July 15, 2008
Child and Family Characteristics Associated With Nonpromotion in Preprimary Education
By Gadeyne, Els Onghena, Patrick; Ghesquiere, Pol
ABSTRACT: Educational research has found few benefits of preprimary school nonpromotion; nevertheless, in many countries the practice remains quite common. The aim of this study is to extend knowledge regarding preprimary nonpromotion via investigation of a wide set of child and family characteristics deemed predictive and discriminative of different postkindergarten nonpromotion alternatives in Flanders, Belgium. The study participants were 3,633 children (3,375 were promoted to first grade, 138 repeated kindergarten, 71 transferred to special education, 49 attended a transition room). Child preacademic scores and psychosocial functioning were relatively strong predictors for promotion versus nonpromotion. Child and family demographic characteristics were more decisive regarding the choice between retention, special education, and a transition program. The study raises questions about nonpromotion practices in terms of selectivity mechanisms, prevention, and educational alternatives.
School nonpromotion practices in primary and preprimary education are widespread, although significant differences are recorded between countries-even within Western Europe (Organization for Economic Cooperation and Development, OECD, 1998; Paul, 1997). Different preprimary nonpromotion practices also exist in the United States, including redshirting or delaying the child's entry into preschool, and the nonpromotion of a student prior to or following kindergarten (Carlton & Winsler, 1999; Mantzicopoulos & Neuharth- Pritchett, 1998; Matthews, May, & Kundert, 1999). In Flanders (the Dutch-language region of Belgium), the present research and that of a birth cohort study (Van Landeghem & Van Damme, 2004) show a delay of at least 1 year for approximately 6% of children entering primary education.
Although many teachers believe that early nonpromotion provides a child with the time needed to naturally develop school readiness (Troncin, 2004), a growing body of literature confirms the negative effects of grade retention in primary education (Holmes, 1989; Jimerson, 2001). The reasons for preprimary nonpromotion differ somewhat from those for later school nonpromotion; therefore, it was thought that its effects might differ as well. To date, however, research on preprimary extra-year programs (primarily conducted in the United States) does not show clear advantages for preprimary nonpromotion (Carlton & Winsler, 1999; Cosden, Zimmer, & Tuss, 1993; Mantzicopoulos, 2003; Matthews, et al., 1999; Owings & Magliaro, 1998; Shepard, 1989; Troncin). Nevertheless, in several countries- including Belgium-this research has not yet produced a significant decrease in retention practices.
Consequently, the focus of research has shifted to examining the reasons that teachers continue to claim that benefits of preprimary nonpromotion do exist, and to exploring the child, family, teacher, and school characteristics that could be associated with nonpromotion practices. This shift has resulted in the identification of several problems. Research indicates that-as is the case for primary students-preschool-aged children, their parents, and their peers often consider nonpromotion to be a stigmatizing consequence of the child's failure at school, and sometimes even as indicating insufficient effort on the part of the child (Shepard & Smith, 1989; Troncin, 2004). International research has revealed that (as in primary education) nonpromotion practices occur more frequently with younger children, boys, and children of nonnative extraction. Additionally, children from lower-income families and families with low parent-school participation levels are overrepresented in the nonpromoted group (Byrd & Weitzman, 1994; Cosden et al., 1993; Delgado & Scott, 2006; Jimerson, Carlson, Rotert, Egeland, & Sroufe, 1997; Mantzicopoulos & Neuharth- Pritchett, 1998; McCoy & Reynolds, 1999; Troncin; Wallingford & Prout, 2000). Although, on average, these children do show more preacademic and psychosocial problems in preprimary education, the decision to hold them back, in turn, produces negative consequences for them (Merle, 1998) and enlarges the educational gap between the children who are held back and those who are promoted.
This study was designed to extend current knowledge about the predictors and antecedents of preprimary nonpromotion in several ways. Internationally, much of the research on educational nonpromotion focuses either on grade retention or on referral to special education. Both trajectories are expected to provide answers to distinct educational problems. Regarding grade retention, it is expected that-to attain general education objectives-some children need (a) more time to mature, or (b) a complete rehearsal of (most of) die subject matter of a particular grade. As to referral to special education, it is assumed that specific, severe, and persistent characteristics (disabilities) hamper a child's development to the extent that specialized treatment and adapted education is required. The authors of this article, however, argue that for many educational systems it is useful to study grade retention, transitional programs, and referral to special education simultaneously. For example, in Belgium the distinct nonpromotion decisions often are based on subjective and overlapping arguments. Findings regarding the prediction of (preprimary) grade retention and on referral to special education (although scarce at the preprimary level) seem to parallel each other to a great extent, but currently are not integrated. To fully understand the phenomenon of nonpromotion, the prediction of all three trajectories should be addressed concurrently.
Considerable overlap is found between educational, demographic, and contextual predictors of nonpromotion; therefore, only a joint study can reveal their unique predictive value. Again, until now this rarely has been done for preprimary nonpromotion (Artiles & Trent, 1994; Mantzicopoulos & Neuharth-Pritchett, 1998).
Particularly when preprimary nonpromotion is considered, psychosocial predictors also should be included. At this educational level "school readiness" often is called upon to justify a nonpromotion decision. This is a poorly defined concept, however, and refers alternatively to preacademic as well as to psychosocial conditions related to a successful start in primary education (Carlton & Winsler, 1999). It currently is unclear to what extent these conditions are interrelated in terms of nonpromotion decisions.
The researchers in the present study also were interested in determining to what degree children facing nonpromotion received extra educational support in kindergarten. We specifically examined the question of whether such additional support is differentially related to the nonpromotion alternatives.
AIMS OP THIS STUDY
Data from a large-scale longitudinal study on the development and school careers of pupils in primary education in Flanders (the northern, Dutchspeaking region of Belgium) were analyzed to address the following research questions:
1. How are differences in preacademic and psychosocial functioning related to distinct nonpromotion alternatives?
2. How are child demographic and family characteristics related to distinct nonpromotion alternatives?
3. To what extent do child functioning characteristics and demographic characteristics overlap in the prediction of nonpromotion alternatives?
