This is a session of Topic 4: Statistics education at the post secondary (tertiary) level
(Monday 12th, 16:30-18:00)
Less parametric methods in statistics
Organizer
- Noël Veraverbeke (Belgium)
Abstract
The traditional core chapters in statistics textbooks taught at most colleges and universities still deal with the classical parametric models for estimation and testing. Statistical inference relies on the assumption that the observations come from a certain distribution depending on some unknown parameter(s). A good example is the common assumption of the underlying normal distribution, which is dependent on the two parameters mean and variance. Such model assumption is not always justifiable and checking it can be problematic, especially in small samples. Therefore several statistical methods have been suggested to handle data sets such as these. They are more versatile since they are valid without these full distributional restrictions and they are called nonparametric. They are less restrictive with respect to the underlying distribution, and require only mild assumptions such as continuity or symmetry. Many of these methods are typically based on the ranks of the observations and they often turn out to be conceptually simple.The ever growing power of computers greatly influenced the area of statistics and in particular nonparametric statistics. The so-called nonparametric smoothing techniques became invaluable instruments in practical data analysis. These include estimation of density functions, distribution functions, regression curves, hazard rate functions, etc. The (nonparametric) bootstrap resampling method provides very useful tools to support nonparametric statistical inference. Also the growing possibilities of graphical tools, Monte Carlo simulations, computer intensive Bayesian techniques,.. gave an important boost to the field of nonparametrics.
The speakers in this session will cover several interesting aspects of this field of nonparametrics and they will illustrate how these could fit in the teaching of post secondary statistics courses. But we also hope to answer the following question: Should all this be added to the traditional theory or should we grow towards a new statistical education with less parametrics and more nonparametrics?
Papers
| Paper | Title | Presenter(s) / Author(s) |
| 4B1 | The use of statistical software to teach nonparametric curve estimation: from Excel to R | Ricardo Cao (Spain) Salvador Naya (Spain) |
| 4B2 | On teaching bootstrap confidence intervals | Joachim Engel (Germany) |
| 4B3 | Exploring data with non- and semiparametric models | Marlene Müller (Germany) |