Content uploaded by Manfred Schmitt
Author content
All content in this area was uploaded by Manfred Schmitt on Mar 28, 2014
Content may be subject to copyright.
International Journal of Public Opinion Research
ßThe Author 2014. Published by Oxford University Press on behalf of The World Association
for Public Opinion Research. All rights reserved.
doi:10.1093/ijpor/edu002
Measuring Traits and States in Public Opinion
Research: A Latent State–Trait Analysis of
Political Efficacy
Frank M. Schneider
1
, Lukas Otto
1
, Daniel Alings
1
and Manfred Schmitt
2
1
Department of Communication Psychology and Media Education, University of
Koblenz-Landau, Germany and
2
Department of Psychology, University of Koblenz-Landau
Abstract
Latent state–trait theory (LSTT) considers the fact that measurement does not take
place in a situational vacuum. LSTT decomposes any observed variable into a latent
state component and a measurement error component, and any latent state into a latent
trait component and a latent state residual representing situational influence and/or
interactional influences. Furthermore, it provides more precise reliability estimates than
common coefficients. This article introduces the basic concepts of LSTT, discusses its
usefulness for public opinion research, and applies LST models to panel data on pol-
itical efficacy from the 2009 German Longitudinal Election Study. The findings show
that internal efficacy is a rather trait-like disposition and external efficacy is significantly
due to situational and/or interactional influences.
The distinction between traits and states has a long history and goes back to
45 B.C., when Cicero distinguished between trait-anxiety and state-anxiety
(Eysenck, 1983). Not only are these concepts long-standing, they are still
relevant for psychology today; in fact, (personality) traits also play a crucial
role in public opinion research. Entire textbooks, such as Measures of Political
Attitudes by Robinson, Shaver, and Wrightsman (1999), are dedicated to the
measurement of relatively enduring and cross-situationally consistent person-
ality characteristics (i.e., traits). With regard to public opinion research, the
role of traits has been investigated in a vast number of studies: Besides the
usual, domain-specific constructs, such as attitudes towards controversial social
All correspondence concerning this article should be addressed to Frank M. Schneider, Institute of Media
and Communication Studies, University of Mannheim, Haus Oberrhein, Rheinvorlandstr. 5, 68159
Mannheim, Germany. E-mail: frank.schneider@uni-mannheim.de.
International Journal of Public Opinion Research Advance Access published March 5, 2014
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
issues (stem cell research; e.g., Becker, Dalrymple, Brossard, Scheufele, &
Gunter, 2010), partisanship (Bartels, 2000), political interest (Stro
¨mba
¨ck,
Djerf-Pierre, & Shehata, 2013), political trust (Catterberg & Moreno, 2006),
political efficacy (PE) (Lee, 2010), and nationalism (Davidov, 2011), public
opinion surveys also often include the Big Five (Gerber, Huber, Doherty,
Dowling, & Ha, 2010;Mondak, Hibbing, Canache, Seligson, & Anderson,
2010), need for cognition (Tsfati & Cappella, 2005), or values (Valenzuela,
2011)—just to name a few striking and recent examples.
The typical measurement of these constructs is usually based on (a) traditional
ideas of the psychological trait concept that encompasses temporal stability and
cross-situational consistency (Steyer, Schmitt, & Eid, 1999) and (b) assumptions of
classical test theory (CTT; Lord & Novick, 1968), especially that an observed
variable can be decomposed into a true score variable and a measurement error
variable. However, traditional approaches to the trait concept, such as the one
described above, have been repeatedly criticized (Allen & Potkay, 1981)andhave
led to several alternative conceptualizations: situationism (e.g., Mischel, 1968), the
use of aggregation to defend the trait concept (e.g., Epstein & O’Brien, 1985), the
moderator approach (e.g., Bem & Allen, 1974), or interactionism (e.g., Bowers,
1973;Endler & Magnusson, 1976). Despite this considerable debate, it is generally
understood that measurement alwaystakes place in a situation and that an observed
score is usually influenced not only by stable person-specific effects and measure-
ment error, but also by effects of the situation and effects of the person–situation
interaction. However, if situational and interactional effects substantially contribute
to a construct, it would be difficult to argue for conceptualizing it as a stable trait.
Therefore, it is important to examine whether constructs like, for instance, political
efficacy (PE) should be best conceptualized as more stable or more temporary
constructions. Moreover, distinguishing systematic situational and/or interactional
influences from unsystematic measurement errors is an important task in order to
determine an appropriate measurement model and estimate reliability coefficients
precisely.
An appropriate way to theoretically and methodologically take these con-
siderations into account is provided by latent state–trait (LST) theory (LSTT;
Steyer, Ferring, & Schmitt, 1992;Steyer et al., 1999) where complete stability
and complete transience are seen as two ends of a continuum rather than two
distinct categories. Thus, the major aim of analyses based on LSTT is to
examine whether an observed variable’s variance is more determined by tem-
porary situational or interactional influence, on the one hand, or more affected
by a perfectly stable trait, on the other hand.
1
1
Of course, there is a considerable amount of studies dealing with the stability of political constructs
(e.g., party identification; Clarke & McCutcheon, 2009). In such cases, sophisticated methods like latent
mixed Markov models have contributed to the debate on stability of partisanship, which is measured as a
categorical variable in unknown subgroups (latent classes) of the electorate. However, in LSTT continuous
variables are analyzed in a homogenous electorate and the distinction of situational and/or interactional
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
2
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
The aims of the present paper are to introduce the basics of LSTT, discuss
its usefulness for public opinion research, and apply it to the construct of PE in
a national panel survey. Although some elaborated studies have already dealt
with the stability of PE (e.g., Aish & Jo
¨reskog, 1990;Semetko & Valkenburg,
1998), the LST approach extends them by explicitly modeling the state–trait
distinction, thereby offering new insights into the properties of the PE construct.
An Introduction to Latent State–Trait Theory
LSTTisanextensionofCTT.InCTT,anobservedvariableY
i
in a test iis
decomposed into a true score variable t
i
and a measurement error variable e
i
(see
Figure 1a). The true score variable is defined as the expectation of the observed
score variable given the person. This definition implies a strictly trait-like perspec-
tive on psychological measurement. Systematic changes in the true score across
occasions of measurement and situations generating intraindividual oscillations of
a true state score around a true trait score are thus not considered in CTT. Rather,
any changes in observed scores are attributed to measurement error.
In CTT, the reliability of a test is defined as the ratio of true score
variance Var (t
i
) to the observed variance Var (Y
i
).
Rel Yi
ðÞ¼Var ti
ðÞ=Var Yi
ðÞ ð1Þ
This definition of reliability does not tell us which proportion of unstable vari-
ance is systematic and which is unsystematic. More specifically, it does not tell
us anything about which proportions of the measurement are determined by
trait, situation, or trait x situation interaction (cf. Deinzer et al., 1995).
