International Journal of Educational Development 46 (2016) 103–113

Contents lists available at ScienceDirect

International Journal of Educational Development

journal homepage: www.elsevier.com/locate/ijedudev

Students’ skipping behavior on truancy items and (school) subjects

and its relation to test performance in PISA 2012
Christine Sälzer *, Jörg-Henrik Heine
TUM School of Education, Center for International Student Assessment (ZIB), Munich, Germany

A R T I C L E I N F O A B S T R A C T

Article history: Most large-scale student assessments focus on cognitive competences which are assumed to be acquired
Received 2 July 2015 at school. This study investigates how skipping lessons on purpose is related to student achievement in
Received in revised form 30 October 2015 the PISA 2012 competence test (mathematics, reading and science). First, a Rasch-residual factor analysis
Accepted 31 October 2015
is carried out on the truancy scale in order to discover the dimensionality of truancy as a latent construct.
Available online 2 December 2015
Second, the PISA proficiency score in each of the three competence domains is predicted using a set of
control variables plus the subject-specific truancy. The main findings are that subject-specific truancy
Keywords:
comprises four dimensions and that skipping certain school subjects goes along with a lower proficiency
Subject-specific truancy
Dimensionality
in the corresponding PISA domains. This is most obvious in mathematics and science, where students
PISA who choose the most extreme response category (skipped ‘3 times and more’) score about 40 or,
Achievement respectively, 38 points less than their non-truant classmates.
ß 2015 Elsevier Ltd. All rights reserved.

Most large-scale student assessments implicitly make two
assumptions: first, they suppose that the proficiency measured by
the achievement tests is mainly acquired at school. This holds
especially true when student assessments focus on competencies
relating to school subjects, such as mathematics or science. Second,
these studies mostly measure cognitive outcomes like student
achievement and assume that this is a valid measure of school
effectiveness. While both assumptions are plausible and true in
many cases, there is more to large-scale student assessments as we
have them today. For one, they measure more than just cognitive
outcomes and the role of context questionnaires is nowadays more
important than it used to be (OECD, 2013a) and will become even
more important in the future (OECD, 2013c). These assumptions are
further limited by the fact that when students skip classes on purpose
(i.e. they play truant), they refrain from taking the opportunity to
learn subject matters at school. Other than days at school missed due
to illness, skipped classes are missed intentionally and the subjects
students select for skipping may be related to students’ achievement
in these subjects, but also to individual characteristics such as
subject-specific self-concept or effort. At the same time, students
who skip one subject may be more likely to skip other subjects as
well. And yet, students who play truant achieve not necessarily lower
than students who do not (Renzulli and Park, 2000).
* Corresponding author.
E-mail address: christine.saelzer@tum.de (C. Sälzer).
So far, the association between class attendance and student
performance has mostly been studied using school grades or
exams as a measure for student achievement (e.g. Malcolm et al.,
1996; Stamm, 2010). Nevertheless, using school exams may
generate biased outcomes because they depend on national or
regional curricula. In our study, we measure student achievement
by the PISA 2012 proficiency scales in mathematics, science and
reading as standardized, cross-curricular scales. Our leading
research question is whether subject-specific truancy is related
to student achievement in the corresponding competence domains
in PISA 2012. First, we determine whether subject-specific truancy
as a scale comprises more than one dimension by a Rasch-residual
factor analysis. Subsequently, we predict student achievement in
the PISA competence domains of mathematics, science and reading
by linear regression models including subject-specific truancy in
the corresponding school subjects.
1. Truancy: declining school classes as an opportunity to learn
Skipping classes or skipping school is generally referred to as
truancy or truant behavior (Malcolm et al., 1996; Veenstra et al.,
2010). This usually involves an active decision of the student
without having an acceptable or legally justified reason such as an
illness or a family emergency. In line with Wilson et al. (Wilson
et al., 2008), we define truancy as ‘‘absences which pupils
themselves indicated would be unacceptable to teachers’’ (p. 3).
Most studies capture student truancy as a global behavior that has

http://dx.doi.org/10.1016/j.ijedudev.2015.10.009
0738-0593/ß 2015 Elsevier Ltd. All rights reserved.
104 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113
a certain degree, e.g. never, occasional or frequent, but they do not
distinguish single school subjects (Vaughn et al., 2013). Consider-
ing that skipping single lessons is quite common among students
at the lower secondary level (Veenstra et al., 2010), the need for
disentangling the broad concept of truancy and taking a closer look
into subject-specific truancy is obvious.
Conceptualizing lessons at school as an offer that can be used by
students, we refer to a model often used in large-scale student
assessments (e.g. Helmke and Schrader, 2013). Key to this concept
is that lessons take place while students have to actively make use
of them and learn. Of course, students have to attend the lessons in
order to be able to do so. This assumption that lessons are an offer
holds true, even though in many countries schooling is compulsory
for a certain age range. Taking into account the legal norm of school
attendance only, truancy should not exist. However, numerous
studies have shown that students skip classes without a proper
leave all over the world (e.g. OECD, 2013b; Reid, 2003; Veenstra
et al., 2010).
As students in the PISA cohort are 15 years old and have been
attending school for about a decade, truancy cannot be reasonably
studied without considering a certain selectivity of students’
truant behavior. This selectivity is prone to have at least two
implications: First, students have developed preferences regarding
their school subjects, they have selected their favorite subjects and
teachers and other subjects which they like less or even dislike.
These preferences may be reflected by students’ academic
achievement or performance at school. Student performance or
academic achievement is a part of students’ school biography that
intuitively seems to be negatively related to truant behavior: The
worse a student does at school, the more he or she will play truant
(Vaughn et al., 2013). Second, this selectivity may bias the results
of studies tackling the association between truancy and academic
achievement or even alter the interpretation of empirical findings.
When truancy is studied empirically by collecting students’
answers to a questionnaire, one obvious problem is that those
students who are absent during the time when the questionnaire is
completed will not be included in the database although their non-
attendance indicates that they are exactly in the focus of the study.
When students with a possibly high score on truancy are not
present in the database, the results are likely to be underestimated
due to an artificially reduced variance. Another issue that needs to
be taken into account is that truancy is a behavior that is socially
undesirable on the one hand and a prevalent part of adolescent
behavior on the other (Sälzer et al., 2012), and may therefore
involve both concealment and exaggeration by the respondents.
Furthermore, the association of truant behavior and student
academic achievement needs to be carefully investigated so that a
possible collinearity does not affect the results. While student
performance can often not be fully separated from immigration
status or SES, the predictive value of academic achievement as an
indicator of truancy is highly dependable on the model and
technique used for data analysis. Especially two studies carefully
examined the relationship of academic achievement and truancy,
controlling for a differentiated background model. Henry and
Huizinga (2007) looked at the characteristics of 8th and 10th graders
who were playing truant and showed that high academic
achievement was related to low truancy rates. In a recent
longitudinal study, Veenstra et al. (2010) investigated a sample of
2230 Dutch students, five percent of which were persistent truants,
their parents and teachers. The authors were able to show that most
of the persistent truants had low academic achievement. Referring
to this finding, we tackle two research questions in the current
study: (1) Is truancy a multidimensional construct that can be
captured by students selectively skipping certain school subjects?
(2) How does subject-specific truancy relate to student achievement
in the corresponding competence domains in the PISA 2012 test?
From a methodological point of view, studies dealing with
student truancy often face two major challenges: first, students’
self-reports are the most reliable source of data since only the
students themselves know when they have played truant. Other
data sources, such as observations or school records, bear quite a
high risk of being biased. This implies that the students to be
surveyed need to be non-truant at the time of the data collection
and researchers have to rely on the students answering honestly.
Second, most studies at hand define truancy quite broadly and
merely use single items to ask whether or not a student has played
truant within a given period of time (e.g. two school weeks, as
implemented in the PISA study in OECD, 2013b). When researchers
want to know more about student truancy and its relatedness to
several factors within and around the student, measuring the
concept of truant behavior becomes difficult. To date, only little is
known both about the subject-related dimensionality of truancy as
a scale and the association of subject-specific truancy and student
achievement in these skipped school subjects. Among the
assessment domains in PISA (Reading, Mathematics and Science),
we find that reading competence is by far more general and
ubiquitous than mathematical or scientific competence. Due to
this broad variance with regard to opportunities to learn inside and
outside school, we chose the PISA domains in order to compare
their relationship with subject-specific truancy in Mathematics,
Science and Reading in the language of test (in our case, German).
2. Sample and sampling procedure
The participants of our study were the PISA 2012 sample of 15-
year-old students in Germany (n = 5001). Students were sampled
as a stratified random sample according to the international
sampling definitions (Heine et al., 2013). First, a so-called sampling
frame consisting of a complete list of all schools possibly attended
by 15-year-old students was established. In order to obtain a
sample representative for the population of 15-year-old students
attending a school in their respective country of test, each country
is divided into several areas, so-called explicit strata. A stratum is a
partition of the population which is defined according to national
specifications that may have an impact on the representativity of
the results (OECD, 2014a). In Germany, the federal structure of the
educational system made it necessary to define a stratum for each
of the 16 federal states. The participating students attended one of
five lower secondary school types: a Gymnasium, which is the only
secondary school type that is prevalent in all 16 federal states in
Germany and represents a merely academic track, a Realschule
(middle school), a Hauptschule (lower secondary school) or one of
two comprehensive school types (integrated or cooperative
comprehensive school with within-school tracking or within-
school streaming).
Conceptually, only Gymnasium schools have a common history
and curriculum within Germany, whereas the other four school
types vary in terms of their history and curricula across the
16 federal states. According to this multi-tier school system, the
explicit strata representing federal states were subdivided into
implicit strata so that in each federal state, all prevalent school
types were present in the sample (OECD, 2014a) and Gymnasium
schools were separated from non-Gymnasium schools.
Based on this stratification, a random sample of schools was
drawn within each stratum (federal state). Within each school,
25 students were randomly selected for participation. All students
were born in the year 1996, as per birth cohort definition of the PISA
2012 assessment framework (OECD, 2013a). 51 percent of the
students were female and 14 percent had an immigrant background.
More than half of the students born in 1996 attended grade 9,
37 percent attended grade 10 and only small proportions attended
grade 7, 8 or 11. This relatively high range recurs to the
 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113
grade-retention policy in Germany, where according to the PISA
2012 data, one in five students at age 15 has at least once repeated a
grade during their school career (OECD, 2014b; Sälzer et al., 2013).
Student participation in the PISA test was completely tracked and
documented. While a student tracking form collected demographic
and school-related information for each student (e.g. grade, sex,
school track), test administrators completed so-called session
reports to document the participation in both the PISA test and
the following student questionnaire. Absent or non-participating
students received a respective code in the session report to indicate
the reason for their non-participation. Such reasons included
student mobility and thus enrollment at a different school, late
arrival, illness or general absence. In cases where students were not
present at the PISA testing day, follow-up sessions were scheduled to
finally include the missing students. Out of the originally drawn
5454 students, 389 were coded as absent both at the PISA testing day
and the follow-up session, 182 girls and 207 boys. Since the risk of
biased results in international large-scale assessments like PISA is
quite high due to several factors related to student non-participa-
tion, the data in PISA are weighted to compensate for non-
participation both at the student and the school level. In doing so,
the fact that a random sample means that students in larger schools
have a lower probability of being drawn into the sample than
students in smaller schools can be controlled for. By weighting the
data, they are adjusted according to the sampling procedure and
the actual participation of schools and students drawn into the
sample. In other words, when less students than originally planned
participate in PISA, selection bias can be minimized by assigning
each of the participating students a weight indicating how many
other students in their stratum they represent (OECD, 2014a). That
way, one can assume that the final database for the German PISA
sample is not biased due to students not participating at the PISA
2012 survey as a whole. However, carrying out these weighting
procedures does not automatically mean that no additional bias
might occur due to students selectively skipping specific parts (e.g.
some items or even whole subscales) of the questionnaires. Thus,
prior to any analyses, one has to examine carefully all possibly
existing patterns of missing values on the scales under study, which
is discussed and outlined in the third part of the subsequent method
section of this paper (Section 3.3).
3. Method
3.1. Procedure
In order to collect self-reported data on truancy, the PISA
2012 student questionnaire that followed the PISA test session was
enhanced by a set of national extra questions on truancy. Trained
test administrators guided the students through the survey
following a standardized manual that is binding for all countries
participating in PISA. Students were informed that their names and
data are at no point retrievable by the PISA project management, so
that they felt free to answer honestly. Teachers that were in the
room to support the test administrator in case of emergency were
asked not to walk around in the classroom and avoid contact with
students in order not to interfere with their completing the test
and questionnaire. Immediately after finishing the questionnaire,
the test administrator collected all documents and packed them for
shipping to the PISA National Center.
3.2. Instruments
The students’ proficiency level in PISA mathematics, reading and
science was measured using the PISA 2012 competence test. Five
plausible values (PV) were assigned to each student (OECD, 2012) in
order to estimate the average competence in mathematics, reading
105
and science according to the international scaling procedures. All
other variables were collected through the student questionnaire
and, where possible, compared to the student tracking form, which
had been filled out by the PISA school coordinator. In addition to the
internationally administered questions asking students about (a)
arriving late for school, (b) skipping single lessons or (c) skipping
school days within the last two full school weeks prior to the PISA
test, the German sample was asked more specifically about their
playing truant. All the truancy questions referred to the current
school year, which at the time of data collection had been going on
for approximately 8 months. This takes into account that in order to
obtain valid data, students must have had a realistic chance to skip
classes before completing the questionnaire. Students were asked to
indicate the number of times they had intentionally skipped certain
school subjects in the current school year. Subject-specific truancy
was measured by questions referring to six subjects. Five of them
correspond to the PISA assessment domains of reading, mathematics
and science. Science in Germany is taught as three separate subjects,
i.e. biology, chemistry and physics. Taking into account that physical
education (PE) lessons are usually the most ‘popular’ subject to
truant from (Fischer, 1989), we also included PE in our list. In order to
operationalize students’ socio-economic status, we used Ganze-
boom’s International Socio-Economic Index (ISEI) (Ganzeboom, 2010),
applying the highest value per family (either mother or father,
HISEI). Since student self-reports on their parent’s occupation bears
some uncertainty with regard to the validity of the data (Wardle,
2002), several instances of control are used in PISA. First, the student
responses are to be coded according to the ISCO system (Interna-
tional Standard Classification of Occupations, Ganzeboom, 2010)
and the coders undergo a thorough training. Each student response
is coded by two coders and then refereed by an expert coder. Second,
in order to take into account that the mentioned validity issues recur
not only to the coding, but to the student responses as well, the
German student questionnaire added two more questions about
what exactly parents do in their professional occupation. This
information can be used to generate EGP-classes (named after
Erikson et al., 1979) expressing social situations depending on
income, education and recognition, but also to validate student
responses about parental occupation that otherwise would be
ambivalent.
As to the immigrant status, we used the students’ response
indicating whether they were born in Germany or abroad and
defined three groups: students without immigrant status, students
with immigrant status and students with unspecified immigrant
status (i.e. missing values).
3.3. Handling of missing data on the truancy scale
Since truancy is a research topic that is sensitive to social
desirability and individual response behavior, a closer look into
possible patterns of missing data is required. After adjusting
students who were fully absent during the PISA test by the
weighting procedure described above, one needs to take into
account that missing values within participating students are likely
to occur. In case there are certain response patterns, they need to be
taken into account for all related data analyses. Therefore, we
screened the data matrix with regard to missing responses, i.e.
truancy items which have been skipped. The analysis of missing
values on the six subject-specific truancy items was conducted
according to two different approaches. Precedent to both
approaches, a dichotomous indicator variable for missing values
on every truancy item was formed and coded ‘1’ for a missing value
and ‘0’ for a valid response. Based on these indicator variables, we
first analyzed the structure of the potential patterns of missing
values following the person-centered statistical approach of the
Configural Frequency Analysis (CFA – cf. Krauth, 1973; Lienert and
106 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113
Krauth, 1975; Stemmler, 2014). Based on the results indicating
the dimensionality of subject-specific truancy, the analysis of the
missing data patterns were conducted in two steps using the
R-package confreq (Heine, 2015) of the statistical software R
(R Development Core Team, 2014). Initially, the missing pattern
for the set of the three science-subject truancy items was analyzed.
Based on these results, the missing pattern for the set of the truancy
items for German, mathematics and physical education followed,
including a new missing indicator variable related to the truancy
(sub-)scale consisting of the three science truancy items (biology,
chemistry and physics).
The second approach to handle the missing truancy data was to
analyze the correlation between the prevalence of missing data on
subject specific truancy and the respective achievement outcome
on the three PISA proficiency scales (mathematics, reading and
science).
Based on the finding that omitting truancy items is inter-
correlated with the outcome on the respective PISA proficiency
scale (cf. Table 1), item omission on the subject-specific truancy
scale was included as an independent variable in the regression
analyses predicting the corresponding PISA proficiency value. In
order to do so, student responses have been dummy-coded,
including a category for item omission. For science, dummy coding
was based on the categorized WLE outcome resulting from a prior
IRT-based scaling approach including the three science truancy
