 0279-6015

134
Effect of Early Versus Later Grade Retention

(Dawson, 1998), but meta-analytic research grade) linked to better short- and long-term
has consistently revealed small to moderate outcomes relative to retention in later grades
mean effect sizes that favored the academic (third through fifth)?
and socioemotional development of promoted
comparison groups over groups of retained Method
children (Holmes. 1989; Holmes & Matthews,
1984; Jimerson, 2001). Participants
Longitudinal research also has failed to Participants were 49 students, from five
demonstrate an overall positive effect for districts in rural and suburban Minnesota, di-
grade retention as an intervention. Short-term vided into two groups. The participating
gains in mathematics skills have been noted, school districts did not have any recorded pol-
but higher absenteeism and lower social- icy or standard procedure for student reten-
emotional rankings among retained children as tions or retention decisions. The participating
compared to a group of promoted children students were all of the retained students
have also been found (Jimerson, Carlson, within those districts with complete data for
Rotert, Egeland, & Sroufe, 1997). Moreover, Grades K-8. Students were assigned to one of
grade retention has been linked to increased two groups based on grade of retention,
risk for dropping out of school to the extent with 27 students who were early retained
that early grade retention has been "one of the (Grades K-2) and 22 who were later retained
most powerful predictors of later school with- (Grades 3-5). The students included 17 (35%)
drawal" (Jimerson, Anderson, & Whipple, females and 32 (65%) males. The ethnicity of
2002, p. 452). i the total sample was 6.1% African Ameri-
Generally speaking, research has not can, 2.0% Asian, 85.7% Caucasian, and 6.1%
supported retention and suggested negative ef- Native American. For the total sample, 29
fects, so why are students retained? Grade (59%) students were eligible for free or re-
retention seems to hold an intuitive appeal duced price lunch for at least 1 school year
despite a lack of empirical support. It also during the study: 16 (59%) from the early-
seems intuitively advantageous to retain a retained group and 13 (59%) from the later-
child earlier (e.g., by second grade) rather than retained group. Finally, 2 (4%) students were
later (e.g., third grade or later). Students are retained in kindergarten, 19 (39%) in first
purportedly retained in early elementary grade, 6 (12%) in second grade, 9 (18%) in
grades to prevent future failure, and are re- third grade, 8 (16%) in fourth grade, and 5
tained in high school to prevent graduation by (10%) in fifth grade.
studetits who lack the basic skills necessary
for post-high school success (Martinez & Procedures
Vandergrift, 1991). Thus, retention before sec-
ond grade is viewed as an early intervention or The school districts that the students at-
a preventative measure. Graue and DiPema tended administered curriculum-based mea-
(2000) found that delayed entry into kinder- sures for reading assessment every fall, winter,
garten led to academic skills consistent with and spring for Grades I- 8, with the exception
peers, and early-retained students were more of fall for first grade. AU administrations were
advanced than students who were retained in a conducted by trained school personnel using
later grade. These data supported early reten- standardized procedures (Shinn, 1989). In to-
tion as prevention hypothesis, but a review of tal, there were 23 data points possible for each
the literature did not reveal any studies that participant, but missing data were possible for
examined timing of retention, rather than re- some participants. Data were compiled over
tention compared to delayed entry. Therefore, the school years from 1996-1997 to 2003-
the current study was conducted to answer the 2004 and student progression through the
following research question: Is retention in grades led to some missing data (e.g., some
early grades (kindergarten through second participants had not yet completed some upper

