Annotated Bibliography Project

The Journal of Educational Research, 105:431–441, 2012 Copyright C! Taylor & Francis Group, LLC ISSN: 0022-0671 print / 1940-0675 online DOI:10.1080/00220671.2012.658456

The Association of Kindergarten Entry Age with Early Literacy Outcomes

FRANCIS L. HUANG MARCIA A. INVERNIZZI University of Virginia

ABSTRACT. The authors investigated whether age at kindergarten entry was associated with early literacy achieve- ment gaps and if these gaps persisted over time. Using the kindergarten age eligibility cutoff date, they created 2 groups of students who represented the oldest and youngest chil- dren in a cohort of students in high-poverty, low-performing schools. The authors followed 405 students from the be- ginning of kindergarten until the end of Grade 2. Results indicated that the youngest students scored lower than their oldest peers at the beginning of kindergarten on various early literacy measures. The early-age achievement gap, however, narrowed over time but did not close completely by the end of Grade 2. Implications for parents and educators are discussed.

Keywords: achievement gap, birthday effects, emergent lit- eracy growth, growth model, kindergarten entry age, school readiness

T he practice of academic redshirting, or intention-ally delaying the entry of an age-eligible child intokindergarten or first grade by a year so a child can mature, has remained a topic of interest in the popular press (e.g., Bronson & Merryman, 2009; Gladwell, 2008; Goot- man, 2006; Paul, 2010; Weil, 2007). In the spring of each year, parenting magazines, radio programs, and television talk shows, in anticipation of a parents’ impending school entry choice for their child, often discuss the pros and cons of beginning kindergarten on time or waiting another year (Graue & DiPerna, 2000). However, even without redshirt- ing, classrooms will always have an oldest and a youngest child. The extreme difference in age between children born on and a day before the eligibility cutoff date is 12 months or a gap of approximately 20% for a kindergarten entry age of 60 months. Age of kindergarten entry, although seemingly innocuous and set by a state’s age-eligibility cutoff date, remains a source of debate (Lincove & Painter, 2006; Na- tional Institute of Child Health and Human Development Early Child Care Research Network, 2007) and has resulted in several empirical studies investigating the association of age and learning over time (e.g., Bedard & Dhuey, 2006; Crone & Whitehurst, 1999; Datar, 2006; Stipek & Byler, 2001).

Although most age-related or birthday effects studies have involved middle-class families (Lin, Freeman, & Chou, 2009; Stipek, 2002) or the larger national population (e.g., Bedard & Dhuey, 2006; Datar, 2006; Lincove & Painter, 2006), only a handful of kindergarten-age enrollment stud- ies have focused on students in high-poverty schools (e.g., Crone & Whitehurst, 1999; Stipek & Byler, 2001). Children in high-poverty schools represent a heterogeneous group with varying abilities and cannot all be labeled as at risk (Cabell, Justice, Konold, & McGinty, 2011). Our study adds to the relatively smaller, but growing, body of knowledge that empirically investigates the association of age and the growth of early literacy outcomes involving students in not just high-poverty, but low-performing schools, following a sample of students from the beginning of kindergarten to the end of Grade 2. Kindergarten entry decisions are even more critical for low-income families if being the youngest child in a class constitutes an additional risk factor (Bedard & Dhuey, 2006; Crosser, 1998; Lin et al., 2009) as children from low-income families are already at greater risk of school failure (Snow, Burns, & Griffin, 1998) and are more likely to begin school on time rather than delaying school entry (O’Donnell, 2008).

Theoretical Background

Trends in delayed enrollment. As children enter kinder- garten for the first time, they differ in what they can and cannot do cognitively, physically, and emotionally (Mal- one, West, Flanagan, & Park, 2006). The basic rationale for delaying kindergarten entry is that if children are older, they are likely to be more mature, well behaved, confident, and ready to learn, compared with their younger classmates (Meisels, 1992; Uphoff & Gilmore, 1985).

Over the past 40 years in the United States, the age of primary school entry has drifted upwards and a large source of the increase has resulted from state-driven legal changes to school entry age (Deming & Dynarski, 2008). In addi- tion, parents of children whose birthdays fall near, although

Address correspondence to Francis Huang, University of Virginia, P.O. Box 800785, Charlottesville, VA 22908-800785, USA. (E- mail: [email protected])

432 The Journal of Educational Research

approaching, the eligibility cutoff date may worry about their child’s ability to cope with the demands of kindergarten, opting instead to delay school entry by a year, not wanting them to be the youngest in the class (Bracey, 1989). An era of high-stakes testing and increased accountability mea- sures may have unintentionally contributed to the practice of redshirting as parents, teachers, and principals focus on the short-term benefits that delayed entry may offer (Deming & Dynarski, 2008).

From 1993 to 1997, based on the nationally representa- tive Early Childhood Longitudinal Study (ECLS) and the National Household Education Survey (NHES), 6%–9% of parents reported delaying the kindergarten entry of their child (Malone et al., 2006; Zill, Loomis, & West, 1995). The NHES 2007 indicated that 7% of parents delayed their child’s kindergarten entry and children who experience de- layed entry were more likely to be White boys who come from non-economically disadvantaged families (O’Donnell, 2008). Children with birthdates closer to the kindergarten entry cutoff dates, making them the youngest ones in the cohort, were also more likely to be redshirted (Cosden, Zimmer, & Tuss, 1993; Graue & DiPerna, 2000).

Findings of age-related studies. On average, the academic performance of much older students upon kindergarten en- try is higher than that of younger children (Oshima & Do- maleski, 2006; Stipek, 2002). As a result, young-for-grade children from economically disadvantaged, minority fami- lies may face even greater challenges in school as a result of an entry-age achievement gap (Cascio, 2008). Stipek’s (2002) meta-analysis of studies investigating the association of age with various outcomes measures (e.g., emergent lit- eracy skills, mathematics, IQ) found little support for the practice of holding children back as early advantages at- tributable to age fade by Grade 3. This finding corroborates earlier age-achievement related studies; the synthesis of re- search on school readiness over a decade earlier by Shepard and Smith (1986) also indicated that although the youngest children will start out with a slight disadvantage, these dif- ferences are not large and disappear by about Grade 3 as well. Although early literacy gaps may be present as an effect of age at school entry, age-related gaps are minor in comparison with other risks associated with race/ethnicity, gender, and socioeconomic status (SES; Jones & Mandeville, 1990).