items. Thus, the resulting dummy variables used for prediction of
the reading, mathematics and science proficiency represent the
following coding scheme. The most frequently chosen category,
dummy (1) – ‘‘not playing truant at all’’ in the respective subject –
was used as the reference category. Dummy (2) meant playing
truant in the respective subject ‘‘once or twice’’, dummy (3) stood
for playing truant in the respective subject ‘‘three times and more’’
and dummy (0) meant omitting the item of the respective subject.
All analyses were undertaken with the PISA replication methodol-
ogy in mind, considering the stratified nature of the sampling
design using the R-package ‘survey’ (Lumley, 2014).
3.3.1. Research question 1
In order to answer our first research question about truancy
being a multidimensional construct that can be represented by
students selectively skipping certain school subjects, we per-
formed an IRT-based (Item Response Theory) pairwise comparative
scaling approach with six subject-specific truancy items. This
comparative scaling approach called ‘pairwise’ was originally
introduced by Choppin (1968, see also McArthur and Wright,
1985) within the context of sample-free calibration of item banks
for large-scale assessments. We used the R-package pairwise
(Heine, 2014), which implemented Choppin’s approach and has
been shown to yield rather unbiased IRT measures under the
condition of higher rates of missing data (Heine and Tarnai, 2015).
Due to the empirically sparse use of the two topmost categories
(third category: ‘‘Three or four times’’; fourth category: ‘‘Five times
and more’’) and the therefore skewed distribution of the responses,
Table 1
Intercorrelations of the PISA proficiency scales and missing indices for subject-specific truancy items.
Variable 1 2
1 Mathematics proficiency – .92
2 Science proficiency .00
3 Reading proficiency .01
4 Omitting truancy item mathematics .03
5 Omitting truancy item science .03
6 Omitting truancy item German .03
Note. Numbers above the diagonal indicate the Pearson correlation (taking into account the stratified sampling design). Numbers below the diagonal represent the respective
standard error.
these two categories were collapsed prior to IRT scaling.
Cronbach’s alpha as an indicator of internal consistency based
on Classical Test Theory (CTT) was computed as well as the
coefficient of the weighted least square estimates’ (WLE) reliability
based on the assumptions of IRT scaling. Following the one-
dimensional scaling approach, we conducted a Rasch-residual
factor analysis (Linacre, 1998) using the scale measuring truancy
by a set of subject-specific items in order to analyze the
dimensionality of truancy. Rasch-residual factor analysis is an
approach looking for patterns in the part of the data which does not
conform to the Rasch-measures and is thus unexpected. The main
objective is to see whether groups of items have the same patterns
of unexpectedness. If this is the case, the respective items probably
also have a substantive attribute in common. The interpretation of
the Rasch-residual factor analysis is based on the contrast between
positive and negative loadings.
Referring to research question 1, we hypothesized that truancy
is a multidimensional construct with regard to different subjects
that are intentionally skipped by students (hypothesis 1). We
assumed that students who skip certain subjects may be more
likely to skip other subjects as well, but not necessarily any subject.
Accordingly, if the truancy scale is multidimensional, this could
mean that students select their subjects to skip according to
different criteria instead of generally choosing to play truant and
then skip whichever subject. Consequently, the latent construct we
suspect behind the students’ selection of subjects to skip is a
subject-specific selectivity that emerges in a multidimensionality
of the truancy scale.
3.3.2. Research question 2
With regard to our second research question referring to the
association of skipping certain subjects and achievement in the
corresponding PISA competence domain, we predicted student
achievement in the PISA 2012 literacy test using three separate
linear regression models. We specified one regression model per
competence domain, i.e. mathematics, reading and science.
Besides a set of control-variables including the school type,
subject-specific truancy was used as a predictor, matching the
respective competence domain. For mathematics and reading,
the collapsed and then dummy-coded responses (see section on
missing data above) on the respective truancy item was used for
prediction. For science, the student responses to the three
science items (biology, chemistry and physics) were aggregated
by conducting an IRT-based scaling approach first, followed by a
categorization of the resulting WLE estimates. This categorical
science truancy scale was dummy-coded considering the
missing data structure (see section on missing data above)
and then in turn used as an explaining variable. Thus, student
achievement in PISA 2012 mathematics was explained by
skipping mathematics classes and so forth. We assumed that
skipping school subjects is negatively associated with student
achievement in the corresponding PISA competence domains
(hypothesis 2).
3 4 5 6
.87 .23 .23 .23
– .89 .36 .26 .26
.00 – .29 .29 .29
.02 .02 – .98 .98
.02 .02 .00 – .98
.02 .02 .00 .00 –
 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113
4. Results
4.1. Descriptives
The proportions of students reporting to have skipped classes in
the current school year revealed that the vast majority of 15 year-
olds has never played truant in this period of time (cf. Table 2).
Apparently, most students who choose to not attend a class, select PE
lessons for skipping (n = 463). The other subjects in our list are almost
equally ‘popular’ with regard to student truancy (191 < n < 243).
Overall, we see for all subject-specific items that most of the students
who indicate to play truant do so only once or twice.
Table 2 indicates that the proportion of missing student
responses within the truancy scale is quite high. Given that
students’ response behavior is likely to be affected by the implied
social desirability of the topic, this is in line with most recent
studies on truancy (e.g. Vaughn et al., 2013; Veenstra et al., 2010).
With regard to the weighting of the data in order to correct for
possible bias due to student non-participation (maybe due to
truant behavior), missing values on the truancy scale indicate that
those students who were present and participated in PISA
2012 and chose not to respond to the items measuring truancy,
did so on purpose. All students have been asked to complete the
truancy scale and according to the test session reports, it is unlikely
that time constraints or overlooking the respective pages of
the questionnaire may have hindered students from completing
the items. Nevertheless, such high rates of missing values require
appropriate handling to avoid bias caused by methodological
imprecision. We chose to explicitly model the ‘missingness’ of
responses on the truancy items by including ‘missing’ as a separate
response category within each subject-specific item for further
analyses. Especially with regard to the fact that each participating
student has had the opportunity to either respond to these items or
not, this seems appropriate. In doing so, each student was assigned
either their selected response code or the code for a missing
response.
Analyses based on CTT (Classical Test Theory, e.g. Kline, 2005)
yielded an internal consistency of ra = .84 for the scale of six items,
which could be increased to ra = .87 by eliminating the truancy
item for physical education. Thus, the results suggest to proceed
with a one-dimensional IRT based scaling approach. However,
calibrating those six items and computing WLE person estimates
(Warm, 1989) using the IRT-based pairwise comparison approach,
as implemented in the R-package pairwise (Heine, 2014), resulted
in a somewhat low coefficient for WLE reliability of rWLE = .37. This
result supports our hypothesis of truancy being a heterogeneous –
or multidimensional – construct. The nature of this construct is
examined in more detail in the following paragraph.
4.2. Hypothesis 1, dimensionality of truancy
The factor analysis of the Rasch-residuals resulting from the
pairwise scaling procedure indicated the assumed (sub-)dimension-
ality of the truancy construct. Fig. 1 shows the item loadings (y-axis)
on the first main component of the Rasch-residual-factor analysis
(Linacre, 1998), plotted against the item difficulties (x-axis) obtained
from the pairwise IRT scaling approach of the six truancy items. The
plot indicates the difficulty of each truancy item, where the leftmost
school subject is the easiest and the rightmost is the hardest to play
truant. In other words, when many students respond positively to a
question asking about truancy in the respective subject, this item
will be classified as easy. When only a small group of students
responds positively, the item will be classified as hard. The loadings
on the y-axis of Fig. 1 indicate the dimensionality of the truancy scale
with regard to the similarity or closeness of the different school
subjects investigated. The greater the distance between two
107
Table 2
Proportions (%) of students who reported to having played truant; n = 5001.
Subjects skipped Missing Score 0 Score 1 Score 2
‘‘Not playing ‘‘Once ‘‘Three times
truant at all’’ or twice’’ and more’’
(collapsed)
Biology 21.6 72.6 4.2 1.6
Chemistry 21.5 71.8 4.9 1.8
Physics 21.9 72.3 3.8 2.0
Mathematics 21.7 71.4 4.3 2.6
Physical 21.4 64.0 9.3 5.3
education
German 21.8 72.6 3.8 1.8
subjects, the clearer is the distinction of several dimensions.
Subjects close to each other indicate a common dimension. When
a subject appears isolated in this kind of plot, this means that
students who skip this subject are unlikely to skip other subjects as
well. Whereas subjects plotted close to each other indicate that a
student who skips one of these has a certain tendency to skip
‘neighboring subjects’ as well.
Fig. 1 indicates that the truancy items differ both in their
loadings and their difficulty. With regard to their loadings, the
items split into four groups suggesting a four-dimensional
structure of the construct ‘subject-specific truancy’. Considering
their difficulty, we see that PE (physical education) is the most
popular subject to play truant from and thus the ‘easiest’ of the six
items for 15-year-old students. Playing truant in PE lessons seems
to be an outstanding student characteristic, since the loading of
this truancy item is plotted far away from the other items and thus
both less difficult than other items and a completely separate
dimension. Considering possible explanations for the relative
isolation of skipping PE lessons compared to the other subjects, one
plausible argument is that in Germany, low achievement in PE does
not affect the students’ advancement into the next grade level at
the end of the school year, so some students may consider it less
important than other subjects. Overall low achievement usually
means that students have to repeat a grade, which is still quite
common in Germany (OECD, 2014b; Sälzer et al., 2013). In contrast,
lessons in German and mathematics form two more dimensions of
truancy. Both subjects are core subjects and therefore considered
to be playing a key role within the students’ schedule. Low
performance in more than one of these core subjects is commonly
followed by grade repetition as well. Interestingly, the three
science subjects of biology, chemistry and physics built up to a
common dimension with regard to subject-specific truancy,
supporting our approach to proceed with an aggregated index
for skipping science lessons in the subsequent regression analyses.
All in all, we find four dimensions within the truancy scale:
three single-subject domains (PE, German, mathematics), where
PE is isolated from the other dimensions and the most popular and
hence the ‘easiest’ among the subjects to be skipped. This means
that, for most students who play truant, PE is the only subject they
avoid. Both German and mathematics in their role as core subjects
form separate dimensions still relatively far away from each other
in the plot, meaning that students who skip German do not
necessarily skip mathematics and vice versa. The fourth dimension
is a conglomerate of the three science-related subjects which are
taught in secondary schools in Germany: physics, chemistry and
biology. This means that students who choose to skip one science
subject tend to skip another science subject as well.
4.3. Missing data analyses on the truancy scale
Investigating our second research question, we needed to be
aware that missing data could bias our regression analyses. In
108 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113
Physical Education
0.6
0.4
0.2
loadings
0.0
−0.2
−0.4
−2.0 −1.5
Fig. 1. Item loadings on the first main component of the Rasch residual factor analysis plotted against the item difficulties obtained from the one-dimensional pairwise scaling
approach; PISA 2012 sample of Germany; n = 5001.
order to take into account that the data set contained missing
values, we handled missing data as follows. In a first step, the
response patterns of the dichotomous indicator variables of
missing values relating to the truancy items in the three science
subjects were included in a first-order Configural Frequency
Analysis (CFA). With regard to eight possible patterns of missing
data, the results yielded a significant global likelihood ratio x2-test
(LR = 9932.366, df = 4, p < .05; x2 = 23,701.53, df = 4, p < .05) with
two significant types (fobserved > fexpected): ‘omitting none of the
three items’ (pattern ‘000’; fobserved = 3891, x2z-approximated ¼ 30:389,
pBonferroni adjusted < .00625) and ‘omitting all of the three items’
(pattern ‘111’; fobserved = 1069, x2z-approximated ¼ 142:516,
pBonferroni adjusted < .00625). This means that the students in our
sample either responded to all of the science truancy items or to
none of them. At the same time, all the other emerging patterns of
item omission were classified to be ‘antitypes’ (fobserved < fexpected)
based on the respective local binomial-test as well as the z-
approximation test. Antitypes are missing data patterns which
appeared less frequently than expected with regard to the basic
model. Response patterns in which students omitted single items
of the science truancy subscale have thus been defined as
antitypes, since they are less prevalent than expected with regard
to the basic model.
According to Victor (1989), the main problem with such a result
is the assumption of total independence with regard to the whole
contingency table within the basic model, implying that these
assumptions are correct for most of all possible configurations
following a common multinomial distribution (see also Krauth and
Lienert, 1973; Lienert and Krauth, 1975). A problem that may occur
is that certain patterns of missing values appear unexpectedly
frequently or rarely. Such cases may be theoretically plausible, e.g.
that students filling in a questionnaire will skip whole scales rather
than single items within a scale. This could result in extreme
groups, producing either none or merely missing values. In order to
account for this problem of ‘structurally’ extreme cell frequencies,
German
Mathematics
Biology
Physics
Chemistry
−1.0 −0.5 0.0 0.5 1.0
logits
possibly biasing the results of significance testing of the other cells,
Victor (1989) suggested to anticipate the existence of certain
configurations as ‘types’ within the definition of the base model.
This approach is generally discussed under the term ‘functional
CFA’ (Eye and von Mair, 2001). In our results, we took into account
the two most prevalent types of missing data patterns (‘omit all
items’ vs. ‘omit no item’) and conducted a subsequent functional
CFA in order to test the assumption of a ‘quasi independence’ in the
presence of the two most prevalent types of missing data patterns.
This analysis revealed no significant global tests (LR = 0.09, df = 2,
p = .96; x2 = 0.09, df = 2, p = .96), while the local tests resulted in no
additional significant types or antitypes (pbinomial-test > .4, for all
patterns; aBonferroni adjusted = .00625) besides the two anticipated
structural types.
Taking into account both these results and those from the
Rasch-residual factor analysis, a one-dimensional truancy WLE-
scale for science was formed, consisting of the three truancy items
referring to the science subjects of biology, chemistry and physics.
Additionally, a variable indicating missing values on this scale (i.e.
skipping all three truancy items) was defined.
The second step of our missing data analyses was to repeat the
first-order and functional CFA for the remaining subject-specific
truancy items. In analogy to the analyses of the science-subjects, a
first-order and functional CFA analysis was conducted including the
missing indicator variables for the three subject specific truancy
items (German, mathematics and PE) and the science-truancy WLE-
scale, derived from the preceding analyses described above. The
first-order CFA now including four variables indicating missing
values and thus accounting for 16 possible patterns confirmed the
two significant types of missing patterns previously found for the
three science-subjects. Students were either omitting no items at all
and had only valid data on the truancy scale (pattern ‘0000’;
fobserved = 3882, x2z-approximated ¼ 45:794, pBonferroni adjusted < .003125)
or they omitted all items and had only missing data on the truancy
scale (pattern ‘1111’; fobserved = 1059, x2z-approximated ¼ 318:108,
 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113 109

