(To view or download this information in .pdf format click here)
Date: Friday, December 17th, 2010, at 10h00.
Place: Universitat Pompeu Fabra. Edifici Roger de Llúria, room 40.109. Carrer de Ramón Trias Fargas 25-27 (metro Ciutadella - Villa Olímpica). Barcelona.
Speakers: Dr. William van der Veld (Radboud University Nijmegen, the Netherlands) and Dr. Toni Toharudin (Padjadjaran University, Indonesia).
Further information: The seminar is free of charge and does not require prior registration. For further questions, please contact us at recsm@upf.edu.
INTRODUCTION:
Given the severe effects of measurement errors on the results of survey research, specific procedures have been developed to evaluate the quality of the measures also for longitudinal data. In this seminar some procedures which are used to evaluate the quality of measures in longitudinal studies will be discussed. Dr. William van de Veld will show that the suggested simplex model does not always work well for satisfaction and income questions. In doing so he will show that impossible results can be obtained. He will also suggest alternatives. This presentation is based on the work he did for his PhD dissertation.
Another issue is that one sometimes does not have the same sets of questions at different points in time in a longitudinal study. The question is then if these sets of questions can be used to measure the same concepts. Dr. Toni Toharudin will show whether the different sets of items still measure the same concept for a longitudinal study done in Belgium with respect to Authoritarianism. His presentation as well is based on his recently finished PhD-dissertation.
For a more elaborate summary of their presentations please refer to the abstracts presented below.
ABSTRACTS:
"Explaining the difference between cross-sectional and panel reliability of satisfaction and income questions"
Dr. William M. van der Veld (Radboud University Nijmegen, the Netherlands)
This study is about the assessment of reliability of single survey questions. We will discuss two of the most common models to assess reliability for such instruments: the quasi-simplex model (panel) and the parallel test model (cross section). The quasi-simplex design should have been the solution for the assumed deficiencies (i.e. recall effects) of the parallel test design. However, the quasi-simplex design seems to have problems of its own, resulting in the underestimation of the reliability. In order to show that there is something wrong with the reliability estimate from the quasi-simplex model we have used those reliability estimates to correct the correlation between different variables for attenuation. For the illustration we have used three measures about income and satisfaction (How satisfied are you with your current life as a whole? How satisfied are you with your family's current financial situation? Do you think you are paid the amount you deserve?). In order to show exactly what is wrong, we made a comparison with the VAS model (Van der Veld and Saris, 2004). The VAS model disentangles the response in four components, an error component, a temporal attitude component (state), a stable attitude component (trait), and an attitudinal change component. For attitude measures where there is little effect of the temporal component, the attitude is said to be crystallized. It follows from the comparison of the VAS model with the quasi-simplex model that the reliability in the latter model is the product of attitude crystallization and instrument reliability. This is a general result, however, it depends on the amount of crystallization whether there will be a difference between the reliability estimated with the quasi-simplex model and the parallel test model. It is very likely that for factual questions this problem does not occur.
"Measuring Authoritarianism with Different Sets of Items in a Longitudinal Study"
Dr. Toni Toharudin (Padjadjaran University, Indonesia)
In the General Election Study in Belgium (Interuniversitair Steunpunt Politieke-Opinieonderzoek K.U. Leuven, 1991, 1995, and 1999), authoritarianism was measured in each of the wave years 1991, 1995, 1999 by seven items. However, only two items were the same in all of the three waves. A series of tests proposed by Jöreskog (1971, 1974) is performed to find out which of the items in each of the waves can be considered to measure the same latent variable and can therefore be used in a longitudinal study to assess the development of authoritarianism over time. We find that three of the items in 1999 did not pass the test convincingly.