AutoCASE is improving the data source for house values of U.S. cities to make better predictions in our models. Using a comprehensive database from Trulia [1] we’re able to extract all reported weekly mean and median values for houses for 2015.

We are using this data to improve the accuracy of our estimate of average house values in AutoCASE so that the house values are based on finer grained city-specific data. Currently, The Business Case Evaluator (BCE) uses state level data house prices as a simplification. This state level data doesn’t properly reflect the difference in house values between cities in the same state. This is because state level information will severely underestimate house prices in a city with expensive housing prices, and overestimate the house prices of a city with much lower housing prices. To see the effect, let’s compare the average house prices of two cities in California: San Francisco with an average of $1,068,771 and Lancaster with an average of $202,471. The BCE uses the median house prices for each city as the state average in California at $584,375. This means that the state average underestimates San Francisco’s average by almost 50%. On the other hand, the state average overestimates Lancaster’s average house price by about 65%.

We will use this data to find the total property value in a city for use in our flood damage models and property uplift calculations. We do this by dividing the population by the people per household to get the number of households. Then we take the number of households and multiply it by the average property value to get the total property value. The mean and median are two different methods to calculate the average house value in a particular area. The simple mathematical property of a mean divides the total property value by the total number of values. Given that the total number of values and the mean are known, rearranging this mathematical property gets you to the total property value. Using the median value instead of the mean would only work if the distribution of data is not skewed in any way. But since the set of data for housing prices exhibits a right skew, the median underestimates the total value of houses in a city. This is true even if we used city level median prices. Thus, the mean provides a relatively inclusive and better prediction in calculating the total property value.

For instance, the median value for property homes in the city of Los Angeles is about $744,000 while the mean value is about $1.44 million. Given the city’s population of roughly 3.9 million [2], we get two very different numbers as the total property value in Los Angeles: approximately $1 trillion using the median and $2 trillion using the mean. Recently we’ve been integrating the mean price into AutoCASE to calculate total property value in a given city more accurately.

At Impact Infrastructure, we work to create precise models for reliable infrastructure analysis. With the new updates in U.S. average house prices, our models for Property Value Uplift and Flood Risk provide better-quality predictions for Green Infrastructure projects and Low Impact Designs.

[1] “Welcome to the Trulia API.” Trulia Real Estate Search. (2015).

[2] United States. U.S. Census Bureau. “State & Country QuickFacts: Los Angeles (city), California.” United States Census Bureau. (2013).