More recently, using the National Educational Longi- tudinal Study (NELS), Lincove and Painter (2006) found no advantage for older children in terms of high school achievement, college enrollment, or graduation rates. Like- wise, Deming and Dynarski (2008) found that older children had no more positive, long-term adult outcomes such as ed- ucational attainment, IQ, or earnings than their younger peers. Using experimental data from the Tennessee Project STAR, Cascio and Schanzenbach (2007) found that relative age did not impact achievement or the likelihood of taking a college-entrance exam. On the other hand, recent studies using the ECLS-K (Datar, 2006; Lin et al., 2009) and inter- national data sets (Bedard & Dhuey, 2006) detected a small

but lasting benefit for older students in terms of academic achievement.

In addition to academic outcome measures, analyses of national datasets suggest that young-for-grade children are more likely to be diagnosed with attention deficit hyperactiv- ity disorder (ADHD; Evans, Morrill, & Parente, 2010) and other learning disabilities (Dhuey & Lipscomb, 2010), have greater probabilities of having behavioral problems (Elder & Lubotsky, 2008), and have a greater likelihood of repeating a grade (Bedard & Dhuey, 2006; Lincove & Painter, 2006). Although relatively older children were more likely to be leaders in high school (Dhuey & Lipscomb, 2008), older students have also been found to have lower levels of edu- cational attainment as a result of being able to drop out of school earlier (Angrist & Krueger, 1991; Meisels, 1992).

Societal consequences of delayed enrollment. As a result of delaying kindergarten enrollment, the age span within a class may increase from 12 months to 24 months, resulting in a graying of kindergarten (Bracey, 1989). This increase in age span might result in unanticipated social consequences. First, an increase in the age span is extremely challenging for teachers who must deal with a wide range of cognitive and social maturity levels (Shepard & Smith, 1986). Sec- ond, given that economically advantaged parents are more likely to hold back their children (Meisels, 1992; O’Donnell, 2008), increased age differences may exacerbate existing achievement gaps associated with race/ethnicity and in- come (Cascio, 2008; Deming & Dynarski, 2008). Finally, high-stakes accountability measures in the upper grades may result in an escalation of the kindergarten curriculum (Cosden et al., 1993) where kindergarten may start to look more like Grade 1.

The Present Study

As only a few age-related studies have focused on low- income children (Stipek, 2002), we studied a longitudinal sample of the oldest and youngest children in a cohort of students in high-poverty schools and investigated their lit- eracy growth over 3 years. First, we explored the magnitude of age effects on early literacy measures collected at the be- ginning of kindergarten. Based on the literature (e.g., Crone & Whitehurst, 1999; Datar, 2006; Stipek & Byler, 2001), we expected the presence of early-age achievement gaps in favor of the older children. Next, we explored the reading growth trajectories of the youngest and the oldest students until the end of Grade 2. We hypothesized, based on previous re- search (e.g., Oshima & Domaleski, 2006; Shepard & Smith, 1986; Stipek, 2002), that gaps would close over time. Finally, as done by Jones and Mandeville (1990), we compared the magnitude of age-based gaps with other achievement-related factors such as gender, race/ethnicity, and economic status. Specifically, our research questions were the following:

Research Question 1: How large are the differences in early lit- eracy measures between the oldest and youngest students in kindergarten?

The Journal of Educational Research 433

Research Question 2: Do the youngest students learn liter- acy skills at faster rates over time compared to the oldest students?

Research Question 3: Are early-age literacy gaps still present by the end of Grade 2?

Research Question 4: How do early-age literacy gaps com- pare to other possible gaps associated with gender, race/ethnicity, and economic status?

Method

Participants

Participants were 405 students (202 boys, 203 girls) who were in high-poverty and low-performing public schools in the Commonwealth of Virginia. High-poverty schools had at least 40% of the students in school who were eligible for free or reduced-price lunch (FRPL; our proxy for eco- nomic disadvantage). Low-performing schools had a passing rate of less than 60% on the end-of-year third-grade state English/reading test. Students were taking part in a larger multiyear study1 related to literacy development and the 405 students represented a subsample of students who were selected specifically based on their age and the availability of assessment scores over 3 years. In the sample, 52% were White, 52% were eligible for FRPL, and 8% had a diag- nosed disability, the majority of which was speech/language impairment.

Procedures

Teachers administered the Phonological Awareness Lit- eracy Screening for Kindergarteners (PALS-K; Invernizzi, Swank, Juel, & Meier, 2003) to their students in the fall of kindergarten. In addition, teachers administered the Stan- ford Reading First assessment (SRF; Harcourt Assessment, 2004) at the end of spring in kindergarten, Grade 1, and Grade 2. Schools provided basic demographic information on the participants.

Variables and Measures

Age. Our independent variable of interest was age and we created a dichotomous variable that indicated whether a student was young for grade or old for grade. In the state, entering kindergarteners must be 5 years old by September 30 and participants in the present study were either born in October (n = 202; YOUNG = 0 or the old group) or 11 months later in September (n = 203; YOUNG = 1 or the young group), representing the naturally occurring oldest and youngest students in the kindergarten cohort, respec- tively.

For our dependent variables (see Table 1), we used se- lected subtasks of the PALS-K assessment as kindergarten outcome measures: lowercase alphabet recognition, letter- sound knowledge, and spelling. The PALS subtasks were chosen specifically because they had a wide amount of vari-

ance, represented a construct of emergent literacy, and were predictive of future reading performance. For example, the rapid recognition and naming of the letters of the alpha- bet, performed in the lowercase alphabet recognition task, is one of the best predictors of early reading achievement (Adams, 1990; Snow et al., 1998). Letter-sound knowledge is highly predictive of children’s reading acquisition (Foy & Mann, 2006). Finally, the application of letter-sound knowledge in the spelling task is an excellent predictor of word recognition in children (McBride-Chang, 1998) and ranks as one of the best predictors of word analysis and syn- thesis (Torgesen & Davis, 1996). For the growth measure over time, we used the written multiple choice scaled score results of the SRF, which is a nationally normed reading assessment.

