We are taught that a comp should be similar, then we make adjustment. Easy.
There are three types of adjustment results. Some are exact, others are approximate, and others are biased but can help us get near the right adjustment number. It is important to understand the basic differences in these three different outcomes. We will consider how these are different, how each occurs, and how to handle each in our appraisals. Future blogs will explore each of these in turn.
Each of the three methods recognizes that there may be some error in measurement. If the subject or a comp is mis-measured or mis-reported, the analytics will not fix that. That has to do with data collection and validation. Also we disrespect any notion that a random sample is taken, so there is no approximation error from the random sample to the “population” being studied.
For our analysis here, the complete data set is the market segment: all the sales which directly compete with the subject as of the effective date of value.
To simplify for now, we will assume only a measure variable, like Sq. Ft. improved area. We will assume the use of only simple regression: One predictor, one predicted variable.
An exact, deterministic result comes when the input data is measured, say to the date each sale closed. If we have 15 sales, we know the date of each sale, and we know the sale price. Run a simple linear regression, and you get an exact number. We can call this the price index. If the trend is linear, and the data is correct, there is no variation in the answer. It is deterministic. 
An estimated result comes when there is random variation in the input variables (predictors), not related (correlated) with any other predictor. An example is capitalization rates of similar properties, of the same or similar age. Variation may come from a number of sources, such as management quality, where the quality of the management is not connected with any other variable.
A biased result comes when there is correlation with another input variable. An example is Sq. Ft. living area, and number of rooms in a house. This is called collinearity. Basically, each measures the same thing in a different way. If we do a simple regression on living area on price, the living area may also pick up the value of extra rooms. From experience, or other analysis, we know that the relationship is positive. We want to use the regression coefficient as a traditional adjustment. But we know that the coefficient is a bit too high per Sq. Ft., because it also picks up the room count.
Whatever regression coefficient we get is too high. But the question is: Is this information useful?
The answer is yes. Of course. We have an indication. We have a number, and we know it is a bit too high. So we are approaching a reliable adjustment. Now we can “adjust the adjustment.” We have a number. We know it is biased. And we know what direction it is biased – too high.
Can we approach the “correct” adjustment even closer? Well yes. Can we use simple judgment now to adjust for the bias? Well, yes. Is that subjective? Well yes. But is it far less subjective than simply applying a subjective adjustment amount in the first place. Of course! But there are technical ways to bring this in even closer. Another blog for another time.
We will consider the three types of adjustments results in future blogs, as we do in the Stats, Graphs, and Data Science1 class, where we do real exercises with real data, and get real results.
May 10, 2017 @ 8:07 am
George, I would couch the complete data set slightly different;, to included all sales {current and historic, listings, pendings and offers to purchase that would have competed with the subject if they were all available on the effective date of the appraisal.
I take my view from out text that defines market data as a sale, listing or offer to purchase, that if properly adjusted, provides a meaningful indicator of value. When we use all of these Tiers there is almost always enough relevant data. Sometimes, in really bad markets where nothing is selling, I reach out and look at expired, cancelled and withdrawn listings to see if they shed any light on the question of market value..
Regarding collinearity with the SF and Bedroom example, I find that Bedrooms are more important to the overall value in smaller homes. Example, an 1100 sf 3-bedroom home verses a 1200 SF, 4-bedroom. or a 975 SF 2-bedroom verses a 1050 SF 3-bedroom.
Regarding Room Count adjustments, the bathrooms make a larger difference in value, than the SF they occupy. Imagine the price difference in a 3-bedroom home, with 1-bath, verses 2. There may only be 40 SF differende in the size of the room, but the value is huge
The value of the bedroom, may in some ways mirror the value of a pool, though one would not think so.
Example, a pool with 380 sf of surface area that costs new today, say, $30,000 in a $200,000 neighborhood, has less value than in a $900,000 one, less demand.
In smaller homes, that are in the affordable neighborhoods, where larger families live, bedrooms are more of importance to the overall value than a pool.
Thanks again for another meaningful post.