Covid-19 Impact on Bicycle Usage

Regression Models

Abdullah Kurkcu explains the choice of Partial Least Squares Regression (PLSR) for analyzing bicycling data. He highlights that PLSR reduces predictors to uncorrelated components, addressing issues of collinearity and high standard errors in ordinary regression methods 1. This approach was chosen over more popular methods like random forests or XGBoost to understand variable importance, such as age, income, and population, rather than just prediction accuracy.

PLS regression is actually a technique that reduces predictors to a similar set of uncorrelated components and then performs lead squares regression on those components rather than the original data.

--- Abdullah Kurkcu

Kyle Polich and Abdullah discuss how the proliferation of mobile devices has enhanced data collection, providing insights into travel behavior and aiding in transportation planning 2.

   

Predictive Trends

The effectiveness of predictive models in capturing cycling trends post-COVID is a dynamic challenge. Abdullah notes that factors like average income and education levels initially influenced cycling rates, but these trends evolved over time, reflecting a dynamic system 3. He emphasizes the importance of infrastructure, suggesting that more bike lanes and trails could increase cycling rates, aligning with the triple convergence theory.

If you build more sidewalks, if you build more bike lanes and trails, more people will use it.

--- Abdullah Kurkcu

In Colorado, Abdullah utilized continuous bike counters to gather data, revealing how the pandemic impacted trail congestion and cycling volumes 4.