In LSTT, the variables Y
it
(where idenotes the test or indicator and
tdenotes the occasion or measurement time) are assumed to be results of
random experiments. This means that a person was randomly drawn from a
population of persons and the situation in which this sampled person is
measured was randomly drawn from a population of situations. Now, the
true score t
it
is the expectation or true mean of the distribution of Y
it
that
refers to a person in a situation (cf. Steyer et al., 1999,p.394) or in other
words: The values of the true score variable are the conditional expectations of
Y
it
given a specific person and a specific situation in which this person is
when Y
it
is measured (see Steyer, Geiser & Fiege, 2012, pp. 292–293). In
LSTT, the true score variable is called latent state variable because it repre-
sents the true score of a person in a situation. It can be decomposed into a
latent trait component x
it
and a situational and/or interactional component z
it
that is called latent state residual (see Figure 1b). The measurement error
variable eit can be defined as the difference between the observed variable Y
it
influences and trait influences is theoretically taken into account. The interested reader is referred to Eid
(2002), who showed that LST models can be integrated into the latent mixed Markov modeling framework.
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 3
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
and the true score variable t
it
. Therefore, eit represents that part of Y
it
‘‘that is
not determined by the person, nor by the situation, nor by the interaction
between person and situation’’ (Steyer et al., 1999,p.395). Some logical
consequences follow from these definitions and require no further assump-
tions. For example, the expectations of measurement error variables and of
latent state residuals are zero. Furthermore, the definitions imply that the
correlations between measurement error variables and other latent variables
(i.e., true score variables, latent trait variables, and latent state variables per-
taining to the same measurement occasion) are zero as well as latent trait
variables and latent state residuals are uncorrelated (for more details on the
formal description of the basic concepts and their implications, see Steyer,
Geiser & Fiege, 2012, pp. 292–293).
As in CTT, the variance of the observed variable is decomposed into the
variance of the true score variable and the variance of the measurement error
variable. Additionally, the variance of the true score variable, which defines
the latent state in LSTT, is decomposed into stable trait variance and sys-
tematic occasion-specific variance. This variance is called the latent state
residual variance and reflects systematic effects of the situation and the per-
son–situation interaction on the observed variable:
Var Yit
ðÞ:¼Var tit
ðÞ þVar eit
ðÞ
¼Var xit
þVar zit
þVar eit
ðÞ ð2Þ
As can be seen from Equation 2, every component is defined by an instru-
ment and an occasion. Hence, to estimate the parameters some restric-
tions must be made. Throughout this paper, we will focus on the most
commonly used model of LSTT, the LST model with method factors
Figure 1
Decomposition of the observed variable Y
i
into a true score variable
i
and a measurement
error variable "
i
according to CTT (a) and decomposition of the observed variable Y
it
into
an error component "
it
and a latent state component that can be further decomposed into a
latent trait component
it
and a latent state residual
it
according to LSTT (b)
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH4
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
(or singletrait-multistate-multimethod model; for further models see Steyer et al.,
2012). This model acknowledges that each method generates unique variance that
is not shared by other methods. The model contains (a) a latent trait xthat
reflects perfectly stable individual differences in the measured construct and is
common to all measurement instruments and occasions (thus without index), (b)
occasion-specific factors z
t
that reflect systematic situational influences on meas-
urement occasion tand are common to all instruments, (c) method-specific fac-
tors x
i
that reflect systematic variance of the measurement instrument iand are
common to all occasions. Readers with an interest in mathematical details are
referred to Steyer et al. (2012).
LSTT defines several coefficients: common consistency (cCon), occasion
specificity (Spe), method specificity (mSpe), and reliability (Rel). Their def-
initions are as follows (Equations 3–6; for more details see Steyer et al., 1992):
cCon Yit
ðÞ¼Var ðxÞ=Var Yit
ðÞ,ð3Þ
Spe Yit
ðÞ¼Var ðztÞ=Var Yit
ðÞ,ð4Þ
mSpe Yit
ðÞ¼Var xi
=Var Yit
ðÞ,ð5Þ
Rel Yit
ðÞ
¼Var x
ðÞ
þVar zt
þVar xi
=Var Yit
ðÞ
¼Var tt
ðÞ
=Var Yit
ðÞ
¼cCon Yit
ðÞþSpe Yit
ðÞþmSpe Yit
ðÞ:
ð6Þ
All coefficients can be directly calculated from observed correlation or covari-
ance matrices (Hagemann & Meyerhoff, 2008;Steyer & Schmitt, 1990).
However, analyzing LST models within the confirmatory factor analysis
(CFA) framework offers the opportunity to test underlying assumptions of
the measurement model and obtain fit indices of competing models (e.g., Does
a restrictive parallel test model with equal effects of the latent trait xand equal
error variances fit the data well, or must we apply a less restrictive congeneric
model where factor loadings and error variances are free to vary in order to
achieve an adequate model fit?). It is important to have a good model fit
because otherwise parameters might be under- or overestimated, thereby lead-
ing to imprecise results and wrong interpretations. After this introduction to
the conceptual and formal propositions of LSTT, we will now describe how
public opinion researchers can benefit from using this theory.
Advantages of LST Modeling in Public Opinion Research
The distinction between states and traits appears to be as old as thinking about
human behavior (Steyer et al., 1999). Nevertheless, in political public opinion
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 5
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
research the trait model seems to be the predominant framework of analyses.
The first reason to apply LSTT to public opinion research is the possible
importance of situational influences on constructs of interest, and the determin-
ation of its stability and occasion specificity. If situational influences with
systematic effects on the construct on an occasion of measurement were con-
sidered irrelevant in survey data, there would be no reason to employ panel
designs.
However, situational influences seem to be important sources of variance
of political beliefs (e.g., Abramson, 1983). Many panel surveys have been
conducted to investigate the ‘‘effects of some specific event or a series of
events’’ (Lazarsfeld, 1948,p.405). Take as an example the PE construct,
which will be examined in more detail later in this article: It is a common
assumption that external efficacy (EE) can be influenced by political events
like elections; at the same time, the reliability of PE instruments has been
found to be rather low especially for EE (Semetko & Valkenburg, 1998).
Election effects and a lack of reliability would both cause a rather low test–
retest correlation of EE measures between two times of measurement. Unlike
models of CTT, appropriate models of LSTT distinguish these two causes of
correlational instability.
As a second example, consider moods and emotions, which are part of
many large-scale surveys (e.g., emotions towards politicians in the American
National Election Study [ANES]). It would be very surprising if these con-
structs showed the same correlational stability as personality traits. The same
is true for interest in election campaigns and the image of a politician. Like
moods and emotions, these constructs seem to be rather sensitive to situational
influences (Hacker, 2004). For other constructs such as political party identi-
fication or political interest, the assumption of substantial correlational stability
over time seems more plausible. For instance, although some scholars have
identified (short-term) influences on political interest like media use
(Boulianne, 2011), a recent analysis of eleven panel studies in four different
countries, spanning over 40 years showed that political interest is ‘‘exception-
ally stable’’ (Prior, 2010,p.747).
