You are on page 1of 16

Received: 21 October 2022 Revised: 23 August 2023 Accepted: 24 September 2023

DOI: 10.1111/jcal.12889

ORIGINAL ARTICLE

Digital re-attributional feedback in high school mathematics


education and its effect on motivation and achievement

Katharina Alexandra Whalen 1 | Alexander Renkl 1 | Alexander Eitel 2 |


3
Inga Glogger-Frey

1
Department of Psychology, Educational &
Developmental Psychology, University of Abstract
Freiburg, Freiburg, Germany
Background: Students often show unfavourable attribution: they attribute poor
2
Department of Psychology & Sports Science,
Educational Psychology, University of Giessen,
school performance to stable factors such as lack of ability and good school perfor-
Giessen, Germany mance to variable factors such as effort. However, attribution can be influenced by
3
Department of Psychology, Educational individualized digital re-attributional feedback leading to positive motivational effects
Psychology, University of Erfurt,
Erfurt, Germany and higher learning outcomes. This is very promising, but it still is unclear, whether
this digital re-attributional feedback can also be successfully integrated in everyday
Correspondence
Katharina Alexandra Whalen, Department of classroom activities.
Psychology, Educational & Developmental
Objectives: The present field experiment investigated how integrating digital re-
Psychology, University of Freiburg,
Engelbergerstr. 41, 79085 Freiburg, Germany. attributional feedback into classroom instruction affects student attribution, motiva-
Email: katharina.whalen@psychologie.uni-
tion and learning outcomes.
freiburg.de
Methods: In the experiment, 8th–10th grade high school students (N = 322) worked
Funding information
with a digital mathematics learning program which was integrated in a three-week
Robert Bosch Stiftung Germany, Grant/Award
Number: 00906424-002 teaching unit. Half the students in each classroom received only standard feedback
after each practice task (SF group); the other half received additionally an individual
re-attributional feedback (RF group) after every third task. Attribution, mathematics
self-concept, and self-efficacy were measured by an online questionnaire twice a
week; learning outcomes in mathematics were measured weekly.
Results and Conclusion: Hierarchical analyses showed that re-attributional feedback
led to a more favourable attribution in case of success on stable factors. Especially
low-performing students benefited from this feedback. No effects on attribution in
case of failure, self-efficacy or learning outcomes could be found. Further research
could investigate if certain adaptations to the digital re-attributional feedback is more
effective in a real classroom setting and has a broader impact on different students.

KEYWORDS
computer-based learning, mathematics self-concept, motivation, re-attributional feedback, self-
efficacy, student learning

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any
medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2023 The Authors. Journal of Computer Assisted Learning published by John Wiley & Sons Ltd.

J Comput Assist Learn. 2023;1–16. wileyonlinelibrary.com/journal/jcal 1


13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 WHALEN ET AL.

1 | I N T RO DU CT I O N However, if the student fails a task, they might explain it with lack
of ability which is stable and uncontrollable. Consequently, the
Learning in Western schools is regularly characterized by the following: student might show less effort when preparing for the next task.
Students who do not perform well in one subject rarely turn the tables Hence, interventions that prevent attributing poor performance
and become overachievers during their school career (Goodwin, 2000). (exclusively) to low ability may increase study motivation, aca-
Why is that? Low-performing students often show problematic demic self-concept, and ultimately performance and learning out-
attribution styles (Pintrich & Schunk, 2002) when explaining their per- comes (Banks & Woolfson, 2008; Möller & Köller, 2004).
formance which can influence learning behaviour (Chodkiewicz & Attribution retraining from instructors might be such an interven-
Boyle, 2014): Especially in the subject of mathematics a student might tion to help students shift their unfavourable attribution style
attribute poor performance to their lack of ability for this subject mat- (Dresel & Ziegler, 2006). Re-attributional feedback suggests, for
ter: “I made this mistake because I simply don't understand math.” example, that the reason for poor performance on a task was (too)
Such attributions reinforce a poor ability self-concept in mathematics low effort rather than (too) low ability (e.g., “If you work in a more
and can be dysfunctional for learning motivation: students may in turn concentrated manner, you will have good results”). By receiving
not invest the effort and persistence that they would need to improve such feedback, the student should gradually shift his or her attribu-
their skills (Hamm et al., 2017). If they do not invest sufficient effort, tion for poor performance from low ability to low effort (e.g., “I
performance is subsequently also likely to be poor, which again con- made this mistake because I didn't concentrate enough.”). Previous
firms the students' beliefs in their low mathematics ability studies have already shown an effect of attribution retraining:
(Akey, 2006; Heller & Ziegler, 1996). The students are stuck in a With more favourable attribution styles, students show increased
vicious circle, and the tables will hardly be turned towards a better self-control (Banks & Woolfson, 2008), self-efficacy (Martinez &
achievement in this subject over the years. Huber, 2019; Zeichner, 2018), and academic self-concept (Dresel &
Haugwitz, 2008; Dresel & Ziegler, 2006). Hence, they should invest
more effort, perform better, and ultimately have higher learning out-
1.1 | Favourable and Unfavourable attribution comes (Horner & Gaither, 2004; Houston, 2016).

How can this vicious circle be broken? Attributions are causal expla-
nations to ascribe a certain outcome of an event. These processes are 1.3 | Motivational effects of re-attributional
malleable and can be changed by attribution retraining (Borkowski feedback
et al., 1988; Haynes et al., 2009) or re-attributional feedback
(Dresel & Ziegler, 2006). In attribution retraining, the students receive Low-performing students often report low motivation when it comes
feedback “on success or failure situations with appropriate causal to studying for an exam or participating in class (Akey, 2006). Motivat-
attributions” (Dresel & Haugwitz, 2008, p. 4). To understand the ing students is essential for a productive classroom environment and
effect of attribution processes on learning outcome it is important to can possibly lead to better learning outcomes (Aronson, 2008;
distinguish between favourable and unfavourable attribution: Based Schiefele et al., 2016). Important predictors of learning motivation are
on Weiner's theory (Weiner, 1985; Weiner et al., 1980) it is consid- ability self-concept and self-efficacy which are more specific concepts
ered favourable if students attribute their performance in case of suc- that not only strongly influence students' learning motivation, but also
cess to stable, controllable and/or internal factors such as ability and through this, their learning outcomes (Bandura, 1993). Ability self-
understanding, and in case of failure to unstable, controllable and/or concept and self-efficacy have similar effects on students' motivation,
external factors such as concentration or conditions in the classroom. emotions, and actions (Martinez & Huber, 2019), but to understand
Favourable attribution in students predicted higher levels of academic their individual impact, it is important to distinguish them. Bong and
performance (Horner & Gaither, 2004; Houston, 2016). In contrast, it Skaalvik (2003) propose the following distinction: “while self-concept
could have a negative effect on students if they attribute in case of represents one's general perceptions of the self in given domains of
success to unstable, uncontrollable, and external factors such as luck functioning, self-efficacy represents individuals' expectations and con-
and in case of failure to stable, uncontrollable, and internal factors victions of what they can accomplish in given situations” (p. 5). As in
such as lack of ability. Students with unfavourable attribution styles our study we concentrate on computer-based re-attributional feed-
are more pessimistic about their future success, show less effort in back in mathematics class, we are interested in measuring mathemat-
perceived difficult tasks (Hamm et al., 2017; Núñez et al., 2005) and ics self-concept as specific domain of functioning (e.g., “I am a good
perform poorer in school (Akey, 2006; Houston, 2016). math student.”).

1.2 | Re-attributional feedback 1.4 | Self-efficacy

Consider a student with an unfavourable attribution style: If this stu- Self-efficacy can be influenced by attributional feedback as Martinez
dent succeeds in a task, they might think it was only out of luck. and Huber (2019) showed in their review. In this present study, we
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WHALEN ET AL. 3

focus on self-efficacy in a mathematical context as in “I feel confident 1.7 | Computer-based re-attributional feedback in
that I can solve this task without help.” In a distant learning environ- regular classroom-learning
ment, Zeichner (2018) found in a sample of 171 high school students
higher reported self-efficacy scores in students who received attribu- Attributional retraining is not new and has been proven to be
tional feedback in comparison to students who only received cogni- effective in laboratory and school settings. Studies in the 20th
tive feedback. Therefore, it is not enough to simply give cognitive century reported promising results (Craven et al., 1991; Heller &
feedback on the student's answer, especially in settings of self- Ziegler, 1996; Morris, 2013): Heller and Ziegler (1996) found effects
regulated learning. Hence, re-attributional feedback might be an in female university students when giving written and verbal re-
important tool for teachers to have a positive effect not only on the attribution training. Before treatment only 50% of the treatment
student's attribution but also on their self-efficacy. group attributed success internally, after the treatment this number
rose to 75%. Furthermore, before treatment in this group 50% attrib-
uted failure to internal factors, after treatment this did not emerge. In
1.5 | Mathematics self-concept a second study (also in 1996), they investigated 126 high school stu-
dents. Forty-two children received verbal and written re-attributional
Ability self-concept (in mathematics) can also be influenced by pro- feedback over 6 weeks in six voluntary afternoon sessions. The treat-
viding re-attributional feedback. Dresel and Ziegler (2006) tested a ment and placebo group participated in the afternoon sessions but
computer-based, re-attributional feedback in a mathematics learning only the treatment group received re-attributional training, the control
software on 140 seventh grade students. Attributing success to first group only had the regular mathematics classes in the morning. Math-
effort and then ability led to an increase in academic (mathematics) ematical and general self-concept improved in the treatment group.
self-concept, self-control and to a decrease in helplessness immedi- The other groups showed a decline over 6 weeks. Following training,
ately after retraining and in follow-up. Positive changes in school the treatment group attributed more favourably in case of success
performance were found half a year later. In a later study in 151 sixth and failure. In addition, the placebo group showed an increase in
grade students, Dresel and Haugwitz (2008) found significant benefi- favourable attribution in case of failure but a decline in favourable
cial effects of computer-based re-attributional feedback on compo- attributions in case of success. However, Craven et al. (1991) had
nents of student motivation as positive attributions, (mathematics) inconsistent results: 162 primary school students were given verbal
self-concept, and reduced helplessness. These effects could be re-attributional feedback over the course of 8 weeks. The students
found not only directly after the training period but also 5 months either received re-attributional feedback from a teacher, or from a
later. These findings suggest a long-lasting effect of re-attributional researcher, or none. Only the researcher-given feedback resulted in
training on attribution and self-concept. The researchers found a an increase in specific areas of self-concept (reading self-concept,
positive effect on mathematical knowledge acquisition directly and mathematics self-concept, school self-concept, academic self-
1
5 months after training. However, after a year no statistically signifi- concept, general self-concept, 0.41 < d < 0.49) and in effort attribu-
cant promotional effect of re-attributional feedback on mathematical tions in success situations, d = 0.40. However, the teacher-given
knowledge could be found. Therefore, the researchers suggest that feedback did not show any effect neither in self-concept nor in attri-
teachers should implement measures to improve attribution and bution behaviour. The researchers suggest that this might be due to
motivational components such as mathematics self-concept continu- differences in re-attributional feedback density as the researcher was
ously throughout the school year to secure enduring effects on working with individuals or small groups only instead of the whole
learning outcome. class as the teacher was.
At best, verbal or written feedback should immediately follow a
task and is tailored to the actual performance in this task; it should be
1.6 | Learning outcomes frequent enough (i.e., multiple times in a classroom hour) and variable
in their formulation (see Dresel & Ziegler, 2006). If these criteria are
Computer-based re-attributional feedback while performing mathe- met, such an intervention may foster motivation and learning out-
matics tasks can have a positive effect on learning outcomes comes, especially for lower performing students. However, in a typical
(Dresel & Haugwitz, 2008; Dresel & Ziegler, 2006). Martinez and mathematics classroom context where a teacher provides their own
Huber (2019) showed in their review that attributional feedback feedback to students, it is impossible to give equally frequent and pre-
not only had a positive effect on self-efficacy but also on learning cise feedback to up to 30 students in 45 or 90 min of teaching time.
outcomes. The positive influence of self-efficacy on learning out- In Heller and Ziegler's study (1996), for example the students received
comes has been widely reported (Bandura, 1993; Schunk & verbal re-attributional feedback about three times per session. Craven
Ertmer, 1999). Based on the previously discussed literature, this et al. (1991) reported about 30 treatment reinforcements by the
study investigates, whether re-attributional feedback might not teacher, which is less than one per day.
only lead to more favourable attribution in students, but further-
1
Effect sizes were computed based on given F values using Wilson's effect size calculator
more, improve students' mathematics self-concept, self-efficacy,
(Wilson, 2023) https://www.campbellcollaboration.org/escalc/html/EffectSizeCalculator-
and learning outcomes. SMD5.php.
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 WHALEN ET AL.

Thus, newer studies used automated, computer-based re- randomly throughout half a school year and not fully integrated in
attributional feedback instead of verbal or written teacher feedback the teaching plan. Other studies implementing a whole-classroom
to increase density and showed promising results (Dresel & training show positive effects but were conducted with a specific
Haugwitz, 2008; Hamm et al., 2017; Lazowski & Hulleman, 2016; group of students (Casserly, 2013). Therefore, these results cannot
Zeichner, 2018). The computer-based approach allows not only giving be applied for the whole class. Furthermore, participating teachers
re-attributional feedback to every student at the same time. It also need a basic understanding of the re-attributional feedback pro-
allows to still be personalized if the computer program takes each stu- cess, as it has been argued that natural feedback responses given
dent's performance into account (Dresel & Ziegler, 2006). A recent by teachers could have a negative effect and promote unfavour-
meta-analysis from Lazowski and Hulleman (2016) showed in 13 stud- able attribution (Chodkiewicz & Boyle, 2014; Haynes et al., 2009).
ies typically moderate effects (d = 0.54) on student motivation in However, no significant impact of these research findings
attributional re-training. Dresel and Haugwitz (2008) found immediate on changing teaching behaviour and integrating computer-based
and long-term positive effects of computer-based re-attribution train- re-attributional training in school teaching practices have been
ing on motivation (d = 0.45, and d = 0.35 in the follow-up 5 months reported (Chodkiewicz & Boyle, 2014). Given these results, the
later) and mathematical knowledge acquisition (d = 0.54, no effect in question remains, how re-attributional feedback fosters students'
the follow-up) in 151 6th-graders. Hamm et al. (2017) showed posi- motivation in the whole classroom? To get a computer-based re-
tive effects in a randomized online treatment study where 806 first- attributional feedback training accepted by teachers and be used
year college students either received attributional retraining (treat- on a regular basis it might help if it is combined with regular class-
ment group) or stress-reduction treatment (control group). Simple room teaching and fully integrated with the school curriculum. The
slope regression analysis showed that low-performing students in the teachers should not get the impression they are doing something
treatment group outperformed low-performing students in the control special, outside the regular classroom activities which requires
group on a class test 5 months after treatment when controlling for extra work and extra time commitments. Computer-based re-
initial test performance. No treatment effects could be found for attributional feedback should become a normal feature within their
high-performing students. Other studies also showed effects only regular schoolwork. The question needs to be addressed whether
with a specific group of students – mainly students with learning dis- computer-based re-attributional feedback fully integrated in class-
abilities (Berkeley et al., 2011; Casserly, 2013; Okolo, 1992). There- room activities is as effective as outside the classroom.
fore, student's initial performance should be taken into account. To
sum it up, recent literature suggests that if the student's initial perfor-
mance is considered, and feedback is given with a high density, re- 1.8 | The present study
attributional feedback might have a positive effect on motivation and
student performance. Thus, in the present study we not only integrated computer-based
However, research on attribution retraining has been criticized feedback into a typical mathematics classroom setting at regular class
for its emphasis on experimental studies far removed from regular hours, we furthermore, developed a lesson plan according to the
classroom-learning (Okolo, 1992). Most studies were performed out- students' curriculum which consists of normal teacher-student
side of the typical classroom situation in a voluntary setting outside activities and computer-based periods. As all participating teachers
regular class hours (Craven et al., 1991; Dresel & Ziegler, 2006; followed the same lesson plan for each grade (grades 8–10), we
Heller & Ziegler, 1996; Morris, 2013). It is not clear whether effects were able to standardize to a higher degree as in previous studies
found under these special circumstances would hold in mainstream what happened in the classroom throughout the whole training
education. Casserly (2013) discovered that effects of re-attributional period of 3 weeks. Furthermore, the lesson plan ensured that the
feedback found in small group training could not be maintained in the study periods with the mathematics-learning program were
regular classroom. Unfortunately, due to monetary and personal allo- embedded at didactically useful moments during non-digital in-
cation difficulties it is hardly ever possible for schools to implement person classroom learning, were fully integrated, and not separate
attributional retraining on an individual basis or in small groups from other classroom instruction activities. We collected data
(Horner & Gaither, 2004). An effective whole-classroom training led twice a week for the experimental period of 3 weeks plus a week
by the teacher is in most cases the only option. In classrooms with before intervention as baseline measurements. While working with
more than a small group of students, computer-based re-attributional the mathematics program, half the students received computer-
training could support the teachers to increase feedback density. based re-attributional feedback, the other half received the stan-
To investigate whether re-attributional training is equally effec- dard feedback of the mathematics program only. Within class-
tive in the daily teaching routine, further research is needed. All stu- rooms, the students were randomly assigned to the two feedback
dents in a class need to be addressed at regular school hours. Some groups. We wanted to investigate whether in a more realistic set-
researchers have already taken a first step and gave an optimistic out- ting computer-based re-attributional feedback training can also
look: Dresel and Haugwitz (2008) found positive effects on motiva- lead to a change of attribution, self-efficacy, mathematics self-
tion and knowledge acquisition in 151 sixth grade-students after concept, and better learning outcomes as in the studies described
computer-based re-attributional training in the whole classroom. above. Furthermore, as in this setup we addressed all students in
However, the training sessions (five to nine) were placed rather the class not only the ones who sign up for extra hours in the
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WHALEN ET AL. 5

