Further analysis of transmission line impact on residential property values.
Many studies have been done on the effects of high-voltage transmission lines on property values; however, the research results have been mixed. This study looks more deeply into the findings of one such study to see if the findings hold up under greater statistical scrutiny. While the confirmation of prior research results is interesting, equally important are the methods used to confirm or refute the prior results. This demonstrates the importance of asking, "When faced with insignificant statistical results, is the insignificance believable, or is the lack of significance merely due to unrecognized and uncontrolled aspects of the data?" This paper addresses these important analytical questions.
The research reported here is a confirmatory study concerning an article in Right of Way. (1) This study does not replicate the procedures followed by Cowger, Bottemiller, and Cahill in that article. Instead, this study investigates whether or not the results ofCowger, Bottemiller, and Cahill hold when using more rigorous analytical methods. Here the same data is employed and is expanded into a richer data set by adding additional home sales that were collected but not required for the study reported in Right of Way (referred to hereafter as the "original study").
The original study utilized a paired-sale methodology to ascertain any difference in sale price between properties abutting rights-of-way of transmission lines (subjects) in Portland, Oregon; Vancouver, Washington; and Seattle, Washington and those located in the same cities but not abutting transmission line rights-of-way (comps). Subjects sold during the study period were selected first. Then a matching comp was selected from the subject neighborhood that was as similar to the subject as possible. The student's t distribution was used to create 95% confidence intervals for the average percentage of difference between subject and comp sale prices. In each geographic setting (Portland, Vancouver, Seattle) and for all data combined, the 95% confidence interval for mean difference contained zero, indicating no significant difference in sale price between the subjects and the comps. Although the comps were carefully selected, the subjects and comps were not identical. The inability of the original study method to control for differences between subject and comp is a potentially serious shortcoming. The current study attempts to overcome this problem.
The importance of overcoming this shortcoming stems from a lack of consensus among prior researchers. Although numerous studies have been conducted concerning the price effect of proximity of housing to high-voltage transmission lines (HVTL), their results vary. Des Rosiers, in the most recendy published study of the topic, cites twelve prior studies and provides a good source of bibliographical information. (2) 1991-1996 study of the Montreal area found that HVTL proximity did not necessarily affect price, although price was affected by a direct view of either the conductors or the pylons (conditions evident at all of the subject properties in Portland, Vancouver, and Seattle). He also found no impact from the well-publicized 1992 Swedish study reporting a weak correlation between electromagnetic field (EMF) exposure and childhood leukemia. (3)
This article first discusses the data and then presents the analytical methods and results. Finally, it discusses the study's implications and conclusions, including suggestions for further research.
Data was collected beginning in 1992 for home sales occurring in 1989-1992 in Washington (Portland) and Clackamas (Portland) Counties in Oregon as well as Clark (Vancouver) and King (Seattle) Counties in Washington. The original study indicates 296 subject properties and 296 matching comparable sales (592 observations). After going through the original data files and adding four subject properties not used in the original study and comparable sales not required for the original study's analytical method, the number of observations in the current study grew to 712, including 300 subject properties abutting an HVTL. Geographically 374 sales were in King County; Washington; 102 were in Clark County, Washington; 227 were in Washington County, Oregon; and 9 were in Clackamas County, Oregon.
Descriptive statistics for the variables used in the following analysis are reported in Table 1. Seasonal distribution of the sales is fairly uniform, ranging from a low of 151 in Quarter 1 (January-March) to a high of 216 in Quarter 2 (April-June). Fifty-six sales occurred in 1989 (8%), 377 in 1990 (53%), 258 in 1991 (36%), and 21 in 1992 (3%). Hence, the study is primarily focused on the 1990-1991 time frame, prior to dissemination of the previously mentioned Swedish epidemiological study.
Site and site improvement variables included lot size, which averaged .37 acres due to the influence of a number of home sites that were larger than one acre (median lot size was. 19 acres or 8,276 square feet [SF]). Lot configuration was categorized as standard (485 sales), cul-de-sac (135 sales), corner (74 sales), or flag-shaped (18 sales). Standard lots were interior lots that were generally rectangular or trapezoidal in shape. Subjective characterization of the lot as being either level or sloping was determined from photographs of the front yard and house front. Sloping topography was adjudged to have been in existence at 140 of the sales. Above average landscaping characterized 39 sales, while 538 were rated average and 135 were rated below average. Landscaping quality ratings based on the photographs may not be reliable for homes that were new when sold, since the photographs were taken well after many of the sale dates, and the home sales may not have included a finished landscape (i.e., new home landscapes rated "average" or "above average" could have actually been "poor" at the time of sale.)
