The relationships between teachers’ evaluation of children’s academic readiness and children’s later outcomes

Abstract

Kindergarten is emphasized as a critical first entry into the education system, as politicians and pundits believe kindergartners’ success can lead to later academic achievement. Therefore, a comprehensive understanding of kindergarten school readiness data should consider how it is measured and how that affects learning. Using the Early Childhood Longitudinal Studies—Kindergarten (ECLS-K) cohort 2010–2011 data, the author examined the relationship between teacher reports about children’s academic readiness at kindergarten entrance and their later outcomes in reading, math, and science through fifth grade. This study used a latent basis growth model with time-invariant predictors to analyze the relationship's trajectory. Overall, findings determined that teachers’ perceptions of children’s Approaches to Learning, mathematical thinking, and science significantly impacted later achievements in math and science direct assessments throughout elementary education. This research discussed the practice and policy implications on teachers’ perceptions of school readiness and its impact on later academic outcomes.

Introduction

Early learning funding has increased based on the belief that school readiness can address inequity in education. The Preschool for All initiative aimed to expand preschool access to ensure a smooth transition to kindergarten, known as school readiness (The White House, 2013). Similarly, the Common Core State Standards claim to prepare students from kindergarten for future success in academic achievement, college, and careers in the twenty-first century (Common Core State Standards Initiative [CCSSI], http://www.corestandards.org/about-the-standards/). By 2021, 38 states employed readiness assessment in kindergarten using various formats (Yun et al., 2021; Curran et al., 2020). Consequently, early childhood education policies emphasize kindergarten as the “readiness grade,” undergirding the expectation of investment returns (Brown, 2021, p. 93). However, despite high expectations of its impact, school readiness has no congruous definition, as its definitions vary over time and across social contexts (Graue, 1992). Nevertheless, state-readiness assessments and learning standards currently define readiness in specific ways (Regenstein et al., 2018), but there is little understanding of how to apply this data (Curran et al.., 2020). Given the critical role of readiness data, it is important to understand what readiness data signifies and how it is measured. This study uses nationally representative data to examine how children’s readiness, as measured by their teachers, relates to their later academic outcomes.

The effect of school readiness on later achievements

Increasing investments in early learning build upon existing studies that have espoused the benefits of school readiness for children’s success in school, college, and careers. Preschool education, for example, facilitates the transition to kindergarten (Christopher & Farran, 2020). Focusing on the transitional periods, La Paro and Pianta (2000) found that academic skills between preschool and kindergarten correlate with those between kindergarten and first or second grade, implying the continuity of preschool education effects. Preschool education has been shown to improve school readiness, economic benefits, children’s cognitive and socioemotional development, and future school experiences (Atteberry et al., 2019; Barnett & Masse, 2007; Yoshikikawa et al., 2016). Some studies have linked kindergarten academic readiness to later academic achievements (Claessens et al., 2009; Duncan et al., 2007; Goldstein et al., 2017; Hair et al., 2006). Chetty et al. (2011) extended the effect of readiness to economic returns, showing that kindergarteners’ math and reading skills can predict future college attendance, property ownership, earnings, and retirement finances.

Much research has focused on academic skills and knowledge, predominantly in reading and math, as indicators of early skills. This focus aligns with growing investments in early childhood education and has led to implementing kindergarten readiness assessments (Goldstein et al., 2017). Early mathematical skills, such as number recognition and counting, predict later math test scores in elementary school (Bodovski & Farkas, 2007; Jordan et al., 2009). Claessens et al. (2009) used the Early Childhood Longitudinal Study—Kindergarten Cohort (ECLS-K): 1998–1999 data set to demonstrate that attention, math, and reading skills in the fall of kindergarten year significantly affect fifth-grade outcomes. Using the same data, Claessens and Engel (2013) extended this impact to eighth grade, explaining that kindergarten math skills predict later reading, math, and science achievement. Duncan et al. (2007) included attention and socioemotional skills in their analysis to predict later academic outcomes in reading and math across six secondary data sources.

Despite the efforts to include more indicators, focusing on academic readiness plays a central role in the perceptions of readiness, policies, and curriculum. For example, Bassok et al. (2016) showed that teachers increasingly believe school readiness includes knowing the alphabet and formally reading before kindergarten, thereby focusing more on academic readiness. However, some researchers caution against narrowly defining readiness (Graue, 1992; Hair et al., 2006; Meisels, 1999). Nevertheless, kindergartners in 2010 outperformed those in 1998 in math and literacy, indicating the impact of an academic readiness focus (Bassok & Latham, 2017).