afternoon, we wanted to investigate whether especially low- feedback group had a mean age of 14.04 (n = 118; SD = 0.91) and
performing students benefit from this intervention. 55.5% were female. The 164 students in the re-attributional feed-
back group had a mean age of 14.00 (n = 127; SD = 0.79) and
61.4% were female. No differences between experimental groups
1.9 | Research questions and hypotheses in demographic or prerequisite variables were detected (see
Section 3.1 for statistics).
In the present study, we tested whether computer-based Each participating class received 50 Euros to spend on class activ-
re-attributional feedback integrated into a learning software has an ities and each participating teacher received 50 Euros for the school's
effect on attribution, self-efficacy, mathematics self-concept, and mathematics department.
learning outcomes compared to the software's standard feedback.
Moreover, we investigated whether re-attributional feedback exerts a
positive effect especially for low-performing students. They may 2.2 | Materials and manipulations
profit the most from a shift in attribution style in case of failure.
To answer these research questions, we formulated the following 2.2.1 | Lesson plan
hypotheses:
During the 4 weeks of data acquisition, the students had mathematics
H1. Computer-based, individual re-attributional feed- class with their regular teachers. Usually students had either a 90-min
back over a course of 3 weeks (a) leads to a change in lesson on 2 days a week or a 45-min lesson on 4 days a week. In week
attribution, (b) fosters self-efficacy, (c) fosters mathe- 0, teachers taught their regular classes according to their individual
matics self-concept, and (d) leads to better learning lesson plan. The only prerequisite for week 0 was that they did not
outcomes compared to standard feedback (main-feed- teach the subject of probability. In weeks 1–3 (intervention period), all
back-hypotheses). teachers taught the topic probability according to the mathematics
curriculum in 8th to 10th grade classrooms. The lesson plan during
H2. The effects on (a) attribution, (b) self-efficacy, this intervention period was generated for each grade level by the first
(c) mathematics self-concept, and (d) learning outcome author who is a researcher and experienced mathematics teacher. The
of re-attributional feedback over the course of 3 weeks lesson plan was composed of a mixture of front-of-class-teaching,
depend on students' initial mathematics performance. group work, partner work, or independent silent learning. The
Especially students with low initial mathematics teachers introduced the new topics to the students for about two
performance benefit from this computer-based, individ- thirds of the classroom time. For the remaining one third of classroom
ual re-attributional feedback (low-performance-modera- time as well as during homework (about 1–2 h per week) the students
tor-hypotheses). worked on their own with the mathematics learning software. There-
fore, most students worked during class twice a week about 60 min
with their teacher and 30 min on the computer. Re-attributional or
2 | METHODS standard feedback was given during about 150 min of independent
silent work per week (including homework), which would add up to a
2.1 | Participants and design total of 450 min. In Dresel and Haugwitz’ study (2008) students par-
ticipated in 5–9 sessions of 45 min, that is 225–405 min in total. The
2.1.1 | Participants authors found positive effects on students' motivation (attributions,
self-concept, and helplessness), and mathematical performance. Dre-
Participants were 322 students (mean age at start of data acquisition sel and Ziegler's (2006) six sessions of 60 min added up to 360 min
M = 14.02 years [n = 245 as not all participants reported their age]; total. Perry et al. (2010) already found positive results of re-
SD = 0.85 years; 58.5% female) attending 8th to 10th grade at the high- attributional training on academic performance after a 1-h-session.
est German school track (gymnasium; preparing for college) from 16 dif- These findings suggest that a total of about 450 min (150 min per
ferent classrooms. The schools were all situated in the same state of week) should be an adequate time frame for re-attributional training.
Germany and therefore had the same teaching curriculum. After the state Each student had their own computer and therefore was able to work
ministry of education, heads of schools, teachers, and all parents of par- at their own pace.
ticipating students gave their informed consent, data acquisition started. The mathematics learning software allows teachers to select exer-
cises from a large pool of tasks and group them into worksheets.
These worksheets as well as the lesson plan were predetermined by
2.1.2 | Experimental groups the research team. However, the teachers had the flexibility to adjust
the lesson plan and/or select extra worksheets in the mathematics
We randomly assigned half the students of each class to one of the program if their students needed more or less time than the
two experimental conditions: The 158 students in the standard researchers estimated.
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 WHALEN ET AL.

2.2.2 | Mathematics learning software There is no data available about the actual amount of feedback
sentences each student received in the re-attributional or in the
As digital mathematics learning software, we used the internationally standard feedback condition. Data on the Bettermarks server was
disseminated program Bettermarks. The students were able to work lost due to high usage of Bettermarks at the onset of the corona
on their tasks independently as the software gives feedback to pandemic.
whether a task or single steps of the task were solved correctly or not
(knowledge-of-result feedback). In addition, if the students were
not sure what to do, they could ask the program for extra hints. In 2.3 | Measures
case they did not solve the problem correctly after exhausting hints,
the software gave the correct answer, an explanation, and the correct 2.3.1 | Attribution
approach to the problem. An exercise in the learning software can
consist of up to three individual steps where each step leads closer to Attribution was measured eight times during the 4 weeks (t1–t8).
the required solution. The exercises were not timed. Table 1 shows an The means of the two measurements in week 0 (t1 and t2) were
example of a problem solved in three steps. taken to calculate a pretest variable as intervention started after t2.
Attribution was measured by eight items: attribution in case of suc-
cess on (1) effort, (2) attention, (3) ability, and (4) understanding;
2.2.3 | Experimental variation attribution in case of failure on (5) effort, (6) attention, (7) talent, and
(8) understanding (e.g., “If I was [not] able to solve the problems in
All participants received on-screen, standard feedback already imple- mathematics today, it is because I [did not] put a lot of effort in it.”/
mented in the mathematics learning software (e.g., “Well done!” or because I am [not] talented in math”). We used a 6-point answer
“This is not correct!”). The control group (standard feedback condi- scale from I completely agree (1) to I completely disagree (6). Attribu-
tion, SF) received a standard comment after every step. In addition to tion in case of failure on effort and attention is considered more
the standard feedback the experimental group (re-attributional feed- favourable than attribution on talent and understanding (Weiner
back condition, RF) received re-attributional feedback developed by et al., 1980). Effort and attention are variable factors which can be
psychologists (see Dresel & Ziegler, 2006). This re-attributional feed- changed more easily than talent and understanding—stable factors
back popped-up in a window after every third step and considered (Weiner, 1985). However, in case of success attribution on stable
the student's performance in these last steps. factors (here: ability and understanding) are considered more favour-
More specifically, for the re-attributional feedback condition able as they would stand for success in the long term. Variables mea-
(RF) we added 73 re-attributional feedback statements to the program suring variable factors and variables measuring stable factors were
Bettermarks. The feedback was designed to establish attributions to aggregated by calculating the means. As a result, the outcome vari-
ability in the case of “success” and attributions to effort in the case of able attribution consisted of four components: success_stable, suc-
“failure” (see Dresel & Haugwitz, 2008). According to the students' cess_variable, failure_stable, failure_variable; see Table 3.
success rate in the previous three exercise steps, they received a
feedback statement from one of four categories: “success”: in case of
more than 85% correct answers, “partial success”: 65% - 85% correct, 2.3.2 | Self-efficacy
“partial failure”: 45%–65% correct, “failure”: less than 45% correct
(see Table 2 for example statements). Out of each category, the pro- Self-efficacy was assessed with five items on an 11-point answer
gram selected feedback statements randomly. This feedback then was scale from 1: never (0%) to 11: in any case (100%): for example, “I am
provided in a pop-up window. Before data acquisition, we sensitized confident I would be able to solve at least one exercise of today's
all teachers to basic aspects of re-attributional feedback in a half-day topic without help.” The means of t1 and t2 of the self-efficacy mea-
workshop, so they would not counteract the computer-based surements were taken to calculate a pretest variable. The internal con-
feedback. sistency of the self-efficacy items over the eight measuring points
was in the range of Cronbach's α = 0.92–0.96.
T A B L E 1 . Example of a student's exercise in grade 8 with three
individual steps.

Tim rolls two dice. What is the probability 2.3.3 | Mathematics self-concept
Problem that he rolls a double?
Step 1 Choose out of the given sets the right event set. The first (t1) and last questionnaire (t8) of the four-week study period
(answer: E ¼ fð1, 1Þ; ð2, 2Þ; ð3, 3Þ; ð4, 4Þ; ð5, 5Þ; ð6, 6Þg) measured mathematics self-concept with four items (talent, ability,
Step 2 Calculate the cardinality of the sample space Ω. capability, understanding). For each item we used a 6-point answer
(answer: jΩj ¼ 6  6 ¼ 36) scale ranging from I think I am talented in math (1) I doubt that I am
Step 3 Calculate the probability PðEÞ. talented in math at all (6). The Cronbach's α for these four items at
(answer: PðEÞ ¼ jjΩEjj ¼ 36
6
¼ 16)
two measuring points were α = 0.89 (t1) and α = 0.86 (t8).
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WHALEN ET AL. 7

TABLE 2 Re-attributional feedback categories according to grading key in German schools (Trautmann, 2022).

Success Partial success Partial failure Failure


More than 85% correct 65%–85% correct 45%–65% correct Less than 45% correct
answers
For example, “You are an For example, “It is evident that you For example, “You had some correct For example, “Next time try and
expert in this topic!” understand most of the topic!” answers. If you work in a more take more time to write down
concentrated way, you will have each individual step. Then
better results!” surely you will be successful.”

TABLE 3 Dimension reduction to four components and their evaluation (see Weiner et al., 1980).

Component measuring
attribution style Description Measured variables Evaluation
success_stable Attribution in case of success to rather stable factors Ability and understanding in case of success Favourable
success_variable Attribution in case of success to rather variable factors Effort and attention in case of success Unfavourable
failure_stable Attribution in case of failure to rather stable factors Talent and understanding in case of failure Unfavourable
failure_variable Attribution in case of failure to rather variable factors Effort and attention in case of failure Favourable

2.3.4 | Learning outcome TABLE 4 Sample exercises from Test 1 in grades 8–10.