Building improvement variables were floor area of the house (mean = 1,845 SF), age at sale date (mean = 8.1 years), bedrooms (mean = 3.2), and baths (mean = 2.4). Other numerical variables in this category were carport size (mean = 359 SF) and garage size (mean = 497 SF). Several categorical variables were created from the data including condition of the improvements, given their age (115 excellent, 570 good, 25 fair, and 2 poor), existence of a tuck-under garage, patio or deck, greenhouse or sports court, hat tub, and a shed, workshop, or extra garage. Condition was based on analysis of the photographs and most likely is a proxy for "curb appeal." Tuck-under garage was included to account for over-counting house floor area when a tuck-under (basement) garage was present, which was discovered when reviewing sale files. (This could not be corrected on a case-by-case basis because MetroScan source documents were not available in all sale files.)
Sale price averaged $128,890 with a standard deviation of $47,751. Median sale price was $123,500, indicating slight right-skewness. The natural log of sale price averaged 11.698 with a median of 11.724, indicating a slight left-skewness. Investigations of regression residual histograms, with normal probability overlays, and of stem and leaf plots of regression residuals reveal no violations of normality assumptions in the models employed later.
Analysis of covariance (ANCOVA) is used to test for an "abutting transmission line" effect on sale price. Abutting transmission line is the factor of interest and is present at 300 observations, as shown in Table 1 (412 observations do not abut an HVTL).
The ANCOVA regression model, similar to the form employed by Hardin and Wolverton, (4) is:
Y = [alpha] + [tau]I + X[gamma] + [epsilon]
Y = sale price
[alpha] = the constant (intercept) term
I = a dummy variable coded "1" if the sale property abuts an HVTL and "0" otherwise
X = a vector of concomitant covariates (temporal, site and site improvements, location, and building improvements)
[epsilon] = random pricing error
[tau] and [gamma] = regression coefficients
The empirical ANCOVA model employed later contains 31 concomitant variables, which is far less than the maximum of 70, computed as. 10n--(number of groups - 1). In this analysis, n = 712 observations and the number of groups is 2--abutting and not abutting an HVTL. (5)
Correlation Between Abutting an HVTL and the Concomitant Variables
An important consideration in ANCOVA is a lack of correlation between the concomitant variables and the factor of interest (often referred to as the treatment). If elements of X are correlated with I, then the concomitant effects are not fully removed from the effect of the factor of interest, giving a less-than-pure look (in this case, at the effect of abutting a transmission line). A number of tests for and means of dealing with excessive correlation between the treatment and control variables exist. A simple assessment is provided by investigating Pearson correlation coefficients between I and X; these are presented in Table 2. None of the correlations in Table 2 are above the .30 threshold causing multicollinearity concerns per se, (6) although multicollinearity can exist despite low correlations. (7) Variance inflation factors were analyzed as well in the regression models that follow, and none were found to be at levels indicative of excessive multicollinearity. Lastly, alternative regression models were run, entering interaction variables (I times all elements of X) stepwise into the base ANCOVA model, ensuring that the effects of any significant (by the t-test) concomitant variables were blocked from the abutting an HVTL effect. These results are reported later.
ANCOVA Regression Models of Abutting an HVTL Right-of-Way
Four alternative models were estimated. The first two models use the natural log of price as the dependent variable. Model 1 has a single factor of interest--abutting an HVTL. Model 2 disaggregates the abutting a transmission line factor by geography (county). The natural-log-dependent-variable model helps correct for nonlinearities in the relationships between regressors and regressand, and it offers the benefit of easily transforming the coefficients into percentage change in the dependent variable. Models 3 and 4 are identical to the first two models except they use nominal price as the dependent variable. The results of Models 1 and 2 are displayed in Table 3, while results of Models 3 and 4 are displayed in Table 4.
Both models in Table 3 are highly significant (F = 135.04 and 123.05, p-values = .000) and they explain high proportions of the price variance in the data ([R.sup.2] = .864). Regarding the main thrust of this study, none of the measures of the effects of abutting an HVTL are statistically significant in either model. Abutting a transmission line has an absolute value t-statistic of .615 in Model 1 and the absolute value disaggregated t-statistics are .90, .07, .07, and .34 in Model 2. The threshold level for moderate statistical significance (the 10% level) is t = 1.65 for all models, given the relatively large sample size. The threshold for statistical significance at the traditional 5% level is t = 1.96. The t-statistics derived from the data do not approach these significance thresholds.
