CHAPTER 7 NOTES Flashcards

1
Q

Steps for Building MRA Models

Building an MRA model requires a systematic step-by-step approach. The Advanced Computer-Assisted Mass Appraisal text listed as a recommended reading outlines this process in great detail in Chapter 3. There are two main phases in model building: (a) model specification and (b) model calibration. Model specification involves selecting the variables to be considered and defining their relationships to value and to each other. Model calibration means attaching numbers to the specified model, solving for the coefficients attached tothevariables of interest.

A part of model calibration is testing to ensure the model will create the estimated value with the desired accuracy. In this testing, the model’s results are compared to real-world examples to see if the model can provide accurate estimates. If not, then the model must be re-specified and re-calibrated until it produces acceptable results.

In this lesson, the modeling process will be explained in nine steps:

• Steps 1 to 6 are model specification; • Step 7 is model calibration; and

• Steps 8 and 9 are testing.

We will illustrate these steps through their practical application in developing a model. However, before this practical application, we will first briefly outline some background on these steps.

A

Steps for Building MRA Models

Building an MRA model requires a systematic step-by-step approach. The Advanced Computer-Assisted Mass Appraisal text listed as a recommended reading outlines this process in great detail in Chapter 3. There are two main phases in model building: (a) model specification and (b) model calibration. Model specification involves selecting the variables to be considered and defining their relationships to value and to each other. Model calibration means attaching numbers to the specified model, solving for the coefficients attached to the variables of interest.

A part of model calibration is testing to ensure the model will create the estimated value with the desiredaccuracy. In this testing, the model’sresults are comparedto real-world examplestoseeifthemodelcan provide accurate estimates. If not, then the model must be re-specified and re-calibrated until it produces acceptable results.

In this lesson, the modeling process will be explained in nine steps:

• Steps 1 to 6 are model specification; • Step 7 is model calibration; and

• Steps 8 and 9 are testing.

We will illustrate these steps through their practical application in developing a model. However, before this practical application, we will first briefly outline some background on these steps.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Steps for Building MRA Models

As outlined in Lesson 6, the first step is to describe an appropriate general model. The Advanced Computer-Assisted Mass Appraisal textbook describes several general model structures in Chapter 3. In selecting an appropriate model structure, you need to consider:

• the available data and its general format;

• the available software programs that can be used to calibrate the model – i.e., it does not make sense to specify a complex hybrid regression model when all that is available for calibration is a simple spreadsheet program; and

• good economic and appraisal theory.

For most of the uses in this course, we will apply the simple additive regression model as follows:

y = b0 + b1x1 + b2x2 + b3x3 + b4x4 +… + bnxn

The dependent variable (y) is the variable you are trying to estimate – usually market value. The independent variables (x) representtheattributes orcharacteristics of thesold propertiesinthedatabase(e.g.,squarefootage, number of bedrooms). The coefficients (b) are the numbers assigned by the regression to each independent variable (e.g., dollars per square foot, dollars per bedroom).

A

Steps for Building MRA Models

As outlined in Lesson 6, the first step is to describe an appropriate general model. The Advanced Computer-Assisted Mass Appraisal textbook describes several general model structures in Chapter 3. In selecting an appropriate model structure, you need to consider:

• the available data and its general format;

• the available software programs that can be used to calibrate the model – i.e., it does not make sense to specify a complex hybrid regression model when all that is available for calibration is a simple spreadsheet program; and

• good economic and appraisal theory.

For most of the uses in this course, we will apply the simple additive regression model as follows:

y = b0 + b1x1 + b2x2 + b3x3 + b4x4 +… + bnxn

The dependent variable (y) is the variable you are trying to estimate – usually market value. The independent variables (x) representtheattributes orcharacteristics of thesold propertiesinthedatabase(e.g.,squarefootage, number of bedrooms). The coefficients (b) are the numbers assigned by the regression to each independent variable (e.g., dollars per square foot, dollars per bedroom).

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Steps for Building MRA Models

Additive multiple regression is the most common form of regression as it conforms to the concepts of value and expectations of most market participants.It does not require a level of complexity beyond simple transformations and most statistical software packages (e.g., Excel,SPSS, NCSS) are capable of carrying out additive regression analysis.In terms of statistical procedures, additive regression is reasonably straight forward and the coefficients assigned to variables are understandable –in many cases,they can beexplained in dollar terms,such as so many dollars per square foot of living area. Additive regression is the only method of model calibration that will be used in this lesson and the next. Multiplicative regression is another method for calibrating models which is more complex and less easy to explain. It is useful in some situations, but because of its limited application we will leave this topic to the more advanced model building discussed in BUSI 444.