By using LSTT we are able to model such assumptions, determine the
systematic and stable as well as the systematic but unstable differences be-
tween individuals, and thus empirically test whether the examined constructs
are rather transient and sensitive to situational influences or rather stable like
personality traits.
Furthermore, the coefficients provided by LSTT allow for a more precise
estimate of an instrument’s reliability than traditional coefficients used in
public opinion research. As mentioned above, test–retest correlation may not
be appropriate to measure reliability since this estimator is not able to distin-
guish between the stability of a variable and the reliability of a measure.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH6
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
In other words, for constructs that are strongly situation-dependent, i.e., show
high occasion specificities, high test–retest correlations cannot be expected, yet
the measures could still be perfectly reliable (Steyer, Eid, & Schwenkmezger,
1997). One of the strengths of LSTT when compared to CTT is that reli-
ability is not just cross-situational consistency, but state measures can be
highly reliable although their test–retest correlations may be close to zero.
Of course, there are alternative approaches for estimating reliability. The
most popular is Cronbach’s alpha as a measure of internal consistency.
However, Cronbach’s alpha provides a precise measurement of reliability
only if items are essentially tau-equivalent—a precondition often overlooked
and violated in our case (Alwin, 2010). When requirements are not met,
Cronbach’s alpha may over- or underestimate the reliability of a certain
instrument. Moreover, as most scales in public opinion research are very
short due to economic reasons but often reflect heterogeneous content,
Cronbach’s alpha is often rather low, thus LSTT coefficients might help to
overcome some weaknesses of this reliability estimator (Cortina, 1993). In the
second part of this paper we will show that reliability coefficients of PE based
on LSTT differ from traditional reliability estimates like test–retest correlation
or Cronbach’s alpha. Thus, LST models can be a very useful way to inves-
tigate the reliability of instruments in public opinion research. Additionally,
we will empirically demonstrate that each argument mentioned above can be
applied to the measurement of PE in large-scale studies. We will discuss
consistency and occasion specificity of both dimensions of this construct and
the reliability of existing instruments used in national election studies.
Measurement of PE and Research Questions
There are two main reasons to analyze PE by means of LSTT: First, most
researchers agree that PE consists of two dimensions, internal efficacy (IE)
that refers to the self-image of the individual and external efficacy (EE) that
refers to the individual’s image of the democratic system (Lane, 1959,p.149),
however the question about the stability and specificity of these dimensions
remains unclear. Second, although PE has been ‘‘one of the most continuously
examined constructs’’ (Morrell, 2003,p.589), there is no consensus on its
measurement (Caprara, Vecchione, Capanna, & Mebane, 2009).
Initially, researchers believed that EE is more transient and more depend-
ent on current political events, e.g., elections (Campbell, Converse, Miller, &
Stokes, 1960). In contrast, IE is thought to be a more stable construct, which
is ‘‘...lying at a relatively ‘deep’ level in any hierarchy of dispositions ...’’
(Campbell et al., 1960,p.516). By analyzing the 1984 American pre- and
post-election studies, Acock and Clarke (1990)found support for different
stability. However, other researchers came to different results. For instance,
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 7
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
Abramson (1983)analyzed three sets of ANES panel data and concluded that
both EE and IE seemed to be relatively stable. Finally, current research has
shown that also IE may be affected by situational influences like deliberative
decision making or framing (Morrell, 2005;Pedersen, 2012). However, none of
these attempts to clarify stability issues applied analyses that are appropriate for
differentiating between trait, situational, and interactional influences. There
have been even a few researchers who do not clearly distinguish between IE
and EE (for an overview see Morrell, 2003,p.592). Following these arguments,
our first research question aims at exploring the stability of IE and EE:
RQ1: Are IE and EE more stable or more transient constructs?
The second reason to apply LSTT to PE is the measurement of reliability
for common PE-scales. Although the scale developed by Niemi, Craig, &
Mattei (1991)seems to be an appropriate measure to assess IE (Morrell,
2003), the measurement in other countries and the assessment of EE still
remain unclear with regards to psychometric properties. As mentioned earlier,
LST analyses provide us with further estimates of reliability, thus we are able
to precisely estimate the accuracy of current measures of IE and EE. Drawing
on the research mentioned above we address the second research question
concerning the reliability of PE-measures:
RQ2: Can we achieve more precise estimates of IE’s and EE’s reliability if
we apply appropriate measurement models (i.e., LST models)?
In the 2009 German Longitudinal Election Study (GLES) panel survey,
four items measuring IE and three items for EE were used after previous
testing (cf. Vetter, 1997). We will now analyze these items by means of
LST models. We will also take into account the theoretical assumptions con-
cerning the stability of IE and EE mentioned in this section.
Latent State–Trait Analysis of Political Efficacy Data from
the 2009 GLES
Method
Sample. Data were obtained from the short-term campaign panel, which was
one of eleven components of the 2009 GLES (Rattinger, Roßteutscher, Schmitt-
Beck, & Weßels, 2009) and included seven waves—six before and one after the
federal election of the German parliament.
2
The GLES short-term campaign
panel was conducted as an online-survey based on quota-sampled respondents
from an online-access panel and aimed to track individual changes rather than to
make generalizations concerning the whole German electorate. However, for our
2
For further details, please visit the respective website: http://www.gles.eu/design.en.htm.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
8
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
purposes, the most important features of this study are the inclusion of more
than one indicator of PE in more than one wave as well as the within-subjects
design. We assessed IE and EE in Wave 1and Wave 7.
Wave 1(t1) was the first pre-election wave of the GLES short-term
campaign panel and conducted as an online survey between July 10,2009
and July 20,2009. It included data from 3,771 respondents and had a par-
ticipation rate—defined as the number of respondents who provided a usable
response divided by the total number of initial personal invitations requesting
participation (American Association for Public Opinion Research, 2011,
p. 38)—of 30%. Wave 7(t2) was the post-election wave and conducted be-
tween September 29,2009 (two days after the Bundestag election) and
October 7,2009. It included data from 2,658 respondents and had a partici-
pation rate of 55%.
Measures. IE was measured at t1and t2with four items (‘‘People like me
don’t have any say about what government does’’ [NOSAY]; ‘‘Sometimes
politics and government seem so complicated that a person like me can’t
really understand what’s going on’’ [COMPLEX]; ‘‘I consider myself well-
qualified to participate in politics’’ [ACTIV]; ‘‘I feel that I have a pretty good
understanding of the important political issues facing our country’’
[UNDERST]). EE was measured in t1and t2with three items: (‘‘I think
public officials care what people like me think’’ [CARE]; ‘‘There are not many
legal ways for citizens to successfully influence what the government does’’
[NOINFL]; ‘‘Representatives are always looking for a close contact with citi-
zens’’ [TOUCH]). Participants rated these items on 5-point scales ranging
from completely disagree to completely agree. Taken together, 2,070 respondents
answered all IE items, whereas 2,098 respondents answered all EE items.
These are the items used in the GLES. In the German context, they
are widely considered to be valid and reliable instruments to measure PE
(Vetter, 1997).
