The Fourth Industrial Revolution is expected to greatly influence green space, leisure, land use, transportation use and service, and related residential choice. In the face of the Fourth Industrial Revolution that calls on the construction of a new environmental policy and management system and the prediction of changes in the urban environment and human behavior, the purpose of this study is to examine how to improve research validity for a better construction/prediction and to provide considerations for an appropriate use of big data. Using the case of Google Flu Trends, this study argues that the first limitation of big data is that they draw only correlation, not causality, which increases the chance of the misestimation. As the second limitation of big data, their lack of scientific sampling is discussed, using the cases of a Twitter survey about Hurricane Sandy and Boston’s StreetBump (pothole response) program. Then, this study examines which conditions are required to improve research validity, through an analysis of previous studies on the built environment-travel behavior relationship. As for studies on the relationship, the hottest topic in the current literature is residential self-selection: Individual features work as a confounder that makes the relationship spurious. The spurious relationship accordingly causes the built environment effect to be misestimated. In this sense, considering the four conditions for constructing internal validity or causality -- causal mechanism, covariation, nonspuriousness, and time precedence -- this study critically evaluates the methods of previous studies, including regression-based approaches such as OLS and 2SLS regression, longitudinal design including panel analysis, quasi-longitudinal analysis and recursive and nonrecursive structural equation modeling, and consonant-dissonant matching. Also investigated are two recent models that actively measure the level of residential self-selection, propensity score matching and sample selection model. As a result, previous studies turn out to have delivered mixed results regardless of methodology. Specifically, they are not in agreement regarding the magnitude of the confounding effect of residential self-selection. This study ends by highlighting the fact that as a validity threat, selection bias lowers not only internal validity, but also external validity through its interaction with an explanatory variable, suggesting further research on the generalizability of analytical results.