The relatively good overall accuracy of measurements contained within the data set is reflected by the high coefficients of determination ([R.sup.2]) and the intuitive appeal of the signs on the significant variables in the two models.
As expected, price effects are seasonal with sales closing in the base season (Quarter 3) being relatively higher in price, as indicated by the negative signs on Quarters 1, 2, and 4. Additionally, prices rose in each year of the analysis with respect to the base year (1989) and in each successive year. The model shows price declining with age, but at a decreasing rate, as indicated by the positive sign on age squared. Also, price rises as lot size increases. Cul-de-sac lots have insignificant price effects (after controlling for lot size), but corner lots sell for significantly more and flag-shaped lots sell for significantly less. Above average landscaping takes on the expected positive sign. However, below average landscaping is also positively signed, but insignificant. This sign should be negative, but may be reflecting measurement error due to the aforementioned difficulty of identifying landscape condition at the time of sale using post-sale photographs. The probability of the photograph reflecting time of sale conditions is higher with above average landscaping, since it is typically more mature.
The signs and magnitudes of home condition variables are as expected, with excellent condition being positive compared to the base good condition assessment and fair condition being negative compared to good condition. Poor condition is more of a detriment than fair condition, but not statistically significant. The signs on bedrooms and baths are both positive, as is garage floor area. The remaining significant improvement variable, tuck-under garage, has a negative sign. This negative sign is expected, given both the generally lower curb appeal of the homes in this data set having this design feature and the previously discussed error in square footage measurements associated with this particular style of house.
Each model also controls for location at the county level. On average, Clark County housing in the data set sold for about 25.8% less ([e.sup.-.299] - 1) than King County housing. Compared also to King County, Washington County housing sold for 16.6% less and Clackamas County housing sold for 11.4% less. All three of these county-level comparisons are statistically significant (Clackamas County being only moderately so in Model 2) and the signs are reflective of an expected King County (Seattle) price premium.
For completeness and to account for the possibility of results varying due to the functional form of the dependent variable, the ANCOVA models presented in Table 3 were also estimated without the natural log transformation of the dependent variable. Instead, the dependent variable was the untransformed nominal sale price. Coefficients for these untransformed models (Models 3 and 4) are therefore expressed in nominal dollars. The results of the two alternative models are displayed as Models 3 and 4 in Table 4,
Models 3 and 4 are also highly significant (F = 122.86 and 122.89, p-values = .000, [R.sup.2] = .853, adjusted [R.sup.2] = .846 and .845). As in Models 1 and 2, none of the measures of the effect of abutting an HVTL are significant in Models 3 and 4. The abutting transmission line variable absolute value t-statistic is .061 in Model 3, and the separate county absolute value t-statistics are .66, .32, .72, and. 19 in Model 4. Again, these all fall short of the 1.65 and 1.96 thresholds for significance at the 10% and 5% levels, respectively.
The signs on Models 3 and 4 are also intuitively appealing. However, unlike Models 1 and 2, the bathroom coefficient is not significant in these models. It is not unusual, however, for bathroom and bedroom coefficients to be insignificant in housing price models due to the relatively high correlations between room counts and house square footage. Models 3 and 4 fail to demonstrate the significance of lot shape in the sale data compared to Models 1 and 2. Additionally, the fair condition variable is no longer significant. Nevertheless, the general indications of Models 3 and 4 are similar to Models 1 and 2, reinforcing the indicated lack of significance of abutting an HVTL right-of-way.
Neter, Wasserman, and Kutner suggest a procedure for testing for significant interactions between the treatment (factor) and covariates in an ANCOVA model (8) The procedure involves an F test of the degree to which a full interaction model adds information not contained In a model without interaction variables. The F test statistic is:
F = SS[E.sub.1] - SS[E.sub.2] / d[f.sub.1] - d[f.sub.2] x d[f.sub.2] / SS[E.sub.2]
SS[E.sub.1] = the sum of squared errors for the no-interaction model
SS[E.sub.2] = the sum of squared errors for the full-interaction model
d[f.sub.1] and d[f.sub.2] = the respective degrees of freedom
This test was applied to Model 1 and the F-statistic was 1.158. The critical F at a 5% significance level is 1.47, indicating that the information contained in the interaction variable set is insignificant.
