In this workshop we explored how researchers combine multiple data types [i.e. survey data and qualitative data] with qualitative and multi-level modelling. We explored what types of levels exist, and how they interact and interrelate, in specific research contexts. A diagram illustrating nested levels versus non-hierarchical levels was included to illustrate how classical statistics is currently dealing with the multi-level stratified nature of reality (Goldstein, 2003). For a summary of the argument see Archer, et al., 1998.
We argue that the analysis of several levels of cases can benefit from the use of survey data, but like many authors we also recommend the use of methodological pluralism. For this reason the practical part of the workshop illustrated (a) how logistic regression models can be interpreted with reference to qualitative and small-sample data; (b) how log-linear models represent qualitative aspects of society in their contingent conjunctions, ie within complex causal relation, again only sensible whilst referring to non-survey data sources which describe the underlying causal mechanisms and reasons for action. In addition research results by Olsen and Morgan were offered, showing how Bayesian estimation methods can be used under conditions where there are multiple routes to a single outcome (e.g. labour force non-participation). The incapacity of existing Bayesian computer software to deal with causal complexity was contrasted with the true Bayesian epistemological commitment to make optimal use of all available information. Precepts for rational judgement among models were discussed.
Finally we re-assessed sampling strategies by adding to the usual toolkit the following: augmenting existing samples using purposive sampling and theoretical sampling to catch people of interesting categories; using weighting systems to allow for supra-random sampling strategies; and using non-random small samples rather than representative samples to capture tendencies within rather than between groups. A concluding discussion of the epistemology of non-parametric statistics enabled participants to engage with a growing literature on qualitative statistics.
All discussion were recorded and processed interpretatively using NVIVO as the basis of our report on the workshop. A task was set as preparation for the final workshop: reflectively studying an extract from the transcripts of discussions.
By following the links below, you can work through the course presentations and, by accessing the spss files on the left, you can then try out some of the techniques for yourself. The output file has been included as an example.
COURSE MATERIALS: