Results and Discussion

Results and Discussion

In our analysis, we found evidence suggesting that house values grow differently for different growth segments post-2014. Especially for the high growth segment, the increase in the house prices outweighs the increase in the medium and low growth groups as well as control areas. Such drastic changes can be the result of the market recovery post-2018 economic downturn as well as the technology innovations and growth happening in the area (Carlisle 2021). This is consistent with our hypothesis that higher company growth is associated with higher house values. Aside from the longitudinal effect of tech growth and other socioeconomic factors on Bay Area house prices over the years, there is also a spatial correlation in each year, meaning that house prices in neighboring areas of a census tract can inform the house price of that area.

In addition, income and whether the census tract has a majority white population are positively correlated with higher house values. Perhaps this is because higher income allows people to afford more expensive houses. Also, the higher the proportion of white residents in a census tract, the higher house values are likely to be. This could be due to the effect of redlining, where people of color have been historically discriminated against in the housing market.

On the other hand, the median household size in a census tract and whether there is a high proportion of people born in California negatively correlate with house prices. Such a trend in household size can be explained in terms of race and age. White households are usually smaller in size, while people of color households tend to be bigger due to living with extended families or having more children. Also, immigrants are more likely to live with extended families who immigrate with them. Regarding the negative effects of the proportion of Californian residents, that can be explained by foreign money. If there is a higher proportion of people from outside of California coming to the area, there will be more wealth flowing into the area, potentially through tech investments from other states or countries.

Finally, the house prices of nearby census tracts impact one another. We can see that Silicon Valley and the neighboring census tracts experience relatively similar house values. Other residential census tracts that are further away or in a county that is not affiliated with big techs like Marin or Contra Costa experience much lower house prices.

Moving forward, there are many directions we can take to improve the analysis. Firstly, we can consider using Uber data to determine the distance between different census tracts and tech companies in terms of minutes. As we hypothesize that house prices are in high demand as high-income tech employees want to live closer to their companies, this can allow us to identify which areas can be desirable for tech employees. Secondly, natural terrain can be a factor in determining house prices. As we have mentioned, areas with national parks or right beside mountains might have cheaper houses; therefore, incorporating natural geographic data into the analysis can improve our models. Thirdly, we hope to combine our longitudinal and spatial models into one as we potentially have to write a new library for the joined model. Finally, we would like to control for interest rates, considering that lower interest rates could lead to higher demand for housing due to better mortgage deals.

Additionally, as our ultimate goal is to understand the house prices in emerging tech hubs like Austin, Texas, and Salt Lake City, Utah, we want to obtain similar data in these areas to verify the insights in this analysis and potentially make a predictive model for these areas.

One limitation to our census data is that there is a ceiling, or maximum, value for house prices. Before 2014, the house value is capped at $1M, whereas the cap is $2M after 2014. There might have been some wording changes on the census survey, which affects the Bay Area census data.