Procedure. To test the models described above, two parallel test halves
on two measurement points are required. For IE, after recoding NOSAY and
COMPLEX, both items were used to construct Test Half 1(IE
1
); Test Half 2
(IE
2
) was constructed by averaging ACTIV and UNDERST. For EE, the
item NOINFL was recoded. The first test half (EE
1
) comprises items
NOINFL and CARE. Item TOUCH defines the second test half (EE
2
).
For further information about the means, standard deviations, variances,
covariance, and correlations of the test halves, see Appendix A (available as
supplementary data).
Results
Test–retest correlations were .57 for IE
1
,.68 for IE
2
,.50 for EE
1
,and .47 for
EE
2
respectively. In addition, Cronbach’s alphas for the four items of IE on
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 9
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
each occasion were .62 and .66 for t1and t2, respectively. Cronbach’s alphas
for the three items of EE were .50 at t1and .62 at t2. None of these common
reliability estimators allows for differentiating trait, situation, and method
effects.
Model testing and variance components. Four types of models were
tested for both constructs, IE and EE, using EQS 6.1(Bentler & Wu, 2005):
(a) The latent trait model, (b) the latent trait model with method factors,
(c) the LST model, and (d) the LST model with method factors. Different
versions of these models that varied in their constraints on factor loadings and
error variances were tested (two EQS code examples are provided in Appendix
B, available as supplementary data). In the most restrictive model version,
factor loadings were fixed to 1, and error variances were set to be equal.
First, we present goodness of fit indices, variance components, and character-
istics for IE. Afterwards, analogous analyses for EE are provided.
Internal efficacy. The LST model with method factors, equal factor
loadings, and unconstrained error variances (Model IE4, see Figure 2) pro-
vided the best fit for IE (see Table 1).
However, as w
2
difference test indicates, there is no statistically significant
difference between this model and Model IE2(w
2
2
¼5.96,p¼.051). This is
due to small, but not statistically significant, latent state residual variances and
supports the theoretical assumption that IE is a stable, trait-like construct.
Figure 2
Latent state–trait model with method factors and variance for IE components (Model 4)as
indicated by superscript a, and for EE (Model 4) as indicated by superscript b. All vari-
ances are statistically significant at ¼0.05, except where indicated by superscript ns. For
additional information see text and Tables 1and 2
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH10
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
By using Equations 3–6above, common consistency, occasion specificity,
method specificity, and reliability coefficients for the test halves on each
occasion can be computed from the variance components provided in Table 2.
As can be seen in Table 3, reliabilities for the test halves are rather low.
After aggregation across test halves (for more details and formulas, see
Deinzer et al., 1995;Steyer & Schmitt, 1990), they amount to .68 for t1
and .73 for t2, respectively. Both reliability estimates are higher than the
test–retest correlations and the Cronbach’s alphas for test halves, indicating
a more precise measurement when taking the specified variance components
into account. Note that the reliability estimates are even higher than the test–
retest correlation of the IE total score (r
tt
¼.65).
In the present analyses, we examined the reliability and stability of inter-
individual differences in IE. LST coefficients were calculated on the basis of
an LST model with method factors, which turned out to be nearly identical to
the latent trait model with method factors. First, all test halves demonstrate
rather low reliabilities across both measurement occasions. On average, about
70% of the variance of the total scales is systematic variance indicating that
the scales are reliable. Second, reliabilities are largely determined by stable
Table 1
Goodness of Fit Indices for Selected Models for IE and EE
Model w
2
df p w
2
/df RMSEA [90% CI] SRMR CFI
IE
IE1 604.78 8 .00 75.60 0.19 [.18,.20]0.10 0.77
IE27.43 3 .06 2.48 0.03 [.00,.05]0.01 1.00
IE3
a
604.78 6 .00 100.80 0.22 [.21,.23]0.10 0.77
IE41.47 1 .23 1.47 0.02 [.00,.06]0.01 1.00
EE
EE1 171.15 5 .00 34.20 0.17 [.11,.14]0.06 0.92
EE2 122.40 3 .00 40.80 0.14 [.12,.16]0.05 0.94
EE3 130.12 3 .00 43.67 0.14 [.12,.16]0.04 0.94
EE41.75 1 .19 1.75 0.02 [.00,.07]0.01 1.00
Note. N
IE
¼2,070;N
EE
¼2,098.
Model IE1: Latent trait model with equal factor loadings, equal error variances. Model EE1: Latent trait
model with equal factor loadings, unequal error variances. Model IE2/EE2: Latent trait model with method
factors, equal factor loadings, unequal error variances. Model IE3: Latent state–trait model with equal factor
loadings, equal error variances, unequal state residuals. Model EE3: Latent state–trait model with equal
factor loadings, unequal error variances, unequal state residuals (see Appendix B (available as supplementary
data) for an example of the EQS code). Model IE4/EE4: Latent state–trait model with method factors,
equal factor loadings, unequal error variances, unequal state residuals (see Appendix B (available as sup-
plementary data) for an example of the EQS code).
a
EQS automatically constrained the latent state residuals in Model IE 3at lower bound (i.e., fixed to zero to
prevent negative variances) without adjusting the degrees of freedom. Due to the fact that this model had
to be rejected anyways, we will not further discuss its misspecification. The interested reader is referred to
Chen, Bollen, Paxton, Curran, & Kirby (2001)for a discussion on improper solutions in structural equation
modeling.
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 11
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
interindividual differences (i.e., common consistencies were 54% on average).
Third, indicators contain moderate, significant proportions of test-half-specific
variance (i.e., method specificities were 16% on average). That is, the test
halves contain systematic variance in addition to common trait variance.
Table 2
Variance Components of Model 4for IE and EE
tiVar Yit
ðÞ Var tt
ðÞ Var xðÞ Var zt
Var eit
ðÞ Var xi
IE
11 0.81 0.29 0.29 0 0.37 0.16
20.81 0.29 0.29 0 0.27 0.25
21 0.71 0.29 0.29 0 0.27 0.16
20.75 0.29 0.29 0 0.21 0.25
EE
11 0.97 0.30 0.27 0.03 0.55 0.12
20.75 0.30 0.27 0.03 0.38 0.07
21 0.94 0.40 0.27 0.13 0.42 0.12
20.73 0.40 0.27 0.13 0.27 0.07
Note. N
IE
¼2,070;N
EE
¼2,098.t¼measurement point; i¼test half; Var (Y
it
)¼test half variance; Var
(t
t
)¼latent state variance; Var (x)¼latent trait variance; Var (z
t
)¼latent state residual variance; Var
(e
it
)¼measurement error variance; Var (x
i
)¼method-specific variance. Latent state residual variance for IE
was 0.02 for t¼1and 0.01 for t¼2, respectively. Both variance estimates are not statistically significant
(p>.050). All other variances are statistically significant at ¼.05.