4. What extra help did children receive at school before nonpromotion was advised?
Researchers studied the differences between promoted and nonpromoted children in preprimary education, and explored the differences among children in distinct preprimary nonpromotion trajectories.
PREPRIMARY EDUCATION AND NONPROMOTION IN BELGIUM
Before presenting the methodology of the study, this article first describes preprimary education in Belgium as well as nonpromotion practices; this is necessary to contextualize the study.
In Belgium, a rather unique 3-year preprimary program immediately precedes primary education, and includes children who are from 2 1/ 2 to 6 years old. This program ends with preparatory reading, writing, and mathematic skills. The last grade of this preprimary program generally can be compared to kindergarten in the United States or to the reception class in the United Kingdom. This article refers to the third grade of the Belgian preprimary program as "kindergarten." In contrast to the U.S. educational system, however, this grade is not structurally different from the two preceding grades. Additionally, repetition of a preprimary year other than the final year is a rare occurrence.
In Flanders, most of the children who fall behind their class prior to the start of primary education must spend an extra year in preprimary education (usually kindergarten), or the child is referred to special education. Although not officially recognized, the Belgian special education system has a 1-year transitional program that is designed as a widespread alternative for children who do not meet the requirements for a successful start in first grade and who do not have mental retardation. The prerequisites for- and the beginning stages of-the reading, writing, and math process are taught in a well-structured format that also includes enough time for the students to engage in play and other nonacademic activities. Note that after a special education year, however, fewer children than the number intended or claimed actually return to the general education school system. Another alternative that sometimes is available is an analogous transition class that is organized in regular Flemish schools. In this class children receive a specific preparation program for the first grade, but they are not labeled as receiving special education and they do not have to leave their regular schools. This alternative, however, is far less widespread than the use of grade repetition or the special education transition class. It is important to note that, in contrast to other countries, in Belgium schools contextual factors might not be particularly relevant to the issue of nonpromotion. Preliminary analyses in this study indicated that there were no statistically significant differences between schools or classes in terms of the ratio of students not promoted after kindergarten. In part, this might be due to the (relatively) low occurrence of the phenomenon. In most classes, zero, one, or two children do not progress to first grade. Although nonpromotion does not occur in every class every year, there are few schools that have an explicit policy of social promotion. As such, it was decided that this research would not examine teacher or school variables with regard to the prediction of nonpromotion.
As noted, the present study (called "SiBO," which stands for "Schoolloopbanen in het Basisonderwijs" [School careers in primary education]) is part of a large-scale longitudinal study of primary education in Flanders, Belgium. The research group consisted of 3,633 children who were all born in 1997 and who entered kindergarten for the first time in the 2002 to 2003 school year. At the school level, the sample (5% of the school population) is representative of the types of schools (public or private), school size, region, and percentage of children attending who are eligible for governmental funding for school (Verhaeghe, Macs, Gombeir, & Peeters, 2002).
In the year following kindergarten, the researchers identified children who were not promoted to the first grade and recruited as many as possible for the study. As a result, the sample included 138 children who were repeating kindergarten (the K3 group), 71 children who were referred to special education (the SE group), and 49 children who were in a transition class in general education (the TR group). Four children could not be traced because during the 2003 to 2004 school year they no longer were listed in the registration system of the Department of Education, and their former schools had no recent school information or current home address for any of them (presumably their families have left the country). The control group consisted of 3,375 children from the initial group who were promoted to the first grade in 2003 to 2004 (the L1 group). For the SE group, only three types of special education were taken into account: that for children with (a) specific learning disabilities, (b) behavior problems, or (c) minor mental disabilities.
The number of children in the transition classes (noted previously) was not representative of the starting group in kindergarten; only 12 children from that sample were transferred to the two transition rooms. The organization of transition rooms in Flanders appears to be related to low socioeconomic status (SES) and high immigration rates. To include transition rooms reliably they were over-sampled, and two extra school samples were used. Admittedly, this additional selection implies that children in this sample who have low SES as well those who are immigrants are more likely to be referred to a transition room than are the children in the original starting group.
Child Variables. The preacademic outcomes for the children were assessed at the beginning of kindergarten (during last week of September and first week of October) and at the end of kindergarten (during last half of May). Tests were administered to small groups of about eight children. Language abilities were assessed by a 40- item Flemish adaptation of a standardized Dutch test (Cito, 1996, adapted by Ponjaert-Christoffersen, Andries, Celestin-Westreich, & Samay, 2000). The items assess listening comprehension, identifying sounds in words and rhymes, identifying the first or the last word in word sequences, orientation in writing, and auditory synthesis. The correlation of the shortened version to the original 56-item test was 0.97, and Cronbach's alpha was 0.86. The preacademic mathematics test (Conception of Math, Dudal, 2004) contains 40 items examining concepts such as quantities, space and time, ordinal numbers up to 10, and counting. Cronbachs alpha ranged between 0.92 and 0.93. Further, the language and mathematic achievement tests administered at the beginning and the end of kindergarten were calibrated through item response modeling (IRT), which is a means to express scores for different tests using a single scale. In this way, in both academic areas, academic progress between the beginning and the end of kindergarten can be expressed as gain scores (the calibrated End Test score minus the calibrated Begin Test score). This technique is preferable to value-added modeling.
To assess psychosocial functioning of the children in kindergarten, questionnaires were administered to the kindergarten teachers in February (Macs, 2003). For each child in their classroom, the teachers completed a series of scales on prosocial behavior (4 items), aggressive behavior (4 items), hyperactivity- distractibility (4 items), asocial behavior (4 items), troubled/ anxious behavior (4 items), conflict with the teacher (4 items), proximity to the teacher (4 items), integration-popularity in the class (4 items), self-confidence (2 items), school well-being (4 items), cooperative participation in the class (4 items), autonomous participation in the class (4 items), and attitude toward work (3 items; 49 items total). All scales were derived from existing and validated instruments (e.g., the Child Behavior Scale by Ladd & Profilet, 1996, translated by Simoens, 2001; the Student-Teacher Relationship Scale by Pianta, 2001, translated by Roncada, 2001, shortened by Cornelissen & Verschueren, 2001, 2002; the Pupil's Profile by Driessen, van Langen, & Vierke, 2000 and Jungbluth, Roede & Roeleveld, 2001; the Teacher Rating Scale of School Adjustment by Birch & Ladd, 1997, translated by Simoens, 2001, shortened by Cornelissen & Verschueren, 2001, 2002). The item format was a 6- point scale.
