Many features in house price models are collinear; as you vary the value of one factor, the other factor will also change correspondingly, to varying degrees.
For example, increasing the Room Count of a house will typically increase the size or gross living area (GLA). This is a strong tendency. This makes model construction difficult as a choice must be made between one feature or another to explain market price differences. MARS regression tries to find the combination of these two features that best explain price variance in a data set or market area. Indeed, MARS often generates models that appear at odds with each other, at times making GLA dominant and other times room count dominant. This may not be clear to someone considering a particular price model.
Traditional appraisers typically have specific adjustments for GLA and room count, giving one the dominant role. However, MARS is open in this regard: It will typically give adjustments for both, depending on which combination best predicts price in the input and test data sets.
Why would Room Count be given more importance than GLA? Consider that the MLS data for home area consists of:
- Total building area, or for Single Family Residential (SFR) homes, “gross living area” (GLA).
- The total number of bedrooms.
- The total number of bathrooms, giving a partial bathroom a value of .5 (toilet + sink) or .25 (sink).
- The total number of rooms.
What should be noted here is that many other types of rooms are usually in the house. We will have a kitchen, family room, dining room, laundry room, playroom, and so on. So, unsurprisingly GLA is often better in providing a basis for valuation, especially if the bathrooms and bedrooms are of varying sizes for different homes in the market area.
However, the exception to this is if you are in an area where the bedrooms and bathrooms happen to be about the same size and value across the board. At the same time, the other rooms vary greatly in size and quality, – and in addition, these other rooms, aside from the typical kitchen and living room, are not as important to price as the bedrooms and bathrooms. The reason for the latter is that the number of bathrooms and bedrooms limits how many occupants the home can support. So, in some areas, the bedroom and bathroom count can be far more indicative of value than GLA. In such areas, the MARS regression model may, in fact, give an unusually low adjustment for GLA while giving a far greater adjustment to the bedroom and bathroom count.