VI. Evaluate Model Fit Basic questions that modelers must address are: How well does the model fit the data? Do changes to a model, such as reparameterization,

Slides:



Advertisements
Similar presentations
Multiple Regression and Model Building
Advertisements

Lecture (11,12) Parameter Estimation of PDF and Fitting a Distribution Function.
Hypothesis Testing Steps in Hypothesis Testing:
Irwin/McGraw-Hill © Andrew F. Siegel, 1997 and l Chapter 12 l Multiple Regression: Predicting One Factor from Several Others.
Hypothesis: It is an assumption of population parameter ( mean, proportion, variance) There are two types of hypothesis : 1) Simple hypothesis :A statistical.
Regression Analysis Once a linear relationship is defined, the independent variable can be used to forecast the dependent variable. Y ^ = bo + bX bo is.
Correlation and regression
Chapter 12 Simple Linear Regression
© 2010 Pearson Prentice Hall. All rights reserved Least Squares Regression Models.
1-1 Copyright © 2015, 2010, 2007 Pearson Education, Inc. Chapter 25, Slide 1 Chapter 25 Comparing Counts.
Regression Analysis. Unscheduled Maintenance Issue: l 36 flight squadrons l Each experiences unscheduled maintenance actions (UMAs) l UMAs costs $1000.
Chapter 12 Simple Regression
Chapter 7 Sampling and Sampling Distributions
The Simple Regression Model
Sampling Theory Determining the distribution of Sample statistics.
Introduction to Regression Analysis, Chapter 13,
Inferential Statistics
1 1 Slide © 2008 Thomson South-Western. All Rights Reserved Slides by JOHN LOUCKS & Updated by SPIROS VELIANITIS.
AM Recitation 2/10/11.
Inference for regression - Simple linear regression
Regression Analysis (2)
CPE 619 Simple Linear Regression Models Aleksandar Milenković The LaCASA Laboratory Electrical and Computer Engineering Department The University of Alabama.
Copyright © 2013, 2010 and 2007 Pearson Education, Inc. Chapter Inference on the Least-Squares Regression Model and Multiple Regression 14.
Simple Linear Regression Models
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Chapter 26 Comparing Counts.
Copyright © 2010 Pearson Education, Inc. Warm Up- Good Morning! If all the values of a data set are the same, all of the following must equal zero except.
Slide 1 Copyright © 2004 Pearson Education, Inc..
© 2002 Prentice-Hall, Inc.Chap 14-1 Introduction to Multiple Regression Model.
IV. Sensitivity Analysis for Initial Model 1. Sensitivities and how are they calculated 2. Fit-independent sensitivity-analysis statistics 3. Scaled sensitivities.
Estimation of Statistical Parameters
Population All members of a set which have a given characteristic. Population Data Data associated with a certain population. Population Parameter A measure.
1 1 Slide © 2005 Thomson/South-Western Slides Prepared by JOHN S. LOUCKS St. Edward’s University Slides Prepared by JOHN S. LOUCKS St. Edward’s University.
III. Ground-Water Management Problem Used for the Exercises.
Statistical Decision Making. Almost all problems in statistics can be formulated as a problem of making a decision. That is given some data observed from.
POSC 202A: Lecture 12/10 Announcements: “Lab” Tomorrow; Final ed out tomorrow or Friday. I will make it due Wed, 5pm. Aren’t I tender? Lecture: Substantive.
1 1 Slide © 2014 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole.
EQT 373 Chapter 3 Simple Linear Regression. EQT 373 Learning Objectives In this chapter, you learn: How to use regression analysis to predict the value.
1 Chapter 12 Simple Linear Regression. 2 Chapter Outline  Simple Linear Regression Model  Least Squares Method  Coefficient of Determination  Model.
Multiple Regression and Model Building Chapter 15 Copyright © 2014 by The McGraw-Hill Companies, Inc. All rights reserved.McGraw-Hill/Irwin.
Statistical analysis Outline that error bars are a graphical representation of the variability of data. The knowledge that any individual measurement.
Inference for Regression Simple Linear Regression IPS Chapter 10.1 © 2009 W.H. Freeman and Company.
MGS3100_04.ppt/Sep 29, 2015/Page 1 Georgia State University - Confidential MGS 3100 Business Analysis Regression Sep 29 and 30, 2015.
Section 9-1: Inference for Slope and Correlation Section 9-3: Confidence and Prediction Intervals Visit the Maths Study Centre.
1 Chapter 6 Estimates and Sample Sizes 6-1 Estimating a Population Mean: Large Samples / σ Known 6-2 Estimating a Population Mean: Small Samples / σ Unknown.
Section 10.1 Confidence Intervals
Understanding Your Data Set Statistics are used to describe data sets Gives us a metric in place of a graph What are some types of statistics used to describe.
CHEMISTRY ANALYTICAL CHEMISTRY Fall Lecture 6.
Copyright © 2010 Pearson Education, Inc. Warm Up- Good Morning! If all the values of a data set are the same, all of the following must equal zero except.
Correlation & Regression Analysis
The Practice of Statistics, 5th Edition Starnes, Tabor, Yates, Moore Bedford Freeman Worth Publishers CHAPTER 12 More About Regression 12.1 Inference for.
Basic Business Statistics, 10e © 2006 Prentice-Hall, Inc. Chap 15-1 Chapter 15 Multiple Regression Model Building Basic Business Statistics 10 th Edition.
1 Statistics 262: Intermediate Biostatistics Regression Models for longitudinal data: Mixed Models.
1 Probability and Statistics Confidence Intervals.
Comparing Counts Chapter 26. Goodness-of-Fit A test of whether the distribution of counts in one categorical variable matches the distribution predicted.
Chapter 9: Introduction to the t statistic. The t Statistic The t statistic allows researchers to use sample data to test hypotheses about an unknown.
Hypothesis Testing. Statistical Inference – dealing with parameter and model uncertainty  Confidence Intervals (credible intervals)  Hypothesis Tests.
The Practice of Statistics, 5th Edition Starnes, Tabor, Yates, Moore Bedford Freeman Worth Publishers CHAPTER 12 More About Regression 12.1 Inference for.
Statistical Decision Making. Almost all problems in statistics can be formulated as a problem of making a decision. That is given some data observed from.
Chapter 15 Inference for Regression. How is this similar to what we have done in the past few chapters?  We have been using statistics to estimate parameters.
Chapter 14 Introduction to Multiple Regression
CHAPTER 12 More About Regression
Chapter 25 Comparing Counts.
POSC 202A: Lecture Lecture: Substantive Significance, Relationship between Variables 1.
Statistical Methods For Engineers
Simple Linear Regression
CHAPTER 12 More About Regression
Chapter 26 Comparing Counts.
Product moment correlation
CHAPTER 12 More About Regression
MGS 3100 Business Analysis Regression Feb 18, 2016
Presentation transcript:

VI. Evaluate Model Fit Basic questions that modelers must address are: How well does the model fit the data? Do changes to a model, such as reparameterization, actually improve the model fit? What aspects of the model or data need to be changed to improve model fit? The methods described here help answer these questions through evaluation of: Statistical Measures of Overall Model Fit Graphical Analysis of Model Fit and Related Statistics Parameter Statistics Model Linearity

VI. Evaluating Model Fit Part 1: Statistical Measures of Overall Fit Objective function values Calculated error variance, standard error, and fitted error statistics The AIC and BIC statistics

Objective Function Values Values of objective functions, such as the weighted least-squares objective function, are a basic measure of model fit. During regression, the objective is to find the set of parameter values that minimizes the objective function. Ideally, as regression proceeds, the model fit is improved. Two commonly used objective functions are the weighted least-squares objective function, used in this class, and the maximum likelihood objective function. Weighted least-squares objective function: where y represents an observed value or a prior value.

Objective Function Values Maximum likelihood objective function (simplified form): The first term is a function of the total number of observations plus prior values. The second term is a function of the weighting; for a diagonal weight matrix, the determinant is simply the product of the diagonal elements (the weights). The third term is the weighted least-squares objective function. Unlike the weighted least-squares objective function, the maximum likelihood objective function can be negative. DO EXERCISE 6.1a: Examine objective-function values at the bottom of file ex5.2c.#uout. where determinant of the weight matrix. is the

Objective Function Values Values of objective functions, such as the weighted least-squares objective function, are a basic measure of model fit. During regression, the objective is to find the set of parameter values that minimizes the objective function. Ideally, as regression proceeds, the model fit is improved. Two commonly used objective functions are the weighted least-squares objective function, used in this class, and the maximum likelihood objective function. Weighted least-squares objective function: where y represents an observed value or a prior information value. DO EXERCISE 6.1a: Examine objective-function values.

