لمس انگشتان شما! توسعه اندازه گیری مستقیم مقررات رفتاری در دوران کودکی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|37160||2008||18 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Early Childhood Research Quarterly, Volume 23, Issue 2, 2nd Quarter 2008, Pages 141–158
Abstract Behavioral aspects of self-regulation, including controlling and directing actions, paying attention, and remembering instructions, are critical for successful functioning in preschool and elementary school. In recent years, several direct assessments of these skills have appeared, but few studies provide complete psychometric data and many are not easy to administer. We developed a direct measure of children's behavioral regulation, the Head-to-Toes Task, and report performance of participants aged 36–78 months, including a group of Spanish-speaking children, from two different sites (N = 353; N = 92). We examined construct validity, examiner reliability, sources of variation, and associations between task scores and background characteristics. Results showed that the task was valid, reliable, and demonstrated variability in children's scores. A cross-classified hierarchical growth curve analysis indicated that girls, participants assessed in English, and higher-socioeconomic status (SES) children achieved slightly higher average scores than did boys, Spanish-speaking and lower-SES children, but effect sizes were small. Older participants achieved higher scores than did younger children, and there were no effects for site. Results suggest that the Head-to-Toes Task is an informative and easy-to-administer direct assessment of children's behavioral regulation. We discuss implications for its use in early childhood settings.
نتیجه گیری انگلیسی
5. Results 5.1. Variability and psychometric properties of the Head-to-Toes Task Our first goal was to examine variability in children's scores and establish construct validity and reliability of the Head-to-Toes Task. Variability was present for all ages of children, with the widest range emerging in scores of 48–60-month-olds. In addition, data were fairly bimodal, with most children scoring relatively low or high on the task. To assess the normality of the distributions, we calculated skewness and kurtosis within each site. General guidelines suggest kurtosis absolute values greater than 8.0 and skewness absolute values greater than 3.0 indicate severe deviations from normality (Kline, 2005). As shown in Table 3, skewness and kurtosis values were not extreme for children across ages. In addition to these values, percentages of participants scoring at floor or ceiling levels were comparable across sites. Table 3. Distribution of scores on the head-to-toes task (Time 1) Age group (site) N Skewness Kurtosis % scoring at floor % scoring at ceiling 36 months (MI) 37 0.65 −1.67 65 0 42 months (MI) 54 0.53 −0.96 56 2 48 months (MI) 54 .02 −0.44 33 6 48 months (OR) 30 −0.30 −0.30 30 3 54 months (MI) 69 −0.16 1.78 13 7 54 months (OR) 48 −0.35 1.48 17 6 60 months (MI) 44 0.85 0.86 2 20 60 months (OR) 14 −1.57 0.50 21 0 66 months (MI) 36 1.87 1.57 0 17 Note: Percent is based on the number of children in that age group. MI: Michigan; OR: Oregon. Table options Although more than half of the children who were age 36 months at the first task administration in Michigan scored zero on the task, only about a third of 48-month-olds and about a sixth of 54-month-olds in both sites scored zero on the task (see Table 3). At Time 1, the majority of children scored at least one point on either the practice or test portion of the task: 72% of all 3-year-olds, 91% of all 4-year-olds, and all except one 5-year-old. We also graphed developmental trends across 6-month age groups (e.g., 36-month-olds, 42-month-olds, 48-month-olds, etc.) for each time point in Fig. 1. Average scores are shown on the y-axis for different age groups. Note that this differs from the standard practice of graphing time on the y-axis, but more easily shows potential differences in scores across multiple task administrations. Fig. 1 demonstrates relative consistency in scores for children in each age group, across up to four task administrations. An analysis of Head-to-Toes Task scores by task administration number (Times 1–4) and age group demonstrated scores differed by 6-month age group, F(8, 1320) = 29.55, p < 0.01, but not task administration, F(3, 1320) = 1.45, p = 0.23; the