Early Risk Factors for Later Mathematics Difficulties Paul L. Morgan, Ph.D., Population Research Institute, The Pennsylvania State University George Farkas, Ph.D., University of California, Irvine Steve Maczuga, M.S., Population Research Institute, The Pennsylvania State University 1 This work is supported by grant #R324A07270, National Center for Special Education, Institute of Education Sciences No official endorsement should be inferred Illustrative data from NAEP (2005) Those falling below “basic” level of mathematics proficiency 2 Grade level Disabled Non-disabled 4th grade 43% 17% 8th grade 68% 27% 12th grade 83% 36% Why mathematics difficulties is something to prevent Being poorly skilled in mathematics lowers an adult’s employability & wages, over & above poor reading, low IQ, & many other factors (Rivera-Batiz, 1992) Even adults with good literacy skills are more likely to be unemployed (& less likely to be promoted when employed) if they have poor mathematics skills (Parsons & Bynner, 1997) Being poorly skilled in mathematics may have negative “carry over” effects on children’s socio-emotional development Children who are poorly skilled in mathematics can also begin to report internalizing psychopathology (Morgan et al., in review) 3 Field’s limited knowledge about MD A preliminary search of PsychInfo yields the following: “Mathematics disabilities” and “longitudinal” and school children” = 32 citations “Reading disabilities” and “longitudinal” and “school children” = 340 citations Much of the extant work on MD has used small convenience samples identified with varying criteria and followed over fairly short time periods “Figuring this out” has implications for practice Wait & fail vs. Wait & see” 4 Hypothesized or identified risk factors for MD Child’s socio-demographic characteristics Age, gender, SES (e.g., Jordan et al., 2007; Lachance & Mazzocco, 2006; Mazzocco & Thompson, 2005) More “educationally relevant factors” Retention, reading difficulties, inattention & other learning-related behavior problems, disability (Cirino et al., 2007; Fuchs et al., 2006; Miller & Mercer, 1997) Some contradictory results from these investigations Gender: Aunola et al. (2004); Jordan et al. (2006) vs. Lachance & Mazzocco (2006) SES: Jordan et al. (2007) vs. Mazzocco & Thompson (2005) Retention: Hong & Raudenbush (2005) vs. Jordan et al. (2003) 5 Repeated learning failure as a proposed “core characteristic” Identifying children as having MD using low score on a single measure’s administration is likely to identify a heterogeneous group, whose low skill level may result from distinct mechanisms Some children may perform poorly due to neurological-based disorders, while other’s may perform poorly due to economic disadvantage or fewer opportunities to learn mathematics informally in the home (e.g., Jordan et al., 2006; Starkey et al., 2004) Repeated failure to learn mathematics has been put forth as a criterion for mathematics disability, but the predictive utility of this criterion, as well as risk factors for repeated learning failure has yet to be established (Geary et al., 2000; Mazzocco & Myers, 2003) 6 Organization to today’s seminar Investigating two, inter-related questions Research Question #1, Study 1: To what extent does the timing and persistence of mathematics difficulties during kindergarten elevate children’s risk of displaying mathematics difficulties throughout the remainder of elementary school? Research Question #2, Study 2: If the persistence of mathematics difficulties “matters,” then which children are especially at risk for experiencing persistent mathematics difficulties? 7 Study 1’s purpose What factors predict of children’s initial mathematics knowledge & skills growth over time? Study 1 characterized these risk factors as belonging to one of three “blocks” The relative severity of the child’s MD (i.e., its timing & persistency) The child’s socio-demographics (e.g., SES) The child’s more “educationally relevant” factors (e.g., his or her relative reading skill, the frequency of the child’s learning-related behaviors, the child’s disability status) 8 Study 1’s theoretical categorizations of MD “Experiential disadvantage” Some children might display an early onset of MD (i.e., at K entry) due to “or a lack of