Table 3
Reliability (Rel), Common Consistency (cCon), Occasion Specificity (Spe), and Method
Specificity (mSpe) Coefficients for IE and EE
ti Rel cCon Spe mSpe
IE
11 0.55 0.35 0 0.19
20.66 0.35 0 0.31
W0.68 0.52 0 0.16
21 0.62 0.40 0 0.22
20.72 0.38 0 0.34
W0.73 0.56 0 0.17
EE
11 0.44 0.28 0.03 0.13
20.49 0.36 0.04 0.10
W0.59 0.46 0.05 0.08
21 0.55 0.29 0.14 0.13
20.64 0.37 0.17 0.10
W0.70 0.44 0.19 0.07
Note. N
IE
¼2,070;N
EE
¼2,098.W¼total scales (aggregation across test halves iwithin occasions t; for
formulas see Deinzer et al., 1995 and Steyer & Schmitt, 1990)
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
12
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
Therefore, we cannot speak of strictly parallel test halves; they measure dis-
tinct facets of the IE construct. In line with theoretical assumptions, the
variance component that is due to systematic, but unstable effects of the
situation or interaction was small and statistically not significant. Obviously,
this can be interpreted in the sense of neither situational nor interactional
influence on IE efficacy.
External efficacy. While both latent trait models yielded unsatisfactory
model fit indices, the LST model with method factors, equal factor loadings,
and unconstrained error variances (Model EE4, see Figure 2) provided a very
good model fit for EE (see Table 1).
In contrast to IE, the variances of the latent state residuals for both times
of measurement (LSR
1
¼0.03; LSR
2
¼0.13) are significant at the 5% level.
Based on the data reported in Table 2and 3, the common assumption that the
EE dimension is more likely to be influenced by political events such as
elections is supported. It should be noted that reliabilities of the two test
halves are rather low (.68 for t1and .73 for t2, respectively), albeit they are
higher than the test–retest correlations and Cronbach’s alphas.
For EE, the LST model fits the data better than does the latent trait
model. The variances of the latent state residuals are statistically significant
and thus support theoretical assumptions and previous findings demonstrating
that EE is sensitive to political events. By contrast, IE seems to be deeply
rooted in a stable belief system and rather insensitive to political events
(Campbell et al., 1960,p.516; see also Aish & Jo
¨reskog, 1990;Semetko &
Valkenburg, 1998). Although our results clearly show that EE is sensitive to
the event of an election, this sensitivity is not pronounced. The average oc-
casion specificity amounts to .12 and is thus considerably smaller as compared
to the average common consistency of .45. This means that individual differ-
ences in EE were more strongly affected by stable beliefs than by the election
event and the interaction of this event with the trait. Next, the reliabilities
of the EE indicators were lower compared to the reliabilities of the IE indi-
cators. This result might reflect that only three items were available for
measuring EE.
Discussion and Conclusion
In the present paper we provided arguments for the relevance and usefulness
of LSTT and introduced its basic concepts. Using LSTT in public opinion
research has two benefits: 1) The consideration of situational influences, ana-
lyses of stability or specificity of a construct, and 2) precise estimates for the
reliability of an instrument. It was demonstrated that each of these benefits
apply to the measurement of PE.
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 13
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
By analyzing data from the 2009 GLES, we showed how decomposing an
observed measurement not only into trait and error components, but also into
situational and/or interactional as well as method-specific components can
foster our understanding of the construct’s nature. For instance, on the one
hand, our findings support the theoretical assumptions of conceptualizing IE
as a clearly trait-like, enduring construct and EE as a construct more suscep-
tible to situational or interactional influences, especially in times of political
election campaigns. The fact that both constructs vary differently over time
emphasizes the importance to speak about EE and IE as different constructs as
Campbell et al. (1960)assumed over 50 years ago. At this point, it is import-
ant to note that a significant event (the federal election) occurred between the
measurement occasions. Thus, it seems possible that the act of voting is
indeed able to change EE but not IE as assumed by Campbell et al. (1960).
However, due to the fact that our design is limited to only two measurement
occasions (one before and one after the election) it is not possible to separate
state variability from trait change.
3
Thus we cannot say whether the effects are
due to (1) a more transient nature of the construct (e.g., answering EE items
depends more on occasion-specific effects due to varying recent experiences
during exposure to election campaigns, political news etc. that are likely to be
reversed, thus reflecting only temporary changes), (2) trait change (e.g., EE
trait changes between specific periods of time due to political learning while
being stable within other periods of time, thus reflecting enduring changes),
(3) situation-specific traits (e.g., structural differences between the situation
before and the situation after the election or campaign),
4
or (4) a combination
of any of the above.
LST reliability coefficients for IE and EE were higher than common
reliability estimates, such as test–retest correlations or Cronbach’s alphas, be-
cause they were computed from appropriate measurement models. However,
the test halves consisting of only one or two items contained method-specific
3
State variability can be described as ‘‘occasion-specific fluctuations which may be characterized as
‘temporary changes’ ’’ and trait change can be described as ‘‘change between periods of time in latent
variables being stable within periods of time’’ (Eid & Hoffmann, 1998,p.195). Although traits are by
definition not affected by situational influences, traits are still changeable due to learning, critical life events
or genetic programs. Yet, trait changes will occur very slowly, and at least four measuring points and an
extension of the presented LST models would be necessary to analyze such trait changes (for further
explanations of true trait changes see Steyer et al., 1997,1999). Alternatively, a second-order latent
growth curve model would already allow for separating state variability and trait change with only three
measurement occasions (Geiser, Keller, & Lockhart, 2013).
4
The EE items might measure different facets of one general EE trait (i.e., trait-specific to the EE
construct). These specific traits (facets) might vary differently due to structurally different situations. For
example, maybe one facet of EE depends on temporarily accessible information about candidates or persons
involved in the election campaign (e.g., the items CARE and TOUCH). Another facet of EE might depend
on rather stable facts or beliefs about the political system (e.g., the item NOINFL). Thus, whereas the
former facet could be altered by personalized information about the politicians that might be available at the
end, but not at the beginning of the election campaign, the latter is a more stable facet because only changes
in the political system might alter this facet.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
14
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
variance and measured the underlying latent trait imprecisely. Method-specific
variance may be reduced by including and aggregating more items within one
occasion (Steyer & Schmitt, 1990). In line with other researchers, we too
would like to see more items that reliably measure PE to be regularly included
in panel studies (e.g., Morgan, Palmgreen, Stephenson, Hoyle, & Lorch, 2003,
p. 595;Valentino, Gregorowicz, & Groenendyk, 2009,p.315).
5
In addition,
the high ratios of unreliability warrant the search for superior indicators for
PE. Finally, more reliable instruments will also provide more accurate esti-
mates of occasion specificity. This might have been a reason for obtaining
statistically non-significant variances of the latent state residuals for IE.
