ABSTRACT. This study was the first to test a model of the influence of homework on classroom performance using a sample of elementary school students. A total of 28 teachers in Grades 2 and 4 took part in the study, along with 428 students and parents. The authors used structural equation modeling to examine relationships among variables. Student norms were positively related to the elimination of distractions from homework by parents. Positive student norms, higher student ability, and positive parent attitudes toward homework were all related to greater parent facilitation. Student's attitude toward homework was unrelated to home and community factors but was related positively to parent attitudes toward homework. Classroom grades were unrelated to student's attitude toward homework but were predicted by how much homework the student completed (even after the use of homework in grading was controlled), by student ability, and by the amount of parent facilitation. More generally, parent facilitation was an important mediator of the relation between student norms, student ability, and parent attitudes toward homework, and the outcome of classroom grades.

Key words: elementary school students, homework, structural equation modeling

THE COMPLETION OF A HOMEWORK ASSIGNMENT involves the complex interaction of more influences than any other pedagogical technique. Teachers structure and monitor homework assignments in a multitude of ways. Student individual differences play a significant role because--more so than classroom instruction--homework allows students considerable discretion about whether, when, and how to complete an assignment. Parents and siblings often participate in assignments--sometimes voluntarily, sometimes by design. The home influences the process by creating an environment that either facilitates or inhibits study. Students take cues from peers about how much importance to place on homework. Finally, the broader community plays a role by providing other leisure-time activities that compete for the student's time.

Research involving thousands of students shows little correlation between homework and test scores in elementary school, but the relationship grows positive and strong in secondary school (Cooper, 1989). There are several reasons why that might be the case (Mulhenbruck, Cooper, Nye, & Lindsay, in press). First, young children who are struggling in school probably take longer to finish assignments. Second, young children have limited ability to keep their attention focused. Distractions at home entice them away from the books spread out on the kitchen table. Third, young children have not yet learned good study skills. They do not know how to apportion their time between easy and hard tasks or how to engage in effective self-testing.

In light of the complex pattern of influences that can affect the value of a homework assignment, it seems necessary to propose models of the homework process that include multiple variables in sequential relation to one another. Our first purpose in the present article was to examine previous attempts to build and test models of educational outcomes that use homework as an integral part of the achievement process. In carrying out that review, we found that few such models exist and that those that do exist are either untested or have been tested on high school samples only. Therefore, our second purpose was to propose and test a model of homework's influence on the classroom performance of elementary school students.

A Review of Models That Include Homework as an Influence on Achievement

Coulter (1979) presented a temporal model of the homework process that was divided into three phases. In the first phase, teachers acted to motivate, structure, and facilitate the completion of assignments, and greater efforts on the teacher's part were assumed to lead to more successful assignments. Those efforts, along with the student's personal characteristics, determined whether students chose to do homework or to engage in other activities. During the home-community phase, several factors that affect performance--including whether the home learning environment was facilitative, the types of tutoring resources available, and whether community resources were available--combined to influence the student's actual performance on the assignment. Finally, Coulter's results suggested that during classroom follow-up, the amount of teacher feedback on homework, the correspondence of tested material to material included on homework, and the relating of homework assignments to other class work positively affected classroom test results, attitudes, and ultimately academic achievement.

Keith and his colleagues have used several large national data sets to test various models of academic achievement that included homework among the predictors. For example, Keith (1982) used the High School and Beyond data set to test a model that included the amount of time spent on homework as a predictor of high school grades. That model proposed that student's race and family background (i.e., parent's education and father's occupation) were exogenous variables affecting student's ability and field of study (i.e., academic or vocational). Time spent on homework was seen as a function of those three variables, with higher ability students from higher socioeconomic status (SES) families expected to spend more time on homework. Student's grades in high school was the final variable in the model.

Keith, Reimers, Fehrman, Pottebaum, and Aubey (1986) also used the High School and Beyond data set, but in that analysis, student's gender was added as an exogenous variable, parental involvement was added as a second-stage variable, and time spent watching television was added concurrently with time spent on homework. They expected that more parental involvement in homework and less television watching would be positive influences on student achievement. The outcome measure was an average of reading and math scores on tests designed for the study. Keith et al. (1993) applied a similar model to data from the National Educational Longitudinal Study. Keith and Cool (1992) added the variables of quality of instruction and student motivation to a new analysis of High School and Beyond data set.

Keith's models included none of the initial classroom or classroom follow-up variables contained in Coulter's model, undoubtedly because information on those variables was not included in the longitudinal data sets he used. On the other hand, Keith's models are clearly richer in representation of student background factors. Also, Coulter never put his model to empirical test, whereas Keith used sophisticated path analysis and structural equation modeling (SEM) procedures to test his models. In all Keith's analyses, homework was found to have significant direct effects on the performance measure, even after spurious effects (i.e., those effects attributable to homework's and achievement's common association with prior variables in the model) were taken into account.

Cooper (1989) presented a modified and expanded temporal model of the homework process (see Table 1). That model was meant to provide a comprehensive account of the variables mentioned in educational research that might influence homework's impact on a list of educational outcomes. Cooper's (1989) model retained the notion that the homework process could be divided into two classroom phases, with a home-community phase separating them. Borrowing from Keith (1982), Cooper's model viewed student ability and other individual differences as exogenous to the process. However, the student characteristics of race and family background were not among the initial variables because they are typically used in research as proxy variables meant to assess indirectly differences in home environments. In the Cooper model, differences in home environments were directly represented under home-community factors. Homework assignment subject matter and student grade level were also included as exogenous factors.

Neither the Coulter (1979) nor the Keith (1982) models included a set of influences relating to the characteristics of the assignment. The Cooper (1989) model contained such a set, distinguishing assignments according to their length, purpose, skill area, degree of choice and individualization, and completion deadlines, and the social context in which homework was to be performed. Finally, the broad set of outcomes that Cooper suggested are influenced by the homework process are delineated in Table 1.

The Present Study

One glaring omission in research to date is that models of the homework process have yet to be tested on data obtained from elementary school students. Neither Coulter (1979) nor Cooper (1989) tested the models they proposed, and both data sets used in Keith's research included only students who were in the eighth grade or above. Given that the relation between homework and achievement appears to be different at different grade levels, that omission is of great consequence. The present study, then, is the first to test an SEM model of the influences on homework and homework's subsequent influence on classroom performance with a sample of elementary school students.

The data for our analyses were drawn from a larger study of the homework process (Cooper, Lindsay, Nye, & Greathouse, 1998). In choosing the variables to include in the model, we used two considerations to guide our decisions. First, we wanted to include variables that were suggested by previous models. However, including all the variables in Figure 1 would make the model prohibitively large. Therefore, we focused on variables related to student individual differences and parent, home, and community influences. We left out of the model the initial classroom and classroom follow-up factors. That decision also simplified our task because, similar to Keith's model tests, we could conduct the analyses using the student as the unit of analysis. Had classroom factors been included, the classroom would have been the appropriate unit. Second, and also similar to Keith's work, we were constrained by the variables represented in the data set. That constraint primarily influenced the measured variables that were available. There were no instances in which we felt an unobserved factor had to be left out of the model because the data set included no indicator of it.

Following the several models we have described, we included in our model three exogenous variables: student ability level, parent attitudes toward homework, and the homework norm created by the behavior of other students in the class. The second-stage, or home and community variables, included (a) distractions present in the home environment; (b) level of positive parent facilitation, lack of interference, or both (hereinafter referred to as parent facilitation); and (c) amount of time the student spent in other after-school activities. The exogenous home and community variables were posited to influence the student's attitude toward homework, which, in turn, would influence amount of homework the student completed. All the variables were then viewed as influencing the student's classroom performance. The observed variables used to represent each unobserved factor are described in the Method section.

We used SEM to examine relationships among our variables. SEM is similar to multivariate regression in its ability to examine relationships after accounting for relationships with other variables. Moreover, SEM allows for the simultaneous estimation of direct and indirect effects and also provides overall fit indices for the model.

Sample Size and Composition

To have a usable data unit, we obtained complete data from the teacher, at least 1 student in that teacher's class, and 1 of that student's parents or guardians. A total of 28 teachers in Grades 2 and 4 agreed to take part in the study. When the data were analyzed, a total of 570 students and parents completed the questionnaires. However, only 428 respondents had no missing data and could be used in this analysis. Because a student's and that student's parent's responses formed a single data unit (about that student), the actual number of independent data units in the analyses was 214.

The response rate for students and parents was approximately 35%, somewhat lower than that obtained in other homework studies that required parents to complete and return questionnaires (Epstein, 1988; Bents-Hill et al., 1988). There may be several reasons for our lower response rate. First, our questionnaire was considerably longer than those used in previous research; therefore, it required parents to commit more time to complete and return it. Second, our response rate was based on completed questionnaires by both a student and 1 of the student's parents. Thus, our response rate was lowered if either the parent or the student did not participate. Although we would have liked to pursue non-responding parents beyond the second mailing procedure, doing so was beyond the resources of our study. In 85% of the cases, the student's mother completed the parent questionnaire.

The vast majority of the teachers in our sample were White women, and more than half had over 10 years' teaching experience. About half of the teachers had advanced degrees (Master's level or beyond).

