Well, you perhaps haven't gotten the memo, but many federal court judges disagree with you and no longer will accept some boomer appraiser getting on the stand and pulling a number out of his behind as an adjustment.
Finally getting around to responding to this old post. RebelNYC, I am not a boomer. My adjustments are based on economic logic using cost, sales, and income approach principles to find match-pairs or capitalization for differentials, and so forth. These adjustments relate to the language of people in the marketplace, like an investor saying, "Renovated units rent for $150/month more....." Thus, the sales comparison adjustment will reflect this. Yes, there is judgement but no wild guesses or pure feelings. Of course, sometimes the support is weak, and that data suggests a higher adjustment than what makes sense so you reconcile to a lower adjustment amount. If in doubt, I don't adjust and simply explain. If the market data shows adjustments over 20%, I sleep well at night applying it.
Regression analysis is not robotic; it requires extensive judgment. What is the quality of the data (this alone makes garbage of most so-called statistical analysis)?What dependent variable and why (e.g., $ or $/sf)? Which variables do you use? Which you don't? Which you pull out? What is the criteria for dropping a variable? Linear or non-linear? Dummy variables or not? Sampling criteria? As residential appraisal has, maybe, 20 variables to consider, the regression analysis needs
n - 20 degress of freedom > 30, equaling
n>50 homogeneous data points to be considered
statistical. It is worse for commercial. Most of what passes for statistical analysis in appraisal is really just "trend analysis". It is similar to a handful of match-pairs. And if you can't find something worthy of match-pairing, then you won't do better with multiple regression analysis. RebelNYC, start running or requesting Confidence Intervals on such analysis. It will be shocking and disappointing.
When I was on the
TAJ peer review panel, there was an academic article that used "cul de sac" as its sole residential location adjustment while using metropolitan-wide SFR sales. I asked them why this one but not others variables? I am guessing that it was available in their downloaded database so they felt "We have it so it must be important". How many variables should they have considered but did not have data on?
The #1 problem with the academic work is that they ignore correlation and decide to chase the
p-stat. For the unfamiliar reader, a p-stat of a variable nearer to zero is said to have statistical significance. For example, if we sample adult male height, the p-stat will approach zero as we get a larger and larger and larger sample. But it doesn't mean much to say we are 99% certain that the average adult male, if we are to grab one randomly, is between (for sake of conversation) 5'0" and 7'0". Um okay. Here is how it relates to real estate appraisal. One academic study I begged
TAJ to not published had a sample of nearly 10,000s scraper houses, average houses, mansions, and condominiums . They threw it into the same blender and declared victory because the
p-stats were low just like in my height example. Yet the correlation not surprisingly (i.e., the quality of the regression line) was like 20%.
"Do not confuse statistical significance with practical significance. With very large sample sizes it is possible to obtain statistically significant results for same values of b1; in such cases one must exercise care in concluding that the relationship has practical significance."Statistics for Business and Economics, 4th Ed., Anderson, et al., p. 507.
I am glad Federal court judges question and doubt wild *** guessing by appraisers. It is about time. It is probably harder for them to evaluate statistical alchemy. Having said that, Federal court judges are not the marketplace and I wish them luck.