The National Education Goals Panel (NEGP) provides a framework for readiness, including “physical well-being and motor development,” “social and emotional development,” “approaches toward learning,” “language development,” and “cognition and general knowledge” (Kagan et al., 1995, p. 4). Particularly, Approaches to Learning (AtL) consist of “curiosity, creativity, independence, cooperativeness, and persistence” (Kagan et al., 1995, p. 4). This framework is widely used for readiness policy and research, such as the Early Childhood Longitudinal Study—Kindergarten (ECLS-K), and to develop readiness assessments. Framed by the readiness profile of NEGP, Hair et al. (2006) demonstrated readiness gaps between children based on individual characteristics and family backgrounds.

Approaches to learning as school readiness in ECLS-K

Many studies have focused on Approaches to Learning (AtL) as a critical dimension of school readiness to predict academic achievements, addressing readiness gaps. Using ECLS-K: 1998–1999, Smith-Adocock et al. (2019) explained that parental involvement and AtL impact reading in kindergarten. With the same data, Li-Grining et al. (2010) found that AtL is associated with children’s academic growth in the first, third, and fifth grades. This result is consistent regardless of race/ethnicity and SES, while the relationship between AtL and academic growth is distinct between genders. Duncan et al. (2007) employed a meta-analysis across six longitudinal data sets, including the ECLS-K data, and found that school readiness—encompassing academic achievement, attention (AtL), and socioemotional skills—significantly predicts children’s later reading and math scores regardless of gender, socioeconomic status (SES), and race/ethnicity. While the impacts of child characteristics and family backgrounds on readiness gaps are relatively consistent in many studies (Buek, 2019; Hair et al., 2006), research has produced divergent results on how demographic factors mediate the relationship between children’s readiness and later outcomes (Bodovski & Farkas, 2007; Claessens et al., 2009; Duncan et al., 2007; Li-Grining et al., 2010).

Li-Grining et al. (2010) heightened early AtL as a way to benefit children with low academic achievement at kindergarten entry, suggesting that certain aspects of AtL can help children cope with different challenges that result from poverty, discrimination, and related issues. Reardon and Portilla (2016) also analyzed ECLS-K 1998–1999/2010–2011 and ECLS-Birth Cohort data sets, demonstrating that racial and economic gaps have substantially narrowed. They attributed this trend to increased health insurance coverage and cultural changes in parental cognitive involvement, which lessened the effects of demographic factors. Taken together, AtL consistently emerges as an essential indicator of school readiness and a strong predictor of children’s later academic achievement despite mixed findings. This study focuses on AtL as a key readiness indicator to highlight its predictive value of later academic growth.

Teacher’s evaluation

In response to the growing interest in readiness, many states have implemented readiness assessments that primarily rely on teachers’ reports (Regenstein et al., 2018), as traditional paper and pencil-based assessments are developmentally inappropriate for young children (National Research Council, 2008). Given this limitation in directly measuring kindergartners’ abilities, teacher reports are expected to provide a comprehensive understanding of children’s abilities (Tourangeau et al., 2015). For example, Illinois emphasizes teacher observations within diverse contexts to measure kindergarten readiness throughout the year (Illinois State Board of Education [ISBE], 2019). This approach highlights teachers’ roles in identifying children’s readiness as the data reflects how teachers define which children are ready for what grounds.

As government entities have already administered readiness assessments, teachers play a critical role in determining the number of children deemed ready for school (or not). Policy decisions often rely on these data, reflecting teachers’ perceptions. Many researchers attempted to explain the extent to which teachers understand students’ abilities (Burkam et al., 2007; Ferguson, 2003; Ready & Wright, 2011). Specifically, while many studies find that teachers’ judgments about their students are reasonably accurate Ferguson (2003), Ready and Wright (2011) found that teachers in low-income and low-achieving schools tend to underestimate children’s skills. Despite these mixed findings regarding teachers’ validity in measuring children’s abilities, teachers’ perceptions are central to readiness assessment policy, which is why this study focuses on them.