In the present study Cronbach alpha for these scales ranged between 0.79 and 0.92, except for troubled/anxious behavior which had a Cronbach alpha of 0.60. A correlational analysis and a second- order principal-component analysis (PCA) on the scales confirmed their construct validity (Macs, 2003). The sign and magnitude of the pair-wise correlations was in line with theoretical expectations. Scales on positive social functioning (e.g., prosocial behavior, popularity), for example, correlated in a positive way with scales on teacher relationship and work attitude (0.42 = r = 0.66), and in a negative way with scales on negative social and behavioral functioning (e.g., aggressive and asocial behavior; -0.40 = r = - 0.70). Hyperactive behavior clearly correlated in a negative way with scales on work attitude (-0.55 = r = -0.68), and anxious behavior was least correlated with other scales. Further, the PCA grouped more academically oriented scales, externalizing behavior and other social scales.
Researchers assessed extra educational support-beyond the teacher's typical efforts-that was provided to each child in kindergarten by means of a questionnaire that teachers completed at the end of the year (Gadeyne, 2003). Using a class list, each teacher identified the children who had received school-related as well as extracurricular support in the course of the school year. Explanatory notes provided a listing of types of support.
Child demographic information (date of birth and gender) was obtained from an extensive questionnaire that parents completed in February 2003 and which covered a wide range of child and family variables (Reynders, Van Heddegem, Nicaise, & Van Damme, 2004).
Family Variables. The parent questionnaire (Reynders et al., 2004) also included questions about the language spoken at home and questions regarding maternal educational background. The language spoken at home offered an indication of the child's ethnic origin and the child's familiarity with the Dutch language. For the present study, 2 categories were distinguished: families speaking only Dutch (N = 2,957 or 82.2%), and other families (those which use another language or that speak Dutch as well another language, N = 642 or 17.8%). With regard to the maternal educational background, 4 categories were distinguished: mothers who have primary education at most (up to 12 years of school; N = 159 or 4.9%); mothers who completed lower secondary education (up to 15 years of school; N = 496 or 15.3%); mothers who have upper secondary education (up to 18 years of school; N = 1,222 or 37.7%); and mothers who have tertiary education (more than 18 years; N = 1,367 or 42.1%).
The teacher questionnaire also included 7 items on parental school involvement for each child, and these were answered using a 6- point scale (e.g., "The parents show interest in the school activities of their child"; "We have a good contact with the parents of this child"). This scale was derived from two family subscales designed by Driessen, van Langen, and Vierke (2000, 2002). Originally, a 2-scale structure was proposed ("family-school contact" and "involvement of the family in school learning"), but due to an intercorrelation coefficient of 0.79, the scales were combined into a single-scale score. Cronbach's alpha was 0.90. DATA ANALYSIS
For research questions 1, 2, and 3, researchers initially performed separate univariate ANOVAs with a post-hoc test on each of the independent variables. Next tested were one-step logistic regression models with groups of related independent variables (entered simultaneously). The different educational trajectories of pupils (i.e., K3, SE, TR, and L1, described earlier) comprised the categories of the dependent variable, whereas all child-level variables (including family variables) were involved as independent variables. In the case of multi-collinearity, the number of independent variables was reduced. In this context, a tolerance below 0.40 was considered problematic (Allison, 1999). Last, one- step logistic models were tested including all variables estimated as relevant in the subgroup models. To interpret the logistic regression coefficients, the exponentiated estimates can be expressed further as the percentage change in odds associated with a one-unit increase in the independent variable (Pampel, 2000). Additionally, the deviance score is reported as a "fit index" reflecting the model improvement going from the intercept-only model to the fitted model. The pseudo-R^sup 2^ reflects the predictive power of the fitted model by comparing the deviance score to the intercept only -2 Log Likelihood value.
Means, Standard Deviations, and One-Way Analyses of Variance (ANOVA) for the Academic Variables in Kindergarten in Relation to the Promotion Decision After Kindergarten
An advantage of using logistic regression instead of discriminant analysis is that the set of predictor variables does not necessarily have to display a multivariate normal distribution (Lea, 1997). Based on Allison (1999), PROC LOGISTIC in the Statistical Analysis Software (SAS) system was used for binary logistic regression (except for the calculation of the chi^sup 2^ for the estimates of the individual parameters, which is performed in PROC GENMOD), and PROC CATMOD for multinomial logistic regression.
Research question 4 was analyzed more qualitatively. The frequency and nature of extra support in kindergarten for each educational trajectory is described below.
PREACADEMIC AND PSYCHOSOCIAL CHILD PREDICTORS TO NONPROMOTION
For each of the four promotion groups, Table 1 shows the means and standard deviations of the Begin Test and End Test scores for language and mathematics in kindergarten, and of the gain scores over the course of the year. The far-right column shows the F- and R^sup 2^-values of a four-group ANOVA performed on each of the preacademic outcome variables. Those means that are found in a row sharing subscripts differ to a statistically significant degree from one another on post-hoc analyses.
All three nonpromotion groups (SE, K3, and TR), had significantly lower test scores for language and preacademic mathematics as compared to those of the promoted group (L1), both at the beginning and at the end of kindergarten. The difference amounted to 1.5 standard deviations (in proportion to the total-sample SD). Additionally, the TR group showed slightly higher language test scores at the end of kindergarten as compared to the K3 group. Similarly, the TR group showed higher language gain scores than those of both the K3 group and the L1 group. Apart from this, however, no differences in language or mathematic gain scores were noted between any of the groups. Likewise, the proportion of explained variance in test scores ranged between 0.13 and 0.15, whereas the explained variance in gain scores was negligible. Apart from the language scores of the TR group, no other test or gain score discriminated between the three nonpromoted groups.
Means, Standard Deviationss, and One-Way Analyses of Variance (ANOVA) for the Psychosocial Variables Assessed by the Kindergarten Teachers in Relation to the Promotion Decision After Kindergarten
On all 13 psychosocial variables measured by the teacher questionnaire (see Table 2), the promoted group outperformed the three nonpromoted groups. Members of the promoted group had better scores on prosocial behavior, proximity to the teacher, integration/ popularity, school well-being, co-operative and autonomous participation in the classroom, hyperactive behavior, asocial behavior, work attitude, and self-confidence. Further, the members of this group showed less troubled/anxious behavior than did those in the SE and K3 groups, and demonstrated less conflict behavior with the teacher than did those in the SE and TR groups. The promoted-group members also were less aggressive than those in the TR group. The difference varied between 0.5 and 1 standard deviation (in proportion to the total-sample SD). Within the nonpromoted groups, the SE group functioned significantly worse than did the K3 and TR groups on measures of hyperactive behavior and work attitude, and members of this group had lower scores than those of the K3 group members on autonomous participation. These differences equal 0.5 standard deviation. The proportion of explained variance in psychosocial variables was found to be below that of the preacademic variables, and ranged between 0.005 and 0.08, except for the scale "autonomous participation" which had an R^sup 2^ of 0.13.
Proportion of Children Within Each Nonpromotion Group Born in the Successive Calendar Months
Taken together, separate preacademic and psychosocial child variables as assessed in kindergarten are more related to promotion versus nonpromotion than to the specific nonpromotion alternative. The differences in the standard deviations and in R^sup 2^ indicate that, in general, individual preacademic variables are stronger predictors than are individual psychosocial variables.
CHILD DEMOGRAPHIC AND FAMILY PREDICTORS TO NONPROMOTION
Regarding the child demographic variables, gender did not have an individual statistically significant effect, chi^sup 2^(3,3628) = 7.504, p = 0.06; with respect to month of birth, the promoted group was older than the non-promoted group, indicating that children born in the last months of the calendar year ran a greater risk of nonpromotion, F(1,3631) = 72, p
The 3 family variables discriminated to a significant degree between the promotion groups (see Table 3). Of the promoted group, 83% of the children came from Dutch-speaking families; in the SE sample this percentage was as great as 93%. Conversely, of the K3 sample only 66% of the children came from Dutch-speaking families, and in the TR group this percentage was a mere 28%, chi^sup 2^(N = 3599, df = 3) = 128.12; p
The scale score on parental school involvement favored the promoted group over each of the nonpromoted groups, F(3,3441) = 46.25, p
In sum, separate child demographic and family variables are related both to promotion versus nonpromotion and to distinct nonpromotion alternatives.
CUMULATIVE PREDICTIVE POWER REGARDING NONPROMOTION
The researchers recorded that, in univariate analyses, nearly all selected child and family variables separately discriminated in some way between the promotion groups; this prompted them to seek to determine the variables' unique as well as joint explanatory power. As noted in the data-analysis section, two 1-step approaches were applied. Logistic regressions first were applied to each of the four groups of child and family variables. All relevant variables emerging from the analyses then were entered in an overall logistic regression function. Separate analyses were performed on promotion versus nonpromotion and on SE versus K3 versus TR within the nonpromoted group. The results of the subgroup analyses are presented in Table 4.
Distribution of Family Variables
Regarding the overall analyses, it is not meaningful to interpret individual significant predictors because they refer to odds changes that are due to the change in one variable-given that all other variables remain constant. Using a large set of interrelated variables means that extreme odds changes-barely related to real- life situations-can appear. In this study, the overall analyses mainly assess the total predictive power of the chosen set of variables. In the subgroup analyses the odds changes of individual significant predictors are reported between brackets. They refer to an increase in the odds of one promotion alternative over another for each 1-point increase in the variable of interest (given equal scores on the other variable[s] being analyzed). This implies that the magnitude of the odds change is relative to the scale of the variable of interest, so that odds changes cannot be compared to one another without taking into account variable scales.
Due to multi-collinearity in the subgroup analysis on the preacademic variables, only two variables were selected-the calibrated End Test scores for language and for preacademic mathematics. For the prediction of promotion versus nonpromotion following kindergarten, this resulted in a pseudo-R^sup 2^ of 0.37. Both predictors showed individual significance (with equal scores on the alternate academic variable, odds of promotion over nonpromotion for each 1-point increase are +26% and +21%). A significant interaction term also indicated that the combination of high language and mathematic test scores still slightly increased these odds changes. Within the nonpromotion groups, the joint preacademic scores revealed a pseudo-R^sup 2^ of 0.02. Again, both variables showed individual significance (with equal scores on the alternate academic variable; for math, the odds of SE over TR for each onepoint increase are +10%, and the odds of SE over K3 are +8%; for language the odds of TR over K3 are +6%). To overcome multi- collinearity problems, 8 of the 13 psychosocial variables were selected. This battery proved to be significant in predicting promotion versus nonpromotion (pseudo-R^sup 2^ = 0.27), but not in the prediction of the specific nonpromotion trajectory (pseudo- R^sup 2^ = 0.03). Further, 3 individual variables seemed particularly interesting with regard to the former prediction: hyperactive behavior, asocial behavior, and autonomous participation. With equal scores on the other psychosocial variables, the odds of promotion over nonpromotion for each 1-point increase were -32%, -28%, and +234% respectively.
Logistic Regressions With Subgroups of Predictor Variables in Relation to Nonpromotion
Logistic Regressions With Subgroups of Predictor Variables in Refation to Nonpromotion
The 3 joint family variables did not reveal any predictive power regarding promotion versus nonpromotion, similar to the preacademic or psychosocial child variables; the pseudo-R^sup 2^ amounted to only 0.09. Both maternal educational level and parental school involvement were found to have unique statistically significant effects (with equal scores on the other family variables), the odds of promotion over nonpromotion for each 1-level or 1-point increase were +90% and +19% respectively. Within the nonpromoted groups, the family variables equally resulted in a pseudo-R^sup 2^ of 0.09. Here, however, only the language spoken at home exhibited a unique statistical significance. (For the variable of speaking another language versus speaking Dutch at home-with equal maternal education and parental school involvement levels-the odds of TR over SE were +5,460%; the odds of K3 over SE were +1,900%; and the odds of TR over K3 were +194%). It is important to note that these extreme odds are caused by the differences in the mothers' education between Dutch- and foreign language-speaking families, and thus are imprecise.
