Graduate Training Programme

————————————————————————————

 

Computer Intensive Analysis of Data and Models

 

The course comprises three main components:

 

(i)      Working Notes: These contain the material covered in the lectures. This has a strong practical emphasis.

 

(ii)    Worked Examples: These are spreadsheet examples. They are embedded in the Working Notes.

 

(iii)   Technical Notes: These cover the theoretical foundations of bootstrapping.

 

————————————————————————————

 

 

The Working notes cover the following:

 

I        Introduction

 

1.      Introduction

2.      Statistical Metamodels

 

II       Classical Methods

 

3.      Random Variables

4.      Fitting Parametric Distributions to Random Samples; Input Modelling

5.      Maximum Likelihood Estimation

6.      Accuracy of MLEs

 

 

III     Computer Intensive Methods

 

7.      Empirical Distribution Functions

8.      Basic Bootstrap Method

9.      Evaluating the Distribution of MLEs by Bootstrapping

10.    Comparing Samples Using the Basic Bootstrap

11.    The Parametric Bootstrap

12     Goodness of Fit Testing

12.1  Classical Goodness of Fit

12.2  Bootstrapping a GOF statistic

13     Comparison of Different Models; Model Selection

14     Final Comments

 

 

You can access the working notes by clicking on the links given below. The Working Notes are meant to be worked through.

 

They contain Examples and Exercises. These illustrate the topic or method being discussed. They are an essential part of the text and need to be studied.

 

Many of the Examples and Exercises come with their own link. (i) Some of the links contain additional notes and more detailed formulas, (ii) The other links are to actual spreadsheets containing data and the worked details using the data.

 

Some of the initial spreadsheets contain elementary exercises connected with generating random variables and simple sampling experiments. The point of these exercises is to remind you of the basic formulas and functions that you will need for the more complicated later examples. You should already be familiar with this material. However you might wish to spend a short time checking that you do know this material well.

 

 

The other spreadsheets contain more substantial problems.

 

These are solved using VBA macros for carrying out more substantial calculations and more extensive analyses. The macros are fairly generic in that they only need minor adjustment to solve other similar problems.

 

The main reason for using such macros is to demonstrate that the structure of many problems follows a similar pattern, depending on the solution of a limited number of standard problems.

 

You are expected to follow the working of the macros in sufficient detail to appreciate this and to be able to make the minor changes to them to solve similar problems.

 

I have tried to make the macros transparent and relatively easy to modify.

 

In the spreadsheets, the following convention for cells is used:

Cells with a Yellow background - Headings, Incidental Information

Cells with a Green background - Input Information used in calculations on that Sheet

Intermediate Results and Calculations are not usually coloured.

 

————————————————————————————

 

The Technical Notes cover the following

 

1.   The Bootstrap

          The Bootstrap Concept

          Basic Method

          The Double Bootstrap and Bias Correction

          Parametric Bootstrap

 

 

2.   Percentiles and Confidence Intervals

          Percentiles

          Confidence Intervals by Direct Bootstrapping

          Studentization

          Percentile Methods

 

3.   Theory

          Convergence Rates

          Asymptotic Accuracy of EDF's

          Asymptotic Accuracy of Confidence Intervals

          Failure of Bootstrapping

 

4.   Monte-Carlo/Simulation Models

          Direct Models

          Metamodels

          Linear Metamodels

          NonLinear Metamodels

          Uses of Metamodels

          Metamodel Comparison and Selection

 

5.   Bootstrap Comparisons

          Goodness-of-Fit and Validation

          Comparison of Different Systems

 

6.   Bayesian Models

 

7.   Time Series Output

          Residual Sampling

          Block Sampling

          Spectral resampling

 

8.   Final Comment

 

   

————————————————————————————

 

Links

 

  Working Notes: Part I

  Working Notes: Part II

  Working Notes: Part III

  Technical Notes

 

References are at the end of the Technical Notes (some references also at the end of Part III of the Working Notes).

 

————————————————————————————

 

Synopsis of Lectures

 

Lecture #1

          W1.  Introduction

          W2. Statistical MetaModels.

          Traffic Queue Length EG

          Moroccan TB Data

          Vaso Constriction Data

          W3.  Random Variables

          W4. Fitting Parametric Distributions to Random Samples; Input Modelling

          Normal Var Generator

          Gamma Var Generator

 

Lecture #2

          W5.  Maximum Likelihood Estimation

          Likelihood Examples

          Nelder Mead.

          NelderMeadDemo

          Gamma MLE

          W6.  Accuracy of ML Estimators

          Gamma MLE

          Regression Fit Morocco Data       

          Vaso Constriction Data

 

Lecture #3

          W7.  Empirical Distribution Functions

          W8.  Basic Bootstrap Method

          Bootstrap Median

          W9.  Evaluating the Distribution of MLEs by Bootstrapping

          Gamma Bootstrap

          Vaso Constriction Data.

 

Lab #1

          Examine examples of Lectures 1,  2 and 3

 

Lecture #4

          T1 The Bootstrap

                   T1.1 The Bootstrap Concept

                   T1.2 Basic Method

                   T1.3  The Double Bootstrap and Bias Correction

          T2 Percentiles

 

Lecture #5

          W11.  The Parametric Bootstrap

          ParametricBS-GammaEG  

          T4.2 Metamodels

          T4.3 Linear Metamodels

          T4.4 Nonlinear Metamodels

 

Lab #2

          Examine examples of Lectures 3 and 5

          Fit a suitable model to the Traffic Queue and Cortisol Assay Data

 

Lecture #6

          W12  Goodness of Fit Testing

          Gamma Fit Toll Booth Data

          Normal Fit Toll Booth Data

          T5.1  Goodness-Of-Fit and Validation

 

Lecture #7

          W10  Comparing Samples Using the Basic Bootstrap

          Law and Kelton EG

          W13  Comparison of Different Models; Model Selection

          Cement Data

          T4.6  Metamodel Comparison and Selection

 

Lab #3

          Examine examples of Lectures 6 and 7

          Particular data sets you may wish to consider are

                   (i) ANOVA analysis of Tyre Wear Data

                   (ii) An analysis of Component Lifetimes. For this problem

there are some accompanying notes you should read first

to help you: Component lifetime Notes

 

Lecture #8

          T3 Theory

          T3.1  Convergence Rates

          T3.2  Asymptotic Accuracy of EDF’s

          T3.3  Asymptotic Accuracy of Confidence Intervals

          T3.4  Failure of Bootstrapping

          W14  Final Comments