Three school districts agreed to take part in the research:

1. A large metropolitan public school district in Tennessee. The district serves nearly 65,000 students. About 58% of the students were White, and 39% were African American. About 45% of the students received free lunch. The participating teachers were drawn from one elementary school.
2. A suburban school district. Adjacent to the urban district, the suburban district served about 12,000 predominately White (94%) middle-class families; 6.5% of the students qualified for free lunch. Participating teachers were drawn from two elementary schools.
3. A rural school district. This district served about 2,200 students, 90% of whom were White and 25% of whom were eligible for free lunch. The rural district participants came from five schools serving Grades K through 8.

When we compared the school-district descriptions with the characteristics of the respondents, we found that the respondents were not a random sample drawn from the districts; they were more likely to be White and less likely to be receiving free lunch than students in the districts as a whole. Therefore, care must be taken in generalizing our results to populations containing percentages of Whites and disadvantaged families that differ markedly from our sample statistics. However, minorities and low-SES families were already somewhat overrepresented in the three school districts when compared with national statistics.

Instruments

The homework survey, which we titled the Homework Process Inventory (HPI), was developed explicitly for this study. The questionnaire is a multi-item survey that has six different versions, one each for lower and upper grade students, their teachers, and their parents. The different versions include parallel questions, so that consistency of responses across the six versions can be examined. More complete descriptions of all the versions can be found in Cooper et al. (1998).

The HPI was pilot tested with small but heterogeneous samples of students, teachers, and parents before the actual data collection began. The pilot testing led to revisions in wording and the addition of questions. Consultants from the Academic Resource Center at the University of Missouri also examined the instrument.

For Grades 2 and 4, the HPI was completed with reference to homework in general, without regard to subject area. At those grade levels, students in the study had a single teacher and probably could not give reliable estimates of the amount of homework they were given or had completed in multiple subject areas. Likewise, we surmised that parents might be able to judge how much time their child spent on homework in general but not be able to distinguish reliably between homework in different subject areas.

Student and family background factors. The presence of positive student norms was assessed with a single teacher item asking, "In general, how many of your students do you think actually finish their homework?" We used a 5-point scale, ranging from all of them to none of them. (All items were coded such that high scores represented the more positive or more frequent end of the dimension or behavior.)

We used a standardized achievement test, the Tennessee Comprehensive Assessment Program (TCAP), to assess student ability. The TCAP was given to all students in the second and fourth grades. The TCAP included both a norm-referenced and a criterion-referenced component. The norm-referenced component consisted of the fourth edition of the Comprehensive Test of Basic Skills (CTBS/4; CTB, 1988). The criterion-referenced component was also developed by CTB but was customized for the Tennessee curriculum. We devised standardized raw score measures of achievement based on total TCAP score, standardized within grade level (with a mean of 0 and a standard deviation of 1). In that way, we were able to combine our scores across grade levels because each student's score was expressed as a departure, in standard deviation units, from the grade-level sample mean. The standardized raw scores ranged from -3.68 to 2.10.

We used three items to assess the availability of economic and time-related parent resources: (a) low or middle SES, which was based on a proxy variable of whether the student was eligible for free school lunch; (b) student and parent reports of whether an adult was usually at home when the student returned home from school; and (c) student and parent reports of the number of other individuals living in the home. However, during model development we found low correlations between SES and whether an adult was at home after school. Furthermore, the initial models revealed no significant relations involving the free-lunch and parent-at-home measures, so we dropped that variable from the analyses reported in the Results section.

We used five parent-report items to assess the extent to which the parent (a) liked homework, on a 5-point scale ranging from like it very much to don't like it at all; (b) thought homework helped their child learn, on a 3-point scale ranging from helps very much to doesn't help at all; (c) thought homework increased versus decreased interest in school, on a 5-point scale ranging from increases it a lot to decreases it a lot; (d) thought homework helped study skills, on a 3-point scale ranging from helps very much to doesn't help at all; and (e) thought homework helped the child learn how to manage time, on a 3-point scale ranging from helps very much to doesn't help at all.

Home and community factors. To assess the availability of a homework environment lacking distractions, we used three parent-report items that tap the extent to which the homework environment is structured so as to minimize distractions. Those included (a) a binary item assessing whether the child was usually alone or with other people when homework was done; (b) an item assessing whether the television was generally on or off while the child did homework, on a 4-point scale ranging from the television is always on to the television is always off; and (c) an item assessing how often the parent made the child set aside quiet time for doing homework, on a 5-point scale ranging from all the time to never. Although we also assessed student reports of those three items, the student items exhibited low interitem correlations among themselves and also with parent responses. An additional student report item that assessed where homework was completed (e.g., at school, in the kitchen, in the bedroom) was dropped because it was not clear how it would relate to the presence or absence of distractions.