high-quality interactions with siblings, parents, & other adults during preschool (e.g., Klibanoff et al., 2006) “Instructional disadvantage” Other children might display a later onset of MD (i.e., at the end of K), which may be due to or a lack of response to typically instruction offered by general classroom teachers (e.g., Jordan et al., 2006) “Doubly disadvantaged” Still other children might display repeated MD, despite their preschool & K experiences, and so may be more likely to have mathematics disabilities 9 Study 1’s categorizations of the timing & persistency of the child’s MD MD as a score in the bottom 10% of scores on the Mathematics 1 0 Test at a particular survey wave D10: Children who displayed MD at the fall but not spring of K D01: Children who displayed MD at the spring but not fall of K D11: Children who displayed MD at both the fall & spring of K D00: Children who displayed MD at neither the fall nor the spring of K Study 1’s major research question Does the onset & relative persistency of MD in K (D10, D01, D11)… 1 1 ….after statistically controlling for sociodemographic & more educationally relevant factors measured during K… …predict children’s initial skill level (spring of 1st grade) & over time growth (spring of 1st grade to spring of 3rd to spring of 5th grade) in mathematics? Database & analytical sample ECLS-K database, a large, nationally representative sample of children entering K in 1998-1999, and tracked through their elementary school years Analytical sample = 7,892 children Sample included those children whose parent-identified race/ethnicity was White, non-Hispanic or Black/African-American, non-Hispanic Other race/ethnic groups excluded because of substantial amounts of missing data on the K administration of the Reading Test Sample also included those children who were retained and those who changed schools between survey waves Socio-demographics of the ECLS-K full & analytical samples fairly equivalent 1 2 Study 1’s measures Mathematics achievement & MD IRT-scaled, untimed & individually administered Mathematics Test (Reliabilities of thetas = .89-.94) Socio-demographic factors Age, SES, Gender Educationally relevant factors 13 K retention K reading difficulties (RD) IRT-scaled untimed & individually administered Reading Test (Reliability of K theta = .91) Difficulties as score <10% Learning-related behavior problems Teacher rating, SSRS Approaches to Learning subscale (e.g., attention, task persistence) Difficulties as score <10% Disability status Study 1’s selected descriptive statistics Fall K Spring K Spring Grade 1st Spring Grade 3rd Spring Grade 5th MD in fall K only N Mean (SD) 326 326 323 259 211 12.96(1.18) 26.18(4.12) 47.12(10.79) 76.45(16.43) 96.93(17.45) MD in spring K only N Mean (SD) 325 325 323 245 180 17.18(2.10) 19.30(1.77) 39.96(10.38) 67.04(14.29) 88.55(20.34) MD in both fall and spring K N Mean (SD) 465 465 461 352 239 11.80 (1.62) 17.09(2.72) 34.99(9.72) 59.86(10.97) 79.35(17.83) No MD in either fall or spring K. N Mean (SD) 6776 6776 6718 5764 4578 25.48(8.40) 36.67(11.03) 62.60(15.54) 98.31(19.06) 118.71(18.42) Note. MD = Mathematics Difficulties. Estimates are weighted by child-level sampling weights. Mean Mathematics Test score, by relative MD groupings 1 5 Relative persistency of these MD categorizations 16 D Spring of 1st grade D Spring of 3rd grade D Spring of 5th grade D10 78/323 (24%) 68/259 (26%) 59/211 (28%) D01 139/323 (43%) 113/245 (46%) 82/180 (46%) D11 314/461 (68%) 245/352 (70%) 156/239 (65%) D00 252/6718 (3%) 236/5764 (4%) 224/4578 (4%) Study 1’s analytical method & models HLM using repeated observations over time as Level 1 & individual child as Level 2 estimating a quadratic growth curve: Level 1: Yti = π0i + π1it + π2it2 + eti Predictor models Model 1: the baseline or unconditional model Model 2: Adds the dummy variables using D10, D01, D11, with D00 as reference group Model 3: Adds to Model 2 the child’s socio-demographic characteristics Model 4: Adds to Model 3 the more educationally relevant characteristics (e.g., child’s reading difficulties, disability status) 17 Study 1’s growth modeling results Model 2 Model 1 Model 3 Model 4 Intercept Slope of time Intercept Slope of Time Intercept Slope of Time Intercept Slope of Time 59.63* 20.82* 