Despite the benefits of LSTT, one limitation should be mentioned. In
order to decompose a latent state into latent trait and latent state residual, it is
necessary to measure the same persons on more than one occasion with more
than one indicator for the same construct. With regard to academic research
endeavors that often have only limited funding, it might be difficult to obtain
large, within-subject data sets. However, LSTT seems to be especially con-
venient for large-scale panel studies, such as those that are regularly conducted
in public opinion research, although they are often subject to strong limita-
tions with regard to the number of questions within one measurement occa-
sion. In sum, from our point of view, the vast number of concepts that are
assumed to be stable and usually included in large-scale panel studies as
mentioned at the beginning of the present paper would all benefit from
LST analyses, presupposing that the same participants are measured with at
least two indicators. Moreover, when designing panel studies in public opinion
research, knowledge about trait and occasion-specific components of a con-
struct might help researchers to choose the appropriate measure for the spe-
cific purpose or to reach a decision concerning the appropriate number of
measurement points. Although, as shown in our data application, two meas-
urement occasions and two indicators might be sufficient to fit a LST model,
estimate variance components, and determine LST coefficients, we strongly
recommend larger designs that include at least three indicators measured on
four measurement occasions or more (see discussion on state variability and
trait change above).
In our outline of LSTT we focused only on the most basic models: The
latent state–trait or single-trait–multi-state model (STMSM) with and without
method factors. Since its introduction in the late 1980s, LSTT has been
applied to numerous research questions, and several modifications have been
proposed. For instance, adding more constructs to an STMSM leads to a
5
Alternatively, when only few items are available and the items rather represent distinct facets of a broad
construct than tap into one general trait, it may be preferable to conduct analyses at the item-level and apply
the LST model with indicator-specific traits (Steyer et al., 2012). As suggested by an anonymous reviewer it
may be necessary to use appropriate estimation methods for ordinal data, when item-level data are used
(e.g., Eid, 1996;Eid & Hoffmann, 1998).
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 15
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
multiconstruct LST (MLST) model (e.g., Eid, Notz, Steyer, &
Schwenkmezger, 1994). Correlations between two or more constructs offer
the opportunity to obtain more accurate estimators of the relationship between
these constructs because they are free from situational or interactional influ-
ences. For example, IE and EE can be simultaneously subjected to a
multiconstruct LST (MLST) model to investigate the relationship of the
latent IE and the latent EE trait. Another benefit of the MLST model is
the possibility to examine the correlation between latent state residuals within
each measurement occasion. For example, a significant substantial correlation
of the latent state residuals of IE and EE at t1would hint to synchronous
occasion-specific effects.
6
Additionally, the inclusion of other methods (e.g.,
face-to-face interview, other-rating) provides further information and extends
the STMSM to a multimethod LST model (Courvoisier, Nussbeck, Eid,
Geiser, & Cole, 2008), which combines the advantages of multitrait–multi-
method (Campbell & Fiske, 1959) and LST analysis. Furthermore, current
research showed that under certain conditions some approaches to model
method effects outperform other approaches and some approaches to model
method effects are not appropriately defined within the LST framework (see
Geiser & Lockhart, 2012, Appendix A, for a thorough explanation why some
method effect models are in line with the LST framework and others are not).
Moreover, it is possible to apply LST models to categorical data. The inter-
ested reader is referred to Eid and Hoffmann (1998)for an application of LST
using continuous latent variables and polytomous indicators and to Eid and
Langeheine (1999)for an application of LST using latent classes and ordinal
indicators. In addition, all LST models can be analyzed in multi-group stu-
dies. This refers to the differences in traits and situational influences between
subpopulations (cf. Steyer et al., 1999). Consider as an example East Germans
and West Germans: With the German reunification in 1990 ‘‘East Germans
were submitted to substantial changes in their economic and social living
conditions, while the living conditions of West Germans remained rather
stable’’ (Steyer et al., 1999,p.402). These changes should be reflected in
different state and trait stabilities in many domains of public opinion research
6
It is theoretically possible to apply a MLST model of IE and EE to the GLES data as one reviewer
recommended. In contrast to the solution of the STMSM, all MLST models we fitted to the data led to
improper solutions as indicated by Heywood cases (i.e., correlations >1.0) or severe misspecifications as
indicated by modification indices and w
2
/df-ratios >9. Improper solutions lead to biased parameter esti-
mates (cf. Chen et al., 2001). Therefore, we do not report latent correlations between IE and EE. The
reasons for improper solutions are manifold (Chen et al., 2001). In our case, one reason might be that the IE
and EE items do not show a simple structure, that is, they show substantial cross-loadings on both factors
(e.g., the item NOSAY). Other reasons might lie (a) in method effects across IE and EE measures that are
due to the negative and positive wording, (b) in the high unreliabilities of the measures, or (c) in the rather
large amount of test-half-specific variances and substantial covariances of the test halves across IE and EE.
In sum, the data set is of limited value for illustrating the proper specification and straightforward inter-
pretation of a MLST model for didactic purposes, given the limited space of this paper. For an illustration
of a MLST model for two constructs, the interested reader is referred to Steyer et al. (2012,p.303).
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
16
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
like political attitudes, values, or living styles (Steyer et al., 1999). Taking
political interest as an example, research has shown that this assumption seems
to hold true for East and West Germans: East Germans exhibited substantial
changes in their political interest after 1990, whereas the interest of West
Germans seemed to be rather stable (van Deth & Elff, 2004). On the basis
of LST models, this can be tested by investigating whether the variances of
latent state residuals for political interest are the same for the subgroups of
East and West Germans, thus indicating different situational influences (Eid
et al., 1994). Finally, when comparing observed scores of different groups or
different occasions it is always important not only to assume measurement
invariance across groups or occasions (as we did in this paper to keep our data
example brief and simple) but also to test for it. Otherwise, trait-change
processes might be masked by non-invariant factor loadings and/or intercepts
and estimated coefficients might be biased (Geiser et al., 2012).
As personality traits become more and more important for understanding
political attitudes, orientation, ideology, civic engagement, or voting behavior
(e.g., Carney, Jost, Gosling, & Potter, 2008;Gerber et al., 2010;Mondak
et al., 2010), it is reasonable that this might be accompanied by debates
about the influences of traits, situations, and person x situation interaction.
In the same vein, the well-known phenomenon of ‘dealignment’ should not
remain unmentioned. While stable factors like partisanship or demographical
variables lose their explanatory power, less stable influences, such as candidate
evaluations, become more important for the explanation of political behavior
(Dalton & Wattenberg, 2000). LSTT provides us with a suitable framework
for facing the challenges of modern-day political public opinion research.
Supplementary Data
Supplementary data are available at IJPOR online.
Acknowledgements
The authors thank Kristin Heybach, Eleonore Hertweck, Rosemarie Morris,
Patrick Bacherle, and Rolf Steyer, as well as the anonymous reviewers, for
their help and useful feedback on earlier versions of this paper.
References
Abramson, P. R. (1983). Political attitudes in America: Formation and change. San
Francisco, CA: W.H. Freeman.
Acock, A., & Clarke, H. (1990). Alternative measures of political efficacy: Models and
means. Quality & Quantity, 24,87–105. doi:10.1007/BF00221386.