As mentioned earlier, one additional procedure was included to mitigate any potential effects of excessive correlation between the factor of interest (abutting an HVTL, or I) and the covariates (X) not revealed by analysis of the correlations in Table 2, variance inflation factor analysis, and the preceding F test. Models 1 and 3 were rerun by entering the covariates and the abutting an HVTL factor into a regression model as Step 1 and then entering all of the possible interactions between I and X stepwise into the same model as Step 2. Any significant interactions were retained by the stepwise procedure, (9) thereby blocking (10) any interaction effects on the factor of interest. The rerun of Model 1 uncovered two significant interactions: "abutting an HVTL x poor condition" (t = -2.09) and "abutting an HVTL x carport square feet" (t= -2.30). After blocking for these two interactions, the t-statistic on the abutting an HVTL factor remained insignificant (-.19). The rerun of Model 3 uncovered one significant interaction: "abutting an HVTL x carport square feet" (t = -2.75). After blocking for this interaction, the t-statistic on the abutting an HVTL factor remained insignificant at .27.
The veracity of the finding of no significant price effect of abutting an HVTL right-of-way is supported by four different modeling approaches and by multiple analyses of potentially influential correlations between the abutting an HVTL factor of interest and the numerous covariates. (11)
Home Price Appreciation
Another aspect of possible HVTL proximity influence on home sales is the rate at which home prices change over time ("appreciate" in this data). The relevant question is: Do prices of homes abutting an HVTL appreciate less than those of comparable homes not abutting an HVTL? The data allows investigation of this question.
Three interaction variables were added to the covariates from Model 1 to develop a comparison of price appreciation for homes abutting an HVTL right-of-way with homes not abutting a right-of-way. The interaction variables were "1990 sale x abutting an HVTL," "1991 sale x abutting an HVTL," and "1992 sale x abutting an HVTL." A significant t-statistic for any of these interactions would indicate that price change from the base 1989 sale year is significantly different from price change for homes not abutting an HVTL right-of-way. Results are shown in Table 5.
As with Model 1, from which this model evolved, the Table 5 model is highly significant and explanatory (F = 127.11 and [R.sup.2] = .865). With regard to the question of differential price change rates, absolute value t-statistics are 1.30, .58, and .25 for the 1990, 1991, and 1992 interaction variables, respectively. The data does not support different price appreciation rates for homes abutting and not abutting an HVTL right-of-way.
Implications and Conclusions
The data does not support a finding of a price effect from abutting an HVTL right-of-way. This finding confirms the results of the original study by Cowger, Bottemiller, and Cahill (12) by controlling for concomitant factors affecting home sale prices, the implication being that there is no evident price sensitivity to abutting an HVTL right-of-way for typical homebuyers and sellers reflected in this data set. This conclusion cannot and should not be generalized outside of the data, however. The caution regarding generalization stems from the data not being a representative random sample from the counties analyzed and the four counties not being representative of other counties and/or locations. The limits on generalizations is a universal problem for real property sale data because analysis is constrained to properties that sell and sold properties are never a randomly drawn representative sample. Hence, generalizations must rely on the weight of evidence from numerous studies, samples, and locations. The foregoing is one such study.
The data also does not support a difference in price appreciation over time for properties abutting and not abutting an HVTL right-of-way. Again, this outcome applies only to the data and it cannot be generalized beyond the home sales in the data set. Nevertheless, this is an important finding, representing a relatively new and unique question. For this reason, future studies should be designed to replicate and confirm or refute this result. A richer data set containing repeat sales of abutting and non-abutting properties would provide another perspective.
Understanding the effects of HVTLs on home prices and appreciation rates is a dynamic process. It is affected by changing public perceptions and different on-site factors such as allowable land uses within the right-of-way, quality of right-of-way ownership (easement v. fee title), type of power line support structure, and right-of-way maintenance procedures. An on-going study process is necessary and could be improved by investigating other cultural and geographic settings, identifying more accurate and reliable sources of data, greater consistency in measurement, and richer data sets allowing more variety in analytical methods. Nevertheless, the original study and this follow-up to it offer one important perspective on the underlying property value and marketability issues.