Building on previous studies about the impact of readiness on later outcomes and teachers’ evaluations, this study attempts to address the research gap. It includes reading, math, science, and Approaches to Learning to address readiness at kindergarten entry and tracks students’ growth from kindergarten to fifth grade. Using teacher evaluations of academic readiness and AtL as a predictor, this study examines the relationship between teacher-reported readiness and later academic achievements assessed directly in children’s elementary education. While the previous studies focus on the impact of preschool education on later outcomes, the present study draws attention to the kindergarten entrance when children’s readiness is actually measured, aligned with many current readiness assessments. In doing so, this study highlights the importance of teacher evaluations in readiness policy and contributes to bridging preschool, kindergarten, and elementary education. By utilizing nationally representative data, ECLS-K, this study deepens the understanding of readiness impacts in a school context.

Current study

The purpose of this research is to examine the prediction of teacher-reported academic readiness for children’s later academic growth throughout elementary school years, highlighting the vital role of teachers in evaluation. In particular, this study utilizes Approaches to Learning, which the NEGP (Kagan et al., 1995) broadly defines as representing children’s learning behaviors that potentially influence their academic outcomes (Duncan et al., 2007). Taken together, I hypothesized that (a) teacher evaluations of Approaches to Learning (AtL) and language predict the growth of children’s direct reading assessment scores from the spring of kindergarten to the spring of the fifth grade; (b) teacher evaluations of AtL and mathematic thinking predict the growth of children’s direct math assessment scores from the spring of kindergarten to the spring of the fifth grade; (c) teacher evaluations about AtL and science assessment scores predict the growth of children’s direct science scores from the spring of kindergarten to the spring of the fifth grade. It is important to note that this study highlights only academic readiness and its related learning behaviors, aligning with current policy interests, although it acknowledges that readiness encompasses multiple domains. For the purpose of this study, emphasis is placed solely on academic readiness, which teacher’s measure.

Methods

This study used nationally representative data, the Early Childhood Longitudinal Studies—Kindergarten (ECLS-K) cohort 2010–2011, developed by the National Center for Education Statistics (NCES). The publicly released version was employed for analysis. In the ECLS-K: 2010–2011, 18,174 children initially participated in the baseline, beginning in the fall of the kindergarten year (Tourangeau et al., 2015). The data were collected at nine timepoints: fall and spring times for each school year from kindergarten to second grade and only springtime for school years from third grade to fifth grade (Tourangeau et al., 2015). To improve the national population, the study applied a sampling weight, W9C29P_2T290, in the analyses as the User’s Manual for the ECLS-K: 2011 recommends (Tourangeau et al., 2019). The ECLS-K data contained missing values of more than 10% in the many variables. In the data set, the missing values included not only non-responses but also legitimate skips when questions were unrelated to the participant. Also, several answers were treated as missing values when respondents did not know how to answer and refused to answer (Tourangeau et al., 2015, p. 7–6). FIML was used to handle the missingness in analysis through Mplus 1.8.6, of which the default was FIML (Muthén & Muthén, 2017). R was employed to analyze the exploratory data in the beginning stage.

Measures

Approaches to learning (AtL)

The ECLS-K data asked teachers to respond to the questionnaires for child level derived from the Social Skills Rating Scale, including AtL (SSRS; Gresham & Elliott, 1990). AtL was designed for the data specifically to identify how often children showed learning behaviors, such as “keeps belonging organized; shows eagerness to learn new things; works independently; easily adapts to changes in routine; persists in competing tasks; pays attention well; and follows classroom rules” (Tourangeau et al., 2015, p. 3–24). The reliability of the AtL scale was 0.91 for each data collection. The AtL variable was calculated as the mean score based on the seven items that the teachers rated on children’s learning behaviors when at least four items were completed. “No opportunity to observe” was included as missing data according to the manual (Tourangeau et al., 2015, p. 3–24).

Academic rating scale

Teachers evaluated individual students’ academic achievement in “language and literacy, science and mathematical thinking” (Tourangeau et al., 2015, p. 3–20) in the first semester of kindergarten. The teachers responded to each item on five scales: “1—Not yet, 2—Beginning, 3—In progress, 4—Intermediate, 5—Proficient, and NA—Not applicable or skill not yet taught” (Tourangeau et al., 2015, p. 3–20). This indirect assessment was designed to supplement what the direct assessment failed to reflect children’s progress (Tourangeau et al., 2015). For this reason, many items measured in each assessment overlapped. As the teachers evaluated their students’ academic progress in language and literacy, mathematical thinking, and science, their report represented what they judged on children’s readiness for school. The scale of this variable contained five categories, which were treated as continuous variables with some limitations.