Calculated Error Variance and Standard Error (Book, p ) Problem with using objective function values to assess model fit: They do not account for the negative effects of increasing the number of parameters, and are of limited use in comparing models with different parameterization schemes. Adding more parameters almost always improves the objective function value, but the parameter estimates become less reliable. The calculated error variance s 2 accounts for the effects of adding more parameters. As NP increases, the denominator decreases, and s 2 increases: The square root of s 2 is s, which is the standard error of the regression. Both s 2 and s are dimensionless, and unlike S, can be used to compare the results of models with different parameterizations (but not models with different weighting schemes).

Calculated Error Variance and Standard Error If the fit achieved by the regression is consistent with the accuracy of the observation data, as expressed by the weighting, then the expected value of both s 2 and s is 1.0. This can be demonstrated using the exercise of Hill and Tiedeman (2007, p , exer. 6.1b), which we will not go over in class. Given that we expect s 2 and s to be 1.0 if the model fit is consistent with the observation errors as represented in the weight matrix, deviations from 1.0 can be interpreted in the context of observation error and model error. This insight into model error can be very useful. The following slides explain how this is done.

Calculated Error Variance and Standard Error In practice, values for s 2 and s often deviate from 1.0. Significant deviations from 1.0 indicate that the model fit to the observation data is inconsistent with the statistics used to calculate the weights. This doesn’t necessarily mean these statistics are wrong. Step 1: Test whether s 2 significantly deviates from 1.0. Construct a confidence interval for the true error variance: tail values of a chi-square distribution with n degrees of freedom. and define the upper and lower chi-square distribution Confidence interval on s: take the square root of the limits on s 2.

Calculated Error Variance and Standard Error Interpretation of 95% confidence intervals on s 2 : If the interval includes 1.0, and the weighted residuals are random, then s 2 does not significantly deviate from 1.0. The model fit is consistent with the statistics used to calculate the weights. Expressed in terms of probability, there is only a 5% chance that the model fit to the data contradicts the assumptions that (1) the model is reasonably accurate and (2) the statistics used to calculate the weights correctly reflect the observation errors. If the entire interval is less than 1.0, and the weighted residuals are random, the model fits better than anticipated, based on the weighting used. This is generally not problematic, but is only common in test cases. If the entire interval is greater than 1.0, then s 2 is significantly greater than 1.0, and the model fit is worse than anticipated based on the weighting used. In this situation, the interpretation depends on whether or not the weighted residuals are random.

Calculated Error Variance and Standard Error If the entire interval > 1.0 and the weighted residuals are random: Reevaluate the weighting The weights are calculated using variances, standard deviations, coefficients of variation. Calculate values of these statistics that are consistent with the model fit. Multiply the variances by s 2 ; the standard deviations and coefficients of variation by s. If the model were re-run with the resulting weights, parameter estimates and residuals would be the same, but s 2 would equal 1.0. If the recalculated statistics can be justified (observation error could be larger than originally assumed), no indication of model error. If the recalculated statistics cannot be justified, model error may be as much as s times the observation error. There is some indication (Hill+, 1998) that the model error can be correctly represented with common uncertainty measures, but more work is needed to be sure.

Calculated Error Variance and Standard Error If the entire interval > 1.0 and the weighted residuals are not random: Significant model error is indicated. Try to find and correct the model error. Inspect weighted residuals individual and examine spatial and temporal patterns. Evaluate the model carefully for data input errors and consistency with independent information about the system.

Fitted Error Statistics (Book, p ) s and s 2 are dimensionless. Difficult to convey goodness of fit to others using dimensionless numbers. Fitted error statistic (not standard statistical terminology) reflects model fit in the same units as one type of observation. Calculation: s × (std dev) or (coef of var) used to define weights for a group of observations. Fitted standard deviation, on average, the difference between simulated values and observations for the group. For a few observations, just report the weighted residuals DO EXERCISE 6.1c: Evaluate calculated error variance, standard error, and fitted error statistics. For fitted error to heads, compare to overall head loss in the system.

AIC and BIC Statistics AIC (Akaike’s Index Criteria), AICc, and BIC more strongly account for the negative effect of increasing the number of estimated parameters when comparing alternative models. Smaller values indicate better models. Start with the maximum-likelihood objective function, S’ and add one or two terms that are a function of the number of parameters: DO EXERCISE 6.1d: Examine the AIC and BIC statistics. Use AICc if NOBS/NP<40

New exercise These figures show the value of the added terms given different numbers of observations and parameters. Based on theory: BIC uses additional data to focus in on an existing model with fewer parameters, while AIC and AICc is more likely to choose a model with more parameters when there are more data. How do these graphs support or refute the theory?