Grade 8 Tim rolls two dice. What is the probability that he


Every week the students filled out a mathematical performance test. rolls a double?
It assessed students' weekly learning outcomes using procedural as Grade 9 Two balls are drawn in succession from the urn
well as conceptual items (7–19 items, depending on complexity and shown without putting them back. For a
number of steps per items). Automatic sum scores and error rates corresponding tree diagram, first determine all
branch probabilities and then all path probabilities
were provided by the mathematics learning software Bettermarks. The
(the tree diagram was already given).
questions were different for each grade as the material for the tests
Grade 10 Ninety-nine students were asked what their
were chosen from each grade's curriculum. Every week the questions favourite subjects were. Thirty-five students
got more advanced as the lesson plan progressed. For this test the indicated astronomy, 44 literature, and 34 neither
program did not provide feedback neither in the SF nor in the RF con- astronomy nor literature. One student is randomly
dition. In contrast to the regular work sheets, cooperation among stu- selected. What is the probability that he/she
named astronomy but not literature?
dents or teacher's help was not possible. The questions for the first
test (Test 0) before the intervention period were taken from the cur-
riculum on the subject of probability from the school year before. For demographic data (age, gender, grade level, name of school, mathematics
all four performance tests (Tests 0–3) we measured the error rate grade in last report card). After week 0 the intervention started (weeks
ranging from everything correct (1) to no correct answer (0). As students 1–3): The students were introduced to the new mathematical topics of
from grades 8–10 had different performance tests, we z-standardized probability in normal classroom interaction. They strengthened their
the results for each grade separately to be able to better compare newly acquired skills (conceptual as well as procedural) with the mathe-
them. Table 4 shows exercises out of Test 1 from grades 8–10. Due matics learning software. The software provided students either with the
to the Bettermarks data loss, we could only calculate Cronbach's α for individualized re-attributional or the standard feedback only.
9 out of 12 tests for which we had at least n > 20 (mostly n > 100). To measure the outcome variables attribution and self-efficacy, the
We found Cronbach's α in the range of 0.56–0.74. In the case of students received a digital questionnaire twice a week which they filled
knowledge (learning outcome) measures, these values can be consid- out at home for about 5 min each (eight measurements: t1–t8)—ideally
ered still to be fully acceptable (Field, 2018; Taber, 2018). on the same day of mathematics instruction in school. Mathematics self-
concept was measured in the beginning and in the end of the data
acquisition (t1 and t8). To measure learning outcomes, at the end of each
2.4 | Procedure week the students were given a mathematics performance test in class
which was integrated in the mathematics learning software (Tests 0–3).
We conducted the study over the period of 4 weeks during which Ger- After data collection, the students who received the standard
man students (grades 8–10) worked with a mathematics learning soft- feedback only (SF condition) were able to benefit from possible posi-
ware. Over 4 weeks we collected students' data during and after their tive effects of the re-attributional feedback as all students of the class
classroom time. The first week (week 0) was used for baseline measure- received the re-attributional feedback for eight to 12 weeks (wait-list
ments (t1 and t2). In the first measurement (t1), we furthermore collected control group design).
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 WHALEN ET AL.

2.5 | Data analysis consists of 15–30 students and the number of contexts relative to individ-
ual students within those contexts is important (Twisk, 2006), we could
An alpha level of 0.05 was used for all statistical analyses. argue that 16 contexts (classrooms) would be enough.

2.5.1 | Data verification 2.5.3 | Testing different models

As the students were asked to complete numerous questionnaires To test whether a two- or a three-level model would be more accu-
throughout the study, we wanted to make sure not to include ques- rate we used the χ2-likelihood ratio test for each dependent variable.
tionnaires, where the students most likely just “clicked through” the After determining whether a two or three level model best fit the
answers without reading them. Therefore, all questionnaires which were data, we tested for differences between experimental conditions in
completed in under 70 s were not considered as students would not the development over time as well as in the end (t8) by entering the
have been able to read the questions and answer them properly in that experimental condition (SF and RF) on level 2 (student level; all
short amount of time. Furthermore, we conducted a factor analysis of hypotheses).
the eight items measuring self-efficacy and mathematics self-concept in We coded the time variable from 6 to 0 for the seven measuring
the last questionnaire (t8) to check whether the students still filled out points used to measure attribution and self-efficacy. The time variable
the last questionnaire properly. The Kaiser-Meyer-Olkin measure verified was coded from 3 to 0 for the four measuring points used to mea-
the sampling adequacy for this analysis, KMO = 0.888, well above the sure mathematical performance and from 1 to 0 for mathematics
minimum criterium of 0.5 (Field, 2018). Two factors had eigenvalues >1 self-concept. Using this coding, the intercept represents the measur-
(Kaiser's criterion [Kaiser, 1960]) and combined explained 77.6% of the ing point at the end of the experimental variation, thus differences
variance. As expected, factor analysis loaded the five items measuring between experimental groups at t8 are tested.
self-efficacy and the four items measuring mathematics self-concept Regarding our main-feedback-hypotheses (H1a–d), we tested
onto those two separate factors (Bartlett's measure: p < 0.001, rotated effects (betas) of received feedback (SF and RF) on the students' attribu-
factor loadings >0.747). In addition, to strengthen these findings, we ran tion, self-efficacy, and mathematical performance at the end of the
a factor analysis for the eight items measuring attribution in t8 4-week-period (treatment effect, β0,1) and on the development over
(KMO = 0.617). Four factors had eigenvalues >1 and combined time (interaction of growth and treatment, β1,1) in separate models. All
explained 86.3% of the variance. The eight items were loaded as other betas are less relevant, as a significant effect in growth rate (slope,
expected onto four factors (success_stable, success_variable, failure_st- β1,0) would only mean that students behaviour changed over time (posi-
able, failure_variable, Bartlett's measure: p < 0.001, rotated factor load- tive values stand for increase, negative for decrease). This change would
ings >0.335). These findings suggests that despite the high number of be independent of feedback received and therefore, would not answer
questionnaires the data obtained in the last questionnaire is reasonable. our research questions. The same is true for significant differences in
intercept (β0,0) which would only indicate that students on the individual
level are different from each other at the end of data acquisition (t8).
2.5.2 | Multilevel analysis To test our moderator hypotheses (H2a–d, low-performance-
moderator-hypotheses), we added another predictor (moderator)
To test our hypotheses, we carried out multilevel analyses using IBM SPSS variable at level 2, namely initial mathematical performance in
Statistics (version 27). On a first level, we had repeated measurements week 0 (Test_0). Here, we tested the following two betas: First,
within students to measure development over time (for all hypotheses). the estimated mean moderation effect on the interaction of treat-
As we took the means of t1 and t2 for the attribution and self-efficacy ment and growth rate (β1,3 ) indicates whether students' initial
variables, we had seven measuring points for those variables. There were mathematical performance has a moderating effect on the
four measuring points for mathematical performance (one each week) and interaction of feedback received (experimental condition) and
two for mathematics self-concept (one before and one after intervention). development over time (growth). This can answer the question
The individual students formed the second level. As those students were whether especially low-performing students benefit from the
nested in three different grade levels (grades 8–10) and 16 different class- computer-based, individual re-attributional feedback and have
rooms, we tested whether our model improves by adding a third level better results in attribution and self-efficacy over time. Second,
(grade level or classroom). If a third level is introduced, the model must the estimated mean moderation on treatment effect (β0,3 )
estimate more parameters and a larger sample size is needed (Field, 2018). answers the question whether especially low-performing stu-
Three hundred twenty-two students on level 2 are a large enough sample dents benefit from the re-attributional feedback regarding attri-
size. However, three different grade levels on level 3 would not be consid- bution and self-efficacy at the end of the 4-week-period. The
ered enough contexts on the higher-level variable (Kreft & de other betas as for example the estimated mean moderation effect
Leeuw, 1998). In addition, the calculated variance of grade level was negli- (β0,2 ) is independent of the experimental condition and would
gible (<1% of variance explained by level 3; χ2change = 0.0). Therefore, we only indicate if students' initial mathematical performance would
decided to test classroom affiliation for level 3 instead. As each classroom influence attribution and self-efficacy at the end of intervention.
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WHALEN ET AL. 9

Multilevel models consist of regression equations nested on each attribution in week 0 did not differ between the two groups: suc-
level. Therefore, betas can be interpreted in an analogous way. Note, cess_stable, F(1,240) = 0.420, p = 0.518, success_variable, F
that betas cannot be compared across different models. Reporting (1,240) = 2.632, p = 0.106, failure_variable, F(1,240) = 0.022,
effect size for multilevel models is not as straightforward as for linear p = 0.883, and failure_stable, F(1,240) = 0.328, p = 0.567.
models. Lorah (2018) demonstrated that the use of intraclass correla- Furthermore, there were no significant differences in self-efficacy
tion coefficient (ICC) for random effects and f2 or R2 for fixed effects before the intervention, F(1,240) = 1.829, p = 0.178, and mathe-
are appropriate effect size measures. Therefore, we report ICC as matics self-concept, F(1,226) = 0.526, p = 0.469. That is, the two
effect size which can be used for two- as well as three-level models groups were comparable before the intervention, regarding age,
(Snijders & Bosker, 2012). The ICC can be seen as a proportion of var- gender, attribution, self-efficacy, self-concept, and mathematical
iance in the outcome accounted for by level 3 (class) or level 2 performance.
(student) membership (Kirk, 2013; Snijders & Bosker, 2012). For fixed
effects we report f2 or R2 with f2 = 0.02 indicating a small, 0.15 a
medium, and 0.35 a large effect (Cohen, 1992). 3.1.2 | Finding the best-fitting model

With the help of multilevel analyses, we controlled for classroom affil-


3 | RESULTS iation (level 3) as this was the more appropriate level-3-variable than
grade level. Furthermore, we controlled for individual differences
3.1 | Preliminary analyses (level 2). Level 1 represented the development of students over time
(seven measuring points). In a first step, we decided whether a two-
3.1.1 | Comparing treatment groups before or a three-level model was more fitting. The decision to use a three-
intervention level-model was based on the χ2-likelihood ratio test with the critical
value of 3.84 for 1 df and 5.99 for 2 df (p < 0.05, Field, 2018): A
The experimental groups did not differ significantly in terms of age, F three-level model was a better fit for the following variables: attribu-
(1,243) = 0.182, p = 0.670, grade level, χ2 (2) = 0.117, p = 0.943, tion on variable factors in case of success (success_variable), in case
or gender, χ 2(2) = 1.199, p = 0.549. The four variables measuring of failure (failure_variable), and mathematical performance. For all

TABLE 5 Testing two- against three-level models for all outcome variables with multiple measurements.