The two joint child demographic variables revealed the lowest pscudo-R^sup 2^ (0.04) in respect to their discriminative power between promotion and nonpromotion. Child gender did not contribute on a unique basis. Conversely, the month of birth did contribute. For same-gender children, the odds of promotion over nonpromotion for each successive month were -19%. No interaction effect was recorded between the variables. Within the nonpromoted groups, however, the child variables clearly made more difference; they resulted in a pseudo-R^sup 2^ of 0.35. Further, both variables contributed on a unique basis to the analysis (in same-gender children), for each successive month of birth the odds of K3 over TR were +26%, the odds of K3 over SE were +22%. In the group of children who have the same month of birth, for boys versus girls the odds of SE over K3 were +114%.
An overall logistic regression model for the prediction of promotion versus nonpromotion, including all variables found to be significant in the previous submodels, revealed a pseudo-R^sup 2^ of 0.42. Due to the intercorrelations, the odds ratios of most variables decreased to a certain extent in comparison to the submodels. The same range of independent variables resulted in a pseudo-R^sup 2^ of just 0.21 with respect to the prediction of the specific nonpromotion alternative.
EXTRA SCHOOL SUPPORT PROVIDED PRIOR TO NONPROMOTION
As expected, children facing nonpromotion after kindergarten clearly received more school support in the preceding year than did the children who were to be promoted to primary school. This additional support seems to have occurred because these children exhibited more problems during kindergarten. The information available from the teacher questionnaire at the end of the kindergarten year could have led to some underestimation of the true frequency of support rendered, because the nonresponse rate for the teacher questionnaires was about 20%.
Findings indicate that about 20% of the promoted group received some type of school support, whereas this was the case for 73% of the K3 group, 61% of the SE group, and 43% of the TR group. The drawback of this scenario is that even when nonresponse is taken into account, a substantial group of children (at least 23% of the K3 group, 30% of the SE group, and 53% of the TR group) did not receive any extra support, even though each student finally was not promoted. In general, support most frequently is directed at language, preacademic abilities, and speech.
The findings suggest that, in Flanders, Belgium, there is little difference in the educational functioning of pupils who were referred to three distinct nonpromotion alternatives after kindergarten. Indeed, striking differences relating to the four sets of child and family variables investigated (preacademic test scores, psychosocial variables, child demographic variables, and family demographic variables) appeared between the prediction of promotion versus nonpromotion, and the prediction of the three nonpromotion alternatives.
In the prediction of promotion versus nonpromotion, the preacademic test scores and the psychosocial variables were the most discriminating. Generally, children facing nonpromotion had lower language and mathematic test scores, and did not function as well psychosocially as the children who were promoted to the first grade. For these young children, psychosocial functioning-and especially variables related to work attitude-seems to play an important role in decisions made regarding their promotion. The researchers who conducted the present study hypothesize that these factors comprise an important part of a teacher's reference to the concept of school readiness as a motivation for a nonpromotion advice.
In the prediction of the three nonpromotion groups, less significant differences were found. In this case, the child demographic variables discriminated the most. It therefore seems that subjective-possibly discriminating-nonpromotion practices mainly prevail in the choice of the specific nonpromotion alternative, but that they are less a factor in the nonpromotion decision itself. Conversely, in international (mostly American) research, child and family demographic variables such as age, gender, SES, and parental involvement did relate to the nonpromotion decision itself (Byrd & Weitzman, 1994; Cosden et al., 1993; Jimerson et al., 1997; Mantzicopoulos & Neuharth-Pritchett, 1998; McCoy & Reynolds, 1999). Thus, different selective mechanisms might be at work in the Flemish society and educational system which, in turn, provide different nonpromotion alternatives for young children. Regarding the promotion decision, the Flemish system seems to have rather objective standards (namely the preacademic and psychosocial functioning of the pupils). The allocation of children to a specific nonpromotion alternative-which reflects significant educational differences-does not seem to be based on educational considerations, however, but rather on demographic characteristics of the children and their families. From this perspective, it would be interesting to study the negotiation process among teachers, parents, and school services in relation to those demographic variables. Elements such as the parents' knowledge about different educational alternatives, teachers' views on family resources, and teachers' and parents' position toward each other all could be part of the process. The findings, however, do demonstrate that the joint study of nonpromotion alternatives provides some additional insight into nonpromotion practices.
Closer examination of the child and family differences between the three nonpromotion groups revealed the following. In the SE group the children exhibited slightly more psychosocial problems, slightly more often were boys, and most came from Dutch-speaking families. These predictors, however, were interrelated to a significant degree, because only the home language variable continually contributed on an individual basis when all the family predictors were entered simultaneously.
Note that, in line with international findings (e.g., Cosden et al., 1993), in Belgium children referred to special education later in their school career more frequently come from families that speak a foreign language at home (Departement Onderwijs, 2004). It is hypothesized that the distinct and low-threshold transition program in special education-that explicitly aims at a return to general education-holds a different status than that of the remainder of the special education curriculum. Consequently, this different status might be sensed or better understood by Dutch-speaking parents. An analogous mechanism could be apparent in redshirting (delaying entry into preschool) in the United States (Zill, Loomis, & West, 1997).