Table 1 Descriptive Statistics (n = 712) Variable Type Mean Temporal Date of Sale Variables Quarter 1 Categorical .21 Quarter 2 Categorical .30 Quarter 3 Categorical .26 Quarter 4 Categorical .22 1989 Sale Categorical .08 1990 Sale Categorical .53 1991 Sale Categorical .36 1992 Sale Categorical .03 Site and Site Improvement Variables Lot Size (Acres) Numerical .37 Standard Lot Categorical .68 Cul-de-Sac Lot Categorical .19 Corner Lot Categorical .10 Flag-Shaped Lot Categorical .03 Sloping Topography Categorical .20 Above Average Landscape Categorical .05 Average Landscape Categorical .76 Below Average Landscape Categorical .19 Location Variables King County, WA Categorical .53 Clark County, WA Categorical .14 Washington County, OR Categorical .32 Clackamas County, OR Categorical .01 Building Improvement Variables Floor Area (SF) Numerical 1,844.5 Age (Yrs.) Numerical 8.1 Excellent Condition Categorical .16 Good Condition Categorical .80 Fair Condition Categorical .04 Poor Condition Categorical .003 Bedrooms Numerical 3.2 Baths Numerical 2.4 Garage Floor Area (SF) Numerical 497 Carport Floor Area (SF) Numerical 359 Tuck-Under Garage Categorical .10 Patio or Deck Categorical .45 Greenhouse or Sports Court Categorical .01 Shed, Workshop, or Extra Garage Categorical .01 Hot Tub Categorical .01 Other Variables Price Numerical 128,890 Abutting Transmission Line Categorical .42 Variable Standard Sum Deviation Temporal Date of Sale Variables Quarter 1 .41 151 Quarter 2 .46 216 Quarter 3 .44 188 Quarter 4 .41 157 1989 Sale .27 56 1990 Sale .50 377 1991 Sale .48 258 1992 Sale .17 21 Site and Site Improvement Variables Lot Size (Acres) .51 n.a. Standard Lot .47 485 Cul-de-Sac Lot .39 135 Corner Lot .31 74 Flag-Shaped Lot .16 18 Sloping Topography .40 140 Above Average Landscape .23 39 Average Landscape .43 538 Below Average Landscape .39 135 Location Variables King County, WA .50 374 Clark County, WA .35 102 Washington County, OR .47 227 Clackamas County, OR .11 9 Building Improvement Variables Floor Area (SF) 591.5 n.a. Age (Yrs.) 10.1 n.a. Excellent Condition .37 115 Good Condition .40 570 Fair Condition .18 25 Poor Condition .05 2 Bedrooms .58 n.a. Baths .69 n.a. Garage Floor Area (SF) 133 n.a. Carport Floor Area (SF) 158 n.a. Tuck-Under Garage .30 72 Patio or Deck .50 320 Greenhouse or Sports Court .07 4 Shed, Workshop, or Extra Garage .10 7 Hot Tub .11 9 Other Variables Price 4,7751 n.a. Abutting Transmission Line .49 300 Table 2 Correlation Coefficients (Abutting Transmission Line v. Covariates) Variable Pearson Correlation Coefficient Quarter 1 -.016 Quarter 2 .076 Quarter 3 -.015 Quarter 4 -.053 1989 Sale -.248 1990 Sale .136 1991 Sale .037 1992 Sale -.114 Lot Size (Acres) .019 Standard Lot .053 Cul-de-Sac Lot .003 Corner Lot -.131 Flag-Shaped Lot .081 Sloping Topography -.005 Above Average Landscape -.055 Average Landscape -.024 Below Average Landscape .059 Fair Condition -.008 King County, WA -.045 Clark County, WA -.007 Washington County, OR .054 Clackamas County, OR -.005 Floor Area (SF) -.029 Age (Yrs.) -.010 Bedrooms -.020 Baths .022 Garage Floor Area (SF) -.072 Carport Floor Area (SF) .023 Tuck-Under Garage -.002 Patio or Deck .113 Greenhouse or Sports Court .050 Shed, Workshop, or Extra Garage .002 Hot Tub .057 Excellent Condition -.017 Good Condition .018 Poor Condition .009 Table 3 ANCOVA Regression Models (Dependent Variable = Natural Log of Nominal Price) Model 1 Variable Coefficient |t-statistic| Intercept 10.74 243.4 Covariates Quarter 1 -.078 4.79 Quarter 2 -.042 2.85 Quarter 4 -.026 1.66 1990 Sale .143 6.38 1991 Sale .204 8.79 1992 Sale .279 7.16 Age -.009 7.11 Age Squared .0001 5.34 Lot Size (acres) .057 4.67 Cul-de-Sac Lot -.009 .61 Corner Lot .038 2.07 Flag-Shaped Lot -.090 2.54 Above Average Landscape .069 2.78 Below Average Landscape .020 1.38 Sloping Topography .018 1.19 Floor Area (SF) .0003 19.50 Excellent Condition .153 8.16 Fair Condition -.100 2.97 Poor Condition -.133 1.24 Bedrooms .040 3.31 Baths .038 2.95 Tuck-Under Garage -.062 3.11 Carport Floor Area (SF) .0001 1.32 Garage Floor Area (SF) .0003 7.19 Patio or Deck .014 1.13 Greenhouse or Sports Court -.103 1.39 Shed, Workshop, or Extra Garage .047 .83 Hot Tub -.021 .44 Clark County Location -.299 17.00 Washington County Location -.182 12.90 Clackamas County Location -.121 2.47 Factors Abutting Transmission Line -.007 .615 Abutting Transmission Line (King Co.) Abutting Transmission Line (Wash. Co.) Abutting Transmission Line (Clark Co.) Abutting Transmission Line (Clackamas Co.) Sample Size (n) 712 Model R Squared .864 Model Adjusted R Squared .858 Model F-statistic 135.04 Model 2 Variable Coefficient |t-statistic| Intercept 10.74 242.6 Covariates Quarter 1 -.077 4.73 Quarter 2 -.042 2.86 Quarter 4 -.027 1.69 1990 Sale .143 6.36 1991 Sale .204 8.75 1992 Sale .279 7.11 Age -.009 7.08 Age Squared .0001 5.34 Lot Size (acres) .057 4.67 Cul-de-Sac Lot -.009 .64 Corner Lot .038 2.04 Flag-Shaped Lot -.091 2.54 Above Average Landscape .069 2.77 Below Average Landscape .020 1.38 Sloping Topography .017 1.16 Floor Area (SF) .0003 19.5 Excellent Condition .153 8.16 Fair Condition -.100 2.97 Poor Condition -.138 1.28 Bedrooms .040 3.32 Baths .039 2.95 Tuck-Under Garage -.061 3.08 Carport Floor Area (SF) .0001 1.33 Garage Floor Area (SF) .0003 7.20 Patio or Deck .013 1.01 Greenhouse or Sports Court -.103 1.40 Shed, Workshop, or Extra Garage .047 .83 Hot Tub -.021 .43 Clark County Location -.306 13.7 Washington County Location -.187 10.8 Clackamas County Location -.112 1.71 Factors Abutting Transmission Line Abutting Transmission Line (King Co.) -.014 .90 Abutting Transmission Line (Wash. Co.) .001 .07 Abutting Transmission Line (Clark Co.) .002 .07 Abutting Transmission Line (Clackamas Co.) -.033 .34 Sample Size (n) 712 Model R Squared .864 Model Adjusted R Squared .857 Model F-statistic 123.05 Table 4 ANCOVA Regression Models (Dependent Variable = Nominal Price) Model 3 Variable Coefficient |t-statistic| Intercept 13,917 2.40 Covariates Quarter 1 -6,121 2.86 Quarter 2 -4,606 2.36 Quarter 4 -2,301 1.11 1990 Sale 16,600 5.65 1991 Sale 22,703 7.43 1992 Sale 31,584 6.17 Age -894 7.11 Age Squared 10.83 3.36 Lot Size (Acres) 6,669 4.17 Cul-de-Sac Lot -2,765 1.47 Corner Lot 2,808 1.16 Flag-Shaped Lot -3,302 0.71 Above Average Landscape 14,104 4.30 Below Average Landscape 1,969 1.02 Sloping Topography 2,254 1.15 Floor Area (SF) 42.19 20.0 Excellent Condition 31,910 13.0 Fair Condition -3,465 .79 Poor Condition -5,928 .42 Bedrooms 3,143 1.99 Baths 93 0.05 Tuck-Under Garage -11,841 4.53 Carport Floor Area (SF) 9.20 .65 Garage Floor Area (SF) 42.7 7.43 Patio or Deck 1,961 1.21 Greenhouse or Sports Court -12,911 1.33 Shed, Workshop, or Extra Garage 2,766 .37 Hot Tub -8,268 1.28 Clark County Location -34,970 15.1 Washington County Location -18,733 10.1 Clackamas County Location -14,578 2.27 Factors Abutting Transmission Line -92.67 .061 Abutting