To assess parent facilitation, we used three student and parent items asking how often (a) the student asks for help with homework from a family member, on a 5-point scale ranging from all the time to never; (b) someone else's help made it harder for the student to do homework, on a 5-point scale ranging from all the time to never (and reverse scored); and (c) someone else helped with homework by doing things the student should do him- or herself, on a 5-point scale ranging from every night to never. In addition, the parent facilitation assessment included a parent report of how often the student's homework required help from another, on a 5-point scale ranging from every, night to never, and how often somebody helped so the child could finish faster, on a 5-point scale ranging from all the time to never. During model development, we dropped student reports of the extent to which the student asked for help with homework from a family member because that item had a low and nonsignificant factor loading, and we dropped the item asking how often the student asked for help because he or she did not understand the homework, again because of its low intercorrelation with the other items measuring parent facilitation.

We used two items to assess alternative activities. The first item was a student report of how much time was spent watching television every day, on a 4-point scale ranging from I don't watch TV to I watch five or more shows every day. The second item was a composite of two parent items, including the number of hours per week that the child spent on organized activity groups outside of school, on a 5-point scale ranging from none to more than 10 items, and the number of hours the student watched television every day.

Homework outcome variables. We used a single item to assess the student's attitude toward homework: "Do you think homework helps you learn?" Homework completion was assessed by a single student item, "How much of your homework do you finish?" Although parent reports of homework completion were assessed, those were uncorrelated with the student report (r = .02) and hence were not included in the analyses. We expected that the student would have greater awareness of the amount of homework completed than would the parent.

Finally, teachers provided the class grade the student would receive if the class ended on the day the teacher completed the questionnaire. For all analyses that included the grade variable, the extent to which homework counted in grade computation (on a 4-point scale ranging from it does not count to more than 20%) was controlled statistically.

Procedures

Recruitment of schools and teachers. In October 1994, a meeting was held between the investigators and administrative personnel from each of the three participating school districts. At the meeting, the background and nature of the study were explained. After the meeting, each of the district administrators asked school principals if they were interested in allowing the study to be conducted in their schools. Names and addresses of consenting principals were then forwarded to the research staff, and a letter of consent was sent to each principal.

Principals then sent to the research staff lists of teachers in their school who had agreed to learn more about the study. A letter describing the study and a consent form were then sent to each interested teacher. A $20 honorarium was offered to each teacher who agreed to participate.

Questionnaire administration. The students, their teachers, and their parents completed the HPI in February, March, and April 1995. Materials for completion by parents or guardians of the students were sent home with the child. The packet contained the HPI and a #2 lead pencil inscribed with the words "Thank you from the homework study." About 4 weeks after the initial contact, new surveys were sent to parents who had not returned the survey. Parents who did not respond to the second contact were lost to the study.

To minimize expenses and disruption in classes, school personnel administered the HPI to students. However, a standard, written, procedural description was developed that included a list of students whose parents had agreed to let them participate and instructions on HPI administration and on how to respond to student queries. The procedural description was meant to reduce the variability in how instructions were given and how much time the students were given to answer.

After administering the HPI, the teachers completed an administration survey that asked (a) the name of the administrator; (b) the name of the classroom teacher; (c) the date; (d) how nonparticipating students were treated; (e) whether the instructions were read aloud; (f) which questions, if any, prompted questions by students; (g) the approximate time it took for students to complete the questionnaire; (h) whether all participating students completed the questionnaire; and (i) when the participating teacher completed the questionnaire.

In about 90% of classrooms, the teacher administered the questionnaire. In most classrooms, the HPI was read aloud by the teacher. About 72% of teachers completed the HPI after the students, whereas about 15% completed it while administering it to the students.

Achievement data collection. The TCAP was administered to the students in April 1995, roughly concurrent with the completion of the homework surveys. Teacher-assigned grades were collected in June 1995.

Data Analysis

We used the SEM program MPlus (Muthen & Muthen, 1998), using maximum likelihood estimation. We provide chi-square goodness-of-fit tests and also an information criteria fit index, the Bayesian information criterion (BIC; Schwarz, 1978). Standardized regression coefficients are presented in Figure I. Proportion of variance explained (R²) is also presented for all endogenous variables. Any time a manifest variable item had two or more reporters (but more than the single item per factor), the errors of those parallel items were correlated within the factor.

Theoretical Model of Homework Completion

We expected that the endogenous factors within the home and community stage of the model (homework environment lacking distractions, parent facilitation, and alternative activities) would all be influenced by the three exogenous factors: student norms, student ability, and parent attitude. Therefore, we estimated the relationships between each of our exogenous factors (student and family background) and each of our home and community factors (see Figure l, left side).