pBonferroni adjusted < .003125). Again, all the other patterns where
classified to be ‘antitypes’ (fobserved < fexpected). The subsequent
functional CFA, anticipating these two most prevalent missing
data patterns resulted in no additional significant missing
patterns (0.02 < pbinomial-test > .44, for all patterns; aBonferroni
adjusted = 0.0031).
To summarize the results of the person-centered approach of
the missing data analyses, one can state that the tendency to omit
any of the six truancy items can be assumed to be independent
from the subject addressed in the respective item.
In order to investigate whether the students’ tendency to omit
truancy items is related to their performance in the PISA
competence test, bivariate correlations of omitting truancy items
and the respective PISA proficiency scale were calculated. The
correlations between any of the indices for truancy item omission
and any of the three PISA competence domains range between
r = .23 and r = .36 (see Table 1 in the previous section) and thus
corresponds to a shared variance of at least 5% up to 12%. This can
be interpreted as a substantial relation between the omission of
truancy items and achievement on the PISA competence scales.
Given that we found two types of missing data patterns for each
of the subject-specific truancy items plus substantial correlations
between omitting subject-specific truancy items and student
proficiency in the respective PISA competence domain, we needed
to account for both the rather large proportion of students
systematically skipping the truancy items and the relationship of
skipping these items and achievement. Hence, omitting truancy
items is explicitly modeled in the regression analyses testing our
second hypothesis.
4.4. Hypothesis 2, subject-specific truancy and achievement in the
respective PISA domain
Our regression analyses indicated that for the three PISA
competence domains, all control variables are significant pre-
dictors. Girls outperform boys in reading, while boys do better in
mathematics and science. This latter finding is likely to be
attributable to the inclusion of attending a Gymnasium school
in the regression analysis, since in the overall German PISA sample,
no gender differences in science are found (OECD, 2014b). Students
with an immigrant background achieve significantly lower than
their classmates in both assessment domains and students’ socio-
economic status is positively correlated to their achievement.
Taking into account subject-specific truancy, we see that students
skipping mathematics classes scored between 21 and 39 points
less on the PISA mathematics scale than students who do not skip
mathematics classes (Table 3). According to a German longitudinal
study from PISA 2003, this discrepancy corresponds to one to two
school-years of learning (Ehmke et al., 2006). Given that there is a
wide consensus about schools being the main learning environ-
ment for mathematics, this finding is not surprising.
It becomes, however, highly relevant when compared to the
association of skipping German lessons and reading achievement
in PISA 2012. Assuming that there are numerous opportunities to
acquire and practice reading outside school, we see that even
missing German classes is significantly associated with reading
competence in PISA 2012 (Table 4). However, this correlation is
only significant for occasional truancy in German lessons, not for
more frequent absences (three times and more). With regard to the
PISA-shock that emerged after the results of the first PISA study in
2000 were published and more than one in five 15 year-old
students in Germany was classified ‘‘at risk’’ due to an alarmingly
low competence level in reading (Baumert et al., 2001), this finding
is highly relevant. The development of reading proficiency is
apparently not completed at the primary school level, it is
continued at the secondary level and German lessons may offer
a significant opportunity for improving one’s reading proficiency.
At the same time, students weak in reading may skip German
lessons on purpose in order to avoid failure when asked to read and
work with a text more frequently than accomplished readers. This
differential finding may be explained by the fact that even though
there are numerous opportunities to practice reading and improve
one’s proficiency by reading for pleasure (Aarnoutse and van
Leeuwe, 1998), students skipping German classes on purpose make
differential use of these opportunities. The pattern found in Table 4
indicates that low reading competence is associated with
occasional truancy in German lessons (‘‘once or twice’’), but not
with more frequent truancy (‘‘three times and more’’).
For science lessons as an aggregate of biology, chemistry and
physics, we find that skipping science classes also correlates with
low achievement (Table 5). As opposed to the correlation of
attending German lessons and PISA reading competence, skipping
science lessons on purpose is only negatively associated with
students’ achievement in the PISA science test when they skip
classes at least three times (B = 37.55; SE = 12.20).
In sum, our findings suggest that being present in mathematics,
German and science lessons correlates with a higher competence
in those assessment domains. Skipping lessons in these subjects
purposefully goes along with a lower proficiency level in the
corresponding assessment domains.