135
School Psychology Review, 2006, Volume 35, No. 1

grades by 2003-2004, or participants were LMMs are useful for analyzing longitudinal
already in a later grade in 1996-1997). The data, where the assumption of independence
proportion of waves completed across the en- between measurements cannot be met, be-
tire sample was 47%. cause the measurements are nested within stu-
dents, and in this case, students are nested
Measures within the early-retained and later-retained
groups. The random effects structure allows
The Reading-Curriculum Based Mea-
individual students to vary on ail parameters
surement (R-CBM) measure consisted of three
being estimated, while still providing group-
l-min reading passages for each benchmark
level estimates more appropriate than those
assessment (fall, winter, spring). Passages
generated by traditional regression models. In
were standardized at grade-appropriate diffi-
this analysis, the Level I equation models
culty as suggested by Howe and Shinn (2002),
individual growth over time. For each individ-
and were curriculum independent, grade-level
ual ( at time j , the Level I equation is as
appropriate, and of equal difficulty within
follows:
each grade level. Student data consisted of the
median number of words read correctly across
the three passages for each assessment. Read- ,V,y= + (1)
ing was assessed because most academic re-
ferrals to school psychologists were for read- In the Level 1 equation, quadratic
ing deficits (Bramlett, Murphy, Johnson, growth is modeled for each student, with c,y
Wallingsford, & Hall, 2002), and reading is representing error. The Level 2 equations then
clearly an important skill for subsequent model group differences in each of these beta
knowledge acquisition. Because of the number values. In these equations, g refers to group
of students being assessed within the district membership (early retained or later retained),
assessment plan and the simplicity of the mea- and b is the error term. Level 2 equations are
sure to administer and score, interrater reli- as follows:
ability was not systematically collected during
these district-wide data collection efforts. Past (2)
research on R-CBM has demonstrated test-
o. = Po + +
retest reliability of 0.92 and interrater reliabil-
ity of 0.99 (Marston, 1989). (3)

Analysis + Psg, + (4)

Research methodologies frequently used LMMs also allow for missing data, a
to examine change in variables for individuals common occurrence when conducting longi-
have not allowed for explicit modeling of in- tudinal studies over a large time span. In this
dividual differences in growth, which can best study, data were assumed to be missing at
be addressed with hierarchical linear modeling random. LMMs also assume independence be-
(HLM; Bryk & Raudenbush, 1992). Thus, R- tween subjects, an assumption that can be ad-
CBM growth rates across Grades 1-8 for stu- dressed by sampling randomly from the pop-
dents retained in early grades (K-2) were ulation. Although this sample is admittedly a
compared to those retained in later grades small sample from a single region of the coun-
(3-5). The 23 total waves of data were used to try, we have assumed this premise has been
conduct an HLM analysis using a random met for the sake of these analyses, and recog-
effects covariance structure. nize the limitations of our sample. A final
HLM is an example of a linear mixed assumption of LMMs is that the response vari-
model (LMM), which is particularly well able be normally distributed, which is gener-
suited to the type of problem addressed in this ally the case with R-CBM (Hintze & Silber-
study (Fitzmaurice, Laird, & Ware, 2004). glitt, 2005).
136
Effect of Early Versus Later Grade Retention

Results of performance had a slight tendency to have


a more negative quadratic bend. Correlation
The variance-covariance matrix re- between the linear slope parameter estimates
vealed increasing variance across the waves of and intercept estimates across participants
the study, with decay. As the amount of time was 0.62, indicating a moderate to strong pos-
between R-CBM measurements increased, the itive correlation between beginning level of
degree of covariance between the scores on performance and the linear slope of growth
those measurements decreased. In addition, a
among retained students.
higher amount of variance was found for R-
CBM scores from higher grade levels. Level 1 Level 2 analysis included a single dum-
analysis generated an individual slope (includ- my-coded variable identifying group member-
ing linear and quadratic components) and in- ship in the early (0) versus later (1) retention
tercept for each participant. Correlation be- groups. Regression lines were drawn using
tween the quadratic slope parameter estimates these mean parameter estimates, as shown in
and the intercept estimates across participants Figure 1. Actual parameter estimates and re-
was —0.23, indicating that the growth curves sults of significance tests are shown in Tables
of retained students with a higher initial level I and 2. In each table, parameter estimates and

160

140

120

100

80

60

40

•20 -1 1 1 r -T 1 I 1 r

Gnde and Season of R-CBM

Figure 1, Longitudinal growth curves for early-retained (Grades K-2) and


later^retained (Grades 3-5) students.
137
School Psychology Review, 2006, Volume 35, No. 1

Table 1
Significance Tests of the Differences in Parameter Estimates for
Early'Retained (K-2) Versus Later-Retained (3-5) Groups
Parameter
Parameter Estimate SE

Intercept 24.852 3.028


Linear slope 6.295 1.255
Quadratic .slope -0.023 0.058
Later-retained intercept^ -25.445 5.184 41 4.91 <.ooai
Later-retained linear slope" 3.909 1.832
Later-retained quadratic slope* -0.181 0.082 28.5 -2.21 0.0356

