1 / 85

Introduction to Generalized Linear Models

Introduction to Generalized Linear Models. Prepared by Louise Francis Francis Analytics and Actuarial Data Mining, Inc. www.data-mines.com September 18, 2005. Objectives. Gentle introduction to Linear Models and Generalized Linear Models Illustrate some simple applications

edwardrtaylor
Download Presentation

Introduction to Generalized Linear Models

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Introduction to Generalized Linear Models Prepared by Louise Francis Francis Analytics and Actuarial Data Mining, Inc. www.data-mines.com September 18, 2005

  2. Objectives • Gentle introduction to Linear Models and Generalized Linear Models • Illustrate some simple applications • Show examples in commonly available software • Address some practical issues

  3. Predictive Modeling Family

  4. A Brief Introduction to Regression • Fits line that minimizes squared deviation between actual and fitted values

  5. Simple Formula for Fitting Line

  6. Excel Does Regression • Install Data Analysis Tool Pak (Add In) that comes wit Excel • Click Tools, Data Analysis, Regression

  7. Analysis of Variance Table

  8. Goodness of Fit Statistics • R2: (SS Regression/SS Total) • percentage of variance explained • F statistic: (MS Regression/MS Residual) • significance of regression • T statistics: Uses SE of coefficient to determine if it is significant • significance of coefficients • It is customary to drop variable if coefficient not significant • Note SS = Sum squared errors

  9. Output of Excel Regression Procedure

  10. Other Diagnostics: Residual Plot • Points should scatter randomly around zero • If not, a straight line probably is not be appropriate

  11. Other Diagnostics: Normal Plot • Plot should be a straight line • Otherwise residuals not from normal distribution

  12. Non-Linear Relationships • The model fit was of the form: • Severity = a + b*Year • A more common trend model is: • SeverityYear=SeverityYear0*(1+t)(Year-Year0) • T is the trend rate • This is an exponential trend model • Cannot fit it with a line

  13. Exponential Trend – Cont. • R2 declines and Residuals indicate poor fit

  14. A More Complex Model • Use more than one variable in model (Econometric Model) • In this case we use a medical cost index, the consumer price index and employment data (number employed, unemployment rate, change in number employed, change in UEP rate to predict workers compensation severity

  15. Multivariable Regression

  16. One Approach: Regression With All Variables • Many variables not significant • Over-parameterization issue • How get best fitting most parsimonious model?

  17. Multiple Regression Statistics

  18. Stepwise Regression • Partial correlation • Correlation of dependent variable with predictor after all other variables are in model • F – contribution • Amount of change in F-statistic when variable is added to model

  19. Stepwise regression-kinds • Forward stepwise • Start with best one variable regression and add • Backward stepwise • Start with full regression and delete variables • Exhaustive

  20. Stepwise regression for Severity Data

  21. Stepwise Regression-Excluded Variables

  22. Econometric Model Assessment • Standardized residuals more evenly spread around the zero line • R2 is .91 vs .52 of simple trend regression

  23. Correlation of Predictor Variables: Multicollinearity

  24. Multicollinearity • Predictor variables are assumed uncorrelated • Assess with correlation matrix

  25. Remedies for Multicollinearity • Drop one or more of the highly correlated variables • Use Factor analysis or Principle components to produce a new variable which is a weighted average of the correlated variables • Use stepwise regression to select variables to include

  26. Degrees of Freedom • Related to number of observations • One rule of thumb: subtract the number of parameters estimated from the number of observations • The greater the number of parameters estimated, the lower the number of degrees of freedom

  27. Degrees of Freedom • “Degrees of freedom for a particular sum of squares is the smallest number of terms we need to know in order to find the remaining terms and thereby compute the sum” • Iverson and Norpoth, Analysis of Variance • We want to keep the df as large as possible to avoid overfitting • This concept becomes particularly important with complex data mining models

  28. Regression Output cont. • Standardized residuals more evenly spread around the zero line – but pattern still present • R2 is .84 vs .52 of simple trend regression • We might want other variables in model (i.e, unemployment rate), but at some point overfitting becomes a problem

  29. Tail Development Factors: Another Regression Application • Typically involve non-linear functions: • Inverse Power Curve: • Hoerel Curve: • Probability distribution such as Gamma, Lognormal

  30. Non-Linear Regression • Use it to fit a non-linear function where transformation to linear model will not work • Example • LDF = (Cumulative % Incurred at t+1)/ (Cumulative % Incurred a t) Assume gamma cumulative distribution

  31. Example: Inverse Power Curve • Can use transformation of variables to fit simplified model: LDF=1+k/ta • ln(LDF-1) =a+b*ln(1/t) • Use nonlinear regression to solve for a and c • Uses numerical algorithms, such as gradient descent to solve for parameters. • Most statistics packages let you do this

  32. Nonlinear Regression: Grid Search Method • Try out a number of different values for parameters and pick the ones which minimize a goodness of fit statistic • You can use the Data Table capability of Excel to do this • Use regression functions linest and intercept to get k and a • Try out different values for c until you find the best one

  33. Fitting non-linear function

  34. Using Data Tables in Excel

  35. Use Model to Compute the Tail

  36. Claim Count Triangle Model • Chain ladder is common approach

  37. Claim Count Development • Another approach: additive model • This model is the same as a one factor ANOVA

  38. ANOVA Model for Development

  39. ANOVA: Two Groups • With only two groups we test the significance of the difference between their means • Many years ago in a statistics course we learned how to use a t-test for this

  40. ANOVA: More than 2 Groups

  41. Correlation Measure: Eta

  42. ANOVA Model for Development

  43. Regression With Dummy Variables • Let Devage24=1 if development age = 24 months, 0 otherwise • Let Devage36=1 if development age = 36 months, 0 otherwise • Need one less dummy variable than number of ages

  44. Regression with Dummy Variables: Design Matrix

  45. Equivalent Model to ANOVA

  46. Apply Logarithmic Transformation • It is reasonable to believe that variance is proportional to expected value • Claims can only have positive values • If we log the claim values, can’t get a negative • Regress log(Claims+.001) on dummy variables or do ANOVA on logged data

  47. Log Regression

  48. Poisson Regression • Log Regression assumption: errors on log scale are from normal distribution. • But these are claims – Poisson assumption might be reasonable • Poisson and Normal from more general class of distributions: exponential family of distributions

  49. “Natural” Form of the Exponential Family

  50. Specific Members of the Exponential Family • Normal (Gaussian) • Poisson • Binomial • Negative Binomial • Gamma • Inverse Gaussian

More Related