Journal of Student Research 2018

The Impact Of Uber’s Presence On Taxi Fare of the country where each taxi market is located. The model was estimated using average fares for one-mile, five-mile, and 10-mile rides to estimate the effect of Uber on average taxi fare. The results of each model display different effects for each of the variables, which I expected. The multiple linear regression results are displayed in Table 6.

51

Table 6: Estimation results

1-mile

5-mile

10-mile

Coefficients t Stat

Coefficients t Stat

Coefficients t Stat

6.48***

21.11***

38.54***

Intercept

24.41

38.12

39.25 0.97 -3.09 -8.73 -11.04

2010 Population 0.00001

2.24 0.00001

1.45 0.00001

-1.12*** -3.63*** -3.52*** -3.75***

-2.35*** -6.87*** -6.62*** -6.91***

Uber

-0.172

-0.84 -3.77 -4.93 -4.18

-2.65 -8.17

-0.802*** -0.799*** -0.953***

NE

South

-10.39

MW

-7.89

-8.19

* = 90% significance, ** = 95% significance, *** = 99% significance Table 6: Estimation results

The linear regression model estimated with the 10-mile fare as the dependent variable displayed the highest level of statistical significance in each of the variables. The presence of Uber has an estimated effect of -$2.35 on the average 10-mile taxi fare. The Uber variable is statistically significant at the 99% level. Examining the results for the 10-mile ride specification, each of the variables proved to be statistically significant at the 99% level, excepting 2010 population. The decrease in taxi fare associated with the presence of Uber is consistent with Hypothesis H1: The presence of Uber in a taxi market is associated with decreased average taxi fare in that market. To check for robustness, the results of the parsimonious model and other models were compared. The results for the Uber indicator were not perfectly robust throughout the models. As mentioned, this is because omitted variable bias was present in the early models; control variables were necessary to estimate the true effect of Uber presence on taxi fare. One explanation for the insignificance could be the lack of regional control variables. In the final regression, results show that each regional control variable proves statistically significant and is associated with an average coefficient value of -$6.00. Without controlling for regions, cities located in more expensive regions with Uber are compared to less expensive regions that do not have Uber. The control variables allow for the regional differences to be proportional. Finally, a variance inflation factor (VIF) test was used on each of the variables to test for multicollinearity, which would cause the coefficient estimates to have increased variance. The results of the VIF test did not display any variables that were multicollinear.