interaction of age group by task administration was not significant, F(18, 1320) = 1.30, p = 0.18. This suggests age group, but not number of task administrations, was associated with improvement on the task, and that this pattern was the same across age groups. Developmental trends across age groups for four task administrations (Times 1–4) ... Fig. 1. Developmental trends across age groups for four task administrations (Times 1–4) of Head-to-Toes. Scores are based on the chronological age group into which children fell when they were assessed. Figure options In addition to demonstrating construct validity by showing a mean increase in scores as children developed, we sought to establish convergent validity for the Head-to-Toes Task by comparing task scores to CBRS ratings. At the Michigan site, children who received high scores on the Head-to-Toes Task were rated higher by their teachers on the eight behavioral regulation items of the CBRS at Time 1, r = 0.20, p < 0.01, Time 2, r = 0.15, p < 0.05, Time 3, r = 0.19, p < 0.05, and Time 4, r = 0.15, p < 0.05. At the Oregon site, higher scores on the Head-to-Toes Task were related to higher teacher ratings on the CBRS at Time 1, r = 0.42, p < 0.01, Time 2, r = 0.47, p < 0.01, and Time 3, r = 0.26, p = 0.06, but not Time 4, r = 0.10, p = 0.48. Correlations for the Michigan sample are shown in Table 4, and correlations for the Oregon sample are shown in Table 5. Correlations among background variables, Head-to-Toes Task scores, and CBRS ratings demonstrated a similar pattern across sites. However, the correlation between Oregon Head-to-Toes Task Time 2 and Time 3 scores was lower than other within-task correlations, at r = 0.29, p < 0.05. Table 4. Correlations for the Michigan site (maximum N = 353) 1 2 3 4 5 6 1. Age Time 1 (months) – 0.01 −0.05 0.08 0.10t −0.06 2. Non-Asian minoritya – −0.17** 0.06 −0.02 0.18** 3. Parent education (years) – 0.03 0.17** c 4. Maleb – 0.001 0.004 5. Childcare (months) – −0.06 6. Missing datad – 7. CBRS Year 1 8. CBRS Year 2 9. HTT Time 1 10. HTT Time 2 11. HTT Time 3 12. HTT Time 4 7 8 9 10 11 12 1. Age Time 1 (years) 0.15* 0.10 0.62** 0.57** 0.50** 0.42** 2. Non-Asian minoritya −0.07 −0.20** −0.13* −0.13* −0.13* −0.12t 3. Parent education (years) 0.12t 0.06 −0.01 −0.002 0.07 0.04 4. Maleb −0.23** −0.29** 0.03 −0.05 −0.08 −0.14* 5. Childcare (months) −0.03 −0.07 0.08 0.13* 0.11 0.05 6. Missing data −0.07 0.02 −0.13* −0.13* −0.22** −0.13* 7. CBRS Year 1 – 0.49** 0.20** 0.15* 0.20** 0.14t 8. CBRS Year 2 – 0.20** 0.19* 0.19* 0.15* 9. HTT Time 1 – 0.57** 0.45** 0.36** 10. HTT Time 2 – 0.59** 0.48** 11. HTT Time 3 – 0.61** 12. HTT Time 4 – Note: HTT—Head-to-Toes Task; CBRS—Child Behavior Rating Scale. a Non-Asian minority = 0; minority = 1. b Female = 0; male = 1. c Invalid because missing data were calculated based on parent education. d Complete data = 0; missing data = 1. t p < 0.10. * p < 0.05. ** p < 0.01. Table options Table 5. Correlations for the Oregon site (maximum N = 92) 1 2 3 4 5 6 1. Age Time 1 (months) – 0.30** −0.11 0.003 0.19 −0.08 2. Non-Asian minoritya – −0.70** 0.06 −0.14 0.13 3. Parent Ed. (years) – −0.18 0.33** c 4. Maleb – −0.18 0.33** 5. Childcare (months) – −0.04 6. Missing datad – 7. CBRS Year 1 8. CBRS Year 2 9. HTT Time 1 10. HTT Time 2 11. HTT Time 3 12. HTT Time 4 13. Spanish-speakinge 7 8 9 10 11 12 13 1. Age Time 1 (years) 0.18t 0.09 0.18t 0.03 0.24t 0.21 0.20t 2. Non-Asian minoritya 0.004 −0.04 −0.002 −0.15 −0.23t −0.27* 0.73** 3. Parent Ed. (years) 0.04 0.24t 0.18 0.34** 0.32* 0.32* −0.67** 4. Maleb −0.13 −0.30* −0.11 −0.14 −0.09 −0.01 0.15 5. Childcare (months) −0.11 −0.14 0.20t 0.18 0.15 0.14 −0.18 6. Missing datad −0.07 −0.09 −0.16 −0.17 −0.18 −0.20 0.22* 7. CBRS Year 1 – 0.51** 0.42** 0.47** 0.38** 0.40** −0.15 8. CBRS Year 2 – 0.33* 0.46** 0.26t 0.10 −0.04 9. HTT Time 1 – 0.54** 0.35** 0.51** −0.20t 10. HTT Time 2 – 0.29* 0.39** −0.31** 11. HTT Time 3 – 0.71** −0.35** 12. HTT Time 4 – −0.45** 13. Spanish-speakinge – Note: HTT—Head-to-Toes Task; CBRS—Child Behavior Rating Scale. a Non-Asian minority = 0; minority = 1. b Female = 0; male = 1. c Invalid because missing data were calculated based on parent education. d Complete data = 0; missing data = 1. e English-speaking = 0; Spanish-speaking = 1. t p < 0.10. * p < 0.05. ** p < 0.01. Table options Next we assessed reliability across examiners for the Head-to-Toes Task scores for both