Example 77.5 Robust Diagnostics

This example models the selling price of a house as a function of several covariates. One of these covariates is a classification variable that indicates whether a house is located on a corner lot (called a corner house in this example). Because corner houses are relatively rare, the inclusion of this classification effect in the model introduces a low-dimensional structure (that is, the majority of the observations are located in a lower dimensional hyperplane defined by being non-corner houses) into the design matrix. As discussed in Robust Distance, the presence of this low dimensional structure causes difficulties in the traditional computation of robust distances. This example illustrates how you can use the projected robust distance to address those difficulties and to obtain meaningful leverage diagnostics. It also shows how you can use the RDPLOT and DDPLOT options to illustrate the outlier-leverage relationship.

The following house price data set contains 66 home resale records on seven variables from February 15 to April 30, 1993 (The Data and Story Library, 2005). The records are randomly selected from the database maintained by the Albuquerque Board of Realtors.

 
data house;
   input price sqft age feats ne cor tax @@;
   label price = "Selling price"
         sqft  = "Square feet of living space"
         age   = "Age of home in year" 
         feats = "Number out of 11 features (dishwasher, refrigerator,
                  microwave, disposer, washer, intercom, skylight(s),
                  compactor, dryer, handicap fit, cable TV access)" 
         ne    = "Located in northeast sector of city (1) or not (0)" 
         cor   = "Corner location (1) or not (0)" 
         tax   = "Annual taxes";  
   sum = sqft+age+feats+ne+cor+tax;
   id  = _N_;
   datalines;
2050 2650 13 7 1 0 1639
2150 2664  6 5 1 0 1193
2150 2921  3 6 1 0 1635
1999 2580  4 4 1 0 1732

   ... more lines ...   

 870 1273  4 4 0 0  638
 869 1165  7 4 0 0  694
 766 1200  7 4 0 1  634
 739  970  4 4 0 1  541
;

To illustrate the dependence detection ability of the generalized MCD algorithm, an extra variable sum is created such that all the observations satisfy

     

Adding sum does not change the rank of the original design matrix, so that sum is expected to be ignored in the model and also in the diagnostics. The next statements apply the MM method and the generalized MCD algorithm to the house price data.

ods graphics on;
proc robustreg data=house method=MM plots=all;
   model price= sqft age feats ne cor tax sum/leverage(opc mcdinfo) diagnostics;
run;

As shown in Output 77.5.1 and Output 77.5.2, PROC ROBUSTREG finds the design dependence equation and forces the parameter estimate of variable sum to be zero.

Output 77.5.1 MM Estimates
The ROBUSTREG Procedure

Parameter Estimates
Parameter DF Estimate Standard Error 95% Confidence Limits Chi-Square Pr > ChiSq
Intercept 1 46.4062 79.1714 -108.767 201.5792 0.34 0.5578
sqft 1 0.3809 0.0756 0.2327 0.5291 25.37 <.0001
age 1 -2.6067 1.7610 -6.0582 0.8449 2.19 0.1388
feats 1 8.3627 14.7107 -20.4697 37.1951 0.32 0.5697
ne 1 65.0081 40.1329 -13.6508 143.6671 2.62 0.1053
cor 1 -19.2997 38.1907 -94.1520 55.5526 0.26 0.6133
tax 1 0.4699 0.1260 0.2229 0.7170 13.90 0.0002
sum 0 0.0000 . . . . .
Scale 0 157.5593          

Output 77.5.2 Design Dependence Equations

Note: The following variables have been ignored in the MCD computation because of linear dependence.

sum = sqft + age + feats + ne + cor + tax

Moreover, PROC ROBUSTREG also identifies a robust dependence equation on cor in Output 77.5.3, which holds for of the observations but not for the entire data set.

Output 77.5.3 Robust Dependence Equations

Note: The following robust dependence equations simultaneously hold for 77.27% of the observations in the data set. The breakdown setting for the MCD algorithm is 22.73%.

cor = 0

Another way to represent the low-dimensional structure is to specify the coefficients of the MCD-dropped components on the data (see Output 77.5.4), which form a basis of the complementary space to the relevant low-dimensional hyperplane.

Output 77.5.4 Coefficients for MCD-Dropped Components
Coefficients for MCD-Dropped
Components
Parameter DesignDrop0 RobustDrop1
sqft 0 0
age 0 0
feats 0 0
ne 0 0
cor 0 1.0000
tax 0 0
sum 1.0000 0

By definitions of projected robust distance and leverage point, an observation is called an off-plane leverage point if at least one of the robust or design dependence equations does not apply to the observation. In this example, the observations with cor are all off-plane leverage points. Output 77.5.5 lists the leverage points and outliers along with the relevant distance measurements and standardized residuals.

Output 77.5.5 Diagnostics
Diagnostics
Obs Projected Distance Leverage Standardized
Robust Residual
Outlier
Mahalanobis Robust Off-Plane
1 3.5567 4.0211 0.0000 * 0.8522  
13 4.0034 5.2310 0.0000 * 0.1411  
15 1.3221 1.5219 2.3681 * 0.0226  
16 1.0839 1.0905 2.3681 * 0.4148  
18 1.9452 2.4655 2.3681 * -0.2789  
20 3.6006 4.0771 2.3681 * -0.0150  
22 3.0210 3.4307 2.3681 * 1.1664  
23 1.5920 1.8197 2.3681 * 0.2422  
24 3.4967 4.5154 0.0000 * 0.6464  
26 3.0420 3.6975 0.0000 * -1.7068  
29 2.3264 2.9925 2.3681 * -2.4980  
30 1.2587 1.2714 2.3681 * -1.2558  
38 2.4064 2.7249 2.3681 * -1.0620  
42 1.4722 1.4645 2.3681 * 0.2584  
44 2.8491 3.0019 0.0000   4.5665 *
46 3.9725 5.2271 0.0000 * 3.5835 *
47 2.9431 3.3728 2.3681 * 0.1365  
55 2.2325 2.9590 2.3681 * 0.3217  
56 1.7999 1.8119 2.3681 * 0.1715  
65 1.8831 2.1822 2.3681 * -0.1990  
66 2.2483 2.5673 2.3681 * 0.4134  

From Output 77.5.6 and Output 77.5.7, you can see that there is no apparent corner-related difference for the houses in terms of standardized robust residual and projected MD versus projected RD, although all the corner houses are defined as off-plane leverage points.