Table 3
Regression model: covariates, truancy in mathematics and PE (dummy coded), predicting PISA mathematics competence.

Variable Model 1: individual Model 2: subject-specific Model 3: subject-specific

characteristics truancy truancy and skipping PE

B SE B SE B SE

Intercept 424.96*** 4.93 440.18*** 4.37 440.21*** 4.36

Gymnasium 105.51*** 3.89 101.17*** 3.88 101.18*** 3.89
Male 21.22*** 2.22 23.01*** 2.16 23.02*** 2.14
Immigrant status 23.38*** 5.00 27.23*** 4.24 27.43*** 4.24
HISEI 0.82*** .08 0.75*** .07 0.75*** .07
Truancy Math omitted 39.91*** 4.31 26.99* 11.87
Truancy Math (‘‘Once or twic’’) 20.97*** 5.62 21.05*** 6.08
Truancy Math (‘‘Three times and more’’, collapsed) 39.68*** 7.55 39.93*** 8.16
Truancy PE omitted 13.21 12.15
Truancy PE (‘‘Once or twic’’) 0.22 4.14
Truancy PE (‘‘Three times and mor’’, collapsed) 0.74 5.93
R2 .52 .59 .64

Note: *PE = physical education.

*
p < .05.
**
p < .01.
***
p < .001.
HISEI values are z-standardized.
110 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113

Table 4
Regression model: covariates, truancy in German and PE (dummy coded), predicting PISA reading competence.

Variable Model 1: individual Model 2: subject-specific Model 3: subject-specific

characteristics truancy truancy and skipping PE

B SE B SE B SE

Intercept 454.11*** 4.90 471.09*** 4.08 471.30*** 4.20

Gymnasium 95.89*** 3.82 91.33*** 3.68 91.22*** 3.70
Male 37.10*** 2.12 35.17*** 2.04 35.19*** 2.06
Immigrant status 17.62*** 4.28 23.18*** 4.10 23.32*** 4.11
HISEI 0.77*** .08 0.69*** .07 0.69*** .07
Truancy German omitted 46.89*** 4.33 25.88* 11.47
Truancy German (‘‘Once or twic’’) 23.18*** 5.54 22.71*** 5.61
Truancy German (‘‘Three times and more’’, collapsed) 22.13** 9.44 20.06 10.19
Truancy PE omitted 21.55 11.76
Truancy PE (‘‘Once or twic’’) .81 3.54
Truancy PE (‘‘Three times and more’’, collapsed) 3.01 5.03
R2 .56 .65 .74

Note: *PE = physical education.

*
p < .05.
**
p < .01.
***
p < .001.
HISEI values are z-standardized.

