Professional Documents
Culture Documents
Texto+19.1 Pendiente
Texto+19.1 Pendiente
Texto+19.1 Pendiente
Constructivist ideas have influenced recent major innovations in Dutch secondary education and
new curricula for reading and math in primary education, for example, pay much more attention to
metacognition than before. In our study, we compared the growth of student metacognition in
varying learning environments, direct instruction, and cognitive apprenticeship in primary school.
The study also included a control group of teachers. In order to measure metacognition we
developed a questionnaire, with separate parts for metacognitive skills and metacognitive
knowledge. In the item selection procedure we made use of item response modeling. It was
found that in the direct instruction and the cognitive apprenticeship group the pupils had higher
scores on metacognitive skills and metacognitive knowledge compared to the control group pupils.
No clear differences were found between direct instruction and cognitive apprenticeship.
Interactions of learning environment and student intelligence were non-significant for both
output measures.
Introduction
Constructivism has changed the traditional view of learning as knowledge absorption
into a view of learning as active knowledge construction. Students actively process
information, using prior knowledge, skills, and strategies (Resnick, 1989). Learning is
considered a constructive, cumulative, self-regulated, goal-oriented, situated,
collaborative, and individually different process of knowledge building and meaning
construction (De Corte, 2000). Education is no longer expected to focus solely on the
transfer of knowledge, but also on the development of metacognition.
Background
Metacognition
The concept of metacognition was introduced by Flavell in 1976 and his
characterisations of the main elements of the concept are still in use (Boekaerts &
Simons, 1993; De Jong, 1992; Resnick, 1989; Simons, 2000).
Metacognition, according to Flavell, encompasses two elements: skills and
knowledge. By several authors metacognitive skills, the self-regulating activities
shown by learners, are further subdivided into skills that can be used before, during,
and after learning activities (Bereiter & Scardamalia, 1989). Before starting to work
on a task, orientation and planning are important, while during the task such skills as
monitoring, testing, making a diagnosis, and repairing are necessary skills. After the
completion of a learning task evaluation and reflection come into focus.
Metacognitive knowledge refers to the knowledge of learners about their own
cognition, cognitive functioning, and possibly that of others. This knowledge is
enlarged by reflection on learning experiences and can be used in the planning of
further learning tasks.
Because metacognition does not develop automatically in all students, teachers
play an essential part in its development. Some authors suggest that especially low
achievers need specific teacher support while high achievers develop metacognition
more easily without any teacher interference (Davidson, Deuser, & Sternberg 1995;
Development of Metacognition in Primary School 181
Davidson & Sternberg, 1998). In the absence of teacher support, high achievers will
take more advantage of the education offered to them and extend their lead (Biemans,
Deel, & Simons, 2001; Bolhuis, 2000; Mayer, 2001). It seems evident that teachers at
least should teach students how to regulate their learning processes before they hand
over responsibilities for learning to them (Schoenfeld, 2001), and for obvious reasons
this is especially important for students who do not have metacognition at their
disposal without any help.
Another point is the relation between intelligence and metacognition (Minnaert &
Janssen, 1999; Veenman, 1992). Veenman (1992) discusses possible models for the
relationships between metacognition and intelligence. First, metacognition can be
viewed as an integral part of intelligence. The independence model rejects this
assumption. Here, metacognitive skills and intelligence are considered as indepen-
dent predictors of learning. In the mixed model it is assumed that metacognition and
intelligence overlap.
Research Questions
Our study focused on the following research questions:
Research Design
Sample
In the sampling stage of our study, we contacted all Dutch primary school teachers in
the northern part of The Netherlands who taught seventh grade and who used the
curriculum ‘‘I know what I read’’ (n = 83). The contacts were made by mail and by
telephone. Almost 25% of this group (20 teachers) participated voluntarily in our
study together with all their students in the seventh grade, who were on average 11
years of age. The teachers used the curriculum ‘‘I know what I read’’ (in Dutch: ‘‘Ik
weet wat ik lees’’), which pays attention to the development of metacognition, but
differed in the learning environment which they offered to their students. Assignment
to the experimental and control conditions was also based on voluntary participation
and therefore non-random. The teachers in the direct instruction (DI) and cognitive
apprenticeship (CA) groups received exemplary lessons specifically designed to
enhance the implementation of either DI or CA, as well as a 15-hr training. The
training was given during 5 sessions in which the theory was explained, and practice
and feedback were given. Additionally, there were coaching sessions. The control
group consisted of teachers who had indicated that they practised no specific
instructional model. Teachers in this group received no training. Table 1 shows the
numbers of students and teachers in the research groups.
