Regression Analysis?

[jtmilby]

Does anyone have any knowledge on where I could purchase a book or some type of learning instrument on how to understand the process of Regression Analysis.
I am having problems with this concept and my supervisor (who has been an appraiser for 25 yrs) has never even heard of it.
I just want to know all I can about this profession, being a trainee in a small rural area competing against two established appraisers, I need all the help I can get.

Any help would be greatly appreciated!

 

[TC]

You might want to check out this website for more info on RA.

TC

http://www.mckissock.com/

 

[Austin]

The Appraisal Institute recently published a book on the subject: "A Guide to Appraisal Valuation Modeling" by Mark Linne, M. Steven Rane, and George Dell. This book will give you the basics of the regression math method and a demonstration of how it works. I have worked with regression for about five years and have used it for about the same length of time. There are two things you have to understand about regression that most people can't seem to grasp: First is the math involved. The math is the easy part because the computer does all of the work. You just understand the concept and push the buttons. Second: The hard part is the algorithm. What I have found from using regression methods is that the present appraisal theory is all-wrong. The sequence of adjustments is wrong, the whole adjustment mentality is wrong; the cost approach theory is interesting but wrong, etc. The problem with this is that most people are reactionary and are afraid to touch anything that is not conventional thought. They are waiting for some authority like the AI to say "it is ok." The thing that makes using regression methods difficult is that you have to undersatnd the math concept, know appraisal theory, not be tied to convention, and be an outside the box thinker. Regression is pure science. You don’t start with any perceived ideas about what a property is worth. You follow a logical sequence that leads you through the process and forces you to treat anything that affects value. It is not just a fill in the blanks and hit the enter key process like an AVM. AVM will never work because it starts off assuming it knows what is influencing price. You have to work your way through using different value factors to find out what is influencing price or more importantly what is not. As of this date, I have not seen a book that describes these methods and for a good reason; you can’t teach people to think and use logical reasoning. If you could, we would not need appraisers. I started writing a book on the subject and gave up after 10 pages. You can’t teach it with a book. You have to sit down at a computer and experience it. To write an article on the subject and give the supporting data takes pages of ANOVA tables and graphs, which is not possible. People are afraid of regression and AVM because they don't know what they are capable of. Regression is very complicated and in my opinion will never be widely used. You can learn a lot about appraisal theory by playing with regression, and you can use simple regression based on the new theory. How to present it to the mass of appraisers is the problem.

 

[bradellis]

Austin,

You are right about RA providing interesting insights; however, I would remind everyone that this is a tool- nothing more. No mathematical model will be able to measure the intangible aspects of how both buyers and sellers make their decisions.

You run a model based upon SF only to find that in one data set the higher prices go to the smaller homes. Shocking! Why? The smaller homes in the market are all older and have 10' ceilings making all the rooms feel bigger. Why was this not reflected in the data? Because no one had a data base for it.

So, RA can be a great tool, and if you know how to use it, by all means have at it. Just remember that it is a tool. It just analyzes data just like the appraiser analyzes non-numerical data.

Brad

 

[Neil (Texas)]

Instead of contending with Jameson's theory of random sequences, RA dances nimbly around it.

 

[Austin]

bradellis & Neil: With all due respects, you guys could not be further from the truth. Bradellis said; “No mathematical model will be able to measure the intangible aspects of how both buyers and sellers make their decisions.” Reply: Regression comes closer to doing that than any known methodology, except of course Mrs. Cleo. The only way to deal with intangibles is on a statistical basis. If you think the regression model cannot account for intangibles, then how do you think the present method of using adjustments out of sequence and totally unsupported by data can do it? In a side-by-side comparison regression makes a mockery of the existing model, which is its self, a crude regression model. Each existing method of estimating price is in and of itself a regression model and the entire appraisal process is collectively a crude intuitive regression model. The only difference is that math regression modeling is based on statistical data and the existing method relies on the intuition of the appraiser.
Neil says that regression does not address randomness. To the contrary, nothing addresses it better. If you pick the 30 most comparable sales, the regression predicts a sale prices with a small variance from the actual sale prices, and then what is left is general randomness. Mathematically, the law of numbers says that the present appraisal method cannot even answer the question the definition of market value ask, that being “what is the most probable price.” The present method doesn’t even address general randomness nor does it acknowledge the covariance of independent variables. General market randomness and covariance of variables far exceeds in significance any major adjustment on the existing marketing grid and the existing marketing grid does not address these factors. The test of any model is how well does it explain the correlation of price (dependent variable) and value influencing factors (independent variables). Regression modeling verifies its prediction and does it with a statistically significant amount of data. The present method asks up to just take the appraisers word for it. The present cost approach is based on one theory and the sales comparison is based on a totally different algorithm. As an analogy, the cost approach puts a watch together in one sequence based on one theory, and the sales comparison approach attempts to take it apart in another sequence based on another theory. The income approach recognizes collective obsolescence as reflected by the market generated rental level, and the cost approach attempts to measure it piece meal. How many times have you heard it said: "The cost approach is always higher and sets the upper limit of price." Well that is the reason, you can't measure covariance of variables using a piece meal methodology.
Then too, how can you criticize and condemn something that you haven’t even seen?

 

[Neil (Texas)]

No, Austin, I did not say RA does not address randomness. I said RA dances nimbly around Jameson's theory of random sequences. There is great difference in addressing part of an element and failing to consider the whole (akin to describing, from a stationary point, the entire surface of a stationary three-dimensional object).

I have to concur with Brad that RA is simply a tool. One that can be used or misused. The very term (regression "analysis") patently implies something other than "pure science."

By the way, I hope you reconsider writing the book, or at least a treatise on the subject.

 

[Austin]

Neil: In that case, you have further complicated the problem. The sequences in real estate appraisal are not random at all. The sequence was determined when we did the cost approach. We start of with a basic quality of construction cost factor for the basic building and add value influencing factors to it such as basement, garages, etc. When we do the sales comparison approach we use regression in the reverse order of the cost approach to unravel the sequence, thus the sequence is not random. When you use the conventional marketing grid you are “addressing part of an element and failing to consider the whole” because you can’t measure covariance or get a view of general market randomness which is something totally different. I would write a book but I don’t think I have enough family members to purchase it to justify the effort. I have three cousins with PhD’s and they all gave me a copy of their doctoral thesis. Interesting binding job but that was as far as I got. Architectural history, English Literature, and a statistical study of education methods fail to arouse my interest. Maybe my cousins would be impressed with my book deal. :lol:

 

[jtrotta]

Math is fun, you can have numbers do any variation in the world, but will have to agree with Neil, and others that it is a tool and used improperly it can create as much trouble, as help.

Don't know if they used at Enron, but when you get that many number crunchers together, and the money flows in the wrong direction, it leads one to believe, that there's more than math at stake. People play a big part, their decision comes first, then number crunching begins. Emotional deceisions are the heartbeat of homebuying.

The appraisal field is and will always be interesting for many reasons and as we discover new things now, the future will bring out even more. I'm not saying RA is wrong or any other theory is right, it would be my opinion that there is no one "perfect" way to perform an appraisal. I thnk thats what keeps my interest in the field, along with the ever changing market.

 

[Anonymous]

My suggestion would be to take college level course in statistics. Now-a-days, this class will include computer lab work running regressions as part of your routine home work assignments. This would be a good intro for you if you are sincerely interested. Please remember that regression has been around for nearly 100 years, it is the basis for much of science research: from medicine to rocket science, finance to geography. You don't even need to buy a book to learn much about the theory, you can find this just by surfing the internet. Your Microsoft excell can run regressions for you. Good luck.