The TR group seemed to be at a great sociodemographic disadvantage, showing the lowest proportion of Dutch-speaking families as well as the lowest mean levels for maternal education and for parental school involvement. Indeed, five out of six transition rooms were organized in schools situated within metropolitan areas that accommodate many children who are eligible for extra educational support. The sociodemographic differences between the promotion groups in the total sample could be even greater because the parental nonresponse was not totally random; it equaled about 10% for the promoted group, but amounted to nearly 30% for the K3 and the TR groups. The K3 group was differentiated by the later months of birth of its members, higher parental involvement, and higher proportion of support received in kindergarten. These children did not exhibit more preacademic or psychosocial problems than did the children in the SE or the TR groups, making them seem somewhat privileged. The maternal educational level for this group, however, was not significantly different from that of the SE and the TR groups. As stated, this finding raises some questions about educational selectivity processes. Do children whose parents are more involved in the school receive more help than the children of less-involved parents? And do the involved parents-who, like most parents, are somewhat afraid of the stigmatizing effects of special education and of lessened educational and vocational opportunities for their child-exert more influence to keep the child in the general educational system, despite the preacademic and/or psychosocial problems that the child faces?
This study's findings regarding differences among the three nonpromotion groups should be handled cautiously. Although a large group of predictors was used to explain the promotion and nonpromotion alternatives after kindergarten, critical but nonobserved factors still must play a part in the decision. This is true particularly for the decision on the nonpromotion alternative (see the R^sup 2^ of 0.21 for the nonpromotion alternative, versus the R^sup 2^ of 0.42 for promotion versus nonpromotion). One element that has not yet been incorporated into this study is IQ score. Specifically for one subtype of special education, the IQ factor could add significantly to the decision. Apart from this it should be noted that, for the referral of children to these types of special education following kindergarten, only rarely were childhood disorders or disabilities mentioned for children in this age group.
According to the teacher reports on the educational support provided to the children in the kindergarten year, 23% to 53% of the children who did not progress to the first grade did not receive any additional intra- or extra-school support (apart from extra help provided by the teachers themselves). This is suspected to be due to some degree to the maturational view that is widely held regarding kindergarten development (Carlton & Winsler, 1999); if assuming that more time is the only thing some children require to attain school readiness, then no furdier actions are warranted or worthwhile in kindergarten. The SE and the TR groups received the least help in kindergarten, however, whereas both trajectories precisely imply that those children need some specific educational aid. Perhaps schools give priority to the children who are most likely to stay there. It is the opinion of the researchers who conducted this study that there exists too great a gap between receiving additional aid from a teacher in kindergarten, and receiving a full-possibly specialized-extra year following kindergarten. These researchers also think that educational nonpromotion of any type never should be suggested before other forms of additional educational support are provided and evaluated. It seems that, in Flanders, few intermediate forms of extra educational support-that do not result in nonpromotion-are available.
This study could not examine teacher and school characteristics related to kindergarten nonpromotion, because no significant difference was found in the nonpromotion ratio between teachers or schools. This might be due to the application of multilevel models on a dichotomous dependent variable (namely, promoted versus nonpromoted children). Other researchers apply matching procedures to this issue (Jepma, 2003; Jepma & Meijnen, 2001; Mantzicopoulos & Morrison, 1992), but perhaps a qualitative indepth study is more suited to track teacher and school differences in nonpromotion.
Obviously, it also could be the case that (in Flanders) no significant teacher or school differences exist in kindergarten promotion practices. Although recent educational research stresses the importance of studying school nonpromotion within the school's sociocultural context (Mantzicopoulos & Neuharth-Pritchett, 1998), in Flanders this context might apply to the whole region. Knowing that Flemish schools do differ from one another in their proportion of at-risk children, one hypothesis arising from this study might be that schools use a proportional standard to decide on promotion issues (e.g., retaining no more than about 5% of the kindergarten pupils). Certainly the issue requires further investigation.
Some limitations of this study arise from the instruments used. The psychosocial variables were assessed only by the kindergarten teachers-who have a significant impact on promotion decisions. To some degree this must bias the relation between the psychosocial variables and the nonpromotion decisions. Conversely, psychosocial functioning was assessed in February; therefore, it can be assumed that the teachers' opinions on nonpromotion for specific children had not yet been decided. Further, the assessment of family SES used in this study is rather primitive. Because it used only maternal education and home language, family predictors of nonpromotion could be underestimated.
Another limitation relates to the fact that this study included actual nonpromoted groups. At the end of the school year the teachers were asked to denote the (motivated) school advice for the next year. Quite a few differences in both directions were found between the school advice given at the end of kindergarten and the real educational trajectory chosen in the subsequent school year. Perhaps some parents did not agree with the school advice given, the school advice was not yet official at the time researchers requested it, or events that took place after the school had advised the parents influenced the parents' final decision. The end-of-year teacher questionnaire had a considerable nonresponse rate, therefore it was not advisable to use it for nonpromotion group formation. It also could be that the predictive variables would more strongly, or differently, relate to the advised nonpromotion groups.
Last, it was quite striking that the children from the nonpromoted groups did not exhibit lower preacademic gain scores than those of the promoted group. This seems to indicate that all four groups of children showed a comparable preacademic growth throughout the kindergarten year but, by the end of the year, some absolute criterion was applied to decide who would and would not be promoted. As mentioned, however, formal requirements are not attached to a successful end of kindergarten or to the entry of the first grade; yet, another factor could play a part in this finding. The preacademic gain scores, computed as the (calibrated) End Test score minus the Begin Test score, could be biased. Specifically, the distribution of the End Test scores was skewed to the left and showed a ceiling effect, implicating that children with high Begin Test scores couldn't exhibit similar progress as children with lower Begin Test scores. Indeed, when the preacademic gain scores for the Begin Test score were corrected (by adding it as a co-variate), the promoted group did display significantly greater gain scores than those of the nonpromoted groups. However, the effect of the corrected gain scores still was less than that of the absolute preacademic test scores. Additionally, the validity of the statistical procedure-whereby the preacademic Begin Test scores are used twice in the computation of gain scores-is debatable.
The findings of this study emphasize the need for schools to conduct a transparent promotion policy. In the absence of formal guidelines, schools should articulate what types of problems could lead to nonpromotion advice, how these problems should be assessed, and what procedures should be followed to reach a final decision. Crucial elements in these procedures relate to the provision and evaluation of additional educational support prior to nonpromotion, and to the informing and involvement of parents. Different educational alternatives should be clarified and discussed with the parents. Procedures of this type might reduce the influence of subjective-and perhaps unconscious-mechanisms in promotion decisions.