62.79* 21.10* 59.58* 20.45* 60.80* 20.55* D10 -16.55* -1.57* -12.54* -1.57* -11.74* -1.30* D01 -24.25* -1.90* -18.90* -1.45* -17.51* -1.29* D11 -29.17* -2.91* -22.62* -2.40* -19.69* -1.96* Age in month .71* -.19* .72* -.20* SES 5.69* .55* 5.46* .53* White 6.29* 1.25* 6.24* 1.35* Female -3.43* -.70* -4.19* -.75* Repeat K -4.33 -.59 -3.81 -.62 -.12 -.50 Fall K. Approaches difficulty -6.44* -.55 IEP -6.70* -.06 Intercept Fall K. Reading difficulty Study 1’s contributions Young children displaying persistent-type MD averaged 5th grade Mathematics Test scores over 2 SD below those who had not displayed MD in K (even those displaying variable-type MD averaged 5th grade scores 1-1.5 SD below their typical peers) Indicate the need to systematically monitoring young children’s mathematics knowledge Young children repeatedly displaying MD likely need intensive early intervention to be mathematically skilled by end of 5th grade, to avoid “wait to fail” 19 Study 1’s limitations Study 1 identified RMD as a relatively stable phenomenon, in that children, even those only who were kindergarten-aged, were unlikely to “catch up” with those who were displaying RMD, and that the relative persistency of RMD “mattered” Yet Study 1 fails to identify risk factors that may increase children’s likelihood of experiencing RMD. That is, who are the children who are likely to be displaying RMD Is there a “common core” of risk factors for RMD that might be incorporated into the field’s screening, monitoring, and prevention efforts? 20 Study’s 2’s purpose and suppositions Study 2’s purpose Is there a “common core” of factors that increase a child’s risk of experiencing repeated learning difficulties in mathematics? Study’s suppositions Identifying risk factors “early” is better than identifying these factors “late” Doing so helps guide earlier screening, monitoring, and intervention efforts Children who repeatedly fail to attain mathematical proficiency should be of elevated concern These children are consistently non-responsive to the instructional practices and routines being provided 21 Study 2’s brief overview We used two population-based, longitudinal datasets (i.e., the ECLS-K, the ECLS-B) to identify early risk factors for later, repeated mathematics difficulties (RMD) We estimated the predicted effects for a wide range of risk factors We were particularly interested in potentially malleable and “educationally relevant” factors We statistically controlled for the “autoregressor” and strong confounds in the analyses to more conservatively estimate predicted effects 22 Study’s 2 two datasets Two NCES-maintained datasets Early Childhood Longitudinal Study-Kindergarten Cohort (ECLS-K) Kindergarten-8th grade longitudinal, nationally representative sample Early Childhood Longitudinal Study-Birth Cohort (ECLS-K) Birth-Kindergarten longitudinal, nationally representative sample Both datasets include individually-administered, adaptive measures of: academic achievement direct observation ratings of learning-related behaviors multi-source surveys of the children’s socio-demographic, gestational, and birth characteristics 23 Analytical samples, time periods,measures, operationalizations ECLS-K ECLS-B Analytical samples N=5,838 N=5,650* Time periods Spring of Kindergarten, 3rd, 5th, & 8th grade 24, 48, & 60 months Measures Socio-demographics, birth characteristics, reading and mathematics achievement, & behavior Socio-demographics, gestational & birth characteristics, cognitive functioning, vocabulary, reading and mathematics achievement, & behavior Repeated Mathematics Difficulties (RMD) Score below 25% cut off at spring of 3rd, 5th, & 8th grade administrations of ECLS-K Mathematics Test Score below 25% at both Preschool & Kindergarten administrations of modified ECLS-K Mathematics Test RMD % of analytical samples 16.44% (n=960) 15.68% (n=900*) 24 *Sub-sample rounded to nearest 50 Analytical methods ECLS-K Descriptive statistics Child- and family-level socio-demographics, child-level learner characteristics Logistic regression (Odds Ratios as the effect size metric) Step 1: Dichotomized as “0” & “1,” with “1” as being in the group of children with scores in the lowest 25% of the score distribution of the spring of 3rd, 5th, & 8th grade administrations of the the