Taking into account that all our regression models comprise a
dummy variable to split students attending a Gymnasium school
(merely academic track) from non-Gymnasium schools, the overall
high proportion of explained variance indicated by the R-square is
not surprising. Given that students in Germany are sorted into
different secondary school tracks at age 10 after four years of
elementary schooling due to their prior achievement, the variance
in student achievement between schools is about as high as the
overall variance within schools. Gymnasium schools have very low
between-school variance (6 percent; Sälzer et al., 2013), while the
overall within-school variance is 47 percent. For this reason, the
stratification during the sampling procedure needed to take into
account the difference between Gymnasium schools and non-
Gymnasium schools which are confirmed in our regression
analyses.
5. Discussion
This study aimed on answering two research questions: first,
we investigated the dimensionality of truancy with regard to
subject-specificity of truant behavior and second, we explored
whether subject-specific truancy relates to student achievement in
the corresponding competence domains in the PISA 2012 test. We
found that truancy measured by the truancy scale we administered
is a four-dimensional construct, yielding three single-subject
dimensions (PE, German, mathematics) and one three-subject
dimension (science, composed of biology, chemistry and physics).
This indicates that truancy is indeed a subject-specific and hence a
multidimensional behavior, where PE-lessons are the most
frequently skipped subject. Furthermore, our results point out
that skipping classes in German, mathematics and science is
related to a significantly lower competence level according to the
PISA 2012 reading, mathematics and science test, respectively. In
other words, students selecting themselves into the truancy-group
achieve lower in the respective subject domain than other
students. We are able to describe this group of students as a
self-selective one, since the PISA test is mandatory in Germany and
students who are absent at the PISA testing day will have to
participate in a follow-up session. Those who are finally absent will
be controlled for by a weighting procedure comprising non-
response at both the school and the student level. It can thus be
assumed that our data are not biased by students who are likely to
score high on truancy have missed the test session. Collecting data
on a topic sensitive to be biased by absenteeism in the context of a
publicly renowned and mandatory large-scale assessment is very
promising with regard to the quality of the data.

Table 5
Regression model: covariates, truancy in science and PE (dummy coded), predicting PISA science competence.

Variable Model 1: individual Model 2: subject-specific Model 3: subject-specific

characteristics truancy truancy and skipping PE

B SE B SE B SE
*** *** ***
Intercept 447.97 5.06 464.89 4.40 465.07 4.35
Gymnasium 96.90*** 3.86 92.47*** 3.76 92.44*** 3.75
Male 6.42* 2.46 8.25** 2.44 8.19** 2.41
Immigrant status 33.94*** 4.60 39.04*** 4.38 39.11*** 4.38
HISEI 0.81*** .08 0.73*** .07 0.73*** .07
Truancy Science omitted 46.47*** 4.69 32.96* 14.40
Truancy Science (‘‘Once or twic’’) 9.62* 4.01 9.34* 4.69
Truancy Science (‘‘Three times and more’’, collapsed) 37.87** 11.38 37.55** 12.20
Truancy PE omitted 13.80 14.49
Truancy PE (‘‘Once or twic’’) 0.56 3.89
Truancy PE (‘‘Three times and more’’, collapsed) 0.34 6.58
R2 .49 .57 .64

Note: *PE = physical education.

*
p < .05.
**
p < .01.
***
p < .001.
HISEI values are z-standardized.
 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113
One finding that we had not hypothesized was the role of
omitting truancy items. Modeling omitted items in contrast to the
other response categories in our regression analyses revealed some
noticeable predictive patterns. Our findings suggest that students’
proficiency in PISA mathematics is about 21–39 score points less
for those who truant mathematics classes – and about the same
amount less for students who skip the item measuring truancy in
mathematics classes (26 points, see Table 3). This finding is all the
more prominent since the regression model shows that skipping
the truancy items referring to physical education classes are not
predicting students’ performance in the PISA mathematics test.
Furthermore, we assume that there is a considerable group of
students who skip mathematics, but refuse to answer the
corresponding truancy items. Since students who omit the
mathematics truancy item achieve about as low as those who
choose the lower of two response categories indicating truancy in
mathematics and there can be no overlap between these groups,
this conclusion seems plausible.
With regard to the reading proficiency, skipping German classes
is associated with the students’ reading achievement only for
students who skip German lessons occasionally (once or twice). For
students who chose the most extreme category of the item
measuring skipping German classes, there is no association with
reading competence in the PISA test. Since a key feature of reading
competence is that it is ubiquitous in the sense that it is not limited
to German as a school-subject, this association needs to be
interpreted with care. On the one hand, German lessons offer the
main opportunity to acquire and deepen reading competence,
since both text comprehension and reading aloud is practised. On
the other hand, reading takes place in most other school subjects as
well, so the link between a lower reading proficiency score and
missed German classes is not obvious. A tentative interpretation of
this finding is – along with the research literature on reading
competence – that reading is only partially learned and acquired at
school (e.g. Aarnoutse and van Leeuwe, 1998; Artelt et al., 2007;
Schiefele et al., 2004). Reading is mostly learned and improved
through practice which is sought by students according to their
interest and motivation for reading for pleasure. It seems plausible
that those students who are accomplished readers are not
dependent from attending German lessons in order to develop
their proficiency in reading. As for the other two competence
domains, skipping the truancy item referring to PE lessons is not
predicting students’ reading achievement in the PISA test.
Finally, students skipping science lessons only once or twice have
a similar level of proficiency resulting from the PISA test to their
classmates who never skip science lessons. One very likely
explanation to this finding lies in the structure of science lessons
in Germany. More than in other school subjects, science lessons are
organized in topic-related blocks comprising several weeks.
Students get familiar with conceptual characteristics of such topics
and gain insights at different levels. Learning is frequently
stimulated and scaffolded by student activation through experi-
ments, research tasks or other activities involving interaction with
others. Thus, it seems plausible that students who skip only one or
two science lessons will be updated by their classmates while in the
process of learning together. In such cases, the student can catch up
with what they missed without exposing themselves as truants in
front of the teacher, it is fairly easy to just dive back into the flow of
the lessons. However, with regard to students who skip more than
just one or two science lessons, the association with science
achievement is very strong. Those students who report to have
skipped science lessons to a higher degree in the current school year
achieve significantly lower than students who have not skipped
science lessons at all. Following the assumption above that science
lessons are organized in thematic blocks, this item measuring the
frequency of truancy could have identified a critical threshold
111
between unharmful skipping of lessons and the beginning of truancy
to be a ‘slippery slope’ into a problematic behavior jeopardizing
students’ achievement (Sälzer et al., 2012). While students omitting
the truancy item for science lessons achieve significantly lower than
non-truant students, omitting items related to truancy in PE is not
associated with students’ science achievement. In other words,
students skip the item measuring truancy in PE lessons regardless
their achievement in science. One tentative explanation could be the
differential role of science subjects within the curriculum. In some
school years or within-school streams, science is a core subject and
in others, it is not. Therefore it is quite unlikely to find response
patterns indicating an association of skipping truancy items and
achievement in science-related subjects.
Considering the findings described so far, it must be kept in
mind that truancy is a research topic that is commonly seen as very
relevant and interesting, yet hard to do sound empirical research
on. Hence, representative and large-scale studies hardly exist.
Enhancing the PISA 2012 study, which is representative for
Germany and a mandatory study, yielded an innovative and
differentiated dataset. Information on subject-specific truancy
could be collected and analyzed with regard to student achieve-
ment. As student achievement is measured by an IRT-based
competence test, one can assume to have highly objective, reliable
and valid data at an aggregate level. In order to understand the
meaning of these results, there is a number of limitations to our
study that need to be considered.
First of all, PISA is designed as a cross-sectional study and does
not yield causal results or effects in a sense that one could
determine antecedents and consequences of certain variables. PISA
is a regular educational monitoring with a benchmarking function
and every three years, the cohort of 15 year-olds is tested in
numerous countries. This cyclic design allows for the description of
trends over time. However, the enhancement of the German
student questionnaire has uniquely been part of PISA 2012 and is
not available over time or across countries.
Second, we do not have information about the number of
lessons missed due to illness or other reasons that are not within
students’ control. Actually, when the role of missed lessons for
student achievement is to be investigated, one should differentiate
between classes missed to exogenous factors such as emergencies
or illness and classes missed on purpose, which are defined as
truancy. In our study, we clearly interpret our findings with regard
to intentionally skipped classes that are explicitly classified as
truancy. In future studies, however, other reasons for missing
school classes should be controlled for.
Third, our data lack information about the amount of time
students spend with reading for pleasure. Arguing that the
significant role of attending German lessons for the reading
competence of 15 year-old students is surprising given that there
are numerous opportunities to practice reading outside school, our
study could not control for this aspect statistically.
Fourth, our data have only been collected in one country and
only once. Information from students in several countries would be
useful to find cross-cultural differences and possible variables
affecting students’ choice to skip classes in certain subjects.
Repeated measures and hence a longitudinal design would allow
for causal analyses. There would still be a non-negligible selection
bias of students selecting themselves into the group of truants (as
opposed to other students who do not) that could not be solved by
a repeated survey. Depending on the research questions to be
answered, methods controlling for selection bias (e.g. the bounds
methods or propensity score matching) are appropriate.
In sum, our study sheds light on the dimensionality of truancy
as a construct describing a subject-specific student behavior and
our findings reveal an association between subject-specific truancy
and a significantly lower competence level in the corresponding
112 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113

