The projects in the theme Research Methodology, Measurement and Data analysis focus on exploring possibilities and/or performance of novel statistical methods or novel applications using (advanced) statistical methods. The application area is within the scope of psychological, educational and health research but other application areas are possible. More applied projects can address for instance statistical methods for challenging data or for data obtained by novel data collection methods. An applied project can also highlight the usability of novel statistical software (preferable in R). Projects on new statistical methods can comprehend a simulation study to examine their performance, compare different statistical methods on their performances, or examine the quality of statistical inferences under different conditions.
Contact person: Jean-Paul Fox j.p.fox@utwente.nl
RMMD 1 - Combining experience sampling data with physiological sensor data Supervisor: Dr. Stéphanie van den Berg
More and more data are gathered in psychological and health research. Many researchers do experience sampling, where once or several times a day, participants fill in a short survey. For this kind of research, specific statistical techniques are available to analyse the data. In this project, we would like to explore and compare various methods to combine experience sampling data with sensor data like heart rate and galvanic skin response: data that are much more time-intensive.
We’re looking for students that like data analysis and are curious and creative, and not afraid of R. The aim of the research is to find an optimal way to study the relationship between physiological measures and self-reports. For example: Does people’s physiological stress match the stress that they consciously experience? Project will be carried out together with Peter ten Klooster.
Contact Stéphanie van den Berg (CU B335)
RMMD 2 - Harmonising educational and psychological test scores Supervisor: Dr. Stéphanie van den Berg
Problem Description
To avoid fraud, exams are often done using different exam versions. Psychological traits are measured using different questionnaires. In the event that a trait or ability is measured with various instruments or test versions, one would like to make test scores comparable. There are standard psychometric solutions for this, but in some cases these do not work properly. In this project we explore various alternative approaches, getting inspiration from machine learning techniques.
Profile of the student
Interested in psychometrics and novel statistical and machine learning techniques.
Contact Stéphanie van den Berg (CU B335)
RMMD 3 Covariance structure modeling for the analysis of EEG (trial) data Problem Description
The analysis of EEG (trial) data is that it concerns a large amount of repeated data, where a person is repeatedly measured in the same condition across trials. This leads to clustered data where observations from one trial (and one condition) are more likely to correlate than observations from different trials (and different conditions). Statistical testing whether means are different across conditions and trials is difficult due to the correlation between observations within a trial (caused by sequentially collecting the data). In practice, trial means are used as observations to determine mean differences across conditions, but this leads to a loss of information. In our approach dependencies within each trial are taken into account by modelling the dependencies between trial observations. Then, testing the mean differences across conditions is carried out conditional on the correlated trial data instead of the aggregated trial data. In the study, it is examined whether this new approach leads to better statistical testing of differences between conditions.
Contact Jean-Paul Fox (CU B309)
RMMD 4 Testing the intra-class correlation in an unbalanced design Problem Description
Data is often hierarchically structured. For instance, patient data is nested within the patient, and the patient is nested within the treatment group, and in the same way a child’s observations is nested within the child, and the child is nested within the teacher\class which is again nested within the school. The hierarchical structure of the data leads to dependencies between observations, which needs to be taken into account when doing statistical analysis on the data. To measure the strength of the clustering in a data the intra-class correlation coefficient is often used. The intra-class correlation represents the correlation between clustered observations, and can also be seen as the proportion of variance attributable to the clustering of observations. Thus, when clusters differ, the clustering is known to be informative.
Testing the intra-class correlation coefficient (ICC) for significance is a complex problem, and up till now, there is no known standard statistical test. A new statistical ICC test has been developed, and the performance of this test will be evaluated (type-1 error, power), specifically for the more complex situation of incomplete designs.
Contact Jean-Paul Fox (CU B309)