. skills used before reading, for example ‘‘Before I start reading, I look at the pictures
and the title of the text’’;
. skills used during the reading process, such as ‘‘During reading, I think over how
the text will continue’’;
. skills aimed at repairing misunderstanding, such as ‘‘When I notice that I do not
understand a part of the text, I read difficult parts of the text once more’’;
. skills used after reading, for instance ‘‘When I have finished reading, I try to tell
myself what the text was about’’.
pay more attention to metacognition in their lessons than the control group, because
the materials they were using and the training that was offered to them explicitly
asked them to do so. The main characteristics of DI and CA are in Table 2.
The implementation of the instructional behaviour of the teachers was registered
with high- and low-inference observational instruments, both focusing on the
characteristics of DI and CA. Several significant differences were found between the
control and the experimental groups and between the experimental groups, indicating
a sufficient degree of implementation. More detailed information is given by De Jager
(2002).
Analyses
To scale the items measuring metacognitive skills and knowledge (research question
1), we made use of item response theory, in particular the one parameter logistic
model (OPLM). The idea that item response models have in common is that there is
a single latent variable determining the response behaviour of individual subjects on
the items of the test. All subjects have a different position on the latent scale that can
only be inferred indirectly, from the item responses. The item response function
specifies the probability of a correct answer given the latent ability of the subject. Item
response models differ in the form of the assumed relation between the latent ability
and the item responses. In the Rasch model, the probability of a correct answer is
dependent on only one item characteristic, namely the difficulty (parameter) of the
item, which has to be estimated. In the so-called two-parameter logistic model items
are characterised by a difficulty and a discrimination parameter. As such the second
model is more realistic but the parameters are, theoretically and practically, more
difficult to estimate.
OPLM combines the tractable mathematical properties of the Rasch model with
the greater flexibility of the two-parameter logistic model. In OPLM we have item
difficulty parameters which have to be estimated and discrimination indices with
imputed values (Glas & Verhelst, 1995; Verhelst, Glas, & Verstralen, 1995). The
Rasch model assumes dichotomous items, but OPLM can also be used if the items
are polytomously scored.
With OPLM a set of test items can be calibrated on a common scale, and several
item oriented statistical tests become available if, overall, the OPLM model shows a
reasonable fit. The model for polytomous items, with dichotomous items as a special
case, can be formulated as follows. It is assumed that the response to item i, denoted
by Xi, falls in the score range (0, 1, mi). The probability of observing Xi = j as a
function of y, is given by,
P
expðai ðjy g big ÞÞ
PðXi ¼ jjyÞ ¼ P P
1 þ h expðai ðhy g big ÞÞ
With y, we denote the (latent) ‘‘ability’’, which the test is supposed to measure. For
an item with three response categories, as in our case, we have three characteristic
curves, linking the probability of a response in the category to the latent ability. The
item parameters b correspond to the position on the ability continuum where the
probabilities of responding in successive categories are equal; or in other words,
where the curves of successive categories cross. In case of three categories, there are
two item parameters per item. The discrimination index a governs the steepness of
the curves: the larger the value of a the steeper the curve. An item with a higher
index discriminates better in the ability region around the item parameters than an
item with a lower index. The discrimination indices a are supposed to be integer
constants. This assumption allows for conditional maximum likelihood estimation
of the item category parameters. Secondly, fit measures are available which focus on
the validity of the selected values of the discrimination indices and are informative
with respect to the direction in which they have to be changed in order to obtain a
better fit. The sum of the item scores, weighted by the discrimination indices, is a
sufficient statistic for the ability. This weighted sum is also used to calculate scale
scores for the subjects.
The fit of the model can be assessed by inspecting the global fit-statistic R, and a
number of item-fit statistics. The M statistics are based on a rationale originally
developed by Molenaar (Glas & Verhelst, 1995; Verhelst et al., 1995). The subject
scores are partitioned in a high and a low score group (sometimes also in an
additional medium group). For each score group, the expected number of subjects
giving the correct answer (or scoring in a category of the item) is calculated using the
estimated model, and the differences between the observed and the expected number
are combined. A negative outcome indicates that the item in question discriminates
better than average while a positive value points to a low discriminating item. The
Development of Metacognition in Primary School 187
three Ms use different partitions. Large values suggest up- or downgrading of the
discrimination indices in order to increase the item fit.
To answer research question 2, we used analysis of variance methods. We
corrected the dependent variables (student scores for metacognitive skills and
knowledge at the end of the school year) for their scores at the beginning of the school
year. In the final stage, we added student intelligence as a second factor in addition to
learning environment. Based on their score on the intelligence test, the students were
divided into four groups containing approximately 25% of the students each (lowest
scoring students, students that scored below average, students that scored above
average, highest scoring students).