A

Steps for Building MRA Models

Additive multiple regression is the most common form of regression as it conforms to the concepts of value and expectations of most market participants.It does not require a level of complexity beyond simple transformations and most statistical software packages (e.g., Excel,SPSS, NCSS) are capable of carrying out additive regression analysis.In terms of statistical procedures, additive regression is reasonably straight forward and the coefficients assigned to variables are understandable –in many cases,they can beexplained in dollar terms,such as so many dollars per square foot of living area. Additive regression is the only method of model calibration that will be used in this lesson and the next. Multiplicative regression is another method for calibrating models which is more complex and less easy to explain. It is useful in some situations, but because of its limited application we will leave this topic to the more advanced model building discussed in BUSI 444.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Steps for Building MRA Models

The second step in the modeling process is to review the variables in the database in order to identify which may be suitable to use as independent variables for the model we are building. Up until now, we have used all of the variables provided in the databases, but in real applications you have to sort through many unsuitable variables and choose between alternative variables that may be all seem suitable but cannot be used together (e.g., multicollinear variables). In order to choose an appropriate model, you need to become familiar with the variables available in the database. For example.

• What property characteristics are represented?

• Do these characteristics fit better with an additive or multiplicative model structure?

• Are there several alternate forms for the same characteristic?

• If the form of the variable is not appropriate for the model structure, can a transformation be developed to put the variable into the correct form?

This is the data analysis phase, outlined in steps three and four below.

The third step, and one of the most important, is to examine the data in the database using the techniques described in Lessons 3, 4, and 6. Tools such as graphic analysis, crosstabs, compare means, and correlation analysis can help identify how variables are related to the dependent variable and also relationships between any independent variables. Multicollinearity was touched on in the previous lesson but will be described in more detail later in this lesson.

A

Steps for Building MRA Models

The second step in the modeling process is to review the variables in the database in order to identify which may be suitable to use as independent variables for the model we are building. Up until now, we have used all of the variables provided in the databases, but in real applications you have to sort through many unsuitable variables and choose between alternative variables that may be all seem suitable but cannot be used together (e.g., multicollinear variables). In order to choose an appropriate model, you need to become familiar with the variables available in the database. For example.

• What property characteristics are represented?

• Do these characteristics fit better with an additive or multiplicative model structure?

• Are there several alternate forms for the same characteristic?

• If the form of the variable is not appropriate for the model structure, can a transformation be developed to put the variable into the correct form?

This is the data analysis phase, outlined in steps three and four below.

The third step, and one of the most important, is to examine the data in the database using the techniques described in Lessons 3, 4, and 6. Tools such as graphic analysis, crosstabs, compare means, and correlation analysis can help identify how variables are related to the dependent variable and also relationships between any independent variables. Multicollinearity was touched on in the previous lesson but will be described in more detail later in this lesson.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Steps for Building MRA Models

In the fourth step, we need to identify any variables that may, in our appraisal judgement, be useful in the model, but are not useable in their current form - these require transformation. For example, in the MRA model for selling price developed in Lesson 6, let’s say we had a new variable that indicated the quality of each condo unit. This would likely be of use in the model, since purchasers tend to review quality in determining price. However, this new variable is coded as poor, fair, average, good, very good, and excellent, descriptions that are of no use in a mathematical regression model - e.g. , what might 812.35 times “excellent” work out to? We need to translate these descriptions into numerical equivalents, so we transform poor to 0, fair to 1, average to 2, up to excellent as 5. Thus, 812.35 times 5 (for excellent) makes arithmetic sense and would add over $4,000 to the selling price of the condo.1 This process is called transformation and is the fourth step in the process.