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 17
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
Aish, A.-M., & Jo
¨reskog, K. G. (1990). A panel model for political efficacy and
responsiveness: An application of LISREL 7with weighted least squares. Quality
& Quantity, 24,405–426. doi:10.1007/BF00152013.
Allen, B. P., & Potkay, C. R. (1981). On the arbitrary distinction between states and
traits. Journal of Personality and Social Psychology, 41,916–928. doi:10.1037/0022-
3514.41.5.916.
Alwin, D. F. (2010). Assessing the reliability and validity of survey measures. In
P. V. Marsden & J. D. Wright (Eds.), Handbook of survey research (pp. 405–
434). Bingley, UK: Emerald Group Publishing Limited.
American Association for Public Opinion Research (AAPOR). (2011). Standard defin-
itions: Final dispositions of case codes and outcome rates for surveys (7th ed.). Retrieved
from http://www.aapor.org/AM/Template.cfm.
Bartels, L. M. (2000). Partisanship and voting behavior, 1952-1996.American Journal
of Political Science, 44,35–50.doi:10.2307/2669291.
Becker, A. B., Dalrymple, K. E., Brossard, D., Scheufele, D. A., & Gunther, A. C.
(2010). Getting citizens involved: How controversial policy debates stimulate issue
participation during a political campaign. International Journal of Public Opinion
Research, 22,181–203. doi:10.1093/ijpor/edp047.
Bem, D. J., & Allen, A. (1974). On predicting some of the people some of the time:
The search for cross-situational consistencies in behavior. Psychological Review, 81,
506–520.doi:10.1037/h0037130.
Bentler, P. M., & Wu, E. J. C. (2005). EQS 6structural equations program manual.
Encino, CA: Multivariate Software.
Boulianne, S. (2011). Stimulating or reinforcing political interest: Using panel data to
examine reciprocal effects between news media and political interest. Political
Communication, 28,147–162. doi:10.1080/10584609.2010.540305.
Bowers, K. S. (1973). Situationism in psychology: An analysis and a critique.
Psychological Review, 80,307–336. doi:10.1037/h0035592.
Campbell, A., Converse, P. E., Miller, W. E., & Stokes, D. (1960). The American
voter. Chicago, IL: University of Chicago Press.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by
the multitrait-multimethod matrix. Psychological Bulletin, 56,81–105.doi:10.1037/
h0046016.
Caprara, G. V., Vecchione, M., Capanna, C., & Mebane, M. (2009). Perceived pol-
itical self-efficacy: Theory, assessment, and applications. European Journal of Social
Psychology, 39,1002–1020.doi:10.1111/0162-895X.00141.
Carney, D. R., Jost, J. T., Gosling, S. D., & Potter, J. (2008). The secret lives
of liberals and conservatives: Personality profiles, interaction styles, and the
things they leave behind. Political Psychology, 29,807–840. doi:10.1111/j.1467-
9221.2008.00668.x.
Catterberg, G., & Moreno, A. (2006). The individual bases of political trust: Trends
in new and established democracies. International Journal of Public Opinion Research,
18,31–48. doi:10.1093/ijpor/edh081.
Chen, F., Bollen, K. A., Paxton, P., Curran, P. J., & Kirby, J. B. (2001). Improper
solutions in structural equation models: Causes, consequences, and strategies.
Sociological Methods & Research, 29,468–508.doi:10.1177/0049124101029004003.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH18
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
Clarke, H. D., & McCutcheon, A. L. (2009). The dynamics of party identification
reconsidered. Public Opinion Quarterly, 73,704–728. doi:10.1093/poq/nfp051.
Cortina, J. M. (1993). What is coefficient alpha? An examination of theory and
applications. Journal of Applied Psychology, 78,98–104.doi:10.1037/0021-
9010.78.1.98.
Courvoisier, D. S., Nussbeck, F. W., Eid, M., Geiser, C., & Cole, D. A. (2008).
Analyzing the convergent and discriminant validity of states and traits:
Development and applications of multimethod latent state-trait models.
Psychological Assessment, 20,270–280. doi:10.1037/a0012812.
Dalton, R. J., & Wattenberg, M. P. (Eds.) (2000). Parties without partisans: Political
change in advanced industrial democracies. Oxford: Oxford University Press.
Davidov, E. (2011). Nationalism and constructive patriotism: A longitudinal test of
comparability in 22 countries with the ISSP. International Journal of Public Opinion
Research, 23(1), 88–103.doi:10.1093/ijpor/edq031.
Deinzer, R., Steyer, R., Eid, M., Notz, P., Schwenkmezger, P., Ostendorf, F., &
Neubauer, A. (1995). Situational effects in trait assessment: The FPI, NEOFFI,
and EPI questionnaires. European Journal of Personality, 9,1–23. doi:10.1002/
per.2410090102.
Eid, M. (1996). Longitudinal confirmatory factor analysis for polytomous item re-
sponses. Methods of Psychological Research–Online, 1,65–85.
Eid, M. (2002). A closer look at the measurement of change: Integrating latent state-
trait models into the general framework of latent mixed Markov modeling. Methods
of Psychological Research–Online, 7,33–52.
Eid, M., & Hoffmann, L. (1998). Measuring variability and change with an item
response model for polytomous variables. Journal of Educational and Behavioral
Statistics, 23(3), 193–215. doi:10.3102/10769986023003193.
Eid, M., & Langeheine, R. (1999). The measurement of consistency and occasion
specificity with latent class models: A new model and its application to the
measurement of affect. Psychological Methods, 4,100–116.doi:10.1037/1082-
989X.4.1.100.
Eid, M., Notz, P., Steyer, R., & Schwenkmezger, P. (1994). Validating scales for the
assessment of mood level and variability by latent state-trait analyses. Personality
and Individual Differences, 16,63–76. doi:10.1016/0191-8869(94)90111-2.
Endler, N. S., & Magnusson, D. (1976). Toward an interactional psychology of per-
sonality. Psychological Bulletin, 83,956–974.doi:10.1037/0033-2909.83.5.956.
Epstein, S., & O’Brien, E. J. (1985). The person–situation debate in historical
and current perspective. Psychological Bulletin, 98,513–537.doi:10.1037/0033-
2909.98.3.513.
Eysenck, H. J. (1983). Cicero and the state-trait theory of anxiety: Another case
of delayed recognition. American Psychologist, 38,114–115.doi:10.1037/0003-
066X.38.1.114.
Geiser, C., Keller, B. T., & Lockhart, G. (2013). First- versus second-order latent
growth curve models: Some insights from latent state–trait theory. Structural
Equation Modeling, 20,479–503. doi:10.1080/10705511.2013.797832.
Geiser, C., Keller, B. T., Lockhart, G., Eid, M., Cole, D. A., & Koch, T. (2012).
Distinguishing between state variability and trait change in longitudinal data: The role
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 19
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
of measurement (non)invariance in latent state–trait analyses. Manuscript submitted
for publication.
