Some of this i put together, a good part came from better brains here than mine, for which i thank them.
A data set may have more than one mode. It is one of the measures of central tendency.
The Appraising Residential Properties, 4th Edition, Appraisal Institute, "Other Quantitative Adjustment Techniques”, Page 344 further states: “…In instances where paired sales analysis is not conclusive, the appraiser may apply judgment to resolve the problem." The adjustments resulting from the appraiser's judgment is based on a study and understanding of historic or past buyer preferences. It further suggests that cost and depreciated cost data may be used with the appraiser arriving at the value contribution (not cost new) of certain features. The process of supporting the contribution of individual variables (features) is limited and often difficult to quantify, with adjustment deemed to be qualitatively supported unless otherwise addressed. All methods of supporting adjustments are usually limited by inherent uncertainties within the applications themselves.
Adjustments, in this report, are based on a combination of Paired Analysis with Sensitivity and/or Trend Analysis & on a study and understanding of historic or past buyer preferences. Support for adjustments may be based on multiple applications and rarely do two methods return identical results with a high degree of accuracy. While not always 'strongly' independently supported, collectively, the adjustments serve to narrow the adjusted value range of the comparables in support of the subject's 'most probable selling price' commensurate with the definition of Market Value set forth herein.
Generally, GLA adjustments are treated as contributory value, not cost to build. The GLA adjustment is the last adjustment made on the market grid. The proper SF (GLA- gross living area) adjustment is that which tightens the range of adjusted values, that results in the smallest range of adjusted values, determined by using simple regression & sensitivity.. I followed the ANSI Z765 standard when I measured the house. However, with obstacles around the house that precludes direct measurements, such as junk, debris, landscaping etc, and set-backs on upper levels or steep inaccessible walls with drop-offs, GLA is estimated and not guaranteed. The ANSI Z765 standard is not intended to supercede any locally imposed system of residential measurement. It may, in many places around the country, challenge longstanding paradigms. The sketch is an approximation, and may not be exact.
If you use regression:
Regression is a better indicator of the final value 'range' and less supportive of the contribution of the independent variables. Any regression model is not weighted into the final value opinion, rather is merely a supplemental tool used to cross-check the reasonableness of the results of Sales Comparison Approach to value. A simple linear analysis was used, with regression on appraiser's “curated” data sale set, giving support to the appraised value. Using regression software for appraiser curated data is as reliable as conventional methods, where as I am instead relying on non-parametric statistics, and of course judgement.
The problem is lender, and their tolerance level. The sensitivity of the underwriter, or reviewer, is based on the tolerance level of the lender. Another one of these requests from the same lender, and it's goodbye to them. I have a low tolerance level for constant stupid. But i only speak from my own way of business, and easy big urban life.