A

Steps for Building MRA Models

In the fourth step, we need to identify any variables that may, in our appraisal judgement, be useful in the model, but are not useable in their current form - these require transformation. For example, in the MRA model for selling price developed in Lesson 6, let’s say we had a new variable that indicated the quality of each condo unit. This would likely be of use in the model, since purchasers tend to review quality in determining price. However, this new variable is coded as poor, fair, average, good, very good, and excellent, descriptions that are of no use in a mathematical regression model - e.g. , what might 812.35 times “excellent” work out to? We need to translate these descriptions into numerical equivalents, so we transform poor to 0, fair to 1, average to 2, up to excellent as 5. Thus, 812.35 times 5 (for excellent) makes arithmetic sense and would add over $4,000 to the selling price of the condo.1 This process is called transformation and is the fourth step in the process.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Steps for Building MRA Models

The fifth step is to repeat Step 3 with any transformed variables - re-evaluating the potential variables for the model to see which offer the best suitability. If necessary, variables can be transformed again.

The sixth step is listing the final group of potential independent variables.

A

Steps for Building MRA Models

The fifth step is to repeat Step 3 with any transformed variables - re-evaluating the potential variables for the model to see which offer the best suitability. If necessary, variables can be transformed again.

The sixth step is listing the final group of potential independent variables.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Steps for Building MRA Models

The final three steps (7 through 9) are:

create the model using the specified variables and statistical software, which will determine the coefficients for the regression equation;

test and evaluate the model, finding how well it performs in terms of estimating the dependent variable or describing each variable’s contribution towards explaining the value of the dependent variable; and finally

• state your conclusions as to the quality of the model - in other words, describe (ideally in plain English that a layperson can understand!) what you did and why, and how well your model achieves its intended results.

A

Steps for Building MRA Models

The final three steps (7 through 9) are:

create the model using the specified variables and statistical software, which will determine the coefficients for the regression equation;

test and evaluate the model, finding how well it performs in terms of estimating the dependent variable or describing each variable’s contribution towards explaining the value of the dependent variable; and finally

state your conclusions as to the quality of the model - in other words, describe (ideally in plain English that a layperson can understand!) what you did and why, and how well your model achieves its intended results.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Steps for Building MRA Models

For model calibration, in valuation work models are primarily calibrated to produce either predictive or explanatory results.

Predictive Model: developed to produce the highest quality overall prediction of market value - for example, achieve the best possible estimate of selling price, but not necessarily the most reliable estimates for the individual coefficients.

Explanatory Model: developed primarily to explain the value that each variable contributes to market value - for example, the value per square foot of living area. In other words, rather than focussing on the outcome of the
model overall, this type of model focuses on developing the most accurate possible values for the coefficients.

A

Steps for Building MRA Models

For model calibration, in valuation work models are primarily calibrated to produce either predictive or explanatory results.

Predictive Model: developed to produce the highest quality overall prediction of market value - for example, achieve the best possible estimate of selling price, but not necessarily the most reliable estimates for the individual coefficients.

Explanatory Model: developed primarily to explain the value that each variable contributes to market value - for example, the value per square foot of living area. In other words, rather than focussing on the outcome of the
model overall, this type of model focuses on developing the most accurate possible values for the coefficients.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Steps for Building MRA Models

In a good explanatory model, the variable coefficients could be used as adjustments in a direct comparison approach or as market-derived costs in a cost approach. In an explanatory model, the model builder wants to maximize the accuracy of the values of the coefficients. As discussed in Lesson 6, the significance of the coefficients is indicated by the t-statistic and its associated significance level. Higher t-statistics and lower significance levels increase the reliance the model builder can place on the statistical significance of the coefficients. A high t-statistic leads to the acceptance of the hypothesis that the coefficient is significantly different than zero, meaning you are confident the coefficient number is accurate. As mentioned in the previous lesson, the criteria for this is usually to have a t-statistic over 2 and a significance level of less than .05. This would indicate that the probability of this coefficient being equal to zero is 5 % or less, meaning you are confident it is a reliable result.

A

Steps for Building MRA Models

In a good explanatory model, the variable coefficients could be used as adjustments in a direct comparison approach or as market-derived costs in a cost approach. In an explanatory model, the model builder wants to maximize the accuracy of the values of the coefficients. As discussed in Lesson 6, the significance of the coefficients is indicated by the t-statistic and its associated significance level. Higher t-statistics and lower significance levels increase the reliance the model builder can place on the statistical significance of the coefficients. A high t-statistic leads to the acceptance of the hypothesis that the coefficient is significantly different than zero, meaning you are confident the coefficient number is accurate. As mentioned in the previous lesson, the criteria for this is usually to have a t-statistic over 2 and a significance level of less than .05. This would indicate that the probability of this coefficient being equal to zero is 5 % or less, meaning you are confident it is a reliable result.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Steps for Building MRA Models

On the other hand, a good predictive model can be used to directly estimate sales prices. In a predictive model, the model builder wants to ensure the R-square is high and the standard error of the estimate is minimized.