In terms of the homework outcome constructs (see Figure 1, right side), we hypothesized that homework completion would mediate the relationship between student's attitude toward homework and student's classroom grade. Therefore, we estimated the paths from all student and family background and home and community factors to student's attitude toward homework, and paths from all prior variables, including student's attitude toward homework, to amount of homework the student reported completing. We then estimated the direct paths from student's attitude toward homework and homework completion to classroom grade. In estimating those last relations, we felt it would be important to control for the extent to which teachers gave grades on homework assignments and then used those grades as a portion of the student's class grade.

The full model was estimated, chi²(185,N = 214) = 342.80, p < .001; BIC = 11,076.82. Figure 1 contains the standardized loadings and only significant (p < .05) standardized structural paths between the factors. Table 2 contains a correlation matrix of the observed variables.

Measurement model. All factor loadings on their respective indicators were significant at least p < .05. Standardized factor loadings ranged from .30 to .85 (M = .56) for multiple-indicator factors (single-indicator factors each had a factor loading of 1.0).

Paths between student and family background and home and community factors. Positive student norms, measured by the teacher's assessment of the percentage of students in class who completed homework assignments, were positively related to (a) the extent to which the student had a homework environment that was free of distractions, beta = .27, p < .01, and (b) the degree to which the parent's help was viewed as facilitating homework, beta = .18, p < .05. A student norm that favored homework completion was associated with a positive homework environment and more parent facilitation. Greater parent facilitation was also predicted by higher student ability, beta = .55, p < .01, and parent's attitude toward homework, beta = .18, p < .05. Of the three endogenous factors in this part of the model, the exogenous (student and family background) factors explained more variance in parent facilitation of homework, R² = .39, than in the homework environment, R² =. 10, or in participation in alternative activities, R² = .03 (of which there were no significant predictors).

Paths to homework outcome measures. We examined the relationship of student and family background constructs to student's attitude toward homework, homework completion, and final grade. Parent's attitude toward homework was associated with student's attitude toward homework, indicating that parent and student attitudes were consistent, beta = .25, p < .01. No other constructs significantly predicted student's attitude toward homework. Approximately 14% of the variance in student's attitude toward homework was explained by parent's attitude toward homework. Homework completion was not associated with any of the student and family background or home and community constructs. Therefore, we estimated the final model shown in Figure 1 without including any paths from student and family background constructs to homework completion. Student's classroom grade was significantly associated with student ability, beta = .22, p < .01. More positive parent facilitation of homework was associated with higher classroom grades, beta = .45, p < .01.

Homework outcomes. When other variables were taken into account, student's attitude toward homework did not predict homework completion or classroom grade. In predicting classroom grade, we controlled also for the extent to which homework counted in grade computation (i.e., the teachers report of the percentage of grades that were based on homework). The extent to which homework counted in grade computation fell just short of being significantly associated with classroom grade, p = .06. Classroom grade was positively associated with homework completion, beta = .20, p < .01, again after controlling for the extent to which homework counted in the final grade. The model explained a great deal more variance in classroom grade, R² = .44, than in student's attitude toward homework, R² = .14, or homework completion, R² = .001.

Although we expected homework completion to mediate the relationship between student's attitude toward homework and final grade, the data suggested otherwise. Student's attitude toward homework was not associated with homework completion, beta = .03, nor did it show a direct path to grade, beta = .01.

Mediational analysis. We were interested in the mediational linkages from the set of student and family background and home and community factors (i.e., student norms, student ability, parent resources, parent's attitude toward homework, homework environment, parent facilitation, and alternative activities) to the set of homework outcome variables. Specifically, we tested whether the endogenous parent facilitation factor significantly mediated (accounted for) the relationship between the three exogenous student and family background factors (student norms, student ability, and parent's attitude toward homework) and classroom grade, because parent facilitation was significantly related to each of those constructs. As parent facilitation was not significantly associated with student's attitude toward homework, we did not test the mediational linkages between the three exogenous factors and student's attitude toward homework. Neither of the other endogenous student and family background constructs (i.e., homework environment or alternative activities) was considered as a potential mediator, because neither was associated with homework outcome variables.

We tested the significance of parent facilitation as a mediator by calculating the standard error (SE) for the mediated (indirect) effect via parent facilitation using the multivariate delta method (MacKinnon, 1994; MacKinnon & Dwyer, 1993; Sobel, 1982, 1986). We found that parent facilitation significantly mediated the relationship between student ability and grade (unstandardized mediated effect = 2.10, SE = .70; z = 2.98) and the relationship between parent's attitude toward homework and grade (unstandardized mediated effect = .95, SE = .31; z = 3.04) but not the relationship between student norms and grade (unstandardized mediated effect = 2.39, SE = 1.78; z = 1.34).