Mathematics Test, and “0” as not not being in this group Step 2: Predicted the child’s group membership, using a range of socio-demographic, birth, & learner characteristics, and controlling for the autoregressor, at spring of kindergarten 25 ECLS-B Step 1: Dichotomized as “0” & “1,” with “1” as being in the group of children with scores in the lowest 25% of the score distribution of the 48 & 60 month administrations of the the Mathematics Test, and “0” as not not being in this group Step 2: Predicted the child’s group membership, using a range of socio-demographic, gestational & birth, & learner characteristic, and controlling for a strong confound (i.e., cognitive delay), at 24 months Study’s longitudinal designs Predictors measured by Criterions measured by ECLS-K Spring of Kindergarten Spring of 3rd, 5th, and 8th grade ECLS-B 24 months Preschool (48 months) and Kindergarten (60 months) Datasets 26 Study 2’s ECLS-K measures ECLS-K Mathematics Test Individually-administered, untimed IRT measure measure of a range of age- and grade-appropriate mathematics skills (e.g., identify numbers and shapes, sequence, multiply, use fractions) Reliabilities of the IRT scaled scores ranged from .89 to .94 “Low” score as having a score in the lowest 25% of the score distribution of the spring of kindergarten Mathematics Test distribution ECLS-K Reading Test Individually-administered, untimed IRT measure measure children’s basic skills (e.g., print familiarity, letter recognition, decoding), vocabulary (receptive vocabulary), and comprehension (e.g., making interpretations) Reliabilities of the IRT scaled scores ranged from .91 to .96 “Low” score as having a score in the lowest 25% of the score distribution of the spring of kindergarten Reading Test administration 27 ECLS-K measures (cont.) Modified version of the Social Skills Rating Scale Kindergarten teacher rated the frequency of that the child engaged in the particular behavior Strong split half reliabilities in kindergarten (e.g., .89, learningrelated behaviors) Three sub-scales, using “worst” 25% cut-off criterion Learning-related behavior problems (e.g., displays attentiveness, persists at tasks) Externalizing problem behaviors (e.g., argues, disturbs the class) Internalizing problem behaviors (e.g., seems anxious, lonely) Survey data of children’s socio-demographics, birth characteristics (e.g., low birthweight, mother’s education level) 28 Descriptive statistics for RMD and nonRMD groups, ECLS-K continuous data RMD Non-RMD Mean (SD) Mean (SD) SD Unit Differences Mathematics Test Score 25.63 (5.69) 40.22 (11.53) -1.3 Reading Test Score 36.87 (7.21) 49.47 (14.19) -.89 Approaches to Learning 2.66 (0.67) 3.24 (0.59) -.98 Externalizing Problem Behavior 1.85 (0.68) 1.63 (0.56) .39 Internalizing Problem Behavior 1.65 (0.51) 1.53 (0.46) .26 Kindergarten Predictors 29 Logistic regression of 3rd, 5th, and 8th grade RMD (ORs) using kindergarten predictors Kindergarten Predictors Low Kindergarten Math 30 Model 1 19.79 *** Model 2 Model 3 Model 4 16.90 *** 16.94 *** 9.76 *** Child is Male 0.52 *** 0.52 *** 0.38 *** Child Age at Assessment 1.06 ** 1.06 ** 1.06 ** Mother’s Education, Less than High School Grad. Mother’s Education, High School Grad. 5.00 *** 5.00 *** 4.89 *** 1.94 *** 1.93 *** 1.94 *** Mother’s Age at Birth > 35 years 0.82 0.82 0.84 Black 2.85 *** 2.86 *** 2.75 *** Hispanic 0.76 0.76 0.82 Other 0.92 0.93 0.90 Birth Weight <= 1500 grams 1.23 0.99 Moderately Low Birth Weight 0.89 1.03 Low Kindergarten Reading 2.00 *** Low Approaches to Learning 2.03 ** High Externalizing Behavior 1.61 High Internalizing Behavior 1.28 Study 2’s ECLS-K results Potentially malleable and educationally relevant risk factors by the end of kindergarten for 3rd-8th grade RMD include earlier history of MD, earlier history of RD, and earlier history of learning-related behavior problems These risk factors are not mediated by the child’s or family’s socio-demographics, or the child’s birth characteristics, despite their sometimes strong predicted effects The onset of MD by kindergarten is an especially strong risk factors for MD through the elementary and middle school years 31 Study 2’s ECLS-B measures Modified Bayley Individually-administered measure of children’s age-appropriate cognitive functioning as