sites. At the Michigan site, looking within schools because examiners were assigned to assess children in particular schools, there were no significant differences in average Head-to-Toes Task scores at any task administration. At the Oregon site, no significant differences in Head-to-Toes Task scores across examiners (also looking within schools) were found at Times 1–3, but a significant difference was found at Time 4, F(7, 51) = 2.95; p < 0.05. This was due to one examiner, who only tested Spanish-speaking children. These children scored lower on the Head-to-Toes Task compared to other children in the study (see Table 5). We also tested for examiner differences in scoring self-corrected responses on the Head-to-Toes Task. At the Michigan site, there were no significant differences in the average number of self-corrects scored by examiner within schools at Times 1–3. At Time 4, a significant difference emerged, F(5, 242) = 5.26, p < 0.01, and closer inspection revealed that two out of six raters gave significantly higher average numbers of self-corrected responses, although this did not produce a difference in overall score. At the Oregon site, there were no significant differences by examiner in the average number of self-corrected responses on the Head-to-Toes Task at Times 1–4. Finally, in another study, inter-rater reliability was obtained by having six examiners score videotapes of twelve children, selected at random, completing the Head-to-Toes Task ( Connor et al., 2007). Examiners achieved excellent inter-rater reliability (alpha = 0.95 for self-corrections, 0.98 overall). Overall, these results meet generally established reliability levels ( Landis & Koch, 1977). 5.2. Sources and predictors of variation in behavioral regulation Our second question asked about the amount of variability in behavioral regulation within individuals over time (i.e., variability due to development), among children, and among classrooms, after taking into account background predictors. Our third research question examined the relation of sociocultural variables, gender, and childcare experience, with Head-to-Toes Task scores. We investigated both questions by modeling growth curves using a single cross-classified random effects feature, known as HCM2, in the HLM 6.02 program (Raudenbush & Bryk, 2002; Raudenbush, Bryk, Cheong, & Congdon, 2004). Behavioral regulation data were cross classified, because most children were members of two classrooms during the study. We included an examination of site effects, which demonstrated consistency in our findings across site and provided justification for including both samples in the same analysis. In the growth curve analysis in HCM2, multiple task administrations were modeled at level-1, and child characteristics were modeled at level-2. The program applies a full-maximum likelihood procedure that uses each score to calculate fitted growth curves based on the child's exact age at score attainment. The average score (intercept) as well as an average effect of age, or growth rate (slope), and quadratic trend (acceleration) are modeled. The technique is robust to non-normal data distributions, as long as model residuals follow a normal distribution (Raudenbush & Bryk, 2002). A histogram of the residuals from the final model (see final model below) upheld the assumption of normality. One assumption that holds from traditional regression techniques is homogeneity of level-1 residual variance; we note that variance on the task decreased over time because more children scored near ceiling levels. It is not yet possible to assess the homogeneity assumption with the cross-classified program, but a model representing a good fit to the data likely accounts for any heterogeneous variance across level-1 scores using predictors such as growth rate and acceleration. Finally, even if the variance assumption were violated, coefficients for child background predictors will be relatively unchanged, although the estimation process may be less efficient (Raudenbush et al., 2004). Missing data for parent education and childcare experience were imputed with the expectation maximization (EM) algorithm from SPSS, which is a maximum likelihood procedure that uses all available background (level-2) variables on participants to iteratively calculate missing values (Acock, 2005). A dummy variable was then created and coded “1” for children missing data (20% at each site). Total sample sizes for the analysis were N = 1472, child N = 445, and classroom N = 213. Seventeen children (12 in Michigan; 5 in Oregon) were assessed in the laboratory and assumed to be in their own classroom (17 + 159 Michigan classrooms + 37 Oregon classrooms = 213). Prior to building the final model, we assessed the extent of variation in intercept, growth rate, and acceleration at different levels. We therefore created a level-1 growth model without level-2 predictors (Model 1). The final model (Model 2; see below) then incorporated predictors of the intercept and the acceleration term. These are variables producing deflections from the mean score or mean acceleration at age t (centered at 48 months), holding all other variables constant at their mean or reference value (for dummy variables, the reference value was zero). We added predictors, one at a time, first to the intercept, then to the growth rate, and finally to acceleration, trimming predictors that did not reach significance to maintain parsimony ( Raudenbush & Bryk, 2002). We also ran the model with age centered at the grand mean (63 months) and found highly similar results. We report the final model, shown below, with age centered at 48 months because it is a meaningful reference point for interpreting findings, since many children enter preschool around age 4 years, and undergo significant development in the skills underlying behavioral regulation during this period ( Bronson, 2000 and Diamond et al., 2002; Shonkoff & Phillips, 2000). equation(1) Level 1: Ytjk=π0jk+π1jk(child age)+π2jk(quadratic of child age)+ejkLevel 1: Ytjk=π0jk+π1jk(child age)+π2jk(quadratic of child age)+ejk Turn MathJax on equation(2) View the MathML sourceLevel 2: π0jk=θ0+b00+c00+(γ01) non-Asian minority+(γ02) Spanish version+(γ03) SES (parent education)+(γ04) gender+(γ05) missing dataπ1jk=θ1π2jk=θ2+(γ22) SES (parent education)+(γ23) missing data Turn MathJax on In Eq. (1), a child's expected score at age t is made up of four parts—the intercept (π0jk), slope or growth rate calculated from child age (π1jk), acceleration or quadratic of child age (π2jk), and error for the individual score (ejk). In Eq. (2), the intercept, or average score at age t, is in turn comprised of the grand mean of all scores, plus random child and classroom effects, plus the effect of child-level predictors such as SES and gender. Random effects associated with child j, averaged across all classrooms, are represented by b00, and c00 is the random classroom effect associated with membership in classroom k, averaged across children. The slope, or growth rate (π1jk), is defined as the average growth rate at age t, and the acceleration (π2jk) is defined as the average acceleration at age t, plus the effects of SES and having missing background data. Because of the relatively small number of children in classrooms (on average four), the error terms for growth rate and acceleration were fixed across children and classrooms ( Raudenbush et al., 2004). 5.3. Sources of variation in behavioral regulation: random effects To address our second research question asking about variance at three different levels, we used results from the random effects portion of Table 6. The initial model, Model 1, which included growth and acceleration, explained 11.0% of the variance among individual children's different test scores (level-1 error variance), 21.0% of the variance among children, and 89.5% of the variance among classrooms. The final model, Model 2, explained 18.5% of the variance remaining among children after accounting for growth, and 66.2% of the variance remaining among classrooms. Overall, the final model, including all significant predictors, explained 41.4% of the total variance, and 8.1% of the variance in average Head-to-Toes Task scores after accounting for the effects of maturation. About four percent of the initial classroom-level variation remained, χ2 = 327.46, p < 0.001. Thus, significant differences among classrooms were present even after considering developmental and child characteristics. Table 6. Cross-classified growth curves modeling head-to-toes task scores, age centered at 48 months Fixed effects Model 1a Model 2 Predictor Coeff. d.f. t Coeff d.f. t d b Intercept 13.50 1469 61.96*** 14.65 1462 50.63*** Non-Asian minority −1.04 1462 −1.94t 0.06 Male −1.07 1462 −2.89** 0.08 Spanish version −2.92 1462 −2.43* 