Our modeling of the elementary school data revealed several important findings. First, norms of other students, as measured by teacher reports of how likely students in their classes were to complete homework, were positively related to the elimination of distractions from homework by parents. Second, positive classmate norms, higher student ability, and positive parent attitudes toward homework were all related to greater parent facilitation. Third, student's attitude toward homework was unrelated to classmate norms, student ability, and home and community factors but related positively to parent's attitude toward homework. Finally, classroom grades were unrelated to student's attitude toward homework but were predicted by how much homework the student completed (even after the use of homework in grading was controlled); by student ability; and by parent facilitation. More generally, parent facilitation proved to be an important mediator of the relation between student norms, student ability, and parent's attitude toward homework and classroom grades.

Implications for Practice

Several of our findings are important to educators. Most important, the data revealed the critical role of parents in both the homework process and in the success of elementary school students. Positive parent involvement in homework was the strongest predictor of grades. Furthermore, the effect of student norms and parent's attitude toward homework on grades was mediated through the influence those two variables had on parent facilitation. A significant portion of the relationship between student ability and class grades also was mediated by parent facilitation.

That being said, it is also important to distinguish between parent facilitation and other types of parent involvement that might not be salutary. Recall that our measure used a reverse-scored question concerning how often involvement of others makes homework harder. Furthermore, using this data set and another involving secondary school students, we found different types of facilitation had opposite relationships to achievement (Cooper, Lindsay, & Nye, in press). Specifically, parental support for autonomous student behavior showed a positive relationship to achievement, whereas direct instructional involvement showed a negative relationship. We speculated that parent instructional involvement did not actually cause poor performance in students, but instead that poor performance caused parents to get directly involved in instruction. Thus, parents appear to modify their type of involvement with homework depending on the ability level of their child.

The role of positive parent attitudes toward homework also deserves attention. Positive parent attitudes toward homework not only predicted amount of parent facilitation but also directly related to student's attitude toward homework. During the elementary school years, student's attitude toward homework appears not to play a large role in determining classroom success, evidenced by the lack of a significant path in our data. However, when the student moves into secondary school, attitudes become more strongly linked to grades (Cooper et al., 1998). Thus, developing positive attitudes when students are young might help lay the groundwork for later school success.

The finding involving parent facilitation and attitudes suggests that examination of the differences between elementary and secondary models might improve our understanding of the homework process. It seems reasonable to suggest that as students get older, the role of parent facilitation of homework may have a diminishing influence on grades. That would occur because parents become less able to directly instruct students as the material becomes more difficult. We might also expect that the importance of the student's own attitude toward homework would become stronger. Likewise, the importance of student norms would increase as children move into adolescence due to the increasing role of peers in determining behavior.

To examine those developmental changes in model structure, future research on homework's influence on the lives of young children should be longitudinal. There has yet to be a large-scale or in-depth study of homework that follows a cohort of students from the early grades into adolescence. Many of the questions about homework that remain unanswered require such an approach.

Limitations of the Study

The low response rate, approximately 35%, in this study raises the possibility that respondents were not a random sample drawn from the districts. That proved to be the case. Just over 90% of respondents were White, but only about 64% of patrons of the three school districts were White. Similarly, about 40% of all school district students were eligible for free lunch, but only about 12% of respondents were eligible for free lunch. The differences in respondent backgrounds are probably related to their attitudes toward homework and homework practices. On the basis of self-presentation concerns, we could hypothesize that parents and students who responded to the survey had more positive attitudes toward homework and better homework practices than nonrespondents. Certainly then, care must be taken in generalizing our results to populations containing percentages of Whites and disadvantaged families that differ markedly from our sample statistics. However, minorities and low-SES families were already somewhat overrepresented in the three school districts compared with national statistics.

Also, we were unable to obtain data on the reliability of responses from all three groups of participants. Although efforts were made to word questions in a manner that was clear (through pilot testing) and that would lead to consistent responding over time (by asking about typical or average behaviors), that is no substitute for having multiple questions on the same topic or obtaining repeated measurements.

This research was supported by Office of Educational Improvement Grant No. RI 17E40309 to the first author. The views expressed herein are those of the authors and not necessarily those of the granting agency. Thanks are expressed to the staff of the Center of Excellence for Research in Basic Skills for help with the data collection.