manifested in memory, habituation, preverbal communication, problem-solving and concept attainment. The interviewers ask children to complete specific tasks (e.g., “turn pages in a book,” “look for contents of a box,” “put three cubes in a cup”). IRT reliability coefficient for the BSF-R mental scale at 24 months was .88 (NCES, 2007) “Low” as having a score in the lowest 25% of the score distribution Modified McArthur Communication Development Inventory (CDI) Child’s parents asked if the child is saying each of 50 vocabulary words (e.g., “meow,” “shoe,” “mommy,” “chase”) CDI recently reported to classify children into language status groups with 97% accuracy (Skarakis-Doyle et al., 2009) “Low” as having a total score in the lowest 25% of the score distribution 32 ECLS-B measures (cont.) Learning-related behavior problems Modified version of the Bayley’s Behavior Rating System Field staff administering the Bayley also rated the children’s behavior on a frequency scale (e.g., 1=“constantly off task,” 5=“constantly attends”) Cronbach alpha of .92 for the behavioral items (Raikes et al., 2007) “High” as having a score in the highest 25% of the distribution of total scores for “inattentive,” “not persistent,” “no interest” Birth certificate data and parental survey on a range of socio- demographic, gestational, and birth characteristics (e.g., preterm, low birthweight, congenital anomalies) 33 Descriptive statistics for RMD and nonRMD groups, ECLS-B continuous data RMD Non-RMD Mean (SD) Mean (SD) SD Unit Differences Modified Bayley Score 121.39 (9.01) 128.79 (10.35) -.71 Modified CDI Word Score 23.67 (10.85) 30.35 (11.62) -.57 24 months 34 Logistic regression of 48-60 month RMD using 24 month predictors 24 Month Predictors Low Bayley at 24 Months Child’s Age at 60 month Assessment Male 35 Model 1 3.64 *** Model 2 Model 3 Model 4 3.02 *** 2.95 *** 2.23 *** 0.79 *** 0.79 *** 0.71 *** 1.18 1.22 1.12 African-American 1.35 * 1.32 * 1.34 * Hispanic 1.18 1.21 1.24 Other 1.19 1.16 1.13 Mother’s Education, no diploma Mother’s Education, High School Graduate Mother’s Age over 35 at Child’s Birth Mother Not Married at Child’s Birth 4.66 *** 4.47 *** 4.40 *** 2.28 *** 2.22 *** 2.24 *** 0.89 0.86 0.84 1.24 1.22 1.22 Logistic regression of 48-60 month RMD using 24 month predictors (cont.) 36 24 Month Predictors Model 3 (cont.) Model 4 (cont.) Very Pre-Term 1.15 1.09 Moderately Pre-Term 1.34 1.27 Very Low Birth Weight Moderately Low Birth Weight Labor Complications 1.77 1.65 1.54 * 1.58 ** 0.75 * 0.74 * Medical Risk Factors 1.03 1.01 Behavioral Risk Factors Obstetric Procedures 1.14 1.17 0.93 0.94 Congenital Anomalies 0.80 0.79 Low Word Score at 24 Months High L-R Behaviors at 24 Months 1.58 ** 1.41 ** Study 2’s ECLS-B results Potentially malleable and educationally relevant risk factors by 24 months for 48-60 month RMD include earlier history of cognitive delay, language delay, and learning-related behavior problems These risk factors are not mediated by the child’s or family’s socio-demographics, or the child’s gestational or birth characteristics, despite their sometimes strong predicted effects 37 Collectively, what do Study 1’s and Study 2’s analyses tell us? RMD is a relatively stable condition, in that children who are experiencing MD “early” are likely to still be experiencing MD “late.” A “common core” of factors that increase a child’s risk of RMD may exist, that includes: Prior history of MD or an early onset of cognitive delay Reading or language difficulties Learning-related behavior problems Being raised by a mother with a low level of education Prior history of learning difficulties and learning-related behavior problems may be particularly educationally relevant, and potentially malleable The effects of these risk factors are robust, and can be detected early, by children’s kindergarten or even toddler years Early screening, monitoring, and intervention efforts may need to be “multifaceted” so as to account for the multiple developmental pathways that may result in children experiencing RMD 38 Thank you! For additional questions, please contact: Paul L. Morgan Department of Educational Psychology, School Psychology, and Special Education The Pennsylvania State University University Park, PA 16802 (814) 863-2285 paulmorgan@psu.edu 39