Geiser, C., & Lockhart, G. (2012). A comparison of four approaches to account for
method effects in latent state–trait analyses. Psychological Methods, 17,255–283.
doi:10.1037/a0026977.
Gerber, A. S., Huber, G. A., Doherty, D., Dowling, C. M., & Ha, S. E. (2010).
Personality and political attitudes: Relationships across issue domains and political
contexts. American Political Science Review, 104,111–133. doi:10.1017/
S0003055410000031.
Hacker, K. L. (Ed.) (2004). Presidential candidate images. Lanham, MD: Rowman &
Littlefield.
Hagemann, D., & Meyerhoff, D. (2008). A simplified estimation of latent state-trait
parameters. Structural Equation Modeling, 15,627–650. doi:10.1080/
10705510802339049.
Lane, R. E. (1959). Political life: Why people get involved in politics. Toronto: Free
Press.
Lazarsfeld, P. F. (1948). The use of panels in social research. Proceedings of the
American Philosophical Society, 92,405–410.
Lee, F. L. F. (2010). The perceptual bases of collective efficacy and protest partici-
pation: The case of pro-democracy protests in Hong Kong. International Journal of
Public Opinion Research, 22,392–411. doi:10.1093/ijpor/edq023.
Lord, F. M., & Novick, M. R. (1968). Statistical theories of mental test scores. Reading,
MA: Addison-Wesley.
Mischel, W. (1968). Personality and assessment. New York: Wiley.
Mondak, J. J., Hibbing, M. V., Canache, D., Seligson, M. A., & Anderson, M. R.
(2010). Personality and civic engagement: An integrative framework for the study of
trait effects on political behavior. American Political Science Review, 104,85–110.
doi:10.1017/S0003055409990359.
Morgan, S. E., Palmgreen, P., Stephenson, M. T., Hoyle, R. H., & Lorch, E. P.
(2003). Associations between message features and subjective evaluations of the
sensation value of antidrug public service announcements. Journal of
Communication, 53,512–526. doi:10.1111/j.1460-2466.2003.tb02605.x.
Morrell, M. E. (2003). Survey and experimental evidence for a reliable and valid
measure of internal political efficacy. Public Opinion Quarterly, 67,589–602.
doi:10.1086/378965.
Morrell, M. E. (2005). Deliberation, democratic decision-making and internal political
efficacy. Political Behavior, 27,49–69. doi:10.1007/s11109-005-3076-7.
Niemi, R. G., Craig, S. C., & Mattei, F. (1991). Measuring internal political efficacy
in the 1988 National Election Study. American Political Science Review, 85,
1407–1413.do
i:10.2307/1963953.
Pedersen, R. T. (2012). The game frame and political efficacy: Beyond the spiral of
cynicism. European Journal of Communication, 27,225–240.doi:10.1177/
0267323112454089.
Prior, M. (2010). You’ve either got it or you don’t? The stability of political interest
over the life cycle. The Journal of Politics, 72,747–766.doi:10.1017/
S0022381610000149.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH20
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
Rattinger, H., Roßteutscher, S., Schmitt-Beck, R., & Weßels, B. (2009). German
Longitudinal Election Study – Election campaign panel, 07/10-10/07/2009.
Cologne, Germany: GESIS. doi:10.4232/1.11131.
Robinson, J. P., Shaver, P. R., & Wrightsman, L. S. (Eds.). (1999). Measures of social
psychological attitudes, Vol. 2: Measures of political attitudes. San Diego, CA:
Academic Press.
Semetko, H. A., & Valkenburg, P. M. (1998). The impact of attentiveness on political
efficacy: Evidence from a three-year German panel study. International Journal of
Public Opinion Research, 10,195–210. doi:10.1093/ijpor/10.3.195.
Steyer, R., Eid, M., & Schwenkmezger, P. (1997). Modeling true intraindividual
change: True change as latent variable. Methods of Psychological Research–Online,
2,21–33.
Steyer, R., Ferring, D., & Schmitt, M. (1992). States and traits in psychological
assessment. European Journal of Psychological Assessment, 8,79–98.
Steyer, R., Geiser, C., & Fiege, C. (2012). Latent state-trait models. In H. Cooper, P.
M. Camic, D. L. Long, A. T. Panter, D. Rindskopf, & K. J. Sher (Eds.), APA
handbook of research methods in psychology, Vol. 3: Data analysis and research pub-
lication (pp. 291–308). Washington, DC: American Psychological Association. doi:
10.1037/13621-014.
Steyer, R., & Schmitt, M. (1990). The effects of aggregation across and within oc-
casions on consistency, specificity and reliability. Methodika, 4,58–94.
Steyer, R., Schmitt, M., & Eid, M. (1999). Latent state–trait theory and research in
personality and individual differences. European Journal of Personality, 13,389–408.
doi:10.1002/(SICI)1099-0984(199909/10)13:5<389::AID-PER361>3.0.CO;2-A.
Stro
¨mba
¨ck, J., Djerf-Pierre, M., & Shehata, A. (2013). The dynamics of political
interest and news media consumption: A longitudinal perspective. International
Journal of Public Opinion Research, 25,414–435.doi:10.1093/ijpor/eds018.
Tsfati, Y., & Cappella, J. N. (2005). Why do people watch news they do not trust?
The need for cognition as a moderator in the association between news media
skepticism and exposure. Media Psychology, 7,251–271. doi:10.1207/
S1532785XMEP0703_2.
Valentino, N. A., Gregorowicz, K., & Groenendyk, E. W. (2009). Efficacy, emotions
and the habit of participation. Political Behavior, 31,307–330. doi:10.1007/s11109-
008-9076-7.
Valenzuela, S. (2011). Materialism, postmaterialism and agenda-setting effects: The
values-issues consistency hypothesis. International Journal of Public Opinion
Research, 23,437–463.doi:10.1093/ijpor/edr018.
Van Deth, J. W., & Elff, M. (2004). Politicisation, economic development and pol-
itical interest in Europe. European Journal of Political Research, 43,477–508.
doi:10.1111/j.1475-6765.2004.00162.x.
Vetter, A. (1997). Political efficacy – Reliabilita
¨t und Validita
¨t [Political efficacy –
Reliability and validity]. Wiesbaden: DUV.
LATENT STATE–TRAIT ANALYSIS OF POLITICAL EFFICACY 21
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
Biographical Notes
Frank M. Schneider is a postdoctoral researcher at the University of Mannheim,
Germany. His research interests include research methods, psychological assessment,
and communication processes and effects.
Lukas Otto is a PhD student at the University of Koblenz-Landau, Germany. His
research interests include political psychology, personalization of politics, and media
effects.
Daniel Alings is a PhD student at the University of Koblenz-Landau, Germany. His
research interests include political efficacy, deviance, and prejudice.
Manfred Schmitt is full professor of psychology at the University of Koblenz-Landau,
Germany. His main research interests are in the domains of social justice, social
emotions, altruism, attitude–behavior consistency, and latent state–trait theory.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH22
by guest on March 6, 2014http://ijpor.oxfordjournals.org/Downloaded from
