Model Building
Regression model building is an extremely powerful tool; however, with great power comes great responsibility. These models aren't
an exact science. This process can be compared to the writing process. It can take a while to sequence your ideas, organize,
draft, and revise. This is an iterative process, and the same can be said for supervised modeling building.
The process begins by attempting to gather possible predictor variables or regressors. One can use contextual clues to consider the attributes which impact the price
that a home sells for. Some of these could include number of bedrooms, number of bathrooms, and square feet. Undoubtably, there are
certainly more factors than these three. Would a 1000 square foot condominium with 2 bedrooms cost the same in New York City and
Chicken, Alaska (a small remote town in Alaska with less than 100 people)? Probably not. This is why three different models were
constructed for Ashburn, Sterling, and Leesburg. Despite being nearby towns, there remains an apparent variability in home prices.
Homes of the same size and attributes can sell at very different prices if they are in different locations. This is only
the tip of the iceberg.
When you decide to place your home on the market, an appraiser would likely give you a price that they think your house is worth.
They often do this by looking at nearby houses that are representative of your house in both size and condition.
However, a computer can't go around your neighborhood and do this like a person can. This is where model building comes into play.
The model was built using many similar houses in your area and uses this information to give an informed guess on what your house could sell for. Model
building can be applied to many areas other than real estate. For example, one could construct a model to predict how changes in advertising
spending impacts revenue. Similarly, one could predict the lifespan of asphalt on a highway based on its thickness, quality, and composition.
The possibilities are endless!
Does This Always Work?
The short answer is no. Model building isn't an exact science, and it can be very challenging at times. Sometimes, the data can be mathematically transformed to provide a better fit, but other times it simply won't work. Oftentimes, a major reason of this has to do with the quality of the predictor variables. The variability of the response variable may not be fully accounted for based on the provided regressors. In the case of this project, it is almost impossible to accurately predict the price of every single house in the eastern half of Loudoun County. Most houses in major subdivisions are easy to predict since there are many other homes in the area which are representative of them. In the case of smaller developments with large, extravagant houses, the margin of error is much greater. If you think the prediction of your home price is off, ask yourself if your house is representative of other houses in the area. Maybe your house is on a significantly larger lot or the house is a unique historic house in Downtown Leesburg. Whatever the reason is, one must always remember that regression models aren't an exact science, but they do their best to capture the most amount of variability within the response variable.