Direct cognitive assessment

Children’s cognitive scores in reading, math, and science were measured from the fall of kindergarten to the spring of fifth grade (science assessment was only administered in the spring semesters of first grade). The children individually took each assessment by pointing or telling their answers when trained staff asked questions by showing related images. Specifically, the reading component assessed children’s abilities in language and literacy through questions measuring basic skills, vocabulary knowledge, and reading comprehension; the mathematics component assessed skills in conceptual knowledge, procedural knowledge, and problem-solving; the science component assessed skills in physical sciences, life sciences, environmental sciences, and scientific inquiry. Reliability and validity for the direct cognitive assessment components were derived from several NAEP sources (Tourangeau et al., 2015). The reliability scores for the direct cognitive assessments in the current sample were between 0.75 and 0.99, indicating adequate to excellent internal consistency. All the scores from direct were scaled based on Item Response Theory (IRT). This study used an IRT-based scale score that enabled the comparison throughout the times as a longitudinal analysis (Tourangeau et al., 2015).

Data analyses

Before constructing analytical models, the multivariate kurtosis and skewness were examined, and multivariate non-normality was diagnosed with a significant p value (< 0.001). This study used Maximum Likelihood with Satorra–Bentler Corrected Statistic (MLR) to handle the violation with normality assumption. All the analyses were employed with MLR through Mplus1 version 8 (Muthén & Muthén, 2017). As robust correction with MLR was applied, Chi-square difference testing should not be used in a general way. The study followed the way Satorra and Bentler (2010) introduced to adjust the difference test for MLR. In addition, as the research was interested in predicting children’s readiness, this study chose the academic rating scale and AtL at the beginning of kindergarten as time-invariant predictors. For the outcome variable, this study only included eight timepoints (2: the spring of kindergarten, 3: the fall of first grade, 4: the spring of first grade, 5: the fall of second grade, 6: the spring of second grade, 7: the spring of third grade 8: the spring of fourth grade, 9: the spring of fifth grade). Through five years, each timepoint from 2 to 6 had a semester difference, while the time intervals between 6 and 7, 7 and 8, and 8 and 9 were a year for each.

Confirmatory factor analyses for academic rating scale

This research employed three confirmatory factor analyses (CFA) to construct each latent variable from the academic rating scales, based on relevant theories. This approach was necessary as the data set only included item-level variables. According to Tourangeau et al. (2015), four items measured language skill: Q1 uses complex sentence structure (CMPSEN), Q2 interprets the story read to him/her (STORY), Q3 child names upper and lower case (LETTER), and Q4 predicts what happens in stories (PRDCT); five items measured literacy: Q5 reads simple books independently (READS), Q6 uses different strategies w/unfamiliar words (USESTR), Q7 shows early writing behaviors (WRITE), Q8 child composes simple stories (CMPSTR), and Q9 understands conventions of print (PRINT). Early Childhood Learning and Knowledge Center (ECLKC) defines language and literacy as this separation can be important to understanding children’s language development: language is related to communication, such as “attending and understanding,” “communicating and speaking,” and vocabulary as literacy is defined as the ability to read and write such as “phonological awareness,” “print and alphabet knowledge,” “comprehension and text structure,” and “writing” (The Office of Head Start, https://eclkc.ohs.acf.hhs.gov/school-readiness/effective-practice-guides/language-literacy). Based on this theory, the author hypothesized a two-factor CFA model. To fit the CFA model, the variances of the factors were fixed at 1, and the loadings were freely estimated.

This study hypothesized one-factor analyses for mathematical thinking and science scales. First, all the items that explained the latent variable, mathematical thinking, were used: “sorts math materials by criteria (SORTS),” “orders group of objects by criteria (ORDER),” “understand quantity relationships(RELAT),” “solves problems with numbers/objects (SOLVE),” “understands graphing activities (GRAPH),” “uses instruments for measuring (MEASU),” “uses strategies for math problems (STRAT),” and “models reads compares fractions (FRACT)” based on the manual (Tourangeau et al., 2015). Eight items were measured for science: “uses senses to explore/observe (OBSRV),” “bases explanation on observations (EXPLN),” “groups living & non-living things (CLSSFY).,” “logical scientific predictions (SCIPRD),” “communicates science information (COMSC),” “understand physical science concept (PHYSCI),” “understand life science concepts (LIFSCI),” and “understand earth/space science (ERSPSC).” The latent variables—language, literacy, mathematical thinking, and science—were identified through observed item-level variables through Mplus to analyze their impacts on later outcomes.