Variable on Variance explained in Variance explained in


attribution style two-level model (%) three-level model (%) χ2change (dfchange)
success_stable a
Level 1: 16.9 Level 1: 16.9 2523.0–2523.0 = 0.0 (7–6 = 1)
Level 2: 85.2 Level 2: 85.2
Level 3: <0.1
success_variablea Level 1: 35.5 Level 1: 37.3 2611.2–2605.4 = 5.8* (1)
Level 2: 64.5 Level 2: 58.9
Level 3: 4.8
failure_stablea Level 1: 22.9 Level 1: 22.9 3025.4–3025.4 = 0.0 (1)
Level 2: 77.1 Level 2: 77.1
Level 3: <0.1
failure_variablea Level 1: 33.1 Level 1: 24.6 3232.6–3228.3 = 4.3* (1)
Level 2: 66.9 Level 2: 72.4
Level 3: 3.0
self-efficacya Level 1: 34.5 Level 1: 34.5 5171.2–5171.2 = 0.0 (1)
Level 2: 65.5 Level 2: 65.5
Level 3: <0.1
Mathematics Level 1: 18.1 Level 1: 21.1 1053.7–1053.7 = 0.0 (1)
Self-conceptb Level 2: 81.9 Level 2: 78.9
Level 3: <0.1
Mathematical performancec Level 1: 42.8 Level 1: 53.3 2548.7–2513.3 = 35.4* (1)
Level 2: 57.2 Level 2: 33.8
Level 3: 13.0
a
Seven measuring points.
b
Two measuring points.
c
Four measuring points.
*p < 0.05: a three-level model is better fitting.
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 WHALEN ET AL.

other variables, the two-level-models showed the better fits. For more

6.82 (2.25)
7.54 (2.20)
6.94 (2.07)
6.93 (2.35)
6.63 (2.37)
6.14 (2.30)
6.74 (2.53)
details see Table 5.
In a second step, we tested whether the development over

SF
time was linear (linear growth or decrease over time) or quadratic
(a U-shaped relationship in growth or decrease). The linear time

Self-efficacyb
variable had a better fit than the quadratic time variable in all

7.20 (2.14)
7.43 (2.15)
7.27 (2.21)
7.53 (1.99)
6.75 (2.44)
7.07 (2.26)
7.18 (2.22)
models.

RF

Abbreviations: RF, re-attributional feedback condition; SF, standard feedback/control condition (high values stand for favourable attribution style variables and high self-efficacy).
3.2 | Effects of different feedback-groups (main-

3.18 (1.20)
3.03 (1.24)
3.01 (1.28)
3.14 (1.29)
2.97 (1.16)
3.25 (1.30)
3.10 (1.37)
feedback-hypothesis H1)

Means for attribution style and self-efficacy (standard deviations in parentheses) for each feedback-group at each of the eight measuring points.

SF
3.2.1 | Change in attribution and self-efficacy

Failure_variablea
To test our main-feedback-hypothesis (H1), we added the indepen-

3.21 (1.30)
2.95 (1.35)
3.29 (1.42)
2.87 (1.41)
3.32 (1.29)
3.28 (1.40)
3.26 (1.32)
dent variable (standard feedback condition SF = 0 or re-
attributional feedback condition RF = 1) as a predictor to our

RF
multi-level model to investigate the impact of different forms of
feedback on change of attribution and self-efficacy. Our findings

3.76 (1.49)
4.01 (1.39)
3.66 (1.40)
3.56 (1.28)
3.79 (1.24)
3.69 (1.32)
3.85 (1.41)
indicate feedback effects under certain conditions. Table 6 gives
an overview of the descriptive statistics including the means and

SF
standard deviations of attribution and self-efficacy at all eight
measuring points. Table 7 provides the results of the hierarchical
analysis. We only report betas which answer our research ques- Failure_stablea

3.87 (1.24)
3.99 (1.24)
3.84 (1.30)
4.15 (1.03)
3.92 (1.22)
4.27 (1.02)
3.97 (1.24)
tions: treatment effect on the students' attribution style and self-
efficacy at the end of the 4-week-period (β0,1) or during the
RF

4 weeks (development over time: interaction of growth and treat-


ment, β1,1).
2.60 (0.85)
2.38 (0.82)
2.33 (0.76)
2.38 (0.88)
2.21 (0.88)
2.44 (0.88)
2.39 (0.93)
We expected an improvement of attribution and self-efficacy
in the RF group in comparison to the SF group over time and a sig-
SF

nificant difference between the two groups at the end of interven-


Success_variablea

tion (Hypotheses H1a,b). A significant effect of received re-


attributional feedback on the improvement over time was found
2.42 (0.90)
2.36 (0.84)
2.36 (1.03)
2.25 (0.83)
2.16 (0.76)
2.29 (1.00)
2.33 (0.93)
for attribution in case of success on stable factors (β1,1; Level
2 ICC = 0.831; f2 = 0.002). Students in the experimental condition
RF

(RF) tended to attribute more towards stable factors in case of suc-


cess over the course of intervention than students in the control
4.10 (1.20)
4.03 (1.16)
4.06 (1.11)
3.87 (1.10)
3.95 (1.16)
3.80 (1.12)
3.99 (1.15)

condition (SF). For attribution variables in case of failure and self-


efficacy, students in the experimental condition (RF) show more
favourable attribution and higher self-efficacy scores over time
SF

Range from 1 (never, 0%) to 11 (in any case, 100%).

than students in the control condition (SF; positive parameters for


Success_stablea

Range from 1 to 6 (most favourable attribution).

β1,1 in Table 7). However, these improvements are not statistically


4.00 (1.14)
3.96 (1.20)
3.82 (1.17)
3.97 (0.97)
4.15 (0.95)
4.11 (1.07)
4.02 (1.18)

significant.
RF

3.2.2 | Change of mathematics self-concept and


learning outcome
Pre (mean of t1 and t2)

We expected an improvement in mathematics self-concept in the re-


attributional feedback condition (Hypothesis H1c). However, the cal-
TABLE 6

culated two-level hierarchical model revealed no difference between


Effect

intervention groups ( p = 0.237, see Table 8). Furthermore, we


t3
t4
t5
t6
t7
t8

expected an improvement in learning outcomes (Hypothesis H1d).


b
a
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WHALEN ET AL. 11

Note, that learning outcomes (mathematical performance) were mea-

F(1,261) = 2.324;

F(1,232) = 0.188;
sured with different tests for each grade level. To find treatment

p = 0.129

p = 0.665
group differences in learning outcomes, we used a three-level model.
When adding our independent variable (experimental condition SF or
self-efficacya (SD) RF) we did not find any influence of received feedback type on the
0.434 (0.278) change of learning outcome over time ( p = 0.789, Table 8). Our

0.019 (0.044)
hypotheses (H1c,d) could hence not be supported.

3.3 | Effects on low-performing students (low-


F(1,268) = 0.346;

F(1,229) = 0.801;

performing-moderator-hypotheses H2)
p = 0.557

p = 0.372
Estimation of parameters in multi-level linear growth model for attribution style and self-efficacy with feedback-group affiliation as predictor.