PISA assessment domains. Considering the limitations mentioned truancy is an innovative approach and our data set is of excellent
above, the study is one step toward knowing more about truancy quality. But the limitations reveal that some necessary next steps
as an unwanted student behavior. Applying the Rasch-residual toward understanding truancy are already sketched and need to be
factor analysis to a six-item scale measuring subject-specific addressed.

Appendix 1

Zero-order correlations of all predictors used in the regression model predicting student competence in PISA mathematics.

Mathematics Gymnasium Male Immigrant SES Truancy Truancy Truancy Truancy Truancy Truancy
status Math Math Math PE PE (‘‘once PE (‘‘three
omitted (‘‘once (‘‘three omitted or twic’’) times and
or twic’’) times and more’’)
mor’’)

Mathematics – .59 .07 S.15 .39 S.23 S.02 S.06 S.23 .01 S.03
Gymnasium .59 – S.07 S.07 .39 S.14 S.00 S.03 S.13 .02 S.03
Male .07 S.07 – S.01 S.01 .05 .01 .03 .04 S.09 S.03
Immigrant status S.15 S.07 S.01 – S.20 S.06 .02 .02 S.07 .03 .02
SES .39 .39 S.01 S.20 – S.11 S.00 S.00 S.11 .01 S.00
Truancy Math omitted S.23 S.14 .05 S.06 S.11 – S.11 S.08 .97 S.15 S.11
Truancy Math (‘‘once or twic’’) S.02 S.01 .01 .02 S.00 S.11 – S.03 S.10 .18 .14
Truancy Math (‘‘three times S.06 S.03 .03 .02 S.00 7 S.03 – S.08 .03 .36
and more’’)
Truancy PE omitted S.23 S.13 .04 S.07 S.11 .97 S.10 S.08 – S.16 S.12
Truancy PE (‘‘once or twice’’) .01 .02 S.09 .03 .01 S.15 .18 .03 S.16 – S.07
Truancy PE (‘‘three times and S.03 S.03 S.03 .02 S.00 S.11 .14 .36 S.12 S.07 –
more’’)

Appendix 2

Zero-order correlations of all predictors used in the regression model predicting student competence in PISA reading.

Reading Gymnasium Male Immigrant HISEI Tru. Truancy Truancy Truancy Truancy Truancy
status German German German PE PE (‘‘once PE (‘‘three
omitted (‘‘once or (‘‘three omitted or twice’’) times and
twice’’) times and more’’)
more’’)

Reading – .58 S.24 S.12 .38 S.29 S.03 S.04 S.29 .05 S.01
Gymnasium .58 – S.07 S.07 .39 S.13 S.00 S.03 S.13 .02 S.03
Male S.24 S.07 – S.01 S.01 .04 .02 .04 .04 S.09 S.03
Immigrant status S.12 S.07 S.01 – S.12 S.07 .01 .02 S.07 .03 .02
HISEI .38 .39 S.01 S.20 – S.11 .00 S.00 S.11 .01 S.00
Truancy German omitted S.29 S.13 .04 S.07 S.11 – S.10 S.07 .98 S.16 S.11
Truancy German (‘‘once or twic’’) S.03 S.00 .02 .01 .00 S.10 – S.02 S.09 .20 .15
Truancy German (‘‘three times and mor’’) S.04 S.03 .04 .020 S.00 S.07 S.02 – S.06 .02 .37
Truancy PE omitted S.29 S.13 .04 S.07 S.11 1 S.09 S.06 – S.16 S.12
Truancy PE (‘‘once or twic’’) .05 .02 S.09 .03 .01 S.16 .20 .02 S.16 – S.07
Truancy PE (‘‘three times and more’’) S.01 S.03 S.03 .02 S.00 S.11 .15 .37 S.12 S.07 –

Appendix 3

Zero-order correlations of all predictors used in the regression model predicting student competence in PISA science.

Science Gymnasium Male Immigrant HISEI Truancy Truancy Truancy Truancy Truancy Truancy
stauts Science Science Science PE PE (‘‘once PE (‘‘three
omitted (‘‘once or (‘‘three omitted or twice’’) times and
twice’’) times and more’’)
more’’)

