PTAM 2017-3

ABSTRACTS

Detecting unmotivated individuals with a new model-selection approach for Rasch models
Jochen Ranger & Jörg-Tobias Kuhn

Abstract
In low-stakes tests some test takers do not work with high motivation but respond carelessly. This has serious consequences for item response models as careless responses impair model calibration and trait inference. In this manuscript we describe an approach to data analysis that reduces the negative implications of careless responding and allows for the identification of the poorly motivated test takers. The approach has been inspired by the Rasch model answer tree (also Rasch tree) suggested by Strobl, Kopf, and Zeileis (2015). The Rasch model answer tree subdivides the sample into several strata in a data driven way by means of significance tests and fits a distinct Rasch model to each stratum. In our new approach we build on this idea of partitioning the data into strata via a sequence of splits. Contrary to this approach we determine multi-group Rasch models by enforcing theoretically motivated configurations of hierarchical data splits and select among the configurations via information criteria. By using the response times of the test takers for partitioning, a stratum can be isolated that contains the motivated test takers and allows for unbiased model calibration. The performance of this new approach with respect to parameter recovery and the detection of unmotivated test takers are compared to alternative models for low-stakes tests in a simulation study, namely the latent class model of Meyer (2010) and a finite mixture model for the response times. The simulation study demonstrates that the new approach reduces the bias caused by low motivation under certain circumstances. An empirical application underscores the usefulness of our suggestion.

Keywords: response time, rapid guessing, Rasch model answer tree

Jochen Ranger, PhD
Department of Psychology
Martin-Luther-University Halle-Wittenberg
Emil-Abderhalden-Str. 26-27
06108 Halle (Saale), Germany
jochen.ranger@psych.uni-halle.de

Theoretical framework, factorial structure and measurement invariance of the Video Game Playing Motives Questionnaire (VGPM-Q) for preadolescents
Eva-Maria Schiller, Dagmar Strohmeier & Christiane Spiel

Abstract
The video game playing motives questionnaire for preadolescents (VGPM-Q) was developed and tested regarding construct validity and measurement invariance. The VGPM-Q is theoretically based on a combination of uses and gratifications theory and mood-management theory. Based on these theoretical frameworks, two distinct motive systems were distinguished, (1) motives oriented towards the satisfaction of needs and (2) motives oriented towards the regulation of mood states. Construct validity and measurement invariance across gender were tested in 1297 preadolescents (42 % girls; Mage = 11.57). The theoretical two-factor structure fit the data well, yielding the two correlated factors “uses and gratifications motives” and “mood management motives”. Partial strong measurement invariance across gender was obtained. Theoretical and methodological implications for research on video game playing motives are discussed.

Keywords: measurement invariance, construct validity, video game playing motives

Eva-Maria Schiller, PhD
University of Münster
Fliednerstraße 21
48149 Münster, Germany
eva-maria.schiller@uni-muenster.de

Measurement models for ordered-categorical indicators: A factor analytic approach for testing the level of measurement
Takuya Yanagida, Petra Gradinger & Dagmar Strohmeier

Abstract
In the social sciences, self-reports administered as questionnaires are frequently used to measure psychological constructs. Data stemming from scale items are commonly analyzed using statistical methods for metric dependent variables. However, the assumption of interval level data is not tested but assumed to be fulfilled. One reason for ignoring this assumptions is the lack of an adequate approach for testing this assumption. Thus, we present a factor analytic approach for testing the level of measurement. First, two empirical examples are presented to demonstrate this approach. Second, a simulation study based on several conditions with varying population model and sample size was conducted. Results of the simulation study demonstrate the functioning of this approach. In sum, the factor analytic approach can be used for testing the level of measurement of scale items enabling empirical decision making about choosing appropriate statistical methods instead of relying on untested assumptions.

Keywords: confirmatory factor analysis, measurement model, level of measurement, ordinal data, statistical assumption

Takuya Yanagida, PhD
Department of Applied Psychology:
Work, Education, and Economy
Faculty of Psychology
University of Vienna
Universitätsstraße 7 (NIG)
1010 Vienna, Austria
takuya.yanagida@univie.ac.at

On the relationship between the retrieval of information and learning: the influence of deep processing
Michael Altmeyer, Tengfei Wang & Karl Schweizer

Abstract
This study investigates the relationship between learning and the retrieval of information. In particular, the influence of deep learning in contrast to shallow learning on this relationship is considered. A sample of 183 university students completed a retrieval task (Posner’s Task) as well as tasks tapping associative and complex learning. All tasks were designed to include several treatment levels that enable the separation of the effects of retrieval and learning processes respectively from the effects of auxiliary processes, as for example, perceptual and motor processes that are necessary for completing a task. Results showed that there were substantial relationships between retrieval and complex learning (r = .56) and also associative learning (r = .34). The relationship due to complex learning showing characteristics of deep learning proved to be substantially larger than the relationship attributed to shallow learning.

Key words: retrieval, associative learning, complex learning, deep processing, Posner`s Task

Michael Altmeyer
Goethe University Frankfurt
Theodor-W.-Adorno-Platz 1
60323 Frankfurt am Main, Germany
altmeyer@pvw.uni-frankfurt.de

Optimal design of surveys and experiments
Dieter Rasch & Jürgen Pilz

Abstract
After a general discussion about designing experiments and surveys it is shown how the program package OPDOE can be used to determine minimal sample sizes for confidence estimation and hypotheses testing for means in the one- and two-sample problem. OPDOE is demonstrated by some examples.

Keywords: Testing hypotheses, confidence intervals, minimal sample size, experimental design

Dieter Rasch, PhD
University of Natural Resources and Life Sciences, Vienna
Institute of Applied Statistics and Computing
Peter-Jordan-Str. 82
1190 Vienna, Austria
psychiatrie@klf-net.de