To test our moderator hypotheses that especially low-performing stu-


failure_variableb (SD)

dents benefit from the computer-based, individual re-attributional


feedback, we added initial mathematical performance (Test_0) as a
0.096 (0.163)

0.024 (0.027)

possible moderator variable to our model. The results are displayed in


Table 9. Only betas β0,2, β0,3, and β1,3 are shown, as we concentrate
on the moderation effects. We found for almost all measured vari-
ables (but failure_variable) a significant effect of mathematical perfor-
F(1,281) = 1.128;

F(1,215) = 0.000;

mance before intervention: better performing students tend to


p = 0.289

p = 0.998

attribute more favourably towards stable factors (more in case of suc-


cess and less in case of failure) than students with a lower perfor-
failure_stablea (SD)

mance. Only in case of success, better-performing students attribute


more towards variable factors which is considered unfavourable (neg-
0.162 (0.153)

0.000 (0.022)

ative parameter for β0,2 in Table 9). Better-performing students also


rate higher on self-efficacy and mathematics self-concept at the end
of intervention (β0,2 in Table 9). This was not surprising as we
expected low-performing students to have less favourable attribution
F(1,231) = 0.596;

F(1,189) = 0.016;

and score lower on self-efficacy and mathematics self-concept. We


p = 0.441

p = 0.899

wanted to find out whether the individual, re-attributional feedback


would be able to help low-performing students to overcome this
success_variableb (SD)

disadvantage over time (Hypotheses H2a–d). Therefore, we were


especially interested in the estimated mean moderation effect on the
0.086 (0.111)

0.002 (0.019)

interaction of treatment and growth rate (β1,3 in Table 9). We did find
a significant interaction effect (p = 0.011 in Table 9) for attribution
towards rather stable factors in case of success (Level 2: ICC = 0.798;
f2 = 0.157). In case of failure and for self-efficacy, this benefit could
Note: t8 is the last measuring point at the end of data acquisition.
F(1,287) = 0.773;

F(1,210) = 5.639;

not be found (β1,3 in Table 9). In addition, when only looking at the
end of intervention no moderation of initial mathematical perfor-
p = 0.380

p = 0.018

mance on treatment effect could be found (β0,3 in Table 9). No other


moderation effect was significant.
success_stablea (SD)

To better interpret this interaction, we performed a simple


Calculated with a 3-level hierarchical model.
Calculated with a 2-level hierarchical model.
0.040* (0.017)
0.120 (0.137)

slopes analysis to being able to visualize the effect. Figure 1 provides


this visualization (please note, that the exact values should not be
Abbreviation: SD, standard deviation.

over interpreted as the focus lies here on the direction of the inter-
action). The horizontal axis represents time (measurements at t1 and
treatment effect (β0,1)

t2 were averaged for the baseline measurement). The vertical axis


on growth rate (β1,1)

represents scores for attribution on stable factors in case of success.


treatment effect
Estimated mean

Estimated mean

Higher values represent more favourable attribution. The four lines


represent better-performing (+1 SD, darker lines) and lower-
TABLE 7

performing students (1 SD, lighter lines) in the Standard Feedback


*p < 0.05.
Effect

(SF, dashed lines) and Re-attributional Feedback (RF, solid lines)


condition.
b
a
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 WHALEN ET AL.

T A B L E 8 Means (standard deviations in parentheses) for mathematics self-concept and for learning outcome for each feedback-group and
estimation of parameters in multi-level linear growth model with feedback-group affiliation as predictor.

Math. self-concepta (SD)

RF SF Effect Math. self-conceptb (SD)


Pre (t1) 3.75 (1.26) 3.63 (1.33) Estimated mean treatment effect (β0,1) 0.239 (0.186) F(1,170) = 1.657; p = 0.200
t8 3.73 (1.04) 3.57 (1.31) Estimated mean treatment effect on growth rate (β1,1) 0.180 (0.151) F(1,117) = 1.416; p = 0.237
c
Learning outcome (SD)

RF SF Effect Learning outcomed (SD)


Pre (Test 0.06 0.08 Estimated mean treatment effect (β0,1) 0.041 F(1,180) = 0.138;
0) (1.03) (1.01) (0.110) p = 0.712
Test 1 0.02 0.08 Estimated mean treatment effect on growth rate 0.013 F(1,253) = 0.072;
(0.99) (1.02) (β1,1) (0.048) p = 0.789
Test 2 0.05 0.05 (0.95)
(1.01)
Test 3 0.10 0.15
(1.08) (1.03)

Abbreviations: RF, re-attributional feedback condition; SD, standard deviation; SF, standard feedback/control condition; t1, first measurement at the
beginning of intervention; t8, last measurement at the end of intervention.
a
Range from 1 to 6 (low values stand for high self-concept).
c
Students in grades 8–10 completed different performance tests, therefore, results are z-standardized for each grade separately.
b
Calculated with a 2-level hierarchical model.
d
Calculated with a 3-level hierarchical model.
*p < 0.05.

The visualization in Figure 1 shows that better-performing stu- of success (H2a). They did not show a benefit on attribution in case of
dents independent of feedback received, attributed in general more failure, nor on self-efficacy and mathematics self-concept (H2b–d).
favourably than lower-performing students on stable factors in case
of success (the two darker grey lines lay at all measurement points
above the two lighter grey lines; p < 0.001 in Table 9). For better- 4 | DI SCU SSION
performing students the type of feedback received did not seem to
have an influence on attribution (the solid and dashed dark grey lines The present study tested whether re-attributional feedback added to
are almost identical). Lower-performing students in the SF group over an existing mathematics learning software which was used in regular
time showed a less favourable attribution (declining dashed light grey classroom instruction leads to a change in attribution, and fosters self-
line). At the same time, the lower-performing students who received efficacy, mathematics self-concept, and learning outcome compared
re-attributional feedback did not show this drop. Therefore, the signif- to the software's standard feedback. Moreover, we investigated
icant interaction can be interpreted as follows: at the end of interven- whether re-attributional feedback exerts a positive effect especially
tion (t8) the lower-performing students in the RF group show more for lower-performing students. In contrast to already existing studies,
favourable attribution in case of success on stable factors (higher we integrated the intervention not only in regular classroom instruc-
values on the vertical axis) than the SF group. This is in line with our tion but also into a fully developed lesson plan which was provided to
hypothesis that especially low-performing students benefit from the the teachers. This design, therefore, has a high external validity while
individual, re-attributional feedback in case of success. still holding up internal validity using standardization in procedure and
In summary, in line with our hypothesis (H1a), students who implementing a randomized controlled trial within the limits of class-
received re-attributional feedback in case of success seemed to have room affiliation. In addition, it can easily be adopted for future use in
increasingly attributed towards stable factors as ability and under- other schools or other subjects as the digital re-attributional feedback
standing over the 3 weeks of intervention (more favourable). In case can be integrated in any other digital learning program.
of failure there was no significant change in attribution. We could not We found positive effects of computer-based re-attributional
find significant support for our hypothesis that over the course of feedback on attribution in case of successfully solving mathematical
4 weeks re-attributional feedback improves self-efficacy (H1b), math- tasks. Students more often attributed towards stable factors when
ematics self-concept (H1c) or learning outcome (H1d). Lower- successfully solving a task. However, in case of failure no effects
performing students seemed to have benefited from computer-based could be found. Therefore, the results of this experimental study over
re-attributional feedback leading—over time—to a more favourable the course of 4 weeks only partly confirmed results from previous
attribution towards stable factors (ability and understanding) in case studies outside the regular classroom (Dresel & Ziegler, 2006) and
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WHALEN ET AL. 13

inside the classroom (Dresel & Haugwitz, 2008). We could not find a

F(1,323) = 19.867;
T A B L E 9 Estimation of parameters in multi-level model for outcome variables on attribution, self-efficacy and mathematics self-concept and effect of feedback-group over time moderated by

F(1,323) = 1.472;

F(1,323) = 0.592;
significant effect on self-efficacy, mathematics self-concept and learn-

p = 0.226

p = 0.442
p < 0.001
ing outcome. Therefore, only one of the main-feedback-

Math. self-concepta (SD)


hypotheses (H1a–d) can be accepted for attribution in case of success
(H1a) and rejected in the other cases.

Abbreviations: SD, standard deviation; β0,2, estimated mean moderation effect; β0,3, estimated mean moderation on treatment effect; β1,3, estimated mean moderation effect on the interaction of treatment and growth rate.
Taking the students' mathematical performance before interven-

(0.133)

(0.190)

(0.162)
0.230

0.125
0.592 tion into account, we found a more detailed picture. First, we found
that lower-performing students attributed more towards stable fac-
tors after receiving computer-based re-attributional feedback when
F(1,218) = 32.941;

F(1,219) = 0.024;
F(1,227) = 3.170;

solving a mathematical task correctly than lower-performing students


p = 0.076

p = 0.878
p < 0.001

who received standard feedback only. Lower-performing students in


the standard-feedback-group attributed over time less favourably,
whereas the lower-performing students in the intervention group did
Self-efficacya (SD)

not show this drop (Figure 1). These results are in line with the previ-
1.106 ** (0.193)

ously discussed findings of Hamm et al. (2017) in first year college stu-
(0.265)

(0.049)
0.471

0.007

dents where lower-performing students who received attributional


training outperformed lower-performing students in the control group
in a class test after treatment when controlling for initial performance.
F(1,251) = 0.093;

F(1,248) = 0.103;

F(1,216) = 0.012;

This finding suggests, that over longer learning periods (here 3 weeks),
p = 0.760

p = 0.749

p = 0.913

re-attributional feedback might help lower-performing students to


Failure_variableb (SD)

keep up more favourable attribution even if learning contents get


more challenging (H2a). Second, better-performing students in the re-
(0.124)

(0.028)

attributional feedback group did not differ in attribution after success


0.038

0.003

from better-performing students in the control group. It seemed that


j

the impact of re-attributional feedback was negligible in better-


F(1,246) = 16.138;

performing students. In addition, Hamm et al. (2017) could not find


F(1,246) = 0.013;

F(1,200) = 0.004;

treatment effects on higher-performing students. A possible explana-


p = 0.910

p = 0.951
p < 0.001

tion might be that better-performing students attributed success more


towards stable factors such as ability and understanding than lower-
Failure_stablea (SD)

performing students independent of feedback received (see Figure 1


0.442** (0.110)

and Table 9). These findings are in line with previous studies who
0.001 (0.022)

showed that students with less favourable attribution performed


poorly in school (Akey, 2006; Houston, 2016).
j

Furthermore, we did not find differential effects depending on ini-


F(1,212) = 10.654;

F(1,213) = j3.369;

tial mathematical performance on other outcome variables. Therefore,


F(1,178) = 2.270;

of the low-performance-moderator-hypotheses (H2a–d), only


p = 0.001

p = 0.068

p = 0.134
Success_variableb (SD)

Hypothesis H2a can be partly accepted and the questions remains


how lower-performing students can benefit more comprehensively
from computer-based re-attributional feedback.
(0.084)

(0.115)

(0.020)
0.276**

As we did not find strong effects, we carried out exploratory


0.211

0.030

analyses to search for other potential differential effects. There-


fore, we investigated if students' initial mathematics self-concept,
F(1,257) = 30.119;

F(1,254) = 2.790;

F(1,200) = 6.577;

self-efficacy, or attribution style moderated the intervention's


mathematical performance in week 0.