Science – .56 S.00 S.18 .38 S.26 .04 S.06 S.26 .02 S.02
Gymnasium .56 – S.07 S.07 .39 S.14 .06 S.04 S.13 .02 S.03
Male S.00 S.07 – S.01 S.01 .05 S.01 .03 .04 S.09 S.03
Immigrant status S.18 S.07 S.01 – S.20 S.07 .00 .04 S.07 .03 .02
HISEI .38 .39 S.01 S.20 – S.11 .04 S.02 S.11 .01 S.00
Truancy Science omitted S.26 S.14 .05 S.07 S.11 – S.16 S.07 .98 S.16 S.11
Truancy Science (‘‘once or twic’’) .04 .06 S.01 .00 .04 S.16 – S.04 S.15 .21 .25
Truancy Science (‘‘three times and mor’’) S.06 S.04 .03 .04 S.02 S.07 S.04 – S.06 .05 .28
Truancy PE omitted S.26 S.13 .04 S.07 S.11 .98 S.15 S.06 – S.16 S.12
Truancy PE (‘‘once or twic’’) .02 .02 S.09 .03 .01 S.16 .21 .05 S.16 – S.07
Truancy PE (‘‘three times and more’’) S.02 S.03 S.03 .02 S.00 S.11 .25 .28 S.12 S.07 –
 C. Sälzer, J.-H. Heine / International Journal of Educational Development 46 (2016) 103–113
References
Aarnoutse, C., van Leeuwe, J., 1998. Relation between reading comprehension,
vocabulary, reading pleasure and reading frequency. Educ. Res. Eval. 4,
143–166.
Artelt, C., McElvany, N., Christmann, U., Richter, T., Groeben, N., Köster, J., Ring,
K., 2007. Förderung von Lesekompetenz: Expertise. Bildungsforschung, vol.
17. Bundesministerium für Bildung und Forschung (BMBF), Bonn, Berlin.
Baumert, J., Klieme, E., Neubrand, M., Prenzel, M., Schiefele, U., Schneider, W.
(Eds.), 2001. PISA 2000. Basiskompetenzen von Schülerinnen und Schülern
im internationalen Vergleich. Leske + Budrich, Opladen.
Choppin, B., 1968. Item bank using sample-free calibration. Nature 219 (5156),
870–872, http://dx.doi.org/10.1038/219870a0.
Ehmke, T., Blum, W., Neubrand, M., Jordan, A., Ulfig, F., 2006. Wie verändert sich
die mathematische Kompetenz von der neunten zur zehnten Klassenstufe?
In: Prenzel, M., Baumert, J., Blum, W., Lehmann, R., Leutner, D., Neubrand,
M., Schiefele, U. (Eds.), PISA 2003. Untersuchungen zur
Kompetenzentwicklung im Verlauf eines Schuljahres. Waxmann, Münster,
pp. 63–86.
Erikson, R., Goldthorpe, J.H., Portocarero, L., 1979. Intergenerational class
mobility in three Western European societies: England, France and Sweden.
Br. J. Sociol. 30, 341–415.
Eye, A., von Mair, P., 2001. A Functional Approach to Configural Frequency
Analysis (Technical Report No. 48). Vienna.
Fischer, B., 1989. Einstellungen zum Sport. Eine empirische Untersuchung über
Einstellungen von Schülern zu Lernfeldern des Sportunterrichts in der
Sekundarstufe II. Waxmann, Münster.
Ganzeboom, H.B.G., 2010, May. A New International Socio-economic Index (ISEI)
of Occupational Status for the International Standard Classification of
Occupation 2008 (ISCO-2008) Constructed with Data from the ISSP 2002–
2007; with an Analysis of Quality of Occupational Measurement in ISSP.
Paper Presented at Annual Conference of International Social Survey
Programme, Lisbon. Retrieved from: http://www.harryganzeboom.nl/isco08/
index.htm.
Heine, J.-H., 2014. pairwise: Rasch Model Parameters by Pairwise Algorithm R
Package Version 0.2.5. Retrieved from: http://CRAN.R-project.org/
package=pairwise.
Heine, J.-H., 2015. confreq: R-Project. Retrieved from: http://cran.r-project.org/
web/packages/confreq/index.html.
Heine, J.-H., Sälzer, C., Borchert, L., Sibberns, H., Mang, J., 2013. Technische
Grundlagen des fünften internationalen Vergleichs. In: Prenzel, M., Sälzer, C.,
Klieme, E., Köller, O. (Eds.), PISA 2012. Fortschritte und Herausforderungen
in Deutschland. Waxmann, Münster, pp. 309–346.
Heine, J.-H., Tarnai, C., 2015. Pairwise Rasch model item parameter REcovery
under sparse data conditions. Psychol. Test Assess. Model. 57 (1), 3–36.
Helmke, A., Schrader, F.-W., 2013. Angebots-Nutzungs-Modell. In: Wirtz, M.A.
(Ed.), Dorsch – Lexikon der Psychologie. 16th ed. Huber, Bern, pp. 147–148.
Henry, K.L., Huizinga, D.H., 2007. School-related risk and protective factors
associated with truancy among urban youth placed at risk. J. Prim. Prevent.
28 (6), 505–519.
Kline, T.J.B., 2005. Psychological Testing: A Practical Approach to Design and
Evaluation. Sage, New York.
Krauth, J., 1973. Die Konfigurationsfrequenzanalyse (KFA) und ihre Anwendung
in Psychologie und Medizin: Ein multivariates nichtparametrisches
Verfahren zur Aufdeckung von Typen und Syndromen (Configural Frequency
Analysis (CFA) and its Application in Psychology and Medicine: A
Multivariate Nonparametric Method for Uncovering Types and Syndroms).
Alber Karl, Freiburg.
Krauth, J., Lienert, G.A., 1973. Nichtparametrischer Nachweis von Syndromen
durch simultane Binomialtests. Biom. Z. 15 (1), 13–20.
Lienert, G.A., Krauth, J., 1975. Configural Frequency Analysis as a statistical tool
for defining types. Educ. Psychol. Meas. 35 (2), 231–238.
113
Linacre, J.M., 1998. Detecting multidimensionality: which residual data-type
works best? J. Outcome Meas. 2, 266–283.
Lumley, T., 2014. Analysis of Complex Survey Samples. Retrieved from: http://
cran.r-project.org/package=survey.
Malcolm, H., Thorpe, G., Lowden, K., 1996. Understanding Truancy: Links
Between Attendance, Truancy and Performance. Scottish Council for
Research in Education, Edinburgh.
McArthur, D.L., Wright, B.D., 1985. Bruce Choppin on measurement and
education. Eval. Educ. 9 (1), 1–107, http://dx.doi.org/10.1016/0191-
765X(83)90005-8.
OECD., 2012. PISA 2009 Technical Report. OECD Publishing, Paris.
OECD., 2013a. PISA 2012 Assessment, Analytical Framework: Mathematics, Reading,
Science, Problem Solving and Financial Literacy. OECD Publishing, Paris.
OECD., 2013b. PISA 2012 Results: Ready to Learn: Students’ Engagement, Drive
and Self-beliefs (Volume III). OECD Publishing, Paris.
OECD., 2013c. PISA 2015 Draft Questionnaire Framework. Retrieved from: http://
www.oecd.org/pisa/pisaproducts/PISA-2015-draft-questionnaire-framework.
pdf.
OECD., 2014a. PISA 2012 Technical Report. OECD Publishing, Paris.
OECD., 2014b. PISA 2012 Results: What Students Know and Can Do: Student
Performance in Mathematics, Reading and Science, Revised ed. OECD
Publishing, Paris.
R Development Core Team., 2014. R: A Language and Environment for Statistical
Computing [Computer software], In: http://www.r-project.org/.
Reid, K., 2003. The search for solutions to truancy and other forms of school
absenteeism. Pastor. Care Educ. 21 (1), 3–9, http://dx.doi.org/10.1111/1468-
0122.00248.
Renzulli, J.S., Park, S., 2000. Gifted dropouts: the who and the why. Gifted Child
Quart. 44 (4), 261–271, http://dx.doi.org/10.1177/001698620004400407.
Sälzer, C., Prenzel, M., Klieme, E., 2013. Schulische Rahmenbedingungen der
Kompetenzentwicklung. In: Prenzel, M., Sälzer, C., Klieme, E., Köller, O.
(Eds.), PISA 2012. Fortschritte und Herausforderungen in Deutschland.
Waxmann, Münster, pp. 155–187.
Sälzer, C., Trautwein, U., Lüdtke, O., Stamm, M., 2012. Predicting adolescent
truancy: the importance of distinguishing between different aspects of
instructional quality. Learn. Instr. 22 (5), 311–319, http://dx.doi.org/10.1016/
j.learninstruc.2011.12.001.
Schiefele, U., Artelt, C., Stanat, P., Schneider, W. (Eds.), 2004. Struktur,
Entwicklung und Förderung von Lesekompetenz – Vertiefende Analysen im
Rahmen von PISA 2000. VS Verlag für Sozialwissenschaften, Wiesbaden.
Stamm, M., 2010. Dropouts am gymnasium. Z. Erzieh. 13 (2), 273–291, http://
dx.doi.org/10.1007/s11618-010-0121-z.
Stemmler, M., 2014. Person-Centered Methods. Springer International Publishing,
Cham.
Vaughn, M.G., Maynard, B.R., Salas-Wright, C.P., Perron, B.E., Abdon, A., 2013.
Prevalence and correlates of truancy in the US: results from a national
sample. J. Adolesc. 36 (4), 767–776, http://dx.doi.org/10.1016/
j.adolescence.2013.03.015.
Veenstra, R., Lindenberg, S., Tinga, F., Ormel, J., 2010. Truancy in late elementary
and early secondary education: the influence of social bonds and self-control –
the TRAILS study. Int. J. Behav. Dev. 34, 302–310, http://dx.doi.org/10.1177/
0165025409347987.
Victor, N., 1989. An alternative approach to configural frequency analysis.
Methodika 3, 61–73.
Wardle, J., 2002. Assessing socioeconomic status in adolescents: the validity of a
home affluence scale. J. Epidemiol. Commun. Health 56 (8), 595–599, http://
dx.doi.org/10.1136/jech.56.8.595.
Warm, T.A., 1989. Weighted likelihood estimation of ability in item response
theory. Psychometrika 54 (3), 427–450.
Wilson, V., Malcolm, H., Edward, S., Davidson, J., 2008. ‘Bunking off’: the impact
of truancy on pupils and teachers. Br. Educ. Res. J. 34, 1–17, http://
dx.doi.org/10.1080/01411920701492191.