Exogenous factors

Student

characteristics

Ability

Motivation

Study habits

Subject matter

Grade level

Assignment characteristics

Amount

Purpose

Skill area used

Degree of

individualization

Degree of student choice

Completion deadlines

Social context

Initial classroom factors

Provision of materials

Facilitators

Suggested approaches

Links to curriculum

Other rationales

Home-community factors

Competitors for

student time

Home environment

Space

Light

Quiet

Materials

Others' involvement

Parents

Siblings

Other students

Classroom follow-up

Feedback

Written comments

Grading

Incentives

Testing of related

content

Use in class discussion

Outcomes or effects

Assignment completion

Assignment performance

Positive effects

Immediate academic

Long-term academic

Nonacademic

Parental

Negative effects

Satiation

Denial of leisure time

Parental involvement

Legend for Chart:

A - Variable

B - 1

C - 2

D - 3

E - 4

F - 5

G - 6

H - 7

I - 8

J - 9

K - 10

L - 11

M - 12

N - 13

O - 14

P - 15

Q - 16

R - 17

S - 18

T - 19

U - 20

V - 21

W - 22

X - 23

A B C D E F

G H I J K

L M N O P

Q R S T U

V W X

1. Homework norms (T) --
2. Student ability (S) .01 --
3. Like homework (P) -.07 .13 --
4. Help learn (P) .00 .03 .56 --
5. Interest (P) -.12 .09 .46 .50 --
6. Skills (P) -.03 .05 .47 .62 .46

--

7. Manage time (P) -.07 .23 .37 .39 .44

.63 --

8. Kid alone (P) -.01 -.07 .02 -.00 -.09

.04 -.08 --

9. TV on (P) .26 -.13 -.08 -.03 -.06

-.04 -.14 .30 --

10. Quiet time (P) .09 .07 -.04 -.12 -.12

-.15 -.19 .26 .36 --

11. Ask help (S) .01 -.08 -.09 -.02 -.19

-.00 -.02 .10 .10 .00

--

12. Help harder (S) .24 .30 .02 .03 .07

.01 .10 -.06 .12 -.01

.19 --

13. Help alone (S) -.05 .30 -.00 .02 .06

.01 .08 -.10 -.05 -.08

.23 -.18 --

14. Require help (P) -.07 -.26 -.08 .03 -.05

.01 -.06 .14 .10 -.15

.23 -.17 -.18 --

15. Ask help (P) -.12 -.22 -.04 .02 -.03

.07 -.11 .17 .17 -.04

.01 -.09 -.08 .46 --

16. Help faster (P) -.09 -.22 -.15 -.19 .26

-.13 -.19 .23 .08 .03

.04 -.11 -.20 .43 .33

--

17. Help harder (P) -.03 .20 .10 .05 .02

.00 .07 .10 -.07 -.00

.16 -.07 -.21 .13 .08

.10 --

18. Help alone (P) .10 .40 .12 .07 .21

.06 .19 -.14 -.10 -.10

.14 -.17 -.23 .36 .23

.45 -.29 --

19. TV time (S) .19 -.06 -.04 -.03 -.13

-.08 -.09 -.05 .20 .13

.11 -.00 -.09 -.05 -.16

-.03 -.17 .05 --

20. Extracurricular (P) -.01 .00 .00 -.08 -.03

.01 -.01 -.00 .04 .04

-.01 .02 -.20 .00 .05

.07 -.00 -.07 .23 --

21. Student attitude (S) .03 -.23 .08 .22 .16

.11 .03 .12 .08 -.08

.10 -.09 -.13 .23 .17

.09 .01 .14 -.03 -.05

--

22. Finish homework (S) .01 .02 -.01 .09 .05

.11 .18 -.11 .05 -.08

.09 .24 .03 .02 .00

.01 -.01 -.04 .10 -.01

.03 --

23. Grade (T) .06 .49 .11 .03 .08

-.00 .21 -.14 -.14 -.04

.08 .31 .18 -.29 -.38

-.28 .13 -.35 -.03 .03

-.18 .23 --

24. % homework counts

toward grade .01 -.17 -.18 -.08 -.08

-.06 .01 -.09 -.02 -.10

.02 -.05 .06 -.10 -.11

-.05 -.07 .10 .03 .02

.00 -.13 -.18

Note. Values beyond approximately. 135 in magnitude are

significant at p < .05, and values beyond approximately .175

in magnitude are significant at p < .01.

Legend Reporter:

S = student;

P = parent;

T = teacher.

Homework norms (T) = how many students finish their homework.

Student ability (S) = Tennessee Comprehensive Assessment Program

test of student ability.

Like homework (P) = extent to which parent likes homework.

Help learn (P) = extent to which parent thinks homework helps

child learn.

Interest (P) = extent to which parent thinks homework increases

versus decreases interest in school.

Skills (P) = extent to which parent thinks homework helps study

skills.

Manage time (P) = extent to which parent thinks homework helps

child learn how to manage time.

Kid alone (P) = whether child is usually alone or with other

people when he/she does homework.

TV on (P) = whether television is generally on or off while

child does homework.

Quiet time (P) = how often parent makes child set aside quiet

time for doing homework.

Ask help (S) = extent to which student asks for help with

homework from a family member.

