|
Authors | Models | Software | Pros and cons |
|
Traditional econometric models |
Kuby et al. (2004) [14] | OLS/heteroskedasticity consistent covariances method (HCCM) | LIMDEP | HCCM assumes that the residual error is irrelevant; however, this assumption is not universal |
Cervero (2006) [13] | OLS | — | The log-log form of OLS is applied to describe suitable relations based on the data; however, the multicollinearity is not delivered |
Usvyat et al. (2009) [15] | OLS | GIS/Microsoft excel | The intercept is negative, which does not conform to the actual conditions; additionally, the issue of collinearity has not been addressed |
Sohn and Shim (2010) [16] | OLS/structural equation model (SEM)) | AMOS | The SEM method needs a large data sample for support; in addition, the selection of latent variables depends on technical and empirical knowledge |
Zhao et al (2013) [17] | OLS | SAS | The signs of the four variable coefficients in the final DRM are negative and incorrect, which indicates that multicollinearity is not considered in the analysis |
Ramos-Santiago and Brown (2016) [18] | Negative binomial regression (NBR) | Stata | The application condition of NBR is that the mean value is equal to the variance; the data sample should be large |
Chu (2004) [19] | Poisson regression (PR) | Stata | The application condition of PR is that the predicted variance should be larger than the mean value |
Chakour and Eluru (2016) [20] | Ordered response probit (ORP) model | Matlab | The parameters should be calibrated by the method of composite marginal likelihood (CML) in this paper |
Spatial econometric models |
Cardozo et al. (2012) [10] | GWR | GIS | The model could explain the diversity of results for spatial factors; however, it needs a large data sample |
Pulugurtha and Agurla (2012) [21] | Spatial proximity method (SPM)/spatial weighted method (SWM) | SPSS/GIS | The buffer of a stop has been divided into four bandwidths, and the best catchment can be identified based on SPM, but the weight function of SWM model (1/D2) is not continuous because of the defined set of bandwidths |
Sung et al. (2014) [22] | Spatial error model (SEM)/spatial lag model (SLM) | GeoDa | This model could describe the relationships between the spatial factor and ridership; however, the definition of the spatial connection matrix is sensitive to the result |
Jun et al. (2015) [23] | Multinomial logit model (MNL)/OLS/MGWR (mixed GWR) | GWR4 | The problem of autocorrelation has been addressed; however, the application of GWR needs a larger data sample for