Discussion: View Thread

Hello:

I have 5+ years of experience building logistic regression models, but I'd like to be more broadly informed about modeling, and I'd like to refresh my skills in interpreting models.  I'm seeking recommendations for two kinds of books on modeling.  Ideally, both would be practical in nature and refer the reader to statistical/mathematical formulas stated elsewhere. 

I'd like a book that provides a broad overview of modeling.  The book should describe different kinds of models, model selection techniques, model building strategies, caveats, etc.  It would differentiate building models with survey data from building models with web-based data.  Information on interpreting models would be a plus. 

I also would like a book specific to the field of finance, and more specifically, to modeling risk, such as credit risk.  The book needs to demonstrate how to interpret models.  I wonder if I should get a textbook with problem sets. 

By the way, I searched the posts in the last year for a similar question but didn't see anything.  Does anyone know if CNSL has ever maintained a list of reviews of member-recommended books?

Thank you,
MJ-

-------------------------------------------
Monica Johnston
Statistical Consultant & Instructor
Mostly Math
-------------------------------------------

Hi Monica,

I also will be waiting for group recommendations. But in the mean time, it is beneficial to go to Amazon, put in Credit Risk Modeling, and peruse the various books and the associated reviews.

-------------------------------------------
Patrick Spagon
-------------------------------------------

Hi Monica,

I had hoped to see some responses from others but I haven't yet so I will add my two cents:

My favorite regression book is Frank Harrell's Regression Model Strategies.  It has case studies and theories and is my go to source on this sort of thing.  I also like Applied Predictive Modeling by Kuhn and Johnson.

I also tend to look at software texts as they have the nuts and bolts on doing different procedures and examining the output.  SAS online documentation is great.  I don't use R much but I like the SAS and R books and blog by Horton and Kleinman, the blog often gives me some good ideas on stuff, not just modeling.  

I like Agresti's applied categorical analysis book for GLM models.  And recently have spent a lot of time with Hilbe's Negative Binomial Regression book.  I have basically decided that Poisson regression is a waste of time :)

Here are some things to at least get things started and maybe get others to speak up.

Good luck!

-------------------------------------------
Jason Brinkley
East Carolina University
-------------------------------------------

The following message has been sent to you in response to your eGroup message

Message From: Georgette Asherman

The Kuhn & Johnson book is good.  I heard Max Kuhn discuss it at a local talk and it sounds rigorous and useful.   Every statistician should own The Elements of Statistical Learning (Hastie, Tibshirani, Friedman).  The pdf is available for free download but I find the paper version valuable, much like the Tufte books.    Most likely you will never use more than 1 or 2 approaches.  But so much is written now of predictive models in unexpected places that  the theoretical view of a technique gives a heads-up on what the researcher or entrepreneur is promoting.   Their work at Stamford is the connection between the 'two worlds' of inference and algorithm.  

The problem with other books in this field is that many of the writers work for software companies or have some other agenda.   For example one book I own has a high price-tag for pages of screen shots from their software.  It is a very shallow introduction that doesn't have much value for someone with 5 years experience.    To get examples in vertical applications it might make sense to go to case studies in the software documentation, e.g. SAS Enterpirse Miner and SPSS Clementine.   Even if you use the base software, the front-end applications deal with the data management and model life-cycle.   Also look in the operation research community (INFORMS) because they seem to focus more on business analytics.   

 

-------------------------------------------
Georgette Asherman
-------------------------------------------

I totally agree with the Hastie, Tibshirani, and Friedman choice.  But it's a hard read (he says as though he has actually read it cover to cover).  Our group just bought a copy of An Introduction to Statistical Learning by James, Witten, Hastie and Tibshirani.  