This is normally achieved by including all variables that reduce the model’s standard error, regardless of the t-statistics or significance levels for the variables. This leads to a model with the lowest overall error possible, but it does not necessarily produce reliable individual variable coefficients.

A

Steps for Building MRA Models

On the other hand, a good predictive model can be used to directly estimate sales prices. In a predictive model, the model builder wants to ensure the R-square is high and the standard error of the estimate is minimized.

This is normally achieved by including all variables that reduce the model’s standard error, regardless of the t-statistics or significance levels for the variables. This leads to a model with the lowest overall error possible, but it does not necessarily produce reliable individual variable coefficients.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Steps for Building MRA Models

Depending on which goal is emphasized, the actions taken in the variable selection and calibration processes will vary. As this is an introductory course, the procedures described will tend to follow a middle course which allows for both a good prediction of market value and variable coefficients which are reasonable and rational from an appraisal perspective.

During the modeling process illustrated in this lesson, we will briefly demonstrate how these different approaches vary in procedures and outcomes.

A

Steps for Building MRA Models

Depending on which goal is emphasized, the actions taken in the variable selection and calibration processes will vary. As this is an introductory course, the procedures described will tend to follow a middle course which allows for both a good prediction of market value and variable coefficients which are reasonable and rational from an appraisal perspective.

During the modeling process illustrated in this lesson, we will briefly demonstrate how these different approaches vary in procedures and outcomes.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Illustration of MRA Model Building
Model Data

For this lesson, we will continue to focus on the Regina condominium sales data. The database we will use in this lesson is “Regina3”. This contains the same 120 sales used in previous lessons, but with 15 additional variables. You can download this database from the course website under “Online Readings”
There are 22 variables in the “Regina3” database. These are listed below:

A

Illustration of MRA Model Building
Model Data

For this lesson, we will continue to focus on the Regina condominium sales data. The database we will use in this lesson is “Regina3”. This contains the same 120 sales used in previous lessons, but with 15 additional variables. You can download this database from the course website under “Online Readings”
There are 22 variables in the “Regina3” database. These are listed below:

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Illustration of MRA Model Building
Model Data

STEP 1: Specifying the Model

We will be developing an additive model to estimate the value of condominiums (condos) based on the variables given in the “Regina3” database.

Again, the additive general model that is often applied to residential property is:

MV = LV + BV

where

MV = estimated market value (or selling price); LV = land value; and

BV = building value.

However, because we are dealing with condos, land value will be ignored. Given the list of variables available in our database, we will produce the following general model for the market value:

MV = bo + (bi X LIVINGAREA) +E(bi x LOCATION VARIABLE)+ (b; x AMENITY)

where

MV = is the estimated condo value or the estimated market value (or selling price);

bo = constant;

bi, = coefficients of the independent variables;

LIVINGAREA = total living area in the condo;

LOCATION VARIABLES = any variable associated with the location of the condo (e.g., floor number); AMENITY; = any variable associated with condo amenities (e.g., underground parking or a pool in the complex).

A

Illustration of MRA Model Building
Model Data

STEP 1: Specifying the Model

We will be developing an additive model to estimate the value of condominiums (condos) based on the variables
given in the “Regina3” database. Again, the additive general model that is often applied to residential property is:

MV = LV + BV

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Illustration of MRA Model Building
Model Data

STEP 2: Reviewing the Variables

The next step in the model development process is to review the variables available in the database and group the variables according to the factor the characteristic represents in the general model (e.g., in this case living area, location, or amenities). Very often, some characteristics to be included in the model will be represented by more than one variable and sometimes it is necessary to use a combination of more than one variable to correctly represent the characteristic or factor needed in the model.

A

Illustration of MRA Model Building
Model Data

STEP 2: Reviewing the Variables
The next step in the model development process is to review the variables available in the database and group the variables according to the factor the characteristic represents in the general model (e.g., in this case living area, location, or amenities). Very often, some characteristics to be included in the model will be represented by more than one variable and sometimes it is necessary to use a combination of more than one variable to correctly represent the characteristic or factor needed in the model.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Illustration of MRA Model Building
Model Data

STEP 2: Reviewing the Variables

For our database, the variables can be sorted as follows:

FACTOR VARIABLES

Living Area Total_Area

Location Topflr, Floor#, Abutting

Amenities Directio, Bedrms, Bath#, FullBath, Tqrbath, Halfbath, Patiofl, Balc#, Parkug, Parksurf, Parkdgar, Pool

Before making choices among variables or making changes to the database (Step 4: Transformations), you should familiarize yourself with the characteristics of the variables. In the next step, we will examine the variables to see which are best suited to be included in the modeling process (Step 6: List the Variables for Calibration).