"pLirameter estimate.s for the later-retained group represent the difference between this group and the early-retained
group. The f tests examine ihe significance of these differences.

corresponding standard errors are given first nificant, E(X 22.1) = 7.90, /> < .001, indicaUng
for the early-retained group. Second, the ad- that the specific tests could reasonably be exam-
justments to these estimates are provided for ined. As shown in Table I. a statistically signif-
the later-retained group. For example, in Table 1, icant difference (p < .0001) was found between
the estimate of the intercept for the early-re- the mean intercept parameters for the eitrly- and
tained group was 24.852. The adjustment to this later-retained groups, with the intetcept for the
intercept for the later-retained group was early-^retained group being larger. In practical
-25.445. So, one could calculate the parameter terms, the difference in the intercept estimates
estimate for the later-retained group's intercept for the two groups was approximately 25 words
lo be —0.593. The corresponding p values read correctly per minute, with a standard error
present the results of the statistical significance of approximately 5 words. In addition, a mod-
tests for these adjustments. estly significant difference (/; = .036) was found
Table I provides parameter estimates between the mean quadratic slope parameters for
for the intercept, linear, and quadratic the early- and later-retained groups, with the
growth component. These are also the pa- later-retained group showing a larger negative
rameter estimates used in Figure I, How- bend in the growth curve. As shown in Table 2,
ever, a significance test of this linear com- no significant difference (p = .37) was found
ponent is not appropriate, as significance between the mean linear slope parameters for the
tests of lower order parameter estimates are early- and later-retained groups. The practical
not appropriate in the presence of higher significance of the slope parameters must be
order parameter estimates (Fitzmaurice et considered simultaneously. In practical terms,
al., 2004). A separate analysis was con- the iater-retained group's quadratic growth term
ducted with the quadratic parameter esti- was 0.181 smaller than that of the early retained
mate removed, to generate an interpretable group. However, the linear growth tenii for the
significance test of the linear parameter es- later-retained group was 3.909 larger. Over the
timate. This analysis is presented in Table 2. course of the study (from Grade I winter to
Before specific significance tests of differ- Grade 8 spring), assuming all other terms in each
ences in parameter estimates were examined in group's growth curve were equal, these differ-
Table 1, an omnibus test was conducted for all ences would lead to the early-retained group's
tests of group differences among slope and in- Grade 8 spring score being 1.6 words conxict per
tercept parameters. This omnibus test was sig- minute higher than that of the later-retained

138
Effect of Early Versus Later Grade Retention

Table 2
Significance Tests of the Differences in the Linear Growth Component for
Early'Retained (K-2) Versus Later-Retained (3-5) Groups

Parameter
Parameter Estimate SE df

Intercept 25.030 3.180


Linear slope 6.051 0.590
Later-retained intercept" -14.867 4.962
Later-retained linear slope" 0.756 0.828 29.9 0.91 0.3684

"Parameter estimates for the later-retained group represent the difference between this group and the early-retained
group. The f tests examine the significance of these differences.

group. Although the differences in the linear larger negative bend in the growtb curve, and
component of the slope for the two groups were tbe early-retained group's growtb rate contin-
not statistically significant, they do appear to ued in a linear fasbion. Tbis trend in tbe
offset the statistically significant difference in growth curve suggested that students retained
the quadratic component. In effect, the practical later had a more rapid deceleration of growtb
difference In slope between these two groups is compared to the more consistent progress rate
negligible. of the early-retained students. Rather than per-
ceive this as a benefit for early retentions, it is
1 •'•
possible that these data are a result of a greater
. - Discussion negative effect from later retentions. Research
Growth curves for the early- and later- has consistently found a negative socioemo-
retained students (Grades K-2 and 3-5, re- tional impact from grade retention (Holmes,
spectively) were modeled and compared using 1989; Holmes & Matthews. 1984; Jimerson,
HLM procedures. Reading data from tbe first 2001). It is possible that this effect is stronger
year of the grade in which the student was for students wbo are older, and more emotion-
retained were removed before the growth ally mature, at the time of retention. However,
curves were modeled. The average reading this is only one possible explanation of this
score in the winter of first grade (i.e., inter- finding, and previous researcb literature was
cept) was significantly bigher for the group seemingly void of comparisons between ef-
tbat was retained earlier. This is somewbat fects of early and later grade retention. Tbus,
unexpected, given that one would anticipate a tbese data sbould be interpreted carefully, and
lower level of performance in the early grades replication is needed before confident interpre-
to be more likely to lead to a decision to retain tations can be made.
immediately. These data could support tbe rel-
atively long-held belief that retention deci- Limitations and Suggestions for Future
sions are often made with teacher judgment Research
and otber subjective factors rather than aca-
demic data (Cadigan, Entwisle, Alexander, & The current study has presented data
Pallas. 1988; Gloeckler, 1986). that may be of interest to botb researcbers and
Despite the difference in the intercepts, practitioners; however, several limitations
there was no significant difference in linear should be noted. First, tbe sample size was
slope parameters for the two growth curves, relatively small, with 27 students in the early-
but the later-retained group bad a significantly retained group and 23 in the later-retained
139
School Psychology Review, 2006, Volume 35, No. 1