Further, it should be clear that correct information on nonpromotion alternatives should be based on knowledge about the effectiveness of those nonpromotion alternatives. Regarding early nonpromotion in Flanders, this is the next step in the present research project. International research, however, reveals that something more is needed to fill the gap between the scientific knowledge on nonpromotion effectiveness and the actual nonpromotion school practices (Carlton & Winsler, 1999; Tanner & Galis, 1997; Tomchin & Impara, 1992).
This study addresses the issue of nonpromotion in preprimary education. Until now, international research could not prove its benefits; therefore, it is necessary to gain more insight into who is directed toward each of the different alternatives for promotion. To this end, the researchers investigated a set of child and family characteristics that was related to different nonpromotion alternatives after kindergarten in Flanders, Belgium.
Child preacademic scores and psychosocial functioning appeared to be relatively strong predictors for promotion versus nonpromotion. Child and family demographic characteristics seemed to be more decisive regarding the specific nonpromotion alternative. This suggests that, to some degree, selective-perhaps discriminating- mechanisms might have been at work. Additionally, for a significant proportion of children, the teachers reported no other educational interventions prior to the nonpromotion decision. It is important to note that no significant differences were found regarding the teacher or school rate of nonpromotion. This finding could imply that teachers and schools informally apply proportional standards for nonpromotion. On all 13 psychosocial variables measured by the teacher questionnaire, the promoted group outperformed the three nonpromoted groups.
Child preacademic scores and psychosocial functioning appeared to be relatively strong predictors for promotion versus nonpromotion. Child and family demographic characteristics seemed to be more decisive regarding the specific nonpromotion alternative.
Allison, P. D. (1999). Logistic regression using the SAS system: Theory and application. Cary, NC: SAS Institute.
Artiles, A. J., & Trent, S. C. (1994). Overrepresentation of minority students in special education: A continuing debate. The Journal of Special Education, 27, 410-438.
Birch, S. H., & Ladd, G. W. (1997). The teacher-child relationship and children's early school adjustment (translated by Simoens, 2001, shortened by Cornelissen & Verschueren, 2001, 2002). Journal of School Psychology, 35(1), 61-79.
Byrd, R. S., & Weitzman, M. (1994). Predictors of early grade retention among children in the United States. Pediatrics, 93, 481- 487.
Carlton, M. P., & Winsler, A. (1999). School readiness: The need for a paradigm shift. School Psychology Review, 28, 338-352.
Cito. (1996). Taal voor Kleuters [Language for kindergarten pupils] (adapted by Ponjaert-Christoffersen, Andries, Celestin- Westreich & Samay, 2000). Arnhem: Cito.
Cosden, M., Zimmer, J., & Tuss, P. (1993). The impact of age, sex, and ethnicity on kindergarten entry and retention decisions. Educational Evaluation and Policy Analysis, 15, 209-222.
Delgado, C. E. E, & Scott, K. G. (2006). Comparison of referral rates for preschool children at risk for disabilities using information obtained from birth certificate records. The Journal of Special Education, 40, 28-35.
Departement Onderwijs. (2004). Vlaamse onderwijsindicatoren in internationaal perspectief [Flemish indicators of education in an international perspective]. Retrieved December 22, 2005, from http:/ /www.ond. vlaanderen.be/onderwijsstatistieken/2004-2005/indipub2004/ ind2004.pdf
Driessen, G., van Langen, A., & Vierke, H. (2000). Basisonderwijs: veldwerkvenlag, leerlinggegevens en oudervragenlijsten. Basisrapportage PRIMA-cohortonderzoek. Derde meting 1998-1999 [Primary education: Field notes, pupil information and parent questionnaires. PRIMA-study, third wave 1998-1999]. Nijmegen: Instituut voor Toegepaste Sociologie (ITS).
Driessen, G., van Langen, A., & Vierke, H. (2002). Basisonderwijs: veldwerkverslag, leerlinggegevens en oudervragenlijsten. Basisrapportage PRIMA-cohortonderzoek. Vierde meting 2000-2001 [Primary education: Field notes, pupil information and parent questionnaires. PRIMA-study, fourth wave 2000-2001]. Nijmegen: Instituut voor Toegepaste Sociologie (ITS).
Dudal, P. (2004). Rekenbegrip [Conception of math]. Leuven: VCLB- Service.
Gadeyne, E. (2003). Longitudinaal onderzoek in het basisonderwijs: Eindejaarsbevraging voor de kleuterleidster schooljaar 2002-2003 [Longitudinal research in preprimary and primary education: End-of-year questionnaire for the kindergarten teacher, school year 2002-2003]. Leuven: Steunpunt LOA, Unit Onderwijsloopbanen.
Holmes, C. T. (1989). Grade level retention effects: A meta- analysis of research studies. In L. A. Shepard, & M. L. Smith (Eds.), Flunking grades: Research and policies on retention (pp. 16- 33). London: The Falmer Press.
Jepma, I. J. (2003). De schoolloopbaan van risicoleerlingen in het primair onderwijs [School career of pupils at risk in primary education] (dissertation). Amsterdam: Thela Thesis.
Jepma, I. J., & Meijnen, G. W. (2001). Risicoleerlingen in het regulier basisonderwijs: Handhaving of verwijzing? [Pupils at risk in regular pre-primary and primary education: Maintenance or referral?]. Pedagogische Studien,78, 313-329.
Jimerson, S., Carlson, E., Rotert, M., Egeland, B., & Sroufe, L. A. (1997). A prospective, longitudinal study of the correlates and consequences of early grade retention. Journal of School Psychology, 35, 3-25.
Jimerson, S. R. (2001). Meta-analysis of grade retention research: Implications for practice in the 21st century. School Psychology Review, 30, 420-437.
Jungbluth, P., Roede, E., & Roeleveld, J. (2001). Validering van het PRIMA-leerlingprofie: Secundaire analyses op de PRIMA-cohort bestand [Validation of the PRIMApupil's profile: Secondary analyses on PRIMA] (SCOreport nr. 608). Amsterdam: SCO-Kohnstamn Instituut van de Faculteit der Maatschappij-en Gedragswetenschappen, Universiteit van Amsterdam.