Output 77.5.6 Projected RDPLOT
Projected RDPLOT

Output 77.5.7 Projected DDPLOT
Projected DDPLOT

Output 77.5.8 shows more details of the robust diagnostics. The number of dimensions indicates that six regressors are used in the MCD analysis. Since sum is excluded in model fitting, it is ignored in the MCD analysis. The number of robust dropped components equals 1 due to cor. The number of off-plane points implies the 15 corner-house observations. The reweighted value of H is the number of observations that are finally used to estimate the MCD covariance.

Output 77.5.8 MCD Information
MCD Profile
Number of Dimensions 6
Number of Robust Dropped Components 1
Number of Observations 66
Number of Off-Plane Observations 15
Specified Value of H 51
Reweighted Value of H 47
Breakdown Value 0.2273

MCD Center
ParameterName Parameter Center
sqft sqft 1752.7
age age 12.809
feats feats 4.0426
ne ne 0.6170
cor cor -2E-16
tax tax 895.40
sum sum 2665.6

MCD Covariance
  sqft age feats ne cor tax sum
sqft 248870.3 -853.232 147.0347 88.60083 0 148494.5 396747.3
age -853.232 126.2886 -1.18733 1.229417 0 -1251.44 -1978.34
feats 147.0347 -1.18733 0.99815 0.234043 0 87.0259 361.5814
ne 88.60083 1.229417 0.234043 0.241443 0 45.76688 134.42
cor 0 0 0 0 0 0 0
tax 148494.5 -1251.44 87.0259 45.76688 0 106652.5 255147
sum 396747.3 -1978.34 361.5814 134.42 0 255147 650413.7

MCD Correlation
  sqft age feats ne cor tax sum
sqft 1 -0.15219 0.295009 0.361446 0 0.911462 0.986126
age -0.15219 1 -0.10575 0.222643 0 -0.34099 -0.21829
feats 0.295009 -0.10575 1 0.476749 0 0.266726 0.448759
ne 0.361446 0.222643 0.476749 1 0 0.285206 0.339204
cor 0 0 0 0 0 0 0
tax 0.911462 -0.34099 0.266726 0.285206 0 1 0.968747
sum 0.986126 -0.21829 0.448759 0.339204 0 0.968747 1

You might speculate that the projected MD and projected RD are equal to the regular MD and RD on the same data set without the variable cor. In fact, this is not true. (See Output 77.5.9 and Output 77.5.10 for the RDPLOT and DDPLOT on the data set without cor.) When included in the MODEL, cor is dropped in the distance calculation, but it is still used for the initial orthonormalization step and the h-subset searching. In this example, inclusion of cor causes all the other covariates to be centered separately for corner houses and non-corner houses. However, without cor, the centering process does not distinguish corner houses from non-corner houses, so that the MCD algorithm can still be influenced by cor through the correlation between cor and other covariates. The following statements drop the variable cor and produce the RDPLOT and DDPLOT for the reduced model, which are shown in Output 77.5.9 and Output 77.5.10:

proc robustreg data=house method=MM plots=all;
  model price= sqft age feats ne tax/leverage(mcdinfo) diagnostics;
run;
ods graphics off;

Output 77.5.9 RDPLOT for the Reduced Model
RDPLOT for the Reduced Model

Output 77.5.10 DDPLOT for the Reduced Model
DDPLOT for the Reduced Model

Compared with Output 77.5.8, Output 77.5.11 shows the changes of the MCD information by removing cor from the model. You can see that the corner houses are no longer identified as off-plane points and the reweighted value of H is increased from 47 to 52. The breakdown value is intact because it depends only on the specified value of H and the total number of observations.

Output 77.5.11 MCD Information for the Reduced Model
MCD Profile
Number of Dimensions 5
Number of Robust Dropped Components 0
Number of Observations 66
Number of Off-Plane Observations 0
Specified Value of H 51
Reweighted Value of H 52
Breakdown Value 0.2273

MCD Center
ParameterName Parameter Center
sqft sqft 1710.9
age age 11.173
feats feats 3.9423
ne ne 0.5962
tax tax 858.10

MCD Covariance
  sqft age feats ne tax
sqft 216974.7 681.2327 199.2492 103.0388 107503.1
age 681.2327 64.49887 -0.9506 1.855581 -187.135
feats 199.2492 -0.9506 0.878959 0.152715 114.9076
ne 103.0388 1.855581 0.152715 0.245475 49.98077
tax 107503.1 -187.135 114.9076 49.98077 66558.68

MCD Correlation
  sqft age feats ne tax
sqft 1 0.182102 0.456255 0.44647 0.89457
age 0.182102 1 -0.12625 0.466337 -0.09032
feats 0.456255 -0.12625 1 0.328771 0.475075
ne 0.44647 0.466337 0.328771 1 0.391018
tax 0.89457 -0.09032 0.475075 0.391018 1