Latent basis growth model with time-invariant predictors

To test the three hypotheses above respectively, a latent basis growth model with the time-invariant predictors (the latent constructs and AtL) was constructed for each later outcome, reading, math, and science. As the study intended to see how teachers’ evaluation in the fall of kindergarten would be associated with later outcomes, all the observed variables from each timepoint in the latent basis growth model were selected from the spring of kindergarten. Initially, the author tried to fit the data into a linear growth model. However, there were problems with model fit and convergence. This study ended up with a latent basis growth model with time-invariant predictors. Each timepoint (seven timepoints—from the spring of kindergarten to the spring of fifth grade) was freely estimated except that the first timepoint (the spring of kindergarten) and the last (the spring of fifth grade) were fixed at 0 and 5, respectively.

Results

This study modeled the data as a latent basis growth model with time-invariant predictors: teachers’ evaluation of AtL and literacy, mathematical thinking, and science eight timepoints. The correlations between variables and their means are shown in Tables 1, 2, and 3 for each hypothesis correspondingly: (a) teachers’ evaluation of AtL and language and literacy predict children’s reading assessment scores and their growth from the spring of kindergarten to the spring of the fifth grade; (b) teachers’ evaluation about AtL and mathematic thinking predict children’s math assessment scores and their growth from the spring of kindergarten to the spring of the fifth grade; and (c) teachers’ evaluation about AtL and science predict children’s science assessment scores and their growth from the spring of kindergarten to the spring of the fifth grade. All the results were standardized for easier interpretation.

Table 1 Means and bivariate correlations for the first analysis

Table 2 Means and bivariate correlations of the second analysis

Table 3 Means and bivariate correlations of the third analysis

Reading assessments by AtL and language and literacy

Each latent variable—language and literacy—was constructed by Q1–Q4 items and Q5–Q9 items, respectively (root mean square error of approximation = 0.040, comparative fit index = 0.948, Tucker–Lewis index = 0.928, standardized root mean squared residual = 0.042). Based on the rule of thumb by Hu and Bentler (1999), this two-factor CFA model was tenable; root mean square error of approximation (RMSEA) < 0.06 and standardized root mean squared residual (SRMR) < 0.08, even though Comparative Fit Index (CFI) and Tucker–Lewis index (TLI) were less than 0.95. ${x}^{2}$ is 657.104 (p < 0.001, df = 26), this value was limited to further interpretation due to the sensitivity of Chi-square with the sample size (sample size in this data is 18,174).

The loadings are shown in Fig. 1. All the loadings are standardized, and their p values are less than 0.001, which indicates their statistical significance. Language factor was explained by four items with high loadings ranging between 0.65 and 0.87, and five items from Q5 to Q9 highly loaded on literacy factor. This result is aligned with the ECLKC’s theory: language and literacy are emphasized in early childhood education and school readiness (The Office of Head Start, n.d.). The correlation between the two factors was 0.557 with a significant p value (< 0.001).

Fig. 1

CMPSEN= uses complex sentence structure; STORY= interprets story read to him/her; LETTER= child names upper and lower case; PRDCT=predicts what happens in stories; READS= reads simple books independently; USESTR= uses different strategies w/unfamiliar words; WRITE= shows early writing behaviors; CMPSTR= child composes simple stories; PRINT= understands conventions of print. All the loadings are statistically significant (p<.001). All the values are standardized

Confirmatory factor analysis for language and literacy

With the two latent factors and AtL, a latent basis growth model was constructed for the data. To improve the model fit, modification indices were referenced, and five correlations were included for each pair: Read 6 and 4, Read 5 and 2, Read 7 and 6, Read 6 and 5, and Read 5 and 4. As Table 4 shows the pathway to decide on the final model, adjusted Chi-square difference tests with MLR were conducted for each step until settling on the final model. The final model fit with robust correction for MLR showed that Chi-square was 3414.571 (p < 0.001, df = 128); RMSEA was 0.042 with confidence interval between 0.040 and 0.043; CFI, TLI, and SRMR were 0.936, 0.924, and 0.160 correspondingly. Even though not every criterion was satisfied by the rule of thumb (Hu & Bentler, 1999), this model was tenable for analysis.

