The Impact of Race and Ethnicity on Housing Values

This site extends the work described in our paper, Impact Charts: A Tool for Identifying Systematic Bias in Social Systems and Data presented at ACM FAccT 24. It allows you to look at the impact of race, ethnicity and median income on median home prices in neighborhoods in the 50 largest metro areas in the United States.

Once you make your selection from the two drop-downs, hit the "OK" or "Update" button to see the charts.

The first chart shows the impact of the feature you chose on median home prices, correcting for all other features. So, if you selected Fraction Hispanic or Latino, then the chart shows the impact of the fraction of the population that is Hispanic or Latino on median home prices independent of all other features.

The second chart is a scatter plot of the actual median home value vs. the selected feature. There is less obvious information here because the effects of other features, which may or may not be correlated with the feature of interest, are not removed.

The difference between the two charts shows us the additional value of the impact chart approach used in the first chart.

Please select the area you are interested in:

Please select the feature whose impact you want to plot:

Impact

The impact chart below shows the impact of a single feature in the data the ensemble of 50 models was trained on the prediction is made for each block group. The green dots are ensemble predictions for each block group. Each green dot's x coordinate is the value of the feature for the block group. Its y coordinate is the impact on the ensemble's predicted median home value for that block group.

For each block group there are also 50 gray dots vertically above or below the green dot. These are the predictions of the 50 models that make up the ensemble, and give an idea of the variance of the ensemble prediciton.

Input Distribution

The chart below shows the relationship between a single feature in the data the model was trained and the median home value it was trained to predict. The impact in the chart above is what the model decided the impact of this feature was, taking into account the impact of all other features also. Each dot represents a single block group.