Results
Research Question 1: Measuring metacognition
The questionnaire for metacognition measured metacognitive skills and metacogni-
tive knowledge by separate sets of items. The metacognitive skills part consisted of 22
multiple choice items with three alternatives (Table 3). The items were scored
polytomously, in three successive categories.
A few subjects with missing values for one or more items were left out of the
analysis. The total number of subjects in the analysis was 267. The classical test
analysis resulted in an alpha coefficient of .64, which is fairly low, and we found that
six items had low or even negative item test correlations (2, 4, 8, 16, 20, 22).
In a first OPLM-analysis on all 22 items, we assumed equal discrimination indices
over items (the discrimination index is set to one for each item). Item and global fit
statistics were obtained (R1c = 534.2; df = 129; p = .00). Given the large value of R1c,
the global fit statistic, the Rasch model had to be rejected. In the next step, the model
fit of individual items was inspected. Large positive values of the M-statistic indicate
that an item discriminates less well than average, while items with negative values are
better than average. We found five items (7, 9, 10, 17, 18) discriminating better than
average, but large positive M-values were found for four items in particular (2, 8, 13,
20), indicating that these items discriminate badly. These were items where reversed
coding was used. This finding is not uncommon and it has been suggested in the
literature to place such items in a separate scale. However, inspecting them more
closely, we concluded that they were formulated somewhat ambiguously (in the sense
that ‘‘incorrect’’ answers were also defendable), an additional reason to discard them.
Removing seven items that did not discriminate well (2, 4, 8, 13, 16, 20, 22) resulted
in a global fit statistic of R1c = 105.8 (df = 87; p = .08), which is somewhat better but
still not very good. A less drastical variant where five items (2, 8, 13, 20, 22) were
removed had a global fit statistic of R1c = 188.7 which is not acceptable (df = 99;
p = .00). We then tried to increase the fit to an acceptable level by changing the
discrimination indices, following the suggestions given by the item fit indices. In the
third and last analysis, the item indices were successively adapted. This resulted in a
reasonable fit globally of 17 items (see Table 3).
188 B. de Jager et al.
Table 3. Metacognitive skills: Calibration results of the 17 item test with unequal discrimination
indices (R1c = 111.8; df = 99; p = .18; number of observations = 267)
Item
nr Item content A B SE(B) M
Table 4. Metacognitive knowledge: Calibration results of the 8 item test with unequal
discrimination indices (R1c = 18.6; df = 19; p = .48; number of observations = 271)
Table 5. Metacognitive skills in the beginning and the end of the school year, in three research
groups
Table 6. Covariance analysis (tests of between-subjects effects) with metacognitive skills as the 5
dependent variable
Table 5 makes clear that there were a priori differences with respect to
metacognitive skills. While in the cognitive apprenticeship and the direct instruction
groups the mean scores on metacognitive skills were practically equal, the control
group scored lower. At the end of the school year, the score means were increased in
all groups. The largest gain was observed in the two experimental groups. To test for
the significance of the learning environment effect on metacognitive skills, an analysis
of covariance was performed with instruction group and student intelligence as
factors and pretest scores as the covariate (Table 6).
Table 6 shows a significant effect of learning environment on metacognitive skills.
The effects of intelligence and the interaction of learning environment and
intelligence were non-significant. The decision made earlier to use intelligence as a
blocking variable and not as a covariate may have resulted in some loss of statistical
192 B. de Jager et al.
Table 7. Estimated marginal means for metacognitive skills as the dependent variable
Table 8. Metacognitive knowledge in the beginning and the end of the school year, in three research
groups
power, but the conclusions would have been the same. In a preliminary analysis, we
found very small differences between correlations of the recoded and the raw IQ
scores.
For the cognitive apprenticeship and the direct instruction groups, the 95%
confidence intervals of the estimated means (corrected for the covariate, the pretest
skills measure) for metacognitive skills overlap strongly. The cognitive apprenticeship
and direct instruction groups both have significantly higher means than the control
group (see Table 7).
With regard to metacognitive knowledge, scaled scores were obtained using the
item weights of the OPLM analysis. We performed the same analyses as for
metacognitive skills. Table 8 shows the metacognitive knowledge scores of students at
the beginning of the school year (know1) and at the end (know2) in the three research
groups. Again, the mean scores on the pretest were very similar for the two
experimental groups, while the control group mean was lower. The same pattern was
observed on the posttest scores. All three groups showed an increase in metacognitive
knowledge.