-------------------------------------------
Jason Brinkley
East Carolina University
-------------------------------------------



Original Message:
Sent: 07-14-2014 17:08
From: Georgette Asherman
Subject: ISO: Recommendations for books on modeling

The following message has been sent to you in response to your eGroup message

Message From: Georgette Asherman

The Kuhn & Johnson book is good.  I heard Max Kuhn discuss it at a local talk and it sounds rigorous and useful.   Every statistician should own The Elements of Statistical Learning (Hastie, Tibshirani, Friedman).  The pdf is available for free download but I find the paper version valuable, much like the Tufte books.    Most likely you will never use more than 1 or 2 approaches.  But so much is written now of predictive models in unexpected places that  the theoretical view of a technique gives a heads-up on what the researcher or entrepreneur is promoting.   Their work at Stamford is the connection between the 'two worlds' of inference and algorithm.  

The problem with other books in this field is that many of the writers work for software companies or have some other agenda.   For example one book I own has a high price-tag for pages of screen shots from their software.  It is a very shallow introduction that doesn't have much value for someone with 5 years experience.    To get examples in vertical applications it might make sense to go to case studies in the software documentation, e.g. SAS Enterpirse Miner and SPSS Clementine.   Even if you use the base software, the front-end applications deal with the data management and model life-cycle.   Also look in the operation research community (INFORMS) because they seem to focus more on business analytics.   

 

-------------------------------------------
Georgette Asherman
-------------------------------------------

I, too, recommend An Introduction to Statistical Learning by James, Witten, Hastie, and Tibshirani.  It is the assigned text for a class I am taking on Computational Statistics.  It presents the material in a well-organized manner, uses marginal annotation to highlight important terms, and, has a lot of R code in lab format at the end of each chapter.  Great textbook for class and a great reference, as well.

-------------------------------------------
Beverly Grunden
Statistical Consultant
Wright State University
-------------------------------------------



Original Message:
Sent: 07-14-2014 17:15
From: Jason Brinkley
Subject: ISO: Recommendations for books on modeling

I totally agree with the Hastie, Tibshirani, and Friedman choice.  But it's a hard read (he says as though he has actually read it cover to cover).  Our group just bought a copy of An Introduction to Statistical Learning by James, Witten, Hastie and Tibshirani.  

-------------------------------------------
Jason Brinkley
East Carolina University
-------------------------------------------

I recommend

Frank Harrell, Regression modelling strategies  - for general methods


-------------------------------------------
Peter Flom
-------------------------------------------



Original Message:
Sent: 07-14-2014 17:08
From: Georgette Asherman
Subject: ISO: Recommendations for books on modeling

The following message has been sent to you in response to your eGroup message

Message From: Georgette Asherman

The Kuhn & Johnson book is good.  I heard Max Kuhn discuss it at a local talk and it sounds rigorous and useful.   Every statistician should own The Elements of Statistical Learning (Hastie, Tibshirani, Friedman).  The pdf is available for free download but I find the paper version valuable, much like the Tufte books.    Most likely you will never use more than 1 or 2 approaches.  But so much is written now of predictive models in unexpected places that  the theoretical view of a technique gives a heads-up on what the researcher or entrepreneur is promoting.   Their work at Stamford is the connection between the 'two worlds' of inference and algorithm.  

The problem with other books in this field is that many of the writers work for software companies or have some other agenda.   For example one book I own has a high price-tag for pages of screen shots from their software.  It is a very shallow introduction that doesn't have much value for someone with 5 years experience.    To get examples in vertical applications it might make sense to go to case studies in the software documentation, e.g. SAS Enterpirse Miner and SPSS Clementine.   Even if you use the base software, the front-end applications deal with the data management and model life-cycle.   Also look in the operation research community (INFORMS) because they seem to focus more on business analytics.   

 

-------------------------------------------
Georgette Asherman
-------------------------------------------

It may be hard to find a book that does what you'd like for the whole world of modeling, but you
may be able to find books with a more practical slant in individual areas of modeling.

For example, Collett's "Modelling Binary Data" has served me well in its specific bailiwick.

One book that I have seen that might be worth a look is "Regression Modelling Strategies" by Harrell,
which has a number of case studies for linear and logistic regression and survival analysis.
There's also the classic "Regression Analysis by Example" (Chatterjee).

>>Kathy

-------------------------------------------
Katherine Godfrey
-------------------------------------------