Table 4 Pathway model fit indices for reading

As shown in Fig. 1 shows, all the loadings of the intercept were fixed at 1. The loadings for each READ3 to READ8 on the slope were freely estimated: 0.602, 1.934, 2.407, 3.236, 3.864, and 4.477. This result represents the average reading achievement trajectory over time from the spring of kindergarten to the spring of fifth grade. As the total time interval between READ2 and READ9 was 5 years, READ2 and READ9 were fixed at 0 and 5, respectively. Considering each time interval between observed timepoints (a semester from READ2 to READS6 and a year from READ6 to READ9), the estimated loadings on the slope explained why the linear growth model was not initially fitted.

The correlation between the slope and the intercept indicates that children with high reading scores in the spring of kindergarten are likely to show lower rates of reading growth throughout elementary school. Those children who are evaluated with better language skills are likely to show higher reading assessment scores in the spring of kindergarten with a 0.379-unit change (p < 0.001). Children whose teachers evaluated high literacy skills are likely to show slower change of reading scores across time (− 0.060 with p < 0.01). On the contrary, the prediction of literacy on the intercept is not significant, and that of language on the slope is negligible. AtL does not statistically affect the growth of reading achievement scores over time. However, those whose teachers rated higher AtL tend to show a reading score in the next semester with a 0.229-unit change (p < 0.001) (Fig. 2).

Fig. 2

*** indicates p < .001. ** indicates p<.01

Latent basis growth model with AtL and literacy and language

Math assessments by AtL and mathematical thinking

One factor, mathematical thinking, was constructed by CFA with eight items from the teacher’s questionnaire. The initial model fit indicates Chi-square = 1965. 617 (p < 0.001, df = 20), RMSEA = 0.081, CFI = 0.824, TLI = 0.753, and SRMR = 0.07. Based on the modification indices, each correlation between FRACTN and MEASU, ORDER with SORTS, and GRAPH with SORTS was added one at a time. The model fit was improved as the final model included all three correlations. The model fit shows that Chi-square = 465.794 (p < 0.001, df = 17), RMSEA = 0.042, CFI = 0.959, TLI = 0.933, and SRMR = 0.036. RMSEA, CFI, and SRMR satisfy the criteria from Hu and Bentler (1999). As the loadings are shown in Fig. 3, all the values are higher than 0.6 except for the loadings of SORTS and FRACTN.

Fig. 3

*** indicates p value < .001, ** (p value < .01) and * (p value < .05). All the values are standardized. M.T is mathematical thinking

Latent basis growth model with AtL and mathematical thinking

A latent basis growth model was constructed for this analysis. With the predictors, AtL, and the latent academic variable, mathematical thinking, the model shows that the Chi-square is 2731.575 (p < 0.001, df = 118). Fit indices indicate that RMSEA = 0.039 (< 0.06), CFI = 0.951 (> 0.95), TLI = 0.943, SRMR = 0.097, which makes the model tenable for the analysis. Table 5 shows the pathway until determining the final model. Starting from constructing a latent basis growth model, modification indices were referenced for adding a correlation for each. A Chi-square difference test in an adjusted way for MLR was employed for each. The Chi-square test was not significant when the correlation between AtL and Mathematical thinking was added. Considering the AIC and BIC became bigger than the comparison model, the final mode was decided as it was bolded in Table 5. That is, correlations between FRACTN and MEASU, ORDER with SORTS, and GRAPH and SORTS were added to the final model.

Table 5 Pathway model fit indices

In this latent basis growth model, loadings of Math scores from each timepoint were fixed at 1. The freely estimated times for each Math3 to Math8 are 0.611, 1.637, 1.997, 2.892, 3.890, and 4.499 with Math2 at 0 and Math9 at 5. These estimations represent the average Math trajectory over time from the spring of kindergarten to the spring of fifth grade. Children whose teachers rated them highly in AtL are likely to have higher math direct assessment scores in the spring of kindergarten with 0.356 unit change (p < 0.001). In addition, these children are also likely to have 0.043 unit change in the rate of change in math assessment scores over time (p < 0.05). The latent factor, Mathematical thinking (Math) also significantly predicts the math score (0.161, p < 0.001) in the next semester and the slope of the math scores over time with 0.059 (p < 0.01). The correlation between the intercept and the slope of the math scores is 0.055 (p < 0.01), which indicates their positive relationship.

Science assessments by AtL and science

One factor, Science, was constructed through CFA. Based on the modification indices, the correlation between OBSRV and EXPLN variables was added. The final model for CFA shows that ${x}^{2}$ is 823.326 (df = 19, p < 0.001). The model fit shows that RMSEA = 0.053, C.I for RMSEA = 0.050 to 0.057, CFI = 0.977, TLI = 0.966, and SRMR = 0.022, which meets the criteria of Hu and Bentler. This CFA with a one-factor model fits with the data. The loadings on science are larger than 0.6 with a statistically significant p value (< 0.001), as shown in Fig. 4.