0.08 Missing data −1.56 1462 −3.23** 0.09 Parent education 0.23 1462 2.13* 0.07 Linear growth rate (slope) Intercept 0.59 1469 18.55*** 0.60 1462 20.59*** 0.95c Quadratic trend (acceleration) Intercept −0.01 1469 −7.80*** −0.01 1462 −9.13*** 0.10 Missing data <0.01 1462 2.09* <0.01 Parent education <0.01 1462 −0.70 (ns) Random effects Model 1 Model 2 Variance d.f. χ2 Variance d.f. χ2 Times (e, intercept) 19.66 19.99 Children (b00, intercept) 10.21 444 1126.94*** 8.32 439 1033.10*** Classrooms (c00, intercept) 1.48 211 521.54*** 0.50 207 327.46*** a We also ran the fully unconditional model (without the growth rate or acceleration), and obtained e = 22.08, b00 = 12.92, and c00 = 14.16. b Effect sizes were calculated for a 1 S.D. change in the coefficient. c The effect size for age represents the effect for a 1 S.D. increase in age (11.25 months), so corresponds roughly to the effect size associated with being 1 year older rather than 1 month older (the coefficients shown are for 1 month, as in the actual model results). t p = 0.052. * p < 0.05. ** p < 0.01. *** p < 0.001. Table options 5.4. Predictors of behavioral regulation: fixed effects We also examined the extent to which child characteristics, such as gender and parent education, changed the predicted growth trajectory, controlling for all other variables. These fixed effects are shown in Table 6. The top right portion of Table 6 lists the coefficients for each predictor. The final model describes a fitted or mean growth curve where the predicted outcome for a child at a certain age depends on the intercept, slope, and acceleration. For “Fixed Effects,” coefficients under “Intercept” represent deflections from the average score at age 4 years associated with predictor values. The sample average Head-to-Toes Task score for children at age four (grand mean) was 14.65, which differed significantly from zero, t = 50.63, p < 0.001. Receiving the Spanish version of the Head-to-Toes Task resulted in a significant negative deflection from this average score of almost three points, t = −2.43, p < 0.05, and being a boy was associated with scoring about one point lower than average, t = −2.89, p < 0.01. Being a member of a family missing background data was associated with scoring 1.6 points lower than average, t = −3.23, p < 0.01. Using the standard deviation of the outcome (7.06) to calculate effect size, these effects were small (see Table 6). Having a parent with two more years of education than average was associated with about a 0.5-point score increase, t = 2.13, p < 0.05. Finally, being minority (non-Asian) had a negative effect at a level of marginal significance, t = −1.94, p = 0.052. To explore the nature and extent of the negative effect of the non-Asian minority variable on the intercept, we conducted a post-hoc analysis with an interaction between minority and site. After including this interaction in the model, the initial marginally significant effect of minority status disappeared, t = 0.35, p = 0.59. We graphed scores by site and minority status, and found that children from Oregon scored significantly lower compared to minority children from Michigan and non-minority participants from both groups, but only at Time 4. No predictors affected the growth rate, which was positive for all children, t = 20.59, p < 0.001, such that children's scores increased, on average, about seven points per year with steadily decreasing rates of growth over time (i.e., negative quadratic trend suggests deceleration over time in rates of HTT growth). The missing parent education data dummy variable significantly predicted the quadratic trend. In general, the growth rate for children's scores flattened out (i.e., decelerated; see also Fig. 1) over time, but scores for children with complete data decelerated more rapidly, t = −9.13, p < 0.001, compared to scores for children with missing data, t = 2.09, p < 0.05. Parent education did not predict acceleration, but it was associated with the missing data variable and thus included in Model 2. In sum, significant variability was found in children's behavioral regulation assessed directly with a task demonstrating reliability and validity. Except for age, effect sizes were small for child characteristics including gender, parent education (SES), being administered the task in Spanish, and non-Asian minority status. There were also no significant site differences.