bookbannerv2
udemlogo

 

 

 

 

 

 

Support for Chapter 1

Support for Chapters 4 and 5

Support for Chapter 6

Support for Chapter 7

Supporting materials to chapters in "Key Demographics in Retirement Risk Management"?

Note: This page contains or provides links to various kinds of information that buyers of "Key Demographics ..." are promised in the text of the book. The time window between the news of a psotive decision to publish (by Springer) and the publication date has not allowed us to assemble all of these materials for prompt posting on the Internet. Thus, over the next several weeks new items from the list will become available on the Internet, at this page. The Table of Contents below will show you what is available now. Thanks in advance for your patience.

Table of Contents

" Design of the composite indicators based on the General Social Survey and the Health and Retirement Survey", by Leroy O. Stone

Abstract: This appendix is the technical introduction to two composite indicators. The first indicator is the survey-based measure of overall preparedness to meet retirement’s challenges, based on the 2007 General Social Survey of Statistics Canada. It is used in Chapter Seven. The second pertains to the measure of potential to engage in effective retirement related risk management, based on the Early Release Core of the 2008 Health and Retirement Survey (HRS) of the University of Michigan. A few results from the analysis of HRS data are in the Discussion section of Chapter 7. Click here to get the PDF file with the details, and remember to use the password "CPRM" to gain access to this file.

Click here to get the goodness-of-fit table for the model fitted in Chapter 7.

Note: More support for Chapter 7 will arrive here in the coming weeks.

 

Go to model development details

Go to goodness-of-fit testing details

• “ Developing and testing the prediction model”,  by Leroy O. Stone


Extracts from the full 23-page document

This text is technical appendix C to Leroy O. Stone (Ed.), Key Demographics in Retirement Risk Management (Springer, 2012), available at http://www.springer.com/social+sciences/population+studies/book/978-94-007-4043-3 . …   
This appendix supports Chapter 6 of “Key Demographics  …” by providing essential background information about the rationale for using parameters estimate from the 2007 Desjardins Financial Retirement Survey to impute within the 2007 General Social Survey file the scores of a variable that entered into the measure of performance of retirement risk management.

Our strategy involves developing the best model we could find to predict retirement related risk management within the DFS using its richer measurement of that behaviour (see “Design of the composite indicators based on the General Social Survey  and the Health and Retirement Survey”, by  Leroy O. Stone and available at this website ), under the constraint that the chosen explanatory variables be limited to a list for which theoretically equivalent variables can be defined in the GSS database. Given this constraint, the parameter estimates produced in modeling risk-management behaviour in the DFS sample are used in a prediction model to simulate the positions of GSS respondents among levels of the risk-management indicator.

Before doing so, however, we would want to ensure that there is a reasonably good fit for the model within the DFS sample. Importantly, the goodness of fit test should include fitting the model in a sub-sample containing those respondents whose data were not used in estimating the parameters


The sample was randomly split into two halves -- sub-sample A and sub-sample B. Only sub-sample A is used to estimate the parameters of the model. After confirming that there is substantial coherence between the structure of parameters so estimated and that obtained when we use the entire sample, the model is then used to make predictions within sub-sample B. The results are shown in Table C.3.

Table C.3 part 1

Table C.3 part 2

In order to interpret the results for sub-sample B in Table C.3, two benchmark figures are needed. The first is the contingency coefficient within sub-sample A. This is usually an upper bound for the coefficient obtained in sub-sample B -- usually it is the best that one can expect. The ratio of the contingency coefficient in sub-sample B to that in sub-sample A measures the predictive efficiency of the model -- how well its operation in sub-sample B makes use of the available information. This coefficient is at the 9% level.

The second benchmark is produced by the cross-classification of sex against the observed distribution of the population over levels of the indicator. The ratio of the contingency coefficient for sub-sample B to that of the just mentioned cross-classification is 2.1 -- in other words, in sub-sample B the model is twice as good as predicting using sex alone.

While this is not especially impressive, an inspection of the table shows a very meaningful association of sex with the distribution of the population over levels of the indicator. Men are much more likely than women to score in the upper two quadrants of the indicator. A model that performs twice as well as sex alone (which does a seemingly decent job) has produced a nontrivial level of prediction accuracy, all of this in sub-sample B -- their members were not used to estimate the parameters of the model.

We conclude that within the context of the DFS sample, where the questionnaire has allowed us to construct a indicator of retirement related risk management activities, the prediction model does a tolerably good, though not very impressive, job.

Let us turn now to the question of the cogency of the model. We say the model is cogent when we find meaningful and internally coherent patterns of association between categories of the predictor variables and the probability of scoring high, or scoring low as the case may be, on the indicator. These patterns are found in the odds ratios for categories of the predictor variables, and these are routinely produced by SAS’s PROC LOGISTIC. Table C.4 shows the pertinent odds ratios.

If you have bought the book you will have free access to the full document by sending a request to me at info@retirementresearch.org .  Thanks in advance.

 

> back to the main page>

 

 

“Goodness-of-fit testing by means of sub-population divergences”,  by Leroy O. Stone


Introduction

This text is technical appendix D2012 to Leroy O. Stone (Ed.), Key Demographics in Retirement Risk Management (Springer, 2012), available at http://www.springer.com/social+sciences/population+studies/book/978-94-007-4043-3 . …   

This appendix supports Chapter 6 of “Key Demographics  …” by providing details about the goodness-of-fit testing procedures and their results. … We begin by reviewing the goodness-of-fit computations that are produced routinely in the output of SAS PROC LOGISTIC. Then follows some specialized goodness-of-fit computations that we have devised for the case where the focus is upon the identification of key demographics.

The important distinction between these two contexts is that in the former the question is how well the model works in predicting the outcomes for particular cases (respondents), whereas in the latter the focus is on how well the model performs in predicting the distribution of outcomes for selected population subgroups. The latter approach is consistent with our focus upon the study of populations, a demographic analysis.

Click here to get the full 9-page PDF document.

> back to the main page>

 

• “ Materials to support book chapters: supplementary goodness-of-fit tables for Chapters 4, 5 and 7 ”,  by Leroy O. Stone


Introduction

This document is a supplement to Leroy O. Stone (Ed.), Key Demographics in Retirement Risk Management (Springer, 2012), available at http://www.springer.com/social+sciences/population+studies/book/978-94-007-4043-3 . …   
It supports Chapters 4, 5 and 7 of “Key Demographics  …” by presenting the goodness-of-fit tables that were prepared for these chapters. The only text drafted about these tables comprises their essential “messages”, and these are to be found in the main text of the book. Accordingly, no text accompanies the pages that follow.

Click here to get the full 8-page PDF document. Remember to use the password CPRM to open the document.

Note: More support for Chapters 4 and 5 will arrive here in the coming weeks.

> back to the main page>

(c) 2012 Leroy O. Stone. All rights reserved.

 

 

buy1
buy2