Alphabet recognition. Participants were asked to name 26 lowercase letters presented in random order. A total of 26 points are possible. Interrater reliability is high (r = .99) and test–retest reliability is .92 (Invernizzi, Juel, Swank, & Meier, 2009).

Letter-sound knowledge. Children were presented with 23 uppercase letters (excluding Q and X and the letter M, which was used as a practice item) and three digraphs (Sh, Th, Ch) and asked to say the sound the letter-digraph makes. A total of 26 points are possible. Interrater reliability for the task is high (r = .99) and test–retest reliability is .88 (Invernizzi et al., 2009).

Spelling. Children were asked to write five high- frequency, single-syllable words (e.g., van, job, rug). Words were scored based on the number of phonemes represented with phonetically acceptable choices. A bonus point is awarded for correctly spelled words resulting in a total pos- sible score of 20. The task has excellent interrater reliability (r = .99) and has a Cronbach’s alpha of .90 (Invernizzi et al., 2009).

Stanford Reading First (SRF). The SRF, an edition of the Stanford Achievement Test, Tenth Edition, focuses on phonemic awareness, phonics, vocabulary development, reading fluency, and comprehension. Scaled scores are linked across different grade levels and represent approximately equal units on a continuous scale, making scaled scores “es- pecially suitable for comparing group performance over time” (Harcourt Assessment, 2004, p. 33). The SRF multiple- choice test has a high degree of internal consistency with a KR20 of .92 (Harcourt Assessment, 2004).

Visual inspection of the distribution of the dependent variables indicated that the letter-sound knowledge task and the SRF scores were relatively normally. However, the lowercase alphabet recognition task was negatively skewed (–.83) and exhibited ceiling effects (i.e., there are only 26 letters in the alphabet) although the spelling task was posi- tively skewed (.79).

434 The Journal of Educational Research

TABLE 1. Descriptive Statistics for Dependent Variables

Assessment M SD Minimum Maximum

PALS-K tasks Alphabet recognition 18.41 7.54 0 26 Letter-sound knowledge 10.94 7.59 0 26 Spelling 6.28 5.44 0 20

Stanford Reading First Kindergarten 535.47 34.74 450 658 Grade 1 586.49 35.14 459 713 Grade 2 623.06 31.55 544 742

Note. PALS-K = Phonological Awareness Literacy Screening for Kindergarteners.

Covariates. We used student-level controls for race/ ethnicity (1 = White, 0 = non-White), economic disad- vantage (1 = eligible for FRPL, 0 = not eligible for FRPL), gender (1 = female, 0 = male), and disability status (1 = with an identified disability, 0 = with no identified disability). At the school level (n = 66), we included the proportion of stu- dents eligible for FRPL (M = 61.70, SD = 11.33, minimum = 41.07, maximum = 92.08).

Data Analysis

We first ran a series of chi-square tests of homogeneity to detect any existing differences between the young and old student groups on the covariates. To investigate the association of age with literacy outcomes, we created sev- eral two-level hierarchical linear models (HLM; i.e., HLM representing students nested within schools) using the four literacy outcome measures administered in kindergarten as our dependent variables (including the SRF in spring). At the school level, the proportion of FRPL was grand mean centered. For each dependent variable, we first ran an uncon- ditional model (i.e., a model only using school as a grouping variable) to gauge if using multilevel modeling was necessary (i.e., if the intraclass correlation coefficient or ! was greater than .05) and then a conditional model with the indepen- dent variable and covariates included. Our combined (Level 1 and Level 2 models), fully conditional multilevel model in kindergarten was

Yi j = "00 + "01(%FRPL) + "10(SBij) +

"20(YOUNGi j ) + u0 j + ei j ,

where Yij was the one of the four dependent variables in kindergarten of student i in school j; " 00 represented the estimated mean of Y when all other variables were 0; " 01 was the average slope showing the association be- tween Y and school-level FRPL; " 10 was the slope showing the relationship between Y and SBij, a vector of dummy- coded student background covariates; u01 was the error term

associated with school j; and eij was the error term associated with student i in school j. Our parameter of interest was " 20 which showed the effect of being young for grade (YOUNG) compared with the reference group of old-for-grade students.

To model the growth of scaled scores over time and to detect any significant differences in scores at the end of Grade 2, we used three-level hierarchical linear growth curve modeling (i.e., scores nested within students nested within schools). The Level 1 model was

Yt i j = #0i j + #1i j (GRADE " 2) + etij,

where Ytij was the reading scaled score of student i in school j at time t; # 0ij was the ending status of student i in school j; # 1ij was the slope of student i in school j; and etij 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 Grade 2 although the slope still indicated the growth rate over time (Singer & Willett, 2003).

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

#sij = $s 0 j + $s 1 j (YOUNG) + $s 2 j (FEMALE) +

$s 3 j (WHITE) + $s 4 j (ECONDIS) + $s 5 j (DIS) + r sij,

where rsij was the corresponding Level 2 random effect; YOUNG (1 = yes) represented the dummy code for being born on September 1999; and FEMALE (1 = yes), WHITE (1 = yes), ECONDIS (1 = economically disadvantaged), and DIS (1 = with an identified disability) were dummy coded student covariates.

Finally, at Level 3, the model was $s q j = "s q 0 + "s q 1(%F R P L ) + us q j , where s = 0 was the ending status; s = 1 was the growth slope; q = 0–5 were the $ coefficients from the Level 2 model; and %FRPL was a function of the percentage of students eligible for FRPL. For s = 0, usqj was the Level 3 random effect. For q > 0, all other Level 2 coefficients were fixed: $s 1 j = "s 1 j · · · $s 5 j = "s 5 j . We first

The Journal of Educational Research 435

ran an unconditional growth model and then a full condi- tional growth model. All multilevel models were run using PROC MIXED in SAS 9.2 using full maximum-likelihood estimation.

Selection bias. Investigating the association of age with academic achievement, using a sample of on-time and red- shirted children is problematic because children who are voluntarily held back are not random samples and are likely to be systematically different from children who began on time on a number of characteristics (Stipek, 2002). How- ever, in a regular school year, there is a naturally occurring gap between children born on the state-mandated cutoff date and the children who are born 364 days later. Birth date tar- geting (i.e., timing pregnancy and birth or nonrandom birth dates) of more educated mothers may confound such analysis but such a practice does not seem to be widespread (Bedard & Dhuey, 2006). Bedhard and Dhuey’s (2006) study, using several nationally representative datasets, suggested a very weak pattern that more educated mothers may target giving birth in the summer months, making their child the youngest in the cohort. In our analytic sample, we avoid the problem of comparing on-time and redshirted students by excluding students who were held back.

Sample attrition over time is common in longitudinal studies and the reasons for students leaving the study may not be random. Our original sample of young and old kinder- garteners was comprised of 735 students and our final ana- lytic sample was reduced to 405 students. Results of a t-test analysis showed that the students who left the sample after kindergarten had lower SRF kindergarten scores (M = 514, SD = 38.90) than the students who remained in the sample (M = 535, SD = 34.75), t(666) = "7.64, p < .001. The finding of the high mobility of low-performing students is not uncommon (Rumberger, 2003). In addition, among the sample leavers, young-for-grade students scored lower than the old-for-grade students in the SRF in kindergarten (Ms = 504 vs. 525, SDs = 35.77 vs. 39.06), t(328) = 5.19, p < .001, d = .60, which is in line with our hypothesis that young-for- grade students do not perform as well as old-for-grade stu- dents. If our main results show inconsistent outcomes with those of the leavers (e.g., what if in the main analysis younger children scored higher than older children?), then attrition may be a larger issue and we may adjust sample weights as is commonly done in nationally representative longitudi- nal surveys. However, attrition may be more of an issue if several observed characteristics of the students who left the sample were significantly different from the characteristics of those who remained in the sample and if one age group experienced more attrition than the other.

Of the 330 students that left the sample, an equal number of students exited the sample from the young and the old groups (n = 165 for both) and as a result, attrition was not associated with the child’s age grouping. In addition, chi- square tests showed that the leavers in the young and old groups did not significantly differ in terms of race/ethnicity,

economic status, and disability status (all ps > .05), though more girls left the young group compared with the old group, % 2(1, N = 735) = 7.60, p < .01. Finally, in our main analysis, we ran a chi-square test of homogeneity to test the compa- rability of the young and old student groups in our analytic sample using all of our covariates.

Results

Comparison of Characteristics Between the Youngest and Oldest Students

Demographic differences in the distribution of children between the youngest and oldest groups in our analytic sam- ple were found to be merely a result of chance and varia- tions were not statistically significant. The chi-square tests of homogeneity failed to detect any difference between the young and old groupings based on the covariates used, for race/ethnicity, % 2(1, N = 405) = 0.06; for economic status, % 2(1, N = 405) = 0.30; for gender, % 2(1, N = 405) = 2.07; and for disability status, % 2(1, N = 405) = 0.04, all ps > .05 (see Table 2). Results indicated that the two groups were comparable based on all of the covariates used.

Correlations between student-level covariates are pre- sented in Table 3. The only statistically significant corre- lation was between White and economically disadvantaged students (r& = ".34), indicating that non-White students were more likely to be from economically disadvantaged households. Statistically significant correlations of age with any of the other covariates were not found (all ps > .05).

Association of Age with Early Literacy Outcomes

To answer the first research question regarding the differ- ences in early literacy skills between the youngest and the

TABLE 2. Comparison of Characteristics Between Young and Old Groups

Characteristic Young

(n = 203) Old

(n = 202) % 2(1) p

Race/Ethnicity 0.06 0.81 White 107 104 Non-White 96 98

Economic status 0.30 0.59 Disadvantaged 108 102 Not disadvantaged 95 100

Gender 2.07 0.15 Female 109 94 Male 94 108

Disability status 0.04 0.84 With disability 15 16 With no disability 188 186

Note. Young were born September 1999; old were born October 1998.

436 The Journal of Educational Research

TABLE 3. Correlational Associations Between Student-Level Covariates

Variable 2 3 4 5

1. Young 0.01 0.07 0.03 "0.01 2. White — "0.02 "0.34### 0.05 3. Female — 0.06 "0.08 4. Economically disadvantaged — 0.07 5. With disability —

###p < .001.

oldest students in kindergarten, the emergent literacy skills of the two groups of students were explored using two-level multilevel models using alphabet recognition, letter sounds, spelling, and the SRF in kindergarten as the dependent vari- ables. The unconditional models (not shown) suggested that multilevel modeling was warranted as intraclass correlation coefficients (!) ranged from .07 (for alphabet recognition) to a high of .14 (for spelling), indicating that a nonignor- able source of variance was found between schools (i.e., ! > .05). Our conditional hierarchical models then prop- erly accounted for the clustered nature of the data. Inter- cepts in the models represented the scores of non-White boys who were not economically disadvantaged and had no identified disabilities. Results of the conditional models (see Table 4), although controlling for student- and school-level variables, showed that young-for-grade students consistently scored lower than the old-for-grade students, as indicated by the statistically significant negative coefficients for YOUNG (all ps < .001) on all dependent variables. For example, in the alphabet recognition task, the youngest students in the sample recognized five fewer letters than the oldest students at the beginning of kindergarten.2

Converting the YOUNG estimates to a more comparable measure of effect size (ES) resulted in a moderate ES for all dependent variables, using Cohen’s (1992) guidelines (i.e., small $ .20; moderate $ .50; large $ .80): ES for alphabet recognition was .66, ES for letter sounds was .61, ES for spelling was .62, and ES for SRF was .52.

Returning for a moment to our attrition analysis using the SRF in kindergarten, comparing differences in effect sizes of YOUNG between the sample stayers and leavers shows results that are consistent in directionality and magnitude (ESleavers = .60, ESstayers = .52). In both samples, young-for- grade students did not perform as well as the old-for-grade students and the effect size was moderate. In addition, we reran the fully specified kindergarten multilevel model us- ing the entire kindergarten sample, which included both the stayers and leavers. We included a dummy variable that in- dicated if a student stayed or left after kindergarten (STAY) and we created an interaction term (STAY % YOUNG) to test for differential bias introduced into the model as a result of attrition. Differential attrition was not found as the interaction term was nonsignificant (p = .33).

