Response to Commentaries on “Does Compact Development Make People Drive Less?”

Click to increase image sizeClick to decrease image size

Acknowledgments

I am grateful to my colleagues Jordi Honey-Rosés and Ward Lyles and my mentor Phil Berke for their helpful comments on an earlier version of this article.

Notes

1. Statisticians sometimes make a similar observation about standardized regression coefficients. Some researchers like to use standardized regression coefficients to determine which independent variables from a set has the largest impact on the dependent variable. The limitation with this procedure is that a 1–standard deviation change in one independent variable is not necessarily equivalent to a 1–standard deviation change in a different independent variable: One of the variables might be more costly or difficult to change than the other.

2. I presented this research at the MAER-Net Colloquium at Hendrix College, Arkansas. The annual colloquium attracts international scholars with an interest in meta-analysis and meta-regression analysis.

3. In the original version of my article that I submitted to JAPA for review, I included a dummy variable in my meta-regression model equal to 1 for original studies conducted in the United States and equal to 0 for studies conducted elsewhere. I included this variable so I could determine whether studies from the United States reported different elasticity values on average than studies from outside the United States. I omitted the variable from my model in the published version of the article because of concerns raised by the reviewers that my sample sizes were too small for me to include multiple independent variables in my models.

4. I appreciate Handy’s (2017) observation that my study’s imperfections are “largely because of the limitations of the (original) studies themselves” (p. 27). Meta-regression analysis is very useful for synthesizing a literature, but that literature’s limitations will be at least partially evident in the meta-regression analysis as well.

5. Other conditions must also be met for the use of regression analysis, such as the condition that both the dependent and independent variables are actually variables that have different (rather than constant) 
values.

6. In the models that control for whether or not the original studies account for residential self-selection, the two independent variables in the models are the standard error of the elasticity estimate and a dummy variable equal to 1 for studies that account for residential self-selection and equal to 0 for studies that do not; in the PEESE estimate models, the two independent variables are the standard error of the elasticity estimate and the inverse of the standard error of the elasticity estimate.

Additional information

Notes on contributors

Mark R. Stevens

Mark R. Stevens ([email protected]) is an associate professor at the University of British Columbia.