Calculated with a 3-level hierarchical model.


Calculated with a 2-level hierarchical model.
p = 0.096

p = 0.011

effects on the different outcome variables. Neither of the multi-


p < 0.001

level analyses resulted in significant effects (attribution variables as


moderators (success_stable, success_variable, failure_stable, failur-
Success_stablea (SD)

e_variable): for all p > 0.104; self-efficacy as moderator: p > 0.090;


0.528** (0.096)

mathematics self-concept as moderator: p > 0.257 for all outcome


**p < 0.001; *p < 0.05.
(0.133)

(0.017)
0.043*

variables). Only a moderator effect for initial mathematical perfor-


0.222

mance could be found as discussed above. Further research in


needed to further investigate why re-attributional feedback has a
Effect

different impact on students with different cognitive or motiva-


β0,2

β0,3

β1,3

tional dispositions.
b
a
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14 WHALEN ET AL.

Attribution towards stable the student's mindset has an influence on student's self-regulated
factors in case of success
6.0 learning (Karlen et al., 2021). Individualized re-attributional feedback
could be more helpful when combined with learning strategy
5.0 feedback and/or mindset growth in the future.
A further point important to consider in interpretation of the
4.0 results and in future research is that the standard feedback might
have been relatively strong. Students in both feedback conditions
received the standard feedback, that is task-related feedback on how
3.0
to solve the problem: a knowledge of result (correct/incorrect) feed-
Effect of RF on better-performing students (at +1 SD)
back, hints to help with each step, and an explanation and model
2.0
Effect of SF on better-performing students (at +1 SD) answer. The correct/incorrect feedback in Bettermarks is graphically
Effect of SF on lower-performing students (at -1 SD) (green/red icons) represented during and after working on tasks. It is
1.0
possible that this task-related feedback may have been quite effective
Effect of RF on lower-performing students (at -1 SD)
and/or drawn attention from the re-attributional feedback. Thus, the
0.0
t1/t2 t3 t4 t5 t6 t7 t8 standard feedback may have diminished the extra effect of re-
Measurement points (t1/t2 = means of t1 and t2) attributional feedback. Students' comments underline this possibility,
as several students stated in the comment section that they were
F I G U R E 1 Visualization and descriptive interpretation of the
more focused on the knowledge-of-result feedback and standard
significant interaction effect of initial mathematical performance by
group on attribution to stable factors in case of success: The two feedback than on the re-attributional feedback. Furthermore, several
darker grey lines represent the better-performing students at +1 SD students reported that the computer-based feedback was not diverse
of mathematical performance before intervention in the re- enough and repetitive. Thus, we recommend adaptations to the spe-
attributional feedback (RF, solid line) and standard feedback (SF, cific implementation and design of the computer-based re-
dashed line) condition. The two lighter grey lines represent the lower-
attributional feedback for future research. For example, further
performing students (1 SD). The plots demonstrate that in general
research could investigate if leaving out or reducing the standard
better-performing students attributed more favourably than lower-
performing students in case of success but whereas lower-performing feedback might draw more focus on the re-attributional feedback
students in the SF group attributed less favourably over time itself (see effects at Kuepper-Tetzel & Gardner, 2021). To increase
(declining line, light grey dashed line), the lower-performing students diversity and to add a more personal note, digital re-attributional feed-
in the RF group did not show this drop (light grey solid line). High back could be combined with teacher-given re-attributional feedback.
values on the vertical axis stand for more favourable attribution.
Further research is needed to find the best conditions for effective re-
attributional feedback that enhances student motivation and learning.
4.1 | Limitations and further research Favourable attribution has been shown to positively affect self-
efficacy (Zeichner, 2018) and self-efficacy to positively affect learning
In a field experiment many more factors which cannot easily be con- outcomes (Bandura, 1993; Schunk & Ertmer, 1999). Attribution in
trolled play an important role. This might explain to some extent, why case of failure might be more or as important than attribution in case
in the present field study only some results of previous studies exe- of success to influence motivational aspects and learning outcomes.
cuted outside the regular classroom in a more controlled setting could Therefore, it could be argued that because in our study only a small
be replicated (Dresel & Ziegler, 2006; Hamm et al., 2017; Perry effect of re-attributional feedback on attribution could be found (only
et al., 2010). Not only density and immediacy of re-attributional feed- in case of success), this change in attribution was not large enough to
back could be important (Dresel & Ziegler, 2006) but also the classroom also affect self-efficacy. The same argumentation could be used
climate, teacher's behaviour, feedback given by other students, or how to explain why no effect was found on mathematics self-concept and
failure is generally treated in this class (Maxwell et al., 2017). However, learning outcome. However, as explained above, we did not find any
as we controlled in the linear multilevel model for classroom affiliation, moderation effect of initial attribution style on the intervention
where indicated, effects of classroom affiliation should be minor. effects on any outcome variable in exploratory analyses. With the
Furthermore, an obstacle with using computer-based feedback above discussed possible adaptations, further research could investi-
attributing failure to insufficient effort is the following: Focusing on gate how lower-performing students can benefit more from digital or
more effort in case of failure could be seen as demotivational as stu- personal re-attributional feedback, why re-attributional feedback has
dents often see “more effort” as “working hard” and do not take all a different impact on different students, and why only in case of suc-
other cognitive and metacognitive activities into account. One student cess an intervention effect has been found.
even reported at the end of the study that the comments at the end
of the exercise were “demotivating and did not fit”, whereas other
students reported the exact opposite. This suggests that the same re- 5 | CONC LU SION
attributional feedback could have a different impact on different stu-
dents. Often, students spend a lot of time (effort) studying but not The results obtained in the present study give useful information for
using optimal learning strategies (Dunlosky et al., 2013). Furthermore, researchers and educators: First, we could show that computer-based
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WHALEN ET AL. 15