A

Illustration of MRA Model Building
Model Data

STEP 2: Reviewing the Variables

For our database, the variables can be sorted as follows:

FACTOR VARIABLES

Living Area Total_Area

Location Topflr, Floor#, Abutting

Amenities Directio, Bedrms, Bath#, FullBath, Tqrbath, Halfbath, Patiofl, Balc#, Parkug, Parksurf, Parkdgar, Pool

Before making choices among variables or making changes to the database (Step 4: Transformations), you should familiarize yourself with the characteristics of the variables. In the next step, we will examine the variables to see which are best suited to be included in the modeling process (Step 6: List the Variables for Calibration).

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Illustration of MRA Model Building
Model Data

STEP 3: Examining the Variables

In this section, we will examine the variables in our database, their relationship to sale price, and their relationship to each other. There are a number of reasons for testing these relationships:

To get a sense of the important variables so we can get a feel for what to expect out of the final model.

To exclude variables from the regression model that are of no use.

For example, they may have little or no statistical relationship to the sale price (within the data being analyzed) or they have too few occurrences within the data to be accurately representative of the property characteristic (e.g., if only one property of 120 has a view, then the model would not be able to accurately determine what value a view might have).

To avoid multicollinearity, excluding any variable strongly correlated with another variable.

To find variables that might be useful, but need to be changed into a useable format. For example, a variable that has values of TRUE or FALSE would need to be changed into a numerical form: e.g., 1 for TRUE and 0 for FALSE.

• To fmd variables that might be useful, but are multiplicative in nature. For example, a percent condition variable might range from 80% to 120% indicating 20% better or worse condition than the 100% average. If you wanted to include the effect of condition in the additive regression model, you could multiply this percentage by the total square footage of the home. This would mean homes in better condition would appear to the model to have more square footage and in this way the effect of condition would be brought into the model.

A

Illustration of MRA Model Building
Model Data

STEP 3: Examining the Variables

In this section, we will examine the variables in our database, their relationship to sale price, and their relationship to each other. There are a number of reasons for testing these relationships:

To get a sense of the important variables so we can get a feel for what to expect out of the final model.

To exclude variables from the regression model that are of no use.

For example, they may have little or no statistical relationship to the sale price (within the data being analyzed) or they have too few occurrences within the data to be accurately representative of the property characteristic (e.g., if only one property of 120 has a view, then the model would not be able to accurately determine what value a view might have).

To avoid multicollinearity, excluding any variable strongly correlated with another variable.

To find variables that might be useful, but need to be changed into a useable format. For example, a variable that has values of TRUE or FALSE would need to be changed into a numerical form: e.g., 1 for TRUE and 0 for FALSE.

• To fmd variables that might be useful, but are multiplicative in nature. For example, a percent condition variable might range from 80% to 120% indicating 20% better or worse condition than the 100% average. If you wanted to include the effect of condition in the additive regression model, you could multiply this percentage by the total square footage of the home. This would mean homes in better condition would appear to the model to have more square footage and in this way the effect of condition would be brought into the model.

17
Q

Illustration of MRA Model Building
Model Data

STEP 3: Examining the Variables

Correlation Analysis

In the previous sections, we have seen how some of the variables are related to SalePrice. Correlation analysis will be the final step in this variable examination process, identifying variables that are correlated with each other. We must ensure that our model only includes one of any correlated variables in order to avoid multicollinearity. Some correlations have already been determined: Patiofl and Balco#, Bath# and HalfBath.

Go to Analyze -* Correlate Bivariate…. Select the following variables: SalePrice, Topflr, Floor#,

Total_Area, Bedrms, Halfbath, Parkug, Parksurf, Parkdgar, and Pool. Ensure that Pearson is the Correlation Coefficient selected. We have removed the N and Sig lines for clarity and brevity (using the instructions in Lesson 3).