Development of Metacognition in Primary School 193
Table 9. Covariance analysis (tests of between-subjects effects) with metacognitive knowledge as the
dependent variable
Table 10. Estimated marginal means for metacognitive knowledge as the dependent variable
Conclusions
In our study, we wanted to find out whether we could succeed in measuring
metacognition by means of a questionnaire. Although a questionnaire may not be
the optimal instrument to measure metacognition, it may be necessary to use this
instrument in studies with relatively large samples for pragmatic reasons. Other
more refined methods then may take too much time or may be too expensive. To
scale the items of the questionnaires, we made use of item response theory, in
particular the one parameter logistic model (OPLM). We succeeded in finding an
194 B. de Jager et al.
References
Bereiter, C., & Scardamalia, M. (1989). Intentional learning as a goal of instruction. In L. B.
Resnick (Ed.), Knowing, learning and instruction (pp. 361 – 393). Hillsdale, NJ: Lawrence
Erlbaum.
Biemans, H. J. A., Deel, O. R., & Simons, P. R. J. (2001). Differences between successful and less
successful students while working with the CONTACT-2 strategy. Learning and Instruction,
11(4/5), 265 – 282.
Boekaerts, M., & Simons, P. R. J. (1993). Leren en instructie, psychologie van de leerling en het leerproces
[Learning and instruction, psychology of the pupil and the learning process]. Assen, The
Netherlands: Dekker & Van de Vegt.
Bolhuis, S. M. (2000). Naar zelfstandig leren: Wat doen en denken docenten? [Towards independent
learning: what do teachers do and think?] Apeldoorn, The Netherlands: Garant.
Development of Metacognition in Primary School 195
Brand-Gruwel, S. (1995). Onderwijs in tekstbegrip: Een onderzoek naar het effect van strategisch lees- en
luisteronderwijs bij zwakke lezers [Instruction in text comprehension: A study into the effects of
strategic reading and listening instruction for weak readers]. Ubbergen, The Netherlands:
Uitgeverij Tandem Felix.
Davidson, J. E., Deuser, R., & Sternberg, R. J. (1995). The role of metacognition in problem
solving. In J. Metcalfe & A. P. Shimamura (Eds.), Metacognition: Knowing about knowing (pp.
207 – 227). Cambridge, MA: MIT.
Davidson, J. E., & Sternberg, R. J. (1998). Smart problem solving: How metacognition helps. In D.
J. Hacker, J. Dunlosky, & A. C. Graesser (Eds.), Metacognition in educational theory and practice
(pp. 47 – 69). London: Lawrence Erlbaum.
De Corte, E. (2000). Marrying theory building and the improvement of school practice: A
permanent challenge for instructional psychology. Learning and Instruction, 10, 249 – 266.
De Jager, B. (2002). Teaching reading comprehension. The effects of direct instruction and cognitive
apprenticeship on comprehension skills and metacognition. Groningen, The Netherlands: GION.
De Jong, F. P. C. M. (1992). Zelfstandig leren: Regulatie van het leerproces en het leren reguleren: Een
procesbenadering [Independent learning: Regulation of the learning process and learning to
regulate: A process approach]. Tilburg, The Netherlands: KUB.
Desoete, A., Roeyers, H., Buysse, A., & De Clercq, A. (2002). Dynamic assessment of
metacognitive skills in young children with mathematics-learning disabilities. In G. M. van
de Aalsvoort, W. C. M. Resing, & A. J. J. M. Ruijssenaars (Eds.), Learning potential assessment
and cognitive training: Actual research and perspectives in theory building and methodology (Vol. 7,
pp. 307 – 333). Oxford: Elsevier.
Dominowski, R. L. (1998). Verbalization and problem solving. In D. J. Hacker, J. Dunlosky, & A.
C. Graesser (Eds.), Metacognition in educational theory and practice (pp. 25 – 47). London:
Lawrence Erlbaum.
Ericsson, K. A, & Simon, H. A. (1993). Protocol analysis: Verbal reports as data. Cambridge, MA:
MIT.
Glas, C. A. W., & Verhelst, N. D. (1995). Testing the Rasch model. In G. H. Fischer & I. W.
Molenaar (Eds.), Rasch models: Foundations, recent developments and applications (pp. 69 – 96).
New York: Springer-Verlag.
Hoogendijk, W., & Wolfgram, P. (1995). KEA, halverwege: Projectverslag schooljaar 1994 – 1995
[KEA, halfway: project report school year 1994 – 1995]. Rotterdam, The Netherlands:
CED.
