The homework–achievement relation reconsidered: Differentiating homework time, homework frequency, and homework effort

Study 1: 34,765 grade-9 students (PISA 2000 German extension). Study 2: 2,216 students in 91 intact classes, grade 7 to grade 8 (German TIMSS). Study 3: 483 grade-8 Gymnasium students in 20 classes, Berlin

German secondary students, grades 7-9

Learning and Instruction 17(3), 372-388. Three multilevel and longitudinal analyses of observational data - no experiment and no instrument, hence grade C. What makes it load-bearing for this topic is that it decomposes a construct the rest of the literature treats as one thing. Homework TIME, homework FREQUENCY (how often the teacher sets it) and homework EFFORT (how hard the student says he tries) are measured separately and modelled at both the student and class or school levels. Study 2 controls prior achievement, making it a value-added specification; Study 3 is longitudinal from November to May but relies on self-reported grades within a single selective track.

THE STUDENT-LEVEL AND CLASS-LEVEL COEFFICIENTS HAVE OPPOSITE SIGNS, and conflating them is the single most common error in the homework literature. In PISA 2000, school-level homework time predicts mathematics achievement at B = +24.67 (SE 2.03) before controls and +1.76 (SE 0.58) after school type, cognitive ability and gender are added - a 93% collapse driven almost entirely by track. At the STUDENT level the same variable is NEGATIVE and grows more negative with controls: -3.45 (SE 0.15) then -7.84 (SE 0.18). Trautwein's own words: 'the students who spent more time on their mathematics homework than their schoolmates scored LOWER on the mathematics test.' In the value-added TIMSS analysis the pattern repeats and separates frequency from time: class-level FREQUENCY survives controls at +0.11 (SE 0.05) while class-level TIME does not (-0.02, SE 0.12), and student-level time is -0.08 (SE 0.02). In the longitudinal study, homework time correlates negatively with achievement at every wave (r = -.24 to -.30) while homework effort correlates positively (r = +.32 to +.35), and effort and time correlate NEGATIVELY with each other (r = -.12). Controlling prior grades, effort predicts later outcomes (beta +.16 and +.17) while time predicts them negatively (beta -.14 on grades). The conclusion the field should have absorbed: 'homework time is not a suitable indicator of the effort that students put into their homework.'

High for the student-level estimates, and the paper effectively demonstrates why. Time spent is a behaviour produced jointly by task difficulty, ability and conscientiousness, so a slow worker looks identical in the data to a diligent one. The negative student-level coefficient is best read as reverse causation - struggling children take longer - rather than as harm from homework. The class-level frequency effect is the only estimate in the paper that resembles a teacher decision rather than a child trait, and it is the only one that survives controls.

The time-versus-effort dissociation replicates across German, Swiss and international samples and is corroborated by Cooper's finding that student-reported homework time correlates r = .25 with achievement while PARENT-reported time correlates r = -.02 - the same demonstration that the measure, not the activity, carries the association. It is challenged from one direction: Fernández-Alonso and colleagues find that once prior grades enter the model, homework EFFORT also loses most of its predictive power (0.19 falling to 0.04 in mathematics, and to zero in science), and say explicitly that this 'contradicts the central position of effort in the model of Trautwein et al.'