Transmission Line (King Co.) Abutting Transmission Line (Nash. Co.) Abutting Transmission Line (Clark Co.) Abutting Transmission Line (Clackamas Co.) Sample Size (n) 712 Model R Squared .853 Model Adjusted R Squared .846 Model F-statistic 122.86 Model 4 Variable Coefficient |t-statistic| Intercept 14,160 2.43 Covariates Quarter 1 -5,982 2.78 Quarter 2 -4,685 2.40 Quarter 4 -2,380 1.15 1990 Sale 16,631 5.64 1991 Sale 22,650 7.39 1992 Sale 31,745 6.17 Age -892 7.08 Age Squared 10.86 3.36 Lot Size (Acres) 6,702 4.19 Cul-de-Sac Lot -2,884 1.53 Corner Lot 2,780 1.14 Flag-Shaped Lot -3,386 .72 Above Average Landscape 14,132 4.30 Below Average Landscape 2,010 1.04 Sloping Topography 2,126 1.08 Floor Area (SF) 42.26 20.0 Excellent Condition 31,962 12.9 Fair Condition -3,577 .81 Poor Condition -6,907 .47 Bedrooms 3,163 2.00 Baths 91 .05 Tuck-Under Garage -11,768 4.49 Carport Floor Area (SF) 9.96 .71 Garage Floor Area (SF) 43 7.45 Patio or Deck 1,811 1.10 Greenhouse or Sports Court -13,076 1.35 Shed, Workshop, or Extra Garage 2,713 .37 Hot Tub -8,321 1.28 Clark County Location -36,684 12.5 Washington County Location -19,553 8.59 Clackamas County Location -14,018 1.63 Factors Abutting Transmission Line Abutting Transmission Line (King Co.) -1,365 .66 Abutting Transmission Line (Nash. Co.) 836 .32 Abutting Transmission Line (Clark Co.) 2,801 .72 Abutting Transmission Line (Clackamas Co.) -2,453 .19 Sample Size (n) 712 Model R Squared .853 Model Adjusted R Squared .845 Model F-statistic 122.8 Table 5 Regression Model with Abutting Transmission Line Price Change Effects (Natural Log of Price Dependent Variable) Variable Coefficient |t-statistic| Intercept 10.73 242.8 Covariates Quarter 1 -.077 4.71 Quarter 2 -.041 2.79 Quarter 4 -.026 1.66 1990 Sale .149 6.51 1991 Sale .196 8.14 1992 Sale .276 6.83 Age -.009 7.12 Age Squared .0001 5.36 Lot Size (Acres) .057 4.66 Cul-de-Sac Lot -.008 .58 Corner Lot .038 2.07 Flag-Shaped Lot -.094 2.63 Above Average Landscape .069 2.78 Below Average Landscape .019 1.26 Sloping Topography .017 1.17 Floor Area (SF) .0003 19.5 Excellent Condition .152 8.13 Fair Condition -.101 2.99 Poor Condition -.134 1.24 Bedrooms .040 3.35 Baths .037 2.86 Tuck-Under Garage -.062 3.11 Carport Floor Area (SF) .0001 1.28 Garage Floor Area (SF) .0003 7.23 Patio or Deck -.015 1.22 Greenhouse or Sports Court .101 1.38 Shed, Workshop, or Extra Garage -.046 .81 Hot Tub -.027 .53 Clark County Location -.300 17.0 Washington County Location -.183 12.9 Clackamas County Location -.120 2.46 Factors 1990 Sale and Abutting Transmission Line -.195 1.30 1991 Sale and Abutting Transmission Line .011 .58 1992 Sale and Abutting Transmission Line .028 .25 Sample Size (n) 712 Model R Squared .865 Model Adjusted R Squared .858 Model F-statistic 127.11
(1.) J.R. Cowger, Steven C. Bottemiller, and James M. Cahill, "Transmission Line Impact on Residential Property Values," Right of Way (September/October 1996): 13-17.
(2.) Francois Des Rosiers, "Power Lines, Visual Encumbrance and House Values: A Microspatial Approach to Impact Measurement," Journal of Real Estate Research 23, no. 3 (2002): 275-301.
(3.) For a thorough discussion and listing of EMF-related epidemiological studies, see www.niehs.nih.gov/emfrapid/booklet/home.htm.
(4.) William G. Hardin and Marvin L. Wolverton, "An Introduction to the Analysis of Covariance Model Using an Empirical Test of Foreclose Status on Sale Price," Assessment Journal 6, no. 1 (1999): 50-55.