A

Illustration of MRA Model Building
Model Data

STEP 3: Examining the Variables

Correlation Analysis

In the previous sections, we have seen how some of the variables are related to SalePrice. Correlation analysis will be the final step in this variable examination process, identifying variables that are correlated with each other. We must ensure that our model only includes one of any correlated variables in order to avoid multicollinearity. Some correlations have already been determined: Patiofl and Balco#, Bath# and HalfBath.

Go to Analyze -* Correlate Bivariate…. Select the following variables: SalePrice, Topflr, Floor#,

Total_Area, Bedrms, Halfbath, Parkug, Parksurf, Parkdgar, and Pool. Ensure that Pearson is the Correlation Coefficient selected. We have removed the N and Sig lines for clarity and brevity (using the instructions in Lesson 3).

18
Q

Illustration of MRA Model Building
Model Data

STEP 3: Examining the Variables

The first set of correlations that are of interest are the ones with Sale Price. The matrix confirms our previous analysis of how variables are related to Sale Price: Total Living Area has the strongest correlation with Sale Price (0.770) and will clearly be a significant variable in the final model.

Of more interest at this point in this analysis, however, are the correlations between the potential independent variables. Where variables are highly correlated (e.g., over 0.8), only one should appear in the model -otherwise, multicollinearity problems will surface in the model.

The correlation between Bedrooms and Total Living Area is very strong (0.979) and therefore one of these will have to be removed from the model.

This correlation is not unexpected, because as the number of bedrooms increases in residential property so does the living area, especially in condos. As a result, any effect that number of bedrooms has on value should also be addressed through the Total Living Area variable.

Therefore, we will exclude Bedrms from the model. If they were both included, it would lead to multicollinearity, which is a major problem in MRA models.

Multicollinearity creates errors in the coefficients and generates meaningless results. BE CAREFUL of multicollinear variables!

One other strong (albeit negative) correlation is between Surface Parking Stalls and Underground Parking Stalls (-0.918). It appears that if a condo has underground parking it will NOT have surface parking, and vice versa.

This was suspected from the results of the boxplot analysis. Therefore, to avoid multicollinearity, only one can be included in the model. We will include Parkug. Remember that Parksurf has that single occurrence of 2 parking stalls - using Parkug removes that issue.

No other correlations are of concern.

A

Illustration of MRA Model Building
Model Data

STEP 3: Examining the Variables

The first set of correlations that are of interest are the ones with Sale Price.

The matrix confirms our previous analysis of how variables are related to Sale Price: Total Living Area has the strongest correlation with Sale Price (0.770) and will clearly be a significant variable in the final model.

Of more interest at this point in this analysis, however, are the correlations between the potential independent variables. Where variables are highly correlated (e.g., over 0.8), only one should appear in the model -otherwise, multicollinearity problems will surface in the model.

The correlation between Bedrooms and Total Living Area is very strong (0.979) and therefore one of these will have to be removed from the model.

This correlation is not unexpected, because as the number of bedrooms increases in residential property so does the living area, especially in condos. As a result, any effect that number of bedrooms has on value should also be addressed through the Total Living Area variable.

Therefore, we will exclude Bedrms from the model. If they were both included, it would lead to multicollinearity, which is a major problem in MRA models.

Multicollinearity creates errors in the coefficients and generates meaningless results. BE CAREFUL of multicollinear variables!

19
Q

Summary of Data Analysis

Our data examination has provided a solid foundation for selecting the variables that should be considered in the model. We have found:

some variables can be ignored, e.g., Fullbath and Tqrbath which are constant for all sales;

some variable combinations should be avoided to eliminate multicollinearity issues later, e.g., Bedrms and Total Area; and

some variables will need to be modified or transformed into a suitable form to use in an additive regression model, e.g., Abutting and Directio.

In the next step we will carry out these necessary transformations.

A

Summary of Data Analysis

Our data examination has provided a solid foundation for selecting the variables that should be considered in the model. We have found:

some variables can be ignored, e.g., Fullbath and Tqrbath which are constant for all sales;

some variable combinations should be avoided to eliminate multicollinearity issues later, e.g., Bedrms and Total Area; and

some variables will need to be modified or transformed into a suitable form to use in an additive regression model, e.g., Abutting and Directio.

In the next step we will carry out these necessary transformations.

20
Q
A
21
Q
A
22
Q
A
23
Q
A
24
Q
A
25
Q
A
26
Q
A
27
Q
A
28
Q
A