Using Logit Models to Analyze Multiple Car Ownership in the Philadelphia Region

To: Yassmin Gramian, Secretary of Transportation, Pennsylvania Department of Transportation
From: Elizabeth Wang, Planning Consultant, Penn City Institute
Re: Using logit models to analyze multiple car ownership in the Philadelphia Region
—————————————————————–

Summary

The Delaware River Valley Planning Commission (DRVPC) and the Penn City Institute have been advocating a compact city development plan in the region. A first and foremost objective is to decrease the use private cars and promote alternative commuting modes such as mass public transit and biking. To study the correlation of individual factors of households and their car ownership of people in the Philadelphia Region, a study has been conducted with the Philadelphia Household Travel Survey data. Factors that may affect multiple car ownership have been examined, including demographic features, counties travelers are from, car parking and transit subsidies and commute mode choices. The study reveals significant differences of multiple car ownership among different social and racial groups based on demographic features, and household locations. The study further reveals possible solutions using subsidies to incentivize people to prefer alternative commuting modes to owning an additional car.

Objectives

This study serves the compact city development plan in the Philadelphia region advocated by the Delaware River Valley Planning Commission (DRVPC) and the Penn City Institute, aiming to find out factors associated with additional car ownership. Researchers in both organizations believe that factors behind owning one more car would be of crucial in decreasing regional auto-dependency, which would furtherenhance a balanced multi-model transportation network. The study also attempts to reveal equity issues regarding additional automobile ownership since people with different f eatures in identity and daily routine would result in different decisions made about whetherto own one more car in their households.

Model Development

The study examines the factors behind people’s incentive of owning one more car, therefore, the dependent variable would be car ownership as a dummy variable – with 0 as owning one car and 1 as owning more than one car. Independent variables include demographic features such as gender, age, race and employment status. Gender, race and employment status have been transformed as dummy variables to avoid meaningless ranking for nominal values. Similarly, counties of original for each household were transferred into dummy variables.

The study further examines monetary incentives on people’s decision of multiple care ownership. Parking and transit subsidies have been examined, in both dummy variables and a percentage of subsidies. The subsidies factors were only analyzed on whether a subsidy was received and how much. In the condition when the employees had to pay parking or transit out of their own pocket or free of charge, it had been categorized as non-subsidy, as a value of zero in the dummy variables. Transportation modes were also considered in the analysis where mode aggregated categories were used. Either the dummy variable or the percentage variable was used in the regression analysis to avoid parallel information as captured by the variables which would over/under-estimate the correlation of certain variables in the regression analysis.

Among all the survey data, refusalto answer and unknown answers were excluded from the study data. This is because such categories usually marked with numbers that are much larger than values in other categories. By excluding these large values, the logit regression analysis would better reflect the real trend among factors analyzed.

In the regression analysis, factors were pulled in logit models in groups, such as demographics, race, and household locations. In the end, factors of all aspects were injected into a giant model to reflect a comprehensive analysis.

Findings

Nine logit regression analysis have been conducted in the study, as shown in Tables 1 and 2. The findings will be based on the most complete model that is model 9, because by including more variables, the omitted variable bias in the model would be minimized. Each of the interpretation assume other factors are held constantly and would not interchangeably affect the models.

To reflect on demographic features, males and females have shown significant difference in car ownership – males on average have a oddsratio of 4.7 times to own multiple cars than females do. In terms of household income, with the increase of one income bracket, the car ownership would increase at an odds ratio of 73%. With the increase of one age bracket, the car ownership would decrease at an odds ratio of around 2%, holding other factors constant. It meansmales, higher income groups and younger generations tend to have larger chancesto ownmultiple cars. The statistics revealsthat certain levels of inequality exist among economic advantaged and disadvantaged groups, different generations, as well as males and femalesin terms of the choices of owning multiple cars.

Model 3 as compared to model 9, better reveals the correlation of factors to multiple car ownership among racial groups. Holding other factors constant, white population have an odds ratio of 1.67 times (67% higher) of other racial groups to have multiple car ownership; black population, however, have an odds ratio of 0.47 (53% lower) of other racial groups to have multiple car ownership – both are statistically significant. It seems that white population have large chancesto own multiple cars, while black population have much lower chance of multiple car ownership.

To interpret parking and transit subsidies and its correlation to multiple car ownership using model 9, parking subsidies generate an odds ratio of 1.08 (8% higher) times of owning multiple cars as compared to not providing parking subsidies. Transit subsidies, however, generate an odds ratio of 0.76 (33% lower) times of owning multiple cars as compared to not providing transit parking subsidies – holding other factors constant. This reveals transit subsidies might provide incentives to reduce number of car ownership (or verse versa).

Statistics further reveals locational differencesin multiple car ownership. In particular, people living in Philadelphia, Delaware and Burlington counties show a lower chance of having multiple cars. For example, in Philadelphia, households on average have an odds ratio of 0.09 (91% lower) times of owning multiple cars as compared to other counties, holding other factors constant. It is reasonable to say that Philadelphia and Delaware enjoy better public transit network and more people of color while the population in Burlington is lower and large percent of the land is used for agriculture and green space.

Conclusion

The study has revealed the socio- and racial unequal distribution of car ownership, in particular, likelihood that a household opt to own more than one car. Wealthier groups, younger generations and white populations show higher chances of willingness to pursue additional private cars as compared to the other groups. Policy makers may use monetary instruments to leverage car ownership and the choice of other public transits considering the significant correlation of transit subsidies and people’s willingnessto own multiple cars. For example, subsidizing transit may incentivize people to use public transit systems and as a result, they would delay their desire of owning an additional car.

R file can be downloaded: https://drive.google.com/file/d/1LcYOxhFhT-bViEEugedCN6MczYqBUy3G/view?usp=sharing