Fig. 4

*** indicates p<.001; * indicates p<.05. All the values are standardized except for the loadings on the slope.

Latent basis growth model with AtL and science

The latent basis growth model explains the third hypothesis. The initial model with time-invariant predictors, AtL and Science, shows that the Chi-square is 2227.244 (df = 120, p < 0.001). The model fit indicates that RMSEA is 0.034 and its upper bound of a confidence interval is less than 0.04; CFI and TLI are 0.956 and 0.950; SRMR is 0.099. Except for SRMR, they satisfy the rule of thumb (RMSEA < 0.06 and CFI and TLI > 0.95). To improve the model fit, its modification indices suggest correlating between Sci4 and Sci6. This model fit indices explain RMSEA = 0.033, upper bound of its confidence interval is 0.034, CFI = 0.960, TLI = 0.955, and SRMR = 0.097. Its Chi-square value is 2032.842 (df = 119, p < 0.001). Even though RMSEA, CFI, and TLI are improved in this model, SRMR is still larger than 0.08. The AIC and BIC in this model are 1,061,213.809 and 1,061,586.220, while each value of the previous model is 1,061,871.393 and 1,062,236.204. This indicates that the latter model reduces AIC and BIC. To compare the two models, adjusted Chi-square difference testing for MLR is employed. The p value is smaller than 0.001, which demonstrates that the latter model is significantly different from the initial model. This study determines the model with the correlation between Sci4 and Sci6 as the final model.

Figure 4 shows that the loadings on the slope were freely estimated except for Sci2 and Sci9, while the loadings on the intercept were fixed at 1. The estimated values are 0.406, 1.179, 1.675, 2.386, 3.332, and 4.167, which indicates the average science achievement trajectory in the children’s direct assessments across time. Teachers’ evaluation of children’s learning behaviors and learning science significantly predicts the science achievement in the next semester (the spring of kindergarten) and its rate of change over time. AtL and Science from the teacher’s evaluation show 0.184 (p < 0.001) and 0.038 (p < 0.05) unit changes, respectively, in the slope. Those who show more learning behaviors in the teacher’s eyes are likely to acquire higher science test scores in the spring of kindergarten with a 0.269-unit change (p < 0.001). Also, children who demonstrate stronger science thinking, as rated by their teachers, are likely to achieve science direct assessment scores, with a 0.149-unit change (p < 0.001). The correlation between the intercept and the slope is estimated at 0.251 (p < 0.001). In other words, children who achieve high science test scores in the spring of kindergarten are likely to progress in science direct assessments over time.

Discussion

This study employed nationally representative data to examine three hypotheses using confirmatory factor analyses and a latent basis growth model. The findings of the first hypothesis contradict the previous studies, which suggested that reading and AtL (attention skills) at kindergarten significantly predict later reading achievements (Claessens et al., 2009; Duncan et al., 2007). Specifically, this study found that teachers’ perceptions of students’ language (i.e., communication-related skills) acquisition and AtL predict children’s direct reading assessment scores only within the same year, not later. This result supports the theory that language skills are a precursor to literacy (The Office of Head Start, n.d.), as indicated by initial reading assessment scores. While the previous studies used direct assessment IRT scores as predictors, this study utilized teacher perceptions of children’s language and literacy skills, possibly leading to divergent conclusions. In addition, teacher perceptions of children’s literacy showed no significant relationship with children’s direct reading test scores in kindergarten, possibly due to various factors: teachers’ misjudgment, a lack of information on children’s prior knowledge, misalignment between curriculum and assessment, and the minimal impact of literacy (Engel et al., 2013; Ready & Wright, 2011).

Moreover, children identified by their teachers as having higher literacy skills showed little improvement in their direct reading assessment scores over time. This could be attributed to teachers’ misjudgment of their students’ literacy skills or a ceiling effect where scores are too high to show growth (Koedel & Betts, 2009). Alternatively, Engel et al. (2013) found that kindergarteners in the ECLS-K 1998–1999 data set learned what they already knew and showed a negative impact of math instruction on their math achievements. This suggests that in this study, what children already knew could be taught and assessed, which could explain the observed lower growth in reading test scores. This is particularly relevant considering that children in the ECLS-K 2010–2011 cohort exhibit greater skillfulness in math and literacy compared to the 1998–1999 cohort (Bassok & Latham, 2017).