re-attributional training in the current form showed some positive request. Furthermore, the datasets will be available in the future at
effect in case of success in regular classroom instruction, an experi- Whalen, et al. (2022). http://doi.org/10.5159/IQB_DigiFeed_v1.
mentally far less controllable context than a lab setting. Second, this
study can serve as an example how computer-based re-attributional ET HICS S TAT E MENT
feedback can be integrated in normal classroom activities and can be The Department of Education of the state of Baden-Württemberg,
a groundwork for further re-attributional feedback in schools. How- Germany approved the study.
ever, given the very limited effects found, the specific contents and
modes of the re-attributional feedback delivery must be reconsid- OR CID
ered (e.g., variability, combination with other feedback). Third, the Katharina Alexandra Whalen https://orcid.org/0000-0003-4173-
study demonstrated that better-performing students show more 3959
favourable attribution in case of success than lower-performing Alexander Renkl https://orcid.org/0000-0001-7971-0619
students. Especially low-performing students benefited from re- Alexander Eitel https://orcid.org/0000-0001-6403-3467
attributional feedback regarding attributions in case of success. Inga Glogger-Frey https://orcid.org/0000-0002-1409-2116
This suggests that teachers should have an eye on lower-
performing students especially and give them re-attributional RE FE RE NCE S
feedback of a high density. Akey, T. (2006). School context, student attitudes and behavior, and aca-
demic achievement: An exploratory analysis. Manpower Demonstration
Research Corporation.
AUTHOR CONTRIBUTIONS
Aronson, J. (2008). Improving academic achievement: Impact of psychological
Katharina Alexandra Whalen: Conceptualization; data curation; formal factors on education. Emerald.
analysis; investigation; methodology; resources; software; supervision; Bandura, A. (1993). Perceived self-efficacy in cognitive development and
validation; visualization; writing – original draft; writing – review and functioning. Educational Psychologist, 28(2), 117–148.
Banks, M., & Woolfson, L. (2008). Research section: Why do students
editing. Inga Glogger-Frey: Conceptualization; formal analysis; funding
think they fail? The relationship between attributions and academic
acquisition; investigation; methodology; project administration; self-perceptions. British Journal of Special Education, 35(1), 49–56.
resources; software; supervision; validation; writing – review and edit- Berkeley, S., Mastropieri, M., & Scruggs, T. (2011). Reading comprehension
ing. Alexander Eitel: Conceptualization; methodology; resources; strategies for secondary students with learning & other mild disabil-
ities. Journal of Learning Disabilities, 44(1), 18–32.
writing – review and editing. Alexander Renkl: Conceptualization; pro-
Bong, M., & Skaalvik, E. M. (2003). Academic self-concept and self-efficacy:
ject administration; supervision; validation; writing – review and editing. How different are they really? Educational Psychology Review, 15(1), 1–40.
Borkowski, J. G., Weyhing, R. S., & Carr, M. (1988). Effects of attributional
ACKNOWLEDGEMEN TS retraining on strategy-based reading comprehension in learning-
The authors thank all participating teachers for their support and fruit- disabled students. Journal of Educational Psychology, 80(1), 46–53.
Casserly, A. M. (2013). The socio-emotional needs of children with dys-
ful cooperation. Furthermore, we would like to thank all research assis-
lexia in different educational settings in Ireland. Journal of Research in
tants in alphabetical order: Lisa Flick, Meta Gütgemann, Lars Link, Special Educational Needs, 13(1), 79–91. https://doi.org/10.1111/j.
David Rommelspacher, Paul Sölder, and Michael Wonnerth. Without 1471-3802.2011.01227.x
their help, hard work and commitment, this study would not have been Chodkiewicz, A. R., & Boyle, C. (2014). Exploring the contribution of attri-
bution retraining to student perceptions and the learning process. Edu-
possible. We thank Bettermarks for providing the software and being
cational Psychology in Practice, 30(1), 78–87. https://doi.org/10.1080/
always available for our questions. We thank Michael Wiedmann from 02667363.2014.880048
the Robert Bosch Stiftung, Germany, for all his help and support. Open Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155–159.
Access funding enabled and organized by Projekt DEAL. Craven, R. G., Marsh, H. W., & Debus, R. L. (1991). Effects of internally
focused feedback and attributional feedback on enhancement of aca-
demic self-concept. Journal of Educational Psychology, 83, 17–27.
FUNDING STATEMENT Dresel, M., & Haugwitz, M. (2008). A computer-based approach to foster-
This work was funded by Robert Bosch Stiftung, Germany ing motivation and self-regulated learning. Journal of Experimental Edu-
(00906424-002). cation, 77(1), 3–18.
Dresel, M., & Ziegler, A. (2006). Langfristige Förderung von
Fähigkeitsselbstkonzept und impliziter Fähigkeitstheorie durch
CONF LICT OF IN TE RE ST ST AT E MENT computer-basiertes attributionales Feedback. Zeitschrift für Pädago-
The authors declare no conflict of interest. gische Psychologie, 20(1/2), 49–63.
Dunlosky, J., Rawson, K. A., Marsh, E. J., Mitchell, J. N., & Willingham, D. T.
(2013). Improving students' learning with effective learning tech-
P EE R R EV I E W
niques: Promising directions from cognitive and educational psychol-
The peer review history for this article is available at https://www.
ogy. Psychological Science in the Public Interest, 14(1), 4–58. https://
webofscience.com/api/gateway/wos/peer-review/10.1111/jcal. doi.org/10.1177/1529100612453266
12889. Field, A. (2018). Discovering statistics using IBM SPSS statistics (5th ed.). Sage.
Goodwin, B. (2000). Raising the achievement of low-performing students.
Policy brief (Vol. 5, pp. 3–10). Mid Continent Research for Education
DATA AVAI LAB ILITY S TATEMENT
and Learning.
The datasets generated during and/or analysed during the current Hamm, J. M., Perry, R. P., Chipperfield, J. G., Murayama, K., & Weiner, B.
study are available from the corresponding author on reasonable (2017). Attribution-based motivation treatment efficacy in an online
13652729, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jcal.12889 by Cochrane Germany, Wiley Online Library on [27/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
16 WHALEN ET AL.

learning environment for students who differ in cognitive elaboration. in students with learning difficulties and their relation to self-concept
Motivation and Emotion, 41, 600–616. https://doi.org/10.1007/ and academic goals. Learning Disabilities Research & Practice, 20(2),
s11031-017-9632-8 86–97.
Haynes, T. L., Perry, R. P., Stupnisky, R. H., & Daniels, L. M. (2009). A Okolo, C. (1992). The effects of computer-based attribution retraining on
review of attributional retraining treatments: Fostering engagement the attributions, persistence, and mathematics computation of students
and persistence in vulnerable college students. In J. Smart (Ed.), Higher with learning disabilities. Journal of Learning Disabilities, 25(5), 327–334.
education: Handbook of theory and research (Vol. 24, pp. 227–272). Perry, R., Stupnisky, R., Hall, N., Chipperfield, J., & Weiner, B. (2010). Bad
Springer. https://doi.org/10.1007/978-1-4020-9628-0_6 start and better finishes: Attributional retraining and initial performance
Heller, K. A., & Ziegler, A. (1996). Gender differences in mathematics and in competitive achievement setting. Journal of Social and Clinical Psychol-
the sciences: Can attributional retraining improve the performance of ogy, 29(6), 668–700. https://doi.org/10.1521/jscp.2010.29.6.668
gifted females? The Gifted Child Quarterly, 40(4), 200–210. Pintrich, P. R. & Schunk, D. H. (2002). Motivation in education: Theory,
Horner, S., & Gaither, M. (2004). Attribution retraining instruction with a research and application (2nd ed.). Englewood Cliffs, NJ: Merrill Pren-
second-grade class. Early Childhood Education Journal, 31(3), tice Hall.
165–170. Schiefele, U., Stutz, F., & Schaffner, E. (2016). Longitudinal relations
Houston, D. M. (2016). Revisiting the relationship between attributional between reading motivation and reading comprehension in the early
style and academic performance. Journal of Applied Social Psychology, elementary grades. Learning and Individual Differences, 51, 49–58.
46(3), 191–200. https://doi.org/10.1111/jasp.12356 https://doi.org/10.1016/j.lindif.2016.08.031
Kaiser, H. F. (1960). The application of electronic computers to factor anal- Schunk, D. H., & Ertmer, P. A. (1999). Self-regulatory processes during
ysis. Educational and Psychological Measurement, 20, 141–151. computer skill acquisition: Goal and self-evaluative influences. Journal
Karlen, Y., Hirt, C. N., Liska, A., & Strebner, F. (2021). Mindsets and of Educational Psychology, 91, 251–260.
self-concepts about self-regulated learning: Their relationships with Snijders, T. A. B., & Bosker, R. J. (2012). Multilevel analysis: An introduction
emotions, strategy knowledge, and academic achievement. Frontiers in to basic and advanced multilevel modeling. Sage Publishing.
Psychology, 12, 1–13. Taber, K. S. (2018). The use of Cronbach's alpha when developing and
Kirk, R. E. (2013). Experimental design: Procedures for the behavioral reporting research instruments in science education. Research in
sciences. Sage. https://doi.org/10.4135/9781483384733 Science Education, 48, 1273–1296.
Kreft, I. G. G., & de Leeuw, J. (1998). Introducing multilevel modeling. Sage. Trautmann, D. (2022). Notentabellen und Notenschlüssel für alle Schulfor-
Kuepper-Tetzel, C. E., & Gardner, P. L. (2021). Effects of temporary mark men. https://www.abi-rechner.com/notentabelle/#Gymnasium
withholding on academic performance. Psychology Learning and Teach- Twisk, J. W. R. (2006). Applied multilevel analysis: A practical guide.
ing, 20(3), 405–419. https://doi.org/10.1177/1475725721999958 Cambridge University Press.
Lazowski, R. A., & Hulleman, C. S. (2016). Motivation interventions in edu- Weiner, B. (1985). An attributional theory of achievement motivation and
cation: A meta-analytic review. Review of Educational Research, 86, emotion. Psychological Review, 92, 548–573.
602–640. https://doi.org/10.3102/0034654315617832 Weiner, B., Kun, A., & Benesh-Weiner, R. (1980). The development of mas-
Lorah, J. (2018). Effect size measures for multilevel models: Definition, tery, emotions, and morality from an attributional perspective. In
interpretation, and TIMSS example. Large-Scale Assessment in Educa- W. A. Collins (Ed.), Development of cognition, affect, and social relations
tion, 6, 1–11. https://doi.org/10.1186/s40536-018-0061-2 (pp. 35–48). Erlbaum.
Martinez, A., & Huber, T. (2019). Attributional feedback and its effect on Wilson, D. B. (2023). Practical meta-analysis effect size calculator. https://
self-beliefs and academic achievement: A review of the literature. www.campbellcollaboration.org/escalc/html/EffectSizeCalculator-
International Journal of Current Research, 11(3), 2242–2246. https:// SMD5.php
doi.org/10.24941/ijcr.34610.03.2019 Zeichner, O. (2018). The impact of cognitive and non-cognitive feedback
Maxwell, S., Reynolds, K. J., Lee, E., Subasic, E., & Bromhead, D. (2017). on students' achievement in a distance learning environment. Journal
The impact of school climate and school identification on academic of Educational Technology, 14(4), 13–27.
achievement: Multilevel modeling with student and teacher data. Fron-
tiers in Psychology, 8, 1–21. https://doi.org/10.3389/fpsyg.2017.
02069
Möller, J., & Köller, O. (2004). Die Genese akademischer Selbstkonzepte: How to cite this article: Whalen, K. A., Renkl, A., Eitel, A., &
Effekte dimensionaler und sozialer Vergleiche. Psychologische
Glogger-Frey, I. (2023). Digital re-attributional feedback in
Rundschau, 55, 19–27.
Morris, M. (2013). A naturalistic investigation into the effectiveness of an high school mathematics education and its effect on
attributional retraining program for academic performance. Social Sci- motivation and achievement. Journal of Computer Assisted
ences Directory, 2(2), 20–40. https://doi.org/10.7563/SSD_02_02_02 Learning, 1–16. https://doi.org/10.1111/jcal.12889
Núñez, J., González-Pienda, J., González-Pumariega, S., Roces, C.,
Alvarez, L., & González, P. (2005). Subgroups of attributional profiles

You might also like