| Popis: |
Researchers in the environment and public health can analyse time series rapidly and efficiently using the computer program, NPBats. The annual CO2 emission or the total number of people confronted with overweight are examples. Using these analyses, researchers can 1) detect systematic changes in time (trends), 2) explain the relations between different time series and 3) obtain knowledge on the influence of governmental policy on humans and the environment. NPBats uses Bayesian statistical concepts to describe the correlation between subsequent observations in time, obtained through the so-called 'prior' models. The simplest prior model is the neighbour model. This model assumes that the difference between two subsequent observations has an expectation zero and a fixed, limited variation. Two other models used by NPBats are the linear and the quadratic models. These models enable NPBats to predict a future observation using 1, 2 or 3 of the preceding observations. Therefore NPBats is more flexible than the usual, classical models for analysing time series. NPBats can already be used for time series with at least 8-10 observations. Missing values can be estimated automatically and changes in the variation of the observations can be taken into account. Co-variables might be included in the model too. These variables augment the knowledge of the underlying process and improve the predictions. NPBats automatically generates confidence intervals for the predictions, thus clarifying the statistical significance of an increasing or decreasing trend. NPBats developed at the RIVM within the statistical package, S-PLUS, is easy to use and contains a comprehensive help function. At the RIVM NPBats is available to the S-PLUS user group and to other interested people on request. |
