Dependencies: Modeling Interdependent Observations
Content
Many data analysis procedures assume that observations are independent of each other, i.e., that the observed value of a particular case is not necessarily related to the values of other cases. This assumption is usually unproblematic in the case of experimental work, in survey studies, and so on. But sometimes it is patently implausible. A country’s level of unemployment, for example, is not independent of levels of unemployment at earlier moments, a problem of serial dependency, commonly found in time-series and other repeated measurements. Pupils’ scholastic performances are not independent to the extent that they share teachers, a form of within-cluster dependency that motivates hierarchical or more generally multi-level models. Nearby municipalities are not independent of each other in terms of traffic congestion, a case of spatial dependency. In each of these circumstances, common data analysis methods must be modified or replaced by methods that take the interdependencies between observations into account.
This clinic elaborates the problem of interdependent data, the risks to valid inference if such dependencies are ignored, and the variety of ways of dealing with them. These include hierarchical or multi-level modelling models of panel data and repeated measurements, regression approaches for dealing with serial dependency, and time series models.
The applicability of these models is wide, but in the context of this clinic they will be applied in particular to problems of comparison using surveys and to the analysis of experimental data.
This clinic will be supplemented with online learning materials which can be accessed after the event.
Prerequisites
Understanding and familiarity with specific statistical models, e.g. multiple regression, analysis of variance, (ANOVA) or factor analysis. If you are attending a clinic as part of our Researcher Development Initiative then you automatically fulfil the prerequisites .