Growth of Literacy Skills over Time, by Age Group

Research questions two and three addressed whether age- related differences found at the beginning of kindergarten were still present at the end of Grade 2 and if the youngest students’ growth trajectories showed faster growth compared with the oldest students. The two questions are related be- cause if the youngest students acquired early literacy skills at a faster rate, it would be possible to catch up with the per- formance of the oldest students. Results of the fitted growth curve coefficients are presented in Table 5. The slope co- efficients indicated that the youngest students learned at a faster rate of 3.08 points per year (ES = .10) compared with the oldest students. The intercept (# 0ij) represented SRF reading scaled scores in Grade 2 and estimates indicated that the gap between the youngest and oldest participants still existed, with the youngest students scoring 11.88 points lower than the oldest students (p < .001; ES = .38).

Magnitude of Age Gaps in Comparison to Other Gaps

Our last research question asked how the age-related achievement gap compared to other possible demographi- cally related gaps. None of the other covariates based on SES, gender, or race/ethnicity was statistically significant upon kindergarten entry (all ps > .05). The statistically sig- nificant achievement gap, at the end of both kindergarten and Grade 2, was based on race/ethnicity with White stu- dents having 24.7 points (ES = .78) higher than non-White students (see Table 5) on the SRF. In comparison, by the end of Grade 2, the ES for the age-based achievement gap was approximately half the ES for the race/ethnicity-based achievement gap. Figure 1 illustrates the achievement gaps based on race/ethnicity and age. We tested for possible differ- ential growth patterns by specifying a YOUNG % WHITE interaction but we did not find statistically significant results.

Summary of Findings

The chi-square tests of homogeneity indicated that the oldest and youngest groups of students were not significantly different based on gender, economic status, race/ethnicity, and disability status. We control for these variables in the succeeding models and the statistically significant coeffi- cients for certain covariates (i.e., White and student with disabilities) indicate that these variables were important to control for. Results of the kindergarten multilevel models show that the youngest students have consistently lower scores than the oldest students. Averaging the ES of age for the fall of kindergarten results in an ES of .63. Our longitudi- nal growth model shows that the youngest students acquired literacy skills at a faster rate than the oldest students, re- sulting in a narrowing of the gap associated with age over time. Although the achievement gap narrowed between the two groups of students, gaps associated with age were not eliminated and were still statistically significant at the end

The Journal of Educational Research 437

TABLE 4. Results of Conditional Two-Level Multilevel Models Using Kindergarten Outcomes (n = 405)

Alphabet recognition Letter sounds Spelling SRF

Variables Estimate SE Estimate SE Estimate SE Estimate SE

Intercept 20.76### 0.91 12.84### 0.95 7.42### 0.68 534.73### 4.22 Young "4.99### 0.71 "4.60### 0.71 "3.35### 0.51 "18.12### 3.19 Female 1.31 0.70 0.80 0.70 0.61 0.50 6.04 3.16 White "0.22 0.82 0.16 0.85 0.47 0.61 16.03### 3.77 Economically disadvantaged "0.69 0.76 "0.39 0.76 "0.29 0.55 "3.32 3.44 With disability "2.80# 1.33 "2.20 1.33 "0.99 0.95 "14.97# 6.04 School SES 0.08# 0.04 0.03 0.04 0.01 0.03 0.26 0.20 Student-level pseudo R2a 0.13 0.11 0.11 0.14

Note. SRF = Stanford Reading First; SES = socioeconomic status. aUsing the method outlined by Bryk and Raudenbush (1992). #p < .05. ###p < .001.

of Grade 2. The difference in scores between the two groups was reduced from 18 points in the spring of kindergarten (ES = .52) to 12 points by the end of Grade 2 (ES = .38). In comparison, the ES of the age-related gap was approxi- mately half the size of the achievement gap associated with race/ethnicity.

Discussion

The results of this study demonstrate that there are signif- icant differences in early literacy development between the naturally occurring oldest and youngest students at the start

of kindergarten. Although the youngest students accelerate their literacy development at a faster pace compared to the oldest students, they never completely catch up within the time frame of our study. Statistically significant age-related literacy gaps still exist at the end of Grade 2.

We find, as in other studies (e.g., Crone & White- hurst, 1999; Oshima & Domaleski, 2006; Stipek, 2002) that kindergarten entry age is associated with differential early lit- eracy outcomes, even when children attend low-performing, high-poverty schools. In our study, younger students consis- tently had lower scores, across all the dependent variables, compared with their older peers at kindergarten entry. This

TABLE 5. Results of Unconditional and Conditional Growth Models (n = 405)

Unconditional growth Conditional growth

Fixed effects Estimate SE Estimate SE Effect size

Ending status: Intercept (# 0ij) 625.17### 1.80 619.30### 3.61 School SES 0.25 0.15 Young "11.88### 2.92 .38 Female 1.73 2.92 White 24.70### 3.28 .78 Economically disadvantaged "2.87 3.15 With disability 0.34 5.51

Rate of change: Slope (# 1ij) 43.80### 0.78 40.73### 1.86 School SES 0.00 0.07 Young 3.08# 1.53 .10 Female "1.41 1.54 White 3.72# 1.69 .12 Economically disadvantaged "0.30 1.65 With disability 5.79# 2.90 .18

Note. SES = socioeconomic status. #p < .05. ###p < .001.

438 The Journal of Educational Research

FIGURE 1. Comparison of growth trajectories among race/ethnicities and age (kindergarten to Grade 2). Scores are for boys with no disabilities. Solid lines represent scores for younger students. Dashed lines represent scores for older students.

finding is hardly surprising considering that the oldest chil- dren had a greater period of growth and development of 11 months, which is a significant portion of a 5-year-old’s lifetime (Shepard & Smith, 1986). However, by spring of kindergarten, based on the SRF, early-age achievement gaps begin to diminish.

As concluded by Stipek (2002), younger students learn more over time compared with the oldest students. How- ever, unlike the findings of Crone and Whitehurst (1999), who noted that the early literacy advantage of older children disappeared as early as Grade 1, the early-age literacy gaps in our study did not close completely even by the end of Grade 2. Although it is true that effect sizes diminished between the youngest and oldest participants, even small-to-moderate ef- fect sizes can be meaningful. An optimist may point out that the age-based literacy gap is narrowing over time, but a pes- simist would argue that the gaps continue to exist at the end of Grade 2. Although Jones and Mandeville (1990) stated that in Grade 2, the risk accounted for by age was only one twelfth the size of the risk associated with race/ethnicity, in our study, the effect size of the early-age literacy gap was much larger and half the size of the race/ethnicity literacy gap. Nevertheless, if the growth rate for our youngest par- ticipants remains constant and is maintained beyond our study period, by extrapolation, gaps associated with age are likely to close by the end of Grade 5, similar to what Oshima and Domaleski (2006) reported when they used the reading scores of the nationally representative ECLS-K sample.

For the youngest students, waiting until the end of Grades 3–5 for the age-based achievement gap to close may be a long time to be trailing behind other students and may result in unintended consequences. For example, youngest students may act out or lose motivation if they are constantly strug- gling to catch up. Parents and schools may use special edu- cation classification to provide supplemental services to stu- dents whose disability may just be relative youth (Dhuey & Lipscomb, 2010). Teachers, under increased pressure to im- prove student passing rates may unintentionally focus on the lower achieving and younger children at the expense of the higher achieving and older students (Farkas & Duffett, 2008; Loveless, 2008; Moon, Callahan, & Tomlinson, 2003). If the naturally occurring age span that exists in kindergarten classrooms is associated with differential growth trajecto- ries, outcomes, and possible unintended consequences, the debate should widen to include how best to accommodate these developmental differences.

For students who attend the most disadvantaged schools and come from economically challenged families, delaying school entry may not be a practical option. At the same time, if students are redshirted, this practice will contribute to the escalation of the kindergarten entry age differential. The National Association for the Education of Young Children (1995) does not advocate the practice of redshirting and stated that “schools must be able to respond to a diverse range of abilities within any group of children” (p. 2).

Although the age-based literacy gap exists upon kinder- garten entry, addressing these gaps prior to kindergarten may

The Journal of Educational Research 439

be an option (Coley, 2002). Although targeted pre-K pro- grams often evaluate student eligibility based on risk factors such as poverty (National Institute for Early Education Re- search, n.d.), a child’s age upon kindergarten entry may be another risk factor to consider. Numerous syntheses of the research of preschool effects have documented the immedi- ate and long-term positive effects of preschool attendance (e.g., Barnett, 2008; Huang, Invernizzi, & Drake, 2012; Loeb & Bassok, 2008; Pianta, Barnett, Burchinal, & Thornburg, 2009).

As with all studies, there are several limitations to our analyses that need to be considered when interpreting re- sults. First, this study was correlational in nature and draw- ing causal inferences from the results of this study is not possible, even with the use of a longitudinal dataset. Sec- ond, we had a limited set of student-level covariates and our measure of economic disadvantage, a student’s eligibility for FRPL, although commonly used in education studies (Sirin, 2005), is an imperfect measure (Harwell & LeBeau, 2010). Third, we only looked at reading scores from kindergarten through Grade 2 and research (Phillips, Norris, Osmond, & Maynard, 2002) has suggested that reading growth tra- jectories may change depending on the grade level studied. Finally, unobserved teacher bias (e.g., Rosenthal & Jacob- son, 1968) may introduce some measurement error because PALS and the SRF are teacher-administered assessments. However, teachers have less incentive to bias results of non-high-stakes assessments (Klein, Hamilton, McCaffrey, & Stetcher, 2000) such as PALS and the SRF and the use of teacher-administered assessments is not uncommon. Nev- ertheless, the present study provides important information about the relationship of age at kindergarten entry and the growth of early literacy skills over the first 3 years of school.

Research over the past few decades (Crone & Whitehurst, 1999; Oshima & Domaleski; 2006; Shepard & Smith, 1986; Stipek & Byler, 2001) has pointed out that early-age achievement gaps are small and disappear over time. The growth trajectories from our present study, if kept constant, support such findings. Although it is reassur- ing for the parents of young-for-grade students that studies (Angrist & Krueger, 1991; Cascio & Schanzenbach, 2007; Deming & Dynarski, 2008; Lincove & Painter, 2006) have found no advantages for older students in relation to high school achievement, college enrollment, graduation rates, and wages, those same parents may be more anxious about immediate short-term outcomes. Ultimately, parents will have to decide on what is best for their children and nu- merous factors have to be taken into consideration when deciding whether to delay kindergarten entry or not (for a set of considerations, see Oshima & Domaleski, 2006). Given the other risks encountered by being the youngest in class, such as being higher risk candidates for being re- tained (Bedard & Dhuey, 2006; Graue & DiPerna, 2000; Lincove & Painter, 2006) or being diagnosed with learning disabilities (Dhuey & Lipscomb, 2010; Evans et al., 2010), parents may subscribe to the old adage: When in doubt,

hold them out. However, parents should bear in mind that if students are held back, children should continue to be in a learning-rich environment and that schooling itself has a direct impact on the development of emergent literacy and reading skills (Crone & Whitehurst, 1999) that is not provided by the simple passage of time. Although holding students out of school may be viewed as giving children the gift of time to mature, it may also be seen as what Graue and Di Perna (2000) referred to as a theft of opportunity to learn. Regardless of parental practices, age-related differences will continue to exist so differentiated instruction will be neces- sary to meet the needs of all children. Parents are not alone in facing this issue and schools and educators will continue to have the shared responsibility of equitably addressing such gaps.

NOTES

1. Participants were enrolled in Reading First (RF; U.S. Department of Education, 2009) schools. As an RF grant recipient, schools were required to use a comprehensive reading program, were provided funds for profes- sional development, and had access to a reading coach within the school. However, comparisons based on reading test scores of students attending RF schools and comparable/control schools (i.e., high-poverty, low-performing schools that did not receive RF support) have shown no significant dif- ferences on a local (Huang & Moon, 2008) and a national level (Gamse, Bloom, Kemple, & Jacob, 2008). An explanation of the nonsignificant find- ings is that many non-RF schools were also using a comprehensive reading program aligned with the principles of RF, provided assistance to struggling readers, and non-RF teachers also participated in the same or similar pro- fessional development programs targeted at RF teachers (U.S. Department of Education, 2008). In the end, RF schools and non-RF schools may not have differed much in terms of the factors that the grant provided for.