group, with little assurance of equal child vari- achievement and adjustment, but research re-
ables prior to being retained. Therefore, repli- veals that the practice of grade retention is
cation thai accounts for child-centered vari- "nonsense," in that it appears to do more harm
ables (e.g., socioeconomic status, IQ, prere- tban good (Jimerson, 2001). Neitber grade re-
tention skill level) is recommended for further tention nor social promotion will adequately
insights regarding tbe nature of growth for address tbe needs of students (Jimerson, 2001;
students retained earlier versus later in their Jimerson et al., 2006); rather, specific evi-
educational careers. Another limitation of the dence-based interventions are required to pro-
current study is the lack of examination of mote the cognitive and social competence of
student perspectives of retention and the effect students.
of risk factors upon student developmental
trajectories (e.g., socioemotional outcomes).
Myriad factors influence tbe development of References
retained students (Jimerson et al, 1997; Jimer- Alexander, K. L.. Enlwisle, D. R., & Horsey, C. S. (1997).
son, 2001; Jimerson et al., 2002). and includ- From first grade forward; Early foundations of high
ing estimates of tbese variables would im- school dropout. Sociology of Education, 70(2), 87-
107.
prove the study. Future retention research Bramlett. R. K.. Murphy. J. J.. Johnson. J., Wallingsford,
could also encompass a more comprebensive L., & Hall, J. D. (2002). Contemporary practices in
array of tbe risk and protective factors imping- school psychology: A national survey of roles and
referral problems. Psychology in ihe Schools. 39, 327-
ing on the developmental trajectories of re- 335.
tained students. Furthermore, despite diversity Bryk. A. S.. & Raudenbush. S. W. (1992). Hierarchical
of age, income (47% free and reduced lunch), linear models: Applications and data analysis meth-
and location (rural and suburban), tbis sample ods. Newbury Park. CA: Sage Publications.
Cadigan. D.. Entwisle. D. R.. Alexander, K. L.. & Pallas,
was racially homogenous (>91% White). The A. M. (1988). First-grade retention among low achiev-
generalizability of these data could be en- ing students: A search for significant predictors. Mer-
banced by including more diverse racial and rill-Palmer Quarterly, 34. 71-88.
Dawson. P. (1998). A primer on student grade retention:
cultural groups. Finally, tbe current study used What the research says. Communique. 26(8). 28-30.
reading as the dependent variable, which is the Fitzmaurice, G. M., Laird, N. M., & Ware. J. H. (2004).
most frequent academic deficit area. However, Applied longitudinal analysis. Hoboken. NJ: John
Wiley & Sons.
analyses examining other academic skills Graue, E., & DiPema. J. (2000). Redshirting and early
(e.g., mathematics) and outcomes through retention: Who gets the "gift of time" and what are its
high school would also be of great value in outcomes? American Educational Re.^tearch Joumal,
37, 509-334.
considering the effects of earlier and later Gloeckler, T. (1986). Data-based decision-making proce-
grade retention, especially given researcher dures for determining in-grade retention. Education,
suggestions of tbe effect of early retention 107, 182-186.
Hintze, J- M., & Silberglitt, B. (2005). A longitudinal
(e.g., in first grade) on dropping out of higb exatnination of the diagnostic accuracy and predictive
scbooi (Alexander, Entwisle, & Horsey, validity of R-CBM and high-stakes testing. School
1997). P.iychology Review, 340), 372-386.
Holmes, C. t . (1989). Grade-level retention effects: A
Tbe analyses of longitudinal reading tra- meta-analysis of research studies. In L. A. Shepard &
M. L. Smith (Eds.) Flunking grades: Research and
jectories in this study revealed that early grade policies on retention (pp. 16-33). London: The Falmer
retention {kindergarten througb second grade) Press.
did not yield advantages in reading trajectories Holme.s. C. T.. & Manhews, K. M. (1984). The effects of
nonpromotion on elementary and junior high school
from first to eighth grade, relative to students pupils: A meta-anatysis. Reviews of Educational Re-
retained later (third through fiftb grades), but a search, 54. 225-236.
more linear progress rate was noted. Altbough Howe, K. B., & Shinn. M. M. (2002). Standard reading
assessment passages (RAPs) for u.'ie in General Out-
tbe idea of early retention has intuitive appeal, come Measurement: A manual describing development
tbe data fail to support retention, whether it is and technical features (technical manual), Eden Prai-
early or later. Thus, proponents of grade re- rie, MN: Edformalion.
tention base their position on "common sense" Jackson. G. (1975). The research evidence on the effects
of grade retention. Review of Educational Research,
tbat repeating the year will enhance student ^5,613-635.
140
Effect of Early Versus Later Grade Retention