Ladd, G. W., & Profiler, S. M. (1996). The Child Behavior Scale: A teacher-report measure of young children's aggressive, withdrawn, and prosocial behaviors. Developmental Psychology, 32(6), 1008- 1024.
Lea, S. (1997). Logistic regression and discriminant analysa. Retrieved December 7, 2004, from Exeter University, http:// www.exeter.ac.uk/-SEGLea/multvar2/ disclogi.html
Maes, F. (2003). Longitudinaal onderzoek in het basisonderwijs: Kleutervragenlijst schooljaar 2002-2003 [Longitudinal research in pre-primary and primary education: Kindergartener questionnaire, school year 2002-2003] (Loa-report n[degrees] 12). Leuven: Steunpunt LOA, Unit Onderwijsloopbanen.
Mantzicopoulos, P. (2003). Academic and school adjustment outcomes following placement in a developmental first-grade program. The Journal of Educational Research, 97, 90-105.
Mantzicopoulos, P., & Morrison, D. (1992). Kindergarten retention: Academic and behavioral outcomes through the end of second grade. American Educational Research Journal, 29, 182-198.
Mantzicopoulos, P. Y., & Neuharth-Pritchett, S. (1998). Transitional first-grade referrals: An analysis of school-related factors and children's characteristics. Journal of Educational Psychology, 90, 122-133.
Matthews, L. L., May, D. C., & Kundert, D. (1999). Adjustment outcomes of developmental placement: A longitudinal study. Psychology in the Schools, 36, 495-504.
McCoy, A. R., & Reynolds, A. J. (1999). Grade retention and school performance: An extended investigation. Journal of School Psychology, 37, 273-298.
Merle, P. (1998). Sociologie de l'evaluation scolaire [Sociology of school evaluations]. Paris: Presses Universitaires de France.
Organisation for Economic Cooperation and Development (OECD). (1998). Overcoming school failure. Paris: OECD.
Owings, W. A., & Magliaro, S. (1998). Grade retention: A history of failure. Educational Leadership, 56, 86-88.
Pampel, F. C. (2000). Logistic regression: A primer. Thousand Oaks, CA: Sage.
Paul, J.-J. (1997). Le redoublement a l'ecole: Une maladie universelle? [School retention: A universal disease?] International Review of Education, 43, 611-627.
Pian ta, R. C. (2001). STRS Student-Teacher Relationship Scale. Professional Manual (translated by Roncada, 2001; shortened by Cornelissen & Verschueren, 2001, 2002). Odessa, FL: Psychological Assessment Resources.
Reynders, T., Van Heddegem, I., Nicaise, I., & Van Damme, J. (2004). Longitudinaal onderzoek in bet basisonderwijs: Oudervragenlijst schooljaar 2002-2003 [Longitudinal research in pre- primary and primary education: Parent questionnaire school year 2002- 2003] (LOA-report n[degrees] 22). Leuven: Steunpunt LOA, Unit Onderwijsloopbanen.
Shepard, L. A. (1989). A review of research on kindergarten retention. In L. A. Shepard, & M. L. Smith (Eds.), Flunking grades: Research and policies on retention (pp. 64-78). London: The Palmer Press.
Shepard, L. A., & Smith, M. L. (1989). Escalating kindergarten curriculum. Retrieved January 25, 2005, from http:// readyweb.crc.uiuc.edu/library/prel990/ sheprd89.html
Tanner, C. K., & Galis, S. A. (1997). Student retention: Why is there a gap between the majority of research findings and school practice? Psychology in the Schools, 34, 107-114.
Tomchin, E. M., & Impara, J. C. (1992). Unraveling teachers' beliefs about grade retention. American Educational Research Journal, 29, 199-223.
Troncin, T. (2004). Le redoublement au cours preparatoire [Retention in preprimary education]. Retrieved December 15, 2004, from http://www.ubourgogne.fr/ IREDU/semi.htm
Van Landeghem, G., Van Damme, J. (2004). Geboortecohorten doorheen het Vlaams Onderwijs: Evolutie van 1989-1990 tot 2001-2002 [Birth cohorts in Flemish education: Evolution from 1989-1990 until 2001-2002] (LOA-report nr. 16). Leuven (Belgium): Steunpunt LOA, Unit Onderwijsloopbanen.
Verhaeghe, J. P., Macs, E, Gombeir, D., & Peeters, E. (2002). Longitudinaal onderzoek in het basisonderwijs: Steekproeftrekking [Longitudinal research in primary education: Research sample] (LOA- report nr. 5). Leuven (Belgium): Steunpunt LOA, Unit Onderwijsloopbanen.
Wallingford, E. L., & Prout, H. T. (2000). The relationship of season of birth and special education referral. Psychology in the Schools, 37, 379-387.
Zill, N., Loomis, L S., & West, J. (1997). The elementary school performance and adjustment of children who enter kindergarten late or repeat kindergarten: Findings from national surveys (NCES 98- 097). Washington, DC: U.S. Department of Education & Office of Educational Research and Improvement. Retrieved December 15, 2004, from http://nces.ed.gov/pubs98/ 98097.pdf
Katholieke Universiteit Leuven (Belgium)
ABOUT THE AUTHORS
ELS GADEYNE, Doctor, Centre for Educational Effectiveness and Evaluation, Department of Educational Sciences; PATRICK ONGHENA, Professor, Educational Methodology and Statistics, Centre for Methodology of Educational Research, Department of Educational Sciences; and POL GHESQUIERE, Professor, Learning Disabilities and Special Education, Centre for Disability, Special Needs Education and Child Care, Department of Educational Sciences, Katholieke Universiteit Leuven, Belgium.
Address correspondence to Els Gadeyne, Centre for Educational Effectiveness and Evaluation (CO&E), Dekenstraat 2, B-3000 Leuven, Belgium (Europe) (e-mail: els.gadeyne@ped. kuleuven.be). Manuscript received January 2006; accepted January 2007.
Copyright Council for Exceptional Children Summer 2008
(c) 2008 Exceptional Children. Provided by ProQuest Information and Learning. All rights Reserved.