ABSTRACT
In Part IV, we work with a single demographic series measured with errors. We use one or more datasets, allowing for the possibility that these datasets are incomplete or unreliable. This gives us much more exibility than requiring a single perfect dataset. In return, we need to specify explicit data models. For each dataset, we need to set out a model describing the probability of observing the dataset given the true array of counts or totals. Data models can be used to encode beliefs about the strengths and weaknesses of datasets. A big practical advantage of using data models is that it allows us to use datasets that are less detailed than the true array of counts or totals.
We infer the true counts or totals, along with the super-population rates, probabilities, or means, and the more abstract parameters in the prior model. In addition, we also produce inferences about the data models. These can include probabilities that people or events will be captured by a data source, rates of over-reporting, and estimates of whether coverage is improving or deteriorating over time. The estimates of demographic rates and coverage rates are all mutually consistent.