• '' 1
Jimerson, S. R. (2001). Meta-analysis of grade retention Marston. D. (1989). A curriculum-based measurement
research: Implications for practice in the 21st century. approach to assessing academic performance: What it
School Psychology Review, 30, 420-438. is and why do it. In M. R. Shinn (Eid.), Curriculum-
Jimerson, S., Anderson, G. E.. & Whipple, A. D. (2002). based measurement: Asse.\sing special children (pp,
Winning the battle and losing the war: Examining the 18-78). New York: Guilford Press.
relationship between grade retention and dropping out Martinez, B., & Vandergrift. J, A. (1991). Eailing students:
of high school. Psychology in the Schools, 39. 4 4 1 - Is it worth Ihe cost? (Issue Paper .1). Tempe: Morrison
457. Institute for Puhlic Policy. Arizona Slate University.
Jimerson. S.. Carlson, E.. Rotert, M., Egeland, B., & Shinn. M. R. (1989). Identifying and defining academic
problems: CBM screening and eligibility procedures.
Sroufe, L. A. (1997). A prospective, longitudinal study
In M. R. Shinn (Ed.). Curriculum-based measurement:
of the correlates and consequences of early grade re-
Assessing speciai children (pp. 90-129). New York;
tention. Joumal of School Psychology, 35. 3-25. Guilford.
Jimerson. S. R., Pletcher, S. M., Graydon, K., Schnurr,
B. L., Nickerson. A. B., & Kundeii. D. K. (2(K)6).
Beyond grade retention and social promotion: Promot- Date Received: July 7, 2005
ing the social and academic competence of students. Date Accepted: December 13, 2005
Psychology in the Schooh, 43, 85-97. Actioti Editor: John Hintze •

Benjamin Silberglitt received his PhD in Educational Psychology (School Psychology)


from the University of Minnesota—Twin Cities in 2003. He currently serves as the
Outcomes Manager for the St. Croix River Education District, managing the research,
assessmeni, and evaluation efforts of its tnember districts.

Shane Jimerson, PhD, is a faculty member at the University of California, Santa Barbara,
in the Gevirtz Graduate School of Education with both the Counseling, Clinical, and
School Psychology program and the Child and Adolescent Development program. His
scholai^hip continues to examine the importance of early experiences on subsequent
development and to emphasize the importance of research infonning professional practice
to promote the social and cognitive competence of children.

Matthew K. Bums, PhD, is an Associate Professor of Educational Psychology at the


University of Minnesota. Current research interests include curriculum-based assessment,
response to intervention, problem-solving teams, and consultative service delivery for
special education.

James Appleton is a school psychology doctoral student at the University of Minnesota


working with the Center for Applied Research and Educational Improvement. He is
interested in student engagement with school, social development, and evidence-based
interventions.

141

Вам также может понравиться