(5.) Joseph F. Hair, Rolph E. Anderson, Ronald L. Tatham, and William C. Black, Multivariate Data Analysis, 5th ed. (Upper Saddle River, N.J.: Prentice Hall, 1998).
(6.) Ibid., 188.
(7.) John Neter, William Wasserman, and Michael H. Kutner, Applied Linear Statistical Models, 3d ed. (Homewood, III: Irwin, 1990), 305.
(9.) The authors are aware of the many cautionary comments existing in the literature regarding the misuse of stepwise, data-driven statistical models rather than theory-driven statistical models. Indeed, one of the authors has written some of the cautionary comments and critiques found in the appraisal literature. In this instance, stepwise regression is used as a model-diagnostic tool only, which is not an abuse of theory but merely a further precautionary attempt to uncover any data-resident anomalies that could have influenced the insignificant result on the abutting HVTL factor of interest. This constitutes an appropriate use of the stepwise algorithm.
(10.) The term blocking, as used in this paper, refers to controlling for the systematic effects of variables that are not of interest to the researcher (often referred to as supplemental variables). Generally speaking, the effects of such variables can be managed directly through experimental design or indirectly through statistical means. Indirect statistical means are employed here. For additional discussion of this topic, see the introductory chapter on experimental design in Benjamin J. Winer, Donald R. Brown, and Kenneth M. Michels, Statistical Principles in Experimental Design, 3d ed. (New York: McGraw-Hill, 1991) and Thomas D. Cook and Donald T. Campbell, Quasi Experimentation: Design and Analysis Issues for Field Settings (Boston: Houghton Mifflin, 1979).
(11.) One important question is whether higher priced homes are more subject to price effects from HVTL proximity than lower priced homes. We analyzed the upper price quartile using the same analytical procedures found in Model 1 in Table 3 and Model 3 in Table 4. The t-statistic on the abutting transmission line factor in Model 1 was -1.023 (p-value = .30) and it was -1.01 3 in Model 3 (p-value = .313). Hence, we find no significant HVTL effect for the upper quartile priced homes. The upper quartile of this data consisted of 184 homes that sold for a price greater than or equal to $158,000. The mean price for the upper quartile was $192,324 and the highest priced observation was $320,000, whereas the mean price for the entirety of the data was $128,890 and price ranged from $47,751 to $320,000.
(12.) Cowger, Bottemiller, and Cahill, 13-17.
Marvin k. Wolverton, PhD, MAI, is a visiting associate professor in the Department of Finance at the University of Nevada-Las Vegas (UNLV), where he teaches courses in real estate, economics, and statistics. Prior to joining UNLV, Wolverton held the Alvin J. Wolff Distinguished Professorship in Real Estate at Washington State University. He holds a PhD in business administration, with concentrations in real estate and decision science, from Georgia State University, an MS in economics from Arizona State University, and a BS in mining engineering from New Mexico Tech. Wolverton is the author and developer of Appraisal Institute Course 800, Separating Real and Personal Property from Intangible Business Assets. He is the editor of the Journal of Real Estate Practice and Education and is on the editorial board of the Journal of Real Estate Research. He is currently on the review panel for The Appraisal Journal and has served on The Appraisal Journal's editorial board. Wolverton is widely published in domestic and international professional journals on the topics of real estate finance, economics, and real estate valuation. Contact: E-mail: email@example.com
Steven C. Bottemiller, MAI, is chief appraiser and manager of valuation and forestry for the Bonneville Power Administration of the U.S. Department of Energy. Bottemiller is a graduate of Seattle Pacific University. He has extensive experience in appraising and reviewing elderly care housing properties throughout the United States, and has developed specialties in electrical transmission line/substations, right-of-way, and large farm/ranch appraisals and reviews throughout the Bonneville region. He has previously published articles in Right of Way on the impact of transmission lines on property value. Contact: Bonneville Power Administration, Mail Stop TRV-TPP-4, P.O. Box 61409, Vancouver, WA 98666-1409; T 360-619-6439; E-mail: firstname.lastname@example.org
|Printer friendly Cite/link Email Feedback|
|Author:||Wolverton, Marvin L.; Bottemiller, Steven C.|
|Date:||Jul 1, 2003|
|Previous Article:||Strategic and structural changes in hotel mortgages: a multiple regression analysis.|
|Next Article:||The effect of proximity to a registered sex offender's residence on single-family house selling price.|