Help harder (S) = extent to which someone else's help makes it

harder for student to do homework.

Help alone (S) = extent to which someone else helps with

homework by doing things student should do his/herself.

Require help (P) = how often student's homework requires help

from another.

Ask help (P) = extent to which student asks for help with

homework from a family member.

Help faster (P) = how often somebody helps so the child can

finish faster.

Help harder (P) = extent to which someone else's help makes it

harder for the student to do homework.

Help alone (P) = extent to which someone else helps with

homework by doing things student should do his/herself.

TV time (S) = how much time student spends watching television.

Extracurricular (P) = composite of number of hours per week that

child spends doing extracurricular activities.

Student attitude (S) = Do you think homework helps you learn?

Finish homework (S) = how much of your homework do you finish?

Grade (T) = final grade at the time of the study.

(Unreadable) homework counts toward grade (T) = extent to

which homework counted in grade computation.

DIAGRAM: FIGURE 1. Student and family background model of student's attitude toward homework, homework completion, and final grade.

Address correspondence to Harris Cooper, Department of Psychology, McAlester Hall, University of Missouri, Columbia, MO 65211. E-mail: cooperh@missouri.edu

Bents-Hill, C., Boswell, R., Byers, J., Cohen, N., Cummings, J., & Leavitt, B. (1988,April). Relationship of academic performance to parent estimate of homework time. Paper presented at the annual meeting of the National Association of School Psychologists, Chicago.

Cooper, H. (1989). Homework. White Plains, NY: Longman.

Cooper, H., Lindsay, J. J., & Nye, B. (in press). Homework in the home: How student, family, and parenting style differences relate to the homework process. Contemporary Educational Psychology.

Cooper, H., Lindsay, J. J., Nye, B., & Greathouse, S. (1998). Relationships among attitudes about homework, amount of homework assigned and completed, and student achievement. Journal of Educational Psychology, 90, 70-83.

Coulter, F. (1979). Homework: A neglected area of research. British Educational Research Journal, 5, 21-33.

CTB. (1988). Comprehensive Test of Basic Skills: Technical report (4th ed.). New York: Macmillan/McGraw-Hill.

Epstein, J. L. (1988). Homework practices, achievements, and behaviors of elementary school students (Center for Research on Elementary and Middle Schools Report No. 26). Baltimore, MD: Johns Hopkins University.

Keith, T. Z. (1982). Time spent on homework and high school grades: A large sample path analysis. Journal of Educational Psychology, 74, 248-253.

Keith, T. Z., & Cool, V. A. (1992). Testing models of school learning: Effects of quality of instruction, motivation, academic coursework, and homework on academic achievement. School Psychology Quarterly, 3, 207-226.

Keith, T. Z., Keith, P. B., Troutman, G. C., Bickley, P. G., Trivette, P. S., & Singh, K. (1993). Does parent involvement affect eight-grade student achievement? Structural analysis of national data. School Psychology Review, 3, 474-496.

Keith, T. Z., Reimers, T. M., Fehrmen, P. G., Pottebaum, S. M., & Aubey, L. W. (1986). Parent involvement, homework, TV time: Direct and indirect effects on high school achievement. Journal of Educational Psychology, 78, 373-380.

MacKinnon, D. P. (1994). Analysis of mediating variables in prevention and intervention research. In A. Cazeres and L. A. Beatty, Scientific methods for prevention intervention research (NIDA Research Monograph 139, DHHS Pub. No. 943631, pp. 127-153). Superintendent of Documents, U.S. Government Printing Office: Washington, DC.

MacKinnon, D. P., & Dwyer, J. H. (1993). Estimating mediated effects in prevention studies. Evaluation Review 17, 144-158.

Mulhenbruck, L., Cooper, H, Nye, B., & Lindsay, J. J. (in press). Homework and achievement: Explaining the different strengths of relation at the elementary and secondary school levels. Contemporary Educational Psychology.

Muthen, L. K., & Muthen, B. O. (1998). MPlus: The comprehensive modeling program for applied researchers. Los Angeles: Author.

Schwarz, G. (1978). Estimating the dimension of a model. The Annals of Statistics, 6, 461-464.

Sobel, M. E. (1982). Asymptotic confidence intervals for indirect effects in structural equations models. In S. Leinhardt (Ed.), Sociological methodology (pp. 290-293). Washington, DC: American Sociological Association.

Sobel, M. E. (1986). Some new results on indirect effects and their standard errors in covariance structure models. In N. Tuma (Ed.), Sociological methodology (pp. 159-186). Washington, DC: American Sociological Association.

~~~~~~~~

By Harris Cooper; Kristina Jackson, University of Missouri-Columbia; Barbara Nye, Tennessee State University and James J. Lindsay, Behavioral Health Concepts