Klatter, E. B. (1996). Studievaardigheden in de brugklas. Een vergelijking tussen algemene
en specifieke verwerkingsstijlen van brugklasleerlingen. Pedagogische Studiën, 73(4), 303 – 316.
Kluvers, K., & Simons, P. R. J. (1992). Zelfregulatievaardigheden en COO: Tussenrapportage
[Selfregulation skills and computer-assisted education: interim report]. Nijmegen, The
Netherlands: KU Nijmegen.
Laros, J. A., & Tellegen, P. J. (1991). Construction and validation of the SON-R 512-17, The Snijders-
Oomen non-verbal intelligence test. Groningen, The Netherlands: Wolters-Noordhoff.
Mayer, R. E. (2001). Changing conceptions of learning: A century of progress in the scientific study
of education. In L. Corno (Ed.), Education across a century: The centennial volume (pp. 34 – 76).
Chicago: University of Chicago.
Meijer, J., Elshout-Mohr, M. E., & Van Hout-Wolters, B. H. A. M. (2001). An instrument for the
assessment of cross curricular skills. Educational Research and Evaluation, 7, 79 – 108.
Minnaert, A., & Janssen, P. J. (1999). The additive effect of regulatory activities on top of
intelligence in relation to academic performance in higher education. Learning and Instruction,
9(1), 77 – 91.
Muijs, D., & Reynolds, D. (2001). Effective teaching: Evidence and practice. Gateshead: Athenaeum
Press.
Pintrich, P. R., & De Groot, E. V. (1990). Motivational and self-regulated learning components of
classroom academic performance. Journal of Educational Psychology, 82(1), 33 – 40.
196 B. de Jager et al.
Pressley, M., & Afflerbach, P. (1995). Verbal protocols of reading: The nature of constructively responsive
reading. Hillsdale, NJ: Lawrence Erlbaum Associates.
Pressley, M., & McCormick, C. B. (1995). Cognition, teaching, and assessment. New York:
Longman.
Resnick, L. B. (1989). Introduction. In L. B. Resnick (Ed.), Knowing, learning and instruction (pp.
1 – 25). Hillsdale, NJ: Lawrence Erlbaum.
Roelofs, E., & Visser, J. (2001). Leeromgevingen volgens ouders en leraren: Voorkeuren en
realisatie [Learning environments according to parents and teachers: Preferences and
realisation]. Pedagogische Studiën, 78, 151 – 168.
Rosenshine, B., & Meister, C. (1994). Reciprocal teaching: A review of the research. Review of
Educational Research, 64(2), 201 – 243.
Schoenfeld, A. H. (2001). Mathematics education in the twentieth century. In L. Corno (Ed.),
Education across a century: The centennial volume (pp. 239 – 279). Chicago: University of
Chicago.
Simons, P. R. J. (2000). Review studie leren en instructie [Review study learning and instruction].
Nijmegen, The Netherlands: University of Nijmegen.
Sliepen, S. E., & Reitsma, P. (1993). Training van lom-leerkrachten in directe instructie van
begrijpend leesstrategieën [Training of teachers in special education in direct instruction in
comprehension skills]. Pedagogische Studiën, 70, 420 – 444.
Van Hout-Wolters, B. (2000). Assessing active self-directed learning. In R. J. Simons, J. van der
Linden, & T. Duffy (Eds.), New learning (pp. 83 – 100). Dordrecht, The Netherlands: Kluwer
Academic Publishers.
Van Someren, M. W., Barnard, Y. F., & Sandberg, J. A. C. (1994). The thinking aloud method: A
practical guide to modelling cognitve processes. London: Academic Press.
Veenman, S. A. M. (1992). Effectieve instructie volgens het directe instructiemodel [Effective
instruction based on the direct instruction model]. Pedagogische Studieën, 69, 242 – 269.
Veenman, S. A. M., Leenders, Y., Meyer, P., & Sanders, M. (1993). Leren lesgeven met het directe
instructiemodel [Learning to teach with the direct instruction model]. Pedagogische Studiën,
70, 2 – 16.
Verhelst, N. D., Glas, C. A. W., & Verstralen, H. H. F. M. (1995). The one-parameter logistic model
OPLM. Arnhem, The Netherlands: CITO.
Walraven, A. M. A. (1995). Instructie in leesstrategieën: Problemen met begrijpend lezen en het effect van
instructie aan zwakke lezers [Instruction in reading strategies: Problems with reading
comprehension and the effect of instruction to low-achieving pupils]. Amsterdam/
Duivendrecht: Paedologisch Instituut.