On the contrary, teachers’ ratings of mathematical thinking were significantly related to children’s progress in math direct assessment scores over time, as were ratings of scientific thinking for science test scores. This aligns with previous research indicating that early math skills are an important predictor for children’s later outcomes in math, reading, and science (Claessens & Engel, 2013; Claessens et al., 2009; Duncan et al., 2007). While these studies used direct assessment scores for early math skills, this study particularly highlights teachers’ perceptions of children’s readiness, reaching the same conclusion.

While teachers’ perceptions of mathematical and scientific thinking skills significantly predicted later progress in math and science assessments, perceptions of literacy were negatively related to growth in reading direct assessments. Furthermore, teachers’ perceptions of children’s AtL predicted children’s direct assessment outcomes in math and science, but they only predicted children’s reading scores at the beginning of kindergarten, not their progress over time. These divergent results in reading scores may have several explanations, including misalignment between assessment, instruction, and children’s previous skills or teachers’ inaccurate measurement of reading skills, possibly due to their expectations or bias (Ready & Wright, 2011). This study suggests that future research could explore a comprehensive understanding of teachers’ rating on reading readiness and its impacts on later reading skills by considering instructional and assessment content and other descriptive characteristics.

The findings of this study have significant implications for research and policy. First, this study underscores mathematical and scientific thinking and Approaches to Learning (AtL) in kindergarten for predicting later academic achievements. While literacy and math often receive the most attention in kindergarten (Bassok et al., 2016), this study particularly calls for researchers’ attention to scientific thinking and learning behaviors. Adopting the whole-child approach, which posits that children should grow in diverse domains (National Association for the Education of Young Children [NAEYC], https://www.naeyc.org/resources/position-statements/dap/principles), this study underscores the importance of addressing readiness across various areas, highlighting the significant impacts of learning behaviors. In addition, this research employed teachers’ perceptions of children’s readiness as a predictor to demonstrate the implications of utilizing teachers’ report data. The findings show that teachers’ reports tend to statistically significantly predict children’s later academic test scores, with the exception of reading.

Study limitations

Although the findings of this study illuminate teachers’ perceptions of children’s academic readiness and their relations with children’s later test outcomes, several limitations should be noted. First, the study focuses only on AtL and academic achievements without addressing demographic factors such as family income, race, and gender. However, this study builds on the previous studies that AtL is a stronger predictor of academic growth than demographic factors (Claessens et al., 2009; Duncan et al., 2007; Reardon & Portilla, 2016). By concentrating solely on teachers’ evaluations of readiness, this study specifically examined their relationship with children’s academic growth in assessments within a school context. As the ECLS-K data was designed for a whole population of children (n = 18,174), this study intended to focus on teachers’ judgments of children’s readiness and its prediction of academic growth over five years rather than targeting specific populations. This approach aligns with current readiness assessment policies.

Furthermore, observational measures of school context can lead to a deeper understanding of the relationship between readiness and children’s academic growth, which the ECLS-K data does not include. In addition, this study used secondary data analysis, which limits the ability to understand the specific criteria teachers used to rate AtL and academic readiness. This contextual information is crucial for understanding the impacts of readiness on academic growth at a deeper level. Despite these limitations, this study still draws attention to teachers’ perceptions of children’s readiness and suggests that future research further explores how teachers assess children in various contexts. It would also be meaningful to investigate school context variables such as teachers’ instructions, their interactions with students, and time management for instructions. Understanding these contextual factors would provide deeper insights into children’s academic growth and the impact of their readiness on it.

Conclusion

This study demonstrates that teachers’ perception of children’s academic learning and learning behaviors at the kindergarten entry can predict their math and science test scores through elementary education years. In particular, learning behaviors at the beginning of kindergarten are significantly associated with initial achievements in all three subjects and growth over time in math and science. Thus, teachers’ perceptions of children’s readiness for learning are critical predictors of future academic outcomes. This study, using the data at the entrance of kindergarten, bridges the gap between many studies by addressing readiness at kindergarten entry and the trajectory of children’s academic growth. As teachers’ judgments about children’s readiness are widely used for policy decisions, curriculum, class placements, and more (Curran et al., 2020), this study contributes to the policy discussion about teachers’ understanding of how kindergartners are academically ready.