Shanling Liu • 5 years ago
Hi Vladimir. Thank you so much for presenting an appropriate method, OSL modelling, to investigate the claim that newer and taller buildings can decrease median rents. In this research, the dependent variable and independence variables are well-explained and steps of a standard OSL modelling report are followed (such as checking for normality, heteroskedasticity, multicollinearity, drawing a fitted straight line and observing the scatter plot).
I Just have one comment about the use of OSL model. The regression equations obtained by fitting mathematical expressions are all calculated from limited data within a certain range. Theoretically, its validity is only applicable within this range, but not outside this range, that is, it is only applicable to interpolation calculation, and is not suitable for extrapolation prediction.
Hayley Umayam • 5 years ago
Hi Vladimir,
This seems like a cool idea to explore, I bet it was fun to have such a massive dataset to play around with! I appreciate the work you did to combine 2 datasets and build some variables where necessary, those steps always sound easier than they are! Your work here is quite systematic in checking for associations in your data and subsequently testing assumptions.
In your conclusion you present a lot of ideas for what could have gone wrong, including omitted variable bias or some of the underlying reasons that your data did not meet OLS assumptions. In addition to what you listed, I was also curious about the long list of control variables you tested. I would be curious to hear more about the theoretical justifications for including some of these variables, and if you feel these existing variables adequately capture everything that needs to be controlled, particularly given the mix of variables related to physical structures and those related to demographic features. I was also a little surprised that a study looking at money over time did not include a control for inflation, but perhaps this is already built in to the dataset?
Beyond these control variables, I wonder if it wouldn't also be useful to look for outliers as a source of some of your problems?
Hi Vladimir. Thank you for sharing this very interesting project. In addition to Haley and Shanling's comments below, I wonder how you could replicate this methodology to test the effect of building characteristics on neighborhood rents in cities such as Paris - and perhaps India and Nairobi.