Statistical methods for marginal inference from multivariate ordinal data

Nooraee, N., 2015, [Groningen]: University of Groningen. 205 p.

Research output: ThesisThesis fully internal (DIV)

Copy link to clipboard


  • Title and contents

    Final publisher's version, 153 KB, PDF document

  • Chapter 1

    Final publisher's version, 352 KB, PDF document

  • Chapter 2

    Final publisher's version, 403 KB, PDF document

  • Chapter 3

    Final publisher's version, 337 KB, PDF document

  • Chapter 4

    Final publisher's version, 654 KB, PDF document

  • Chapter 5

    Final publisher's version, 375 KB, PDF document

  • Chapter 6

    Final publisher's version, 330 KB, PDF document

  • Chapter 7

    Final publisher's version, 684 KB, PDF document

  • Chapter 8

    Final publisher's version, 234 KB, PDF document

  • Appendices

    Final publisher's version, 224 KB, PDF document

  • Complete dissertation

    Final publisher's version, 1.73 MB, PDF document

  • Propositions

    Final publisher's version, 21.4 KB, PDF document


  • Nazanin Nooraee
Ordinal scoring is routinely applied in many research areas, like for instance medical sciences, social sciences, quality control, and engineering. This type of data can be considered as manifest variables of underlying continuous variables that cannot be measured directly, possibly due to lack of appropriate devices or financial constraints. A common measurement tool, particularly in social sciences, is a questionnaire that consists of multiple items (possibly with ordinal scales) to obtain information from respondents. Nowadays, data are preferably collected repeatedly over time (longitudinally) to examine the impact of covariates (risk factors) on short and long term time profiles of outcomes. A typical issue in longitudinal data collection is that some of the proposed outcomes may become unobserved (missing).

Ordinal outcomes likely interrupt the use of ordinary statistical models, which are frequently based on the normality assumption. Therefore, analysis ordinal outcomes require more complicated models and methods. Moreover, time related correlations and missing data issues should be taken into account in the analysis model in longitudinal settings.

This thesis provides insight about the performance of a few existing approaches in new settings and presents a few new methods for the analysis of longitudinal ordinal data. Generalized estimating equations, available in popular software packages, are evaluated theoretically and practically. Some missing data methods are assessed for the first time on how well they deal with missing items from longitudinal questionnaire data. A new univariate and multivariate longitudinal ordinal data model was developed and applied to a real medical application.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • van den Heuvel, Edwin, Supervisor
  • Ormel, Hans, Supervisor
  • Balakrishnan, N. (Narayanaswamy), Assessment committee, External person
  • Lesaffre, Emmanuel M. E. H., Assessment committee, External person
  • Snijders, Thomas, Assessment committee
Award date8-Apr-2015
Place of Publication[Groningen]
Print ISBNs978-90-367-7731-5
Electronic ISBNs978-90-367-7730-8
Publication statusPublished - 2015

Download statistics

No data available

ID: 17588928