2. We had used school fixed-effect Tobit (Tobin, 1958) regression anal- yses for the two dependent variables that exhibited partial ceiling and floor effects (i.e., alphabet recognition and spelling, respectively) but results did not change substantially compared with HLM results. To maintain con- sistency with the presentation of results, we retained the use of two-level HLM.

REFERENCES

Adams, M. (1990). Beginning to read: Thinking and learning about print. Cambridge, MA: MIT Press.

Angrist, J. D., & Krueger, A. B. (1991). Does compulsory school attendance affect schooling and earnings? The Quarterly Journal of Economics, 106, 979–1014.

Barnett, W. (2008). Preschool education and its lasting effects: Research and policy implications. Boulder, CO: Education and the Public In- terest Center & Education Policy Research Unit. Retrieved from http://epicpolicy.org/publication/preschool-education

Bedard, K., & Dhuey, E. (2006). The persistence of early maturity: Inter- national evidence of long-run age effects. Quarterly Journal of Economics, 121, 1437–1472.

Bracey, G. (1989). Age and achievement. Phi Delta Kappan, 70, 732. Bronson, P., & Merryman, A. (2009, September 3). Should

children redshirt kindergarten? Newsweek. Retrieved from http://blog.newsweek.com/blogs/nurtureshock/archive/2009/09/03/should- children-redshirt-kindergarten.aspx

Bryk, A., & Raudenbush, S. (1992). Hierarchical linear models: Applications and data analysis methods. Newbury Park, CA: Sage.

Cabell, S. Q., Justice, L. M., Konold, T. R., & McGinty, A. S. (2011). Profiles of emergent literacy skills among preschool children who are at risk for academic difficulties. Early Childhood Research Quarterly, 26, 1–14. doi:10.1016/j.ecresq.2010.05.003

Cascio, E. (2008, August). How and why does age at kindergarten entry mat- ter? Federal Reserve Bank of San Francisco Economic Letter. Retrieved from

440 The Journal of Educational Research

http://www.frbsf.org/publications/economics/letter/2008/el200808–24. pdf

Cascio, E., & Schanzenbach, D. W. (2007). First in the class? Age and the education production function (Working Paper No. 13663). Cam- bridge, MA: National Bureau of Economic Research. Retrieved from http://www.nber.org/papers/w13663

Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155–159. doi:10.1037/0033-2909.112.1.155

Coley, R. (2002). An uneven start: Indicators of inequality in school readiness. Princeton, NJ: Educational Testing Service.

Cosden, M., Zimmer, J., & Tuss, P. (1993). The impact of age, sex, and ethnicity on kindergarten entry and retention decisions. Educational Eval- uation and Policy Analysis, 15, 209–222.

Crone, D., & Whitehurst, G. (1999). Age and schooling effects on emergent literacy and early reading skills. Journal of Educational Psychology, 91, 604–614. doi:10.1037/0022-0663.91.4.604

Crosser, S. (1998). He has a summer birthday: The kindergarten en- trance age dilemma. Champaign, IL: ERIC Clearinghouse on Elemen- tary and Early Childhood Education. Retrieved from http://eric.ed.gov/ PDFS/ED423079.pdf

Datar, A. (2006). Does delaying kindergarten entrance give children a head start? Economics of Education Review, 25, 43–62.

Deming, D., & Dynarski, S. (2008). The lengthening of childhood. Journal of Economic Perspectives, 22, 71–92.

Dhuey, E., & Lipscomb, S. (2008). What makes a leader? Relative age and high school leadership. Economics of Education Review, 27, 173–183.

Dhuey, E., & Lipscomb, S. (2010). Disabled or young? Relative age and special education diagnoses in schools. Economics of Education Review, 29, 857–872.

Elder, T., & Lubotsky, D. (2008). Kindergarten entrance age and children’s achievement: Impacts of state policies, family background, and peers. Journal of Human Resources, 44, 641–683.

Evans, W. N., Morrill, M. S., & Parente, S. T. (2010). Measuring inappro- priate medical diagnosis and treatment in survey data: The case of ADHD among school-age children. Journal of Health Economics, 29, 657–673.

Farkas, S., & Duffett, A. (2008). Results from a national teacher survey: High achieving students in the era of NCLB. Washington, DC: Thomas B. Fordham Institute.

Foy, J., & Mann, V. (2006). Changes in letter sound knowledge are associ- ated with development of phonological awareness in pre-school children. Journal of Research in Reading, 28, 143–161.

Gamse, B., Bloom, H., Kemple, J., & Jacob, R. (2008). Reading First im- pact study: Interim report. Washington, DC: National Center for Edu- cation Evaluation. Retrieved from http://ies.ed.gov/ncee/pdf/20084016. pdf

Gladwell, M. (2008). Outliers: The story of success. New York, NY: Little, Brown.

Gootman, E. (2006, October 19). Preschoolers grow older as parents seek an edge. The New York Times. Retrieved from http://www.nytimes.com/ 2006/10/19/nyregion/19kindergarten.html

Graue, M. E., & DiPerna, J. (2000). Redshirting and early retention: Who gets the “gift of time” and what are its outcomes? American Educational Research Journal, 37, 509–534.

Harcourt Assessment. (2004). Stanford Reading First: Technical data report. San Antonio, TX: Author.

Harwell, M., & LeBeau, B. (2010). Student eligibility for a free lunch as an SES measure in education research. Educational Researcher, 39, 120–131.

Huang, F., & Moon, T. (2008, March). Reading First’s impact on reading outcomes on second grade students: One county’s experience in a Mid-Atlantic state. Poster session presented at the annual meeting of the American Educational Research Association, New York, NY.

Invernizzi, M., Juel, C., Swank, L., & Meier, C. (2009). PALS-K technical reference. Charlottesville, VA: University Printing Services.

Invernizzi, M., Swank, L., Juel, C., & Meier, J. (2003). Phonological Awareness Literacy Screening–Kindergarten. Charlottesville, VA: Univer- sity Printing Services.

Jones, M. M., & Mandeville, G. K. (1990). The effect of age at school entry on reading achievement scores among South Carolina students. Remedial and Special Education, 11, 56–62.

Klein, S., Hamilton, L. S., McCaffrey, D. F., & Stecher, B. M. (2000). What do test scores in Texas tell us? Santa Monica, CA: RAND.

Lin, H. H., Freeman, L., & Chu, K. L. (2009). The impact of kindergarten enrollment age on academic performance through kindergarten to fifth grade. European Journal of Social Sciences, 10, 45–54.

Lincove, J., & Painter, G. (2006). Does the age that children start kinder- garten matter? Evidence of long-term educational and social outcomes. Educational Evaluation and Policy Analysis, 28, 153–179.

Loeb, S., & Bassok, D. (2008). Early childhood and the achievement gap. In H. F. Ladd & E. B. Fiske (Eds.), Handbook of research in education finance and policy (pp. 517–534). London, UK: Routledge.

Loveless, T. (2008). Analysis of NAEP data: High achieving students in the era of NCLB. Washington, DC: Thomas B. Fordham Institute.

Malone, L., West, J., Flanagan, K., & Park, J. (2006). The early reading and mathematics achievement of children who repeated kindergarten and who began school a year late. Washington, DC: U.S. Department of Education, National Center for Education Statistics.

McBride-Chang, C. (1998). The development of invented spelling. Early Education and Development, 9, 147–160. doi:10.1207/ s15566935eed0902 3

Meisels, S. (1992). Doing harm by doing good: Iatrogenic effects of early childhood enrolment and promotion policies. Early Childhood Research Quarterly, 7, 155–174.

Moon, T. R., Callahan, C. M., & Tomlinson, C. A. (2003). Effects of state testing programs on schools with high concentrations of student poverty: Good news or bad news? Current Issue in Education, 6(8). Retrieved from http://cie.asu.edu/volume6/number8/

National Association for the Education of Young Children. (1995). NA- EYC position statement on school readiness. Retrieved from http://www. naeyc.org/files/naeyc/file/positions/PSREADY98.PDF

National Institute of Child Health and Human Development Early Child Care Research Network. (2007). Age of entry to kindergarten and chil- dren’s academic achievement and socioemotional development. Early Education and Development, 18, 337–368.

National Institute for Early Education Research. (n.d.). State pre-K program elegibility requirements. Retrieved from http://nieer.org/pdf/eligibility. pdf

O’Donnell, K. (2008). Parents’ reports of the school readiness of young children from the national household education surveys program of 2007. Washington, DC: National Center for Education Statistics.

Oshima, T. C., & Domaleski, C. S. (2006). Academic performance gap be- tween summer-birthday and fall-birthday children in grades K-8. Journal of Educational Research, 4, 212–217.

Paul, P. (2010, August 20). The littlest redshirts sit out kinder- garten. The New York Times. Retrieved from http://www.nytimes.com/ 2010/08/22/fashion/22Cultural.html

Phillips, L., Norris, S., Osmond, W., & Maynard, A. (2002). Relative reading achievement: A longitudinal study of 187 children from first through sixth grades. Journal of Educational Psychology, 94, 3–13.

Pianta, R., Barnett, W., Burchinal, M., & Thornburg, K. (2009). The effects of preschool education: What we know, how public policy is or is not aligned with the evidence base, and what we need to know. Psychological Science in the Public Interest, 10, 49–88.

Rosenthal, R., & Jacobson, L. (1968). Pygmalion in the classroom. New York, NY: Holt, Rinehart, & Winston.

Rumberger, R. (2003). The causes and consequences of student mobility. The Journal of Negro Education, 72, 6–21.

Shepard, L. A., & Smith, M. L. (1986). Synthesis of research on readiness and kindergarten retention. Educational Leadership, 44, 78–86.

Sirin, S. R. (2005). Socioeconomic status and academic achievement: A meta-analytic review of research. Review of Educational Research, 75, 417–453.

Singer, J., & Willett, J. (2003). Applied longitudinal data analysis. New York, NY: Oxford University Press.

Snow, C. E., Burns, M. S., & Griffin, P. (1998). Preventing reading difficulties in young children. Washington, DC: National Academy Press.

SAS Institute Inc. (2008). SAS/STAT 9.2 user’s guide. Cary, NC: SAS Institute Inc.

Stipek, D. (2002). At what age should children enter kindergarten? A question for policy makers and parents. Society for Research in Child De- velopment Social Policy Report, 16, 1–16.

Stipek, D., & Byler, P. (2001). Academic achievement and social behav- iors associated with age of entry into kindergarten. Journal of Applied Developmental Psychology, 22, 175–189.

Tobin, J. (1958). Estimation of relationships for limited dependent vari- ables. Econometrica, 26, 24–36.

Torgesen, J., & Davis, C. (1996). Individual difference variables that predict responses to training in phonological awareness. Journal of Experimental Child Psychology, 63, 1–21.

The Journal of Educational Research 441

Uphoff, J. K., & Gilmore, J. (1985). Pupil age at school entrance–how many are ready for success? Educational Leadership, 43, 86–90.

U.S. Department of Education. (2008). Reading First implementa- tion and evaluation final report: Report highlights. Retrieved from http://www2.ed.gov/rschstat/eval/other/readingfirst-final/index.html

U.S. Department of Education. (2009). Reading First. Retrieved from http://www2.ed.gov/programs/readingfirst/index.html

Weil, E. (2007, June 3). When should a child start kindergarten? The New York Times Magazine. Retrieved from http://www.nytimes.com/2007/ 06/03/magazine/03kindergarten-t.html?pagewanted=1& r=1

Zill, N., Loomis, L. S., & West, J. (1995). The elementary school performance and adjustment of children who enter kindergarten late or repeat kindergarten: Findings from national surveys. Washington, DC: U.S. Department of Education, National Center for Education Statistics.

AUTHORS NOTE

Francis L. Huang is a Senior Scientist at the Curry School of Education, University of Virginia. His research interests focus on the use of quantitative methods in evaluation and policy studies.

Marcia A. Invernizzi is the Henderson Professor of Ed- ucation at the Curry School of Education, University of Virginia. Her research interests revolve around early in- tervention and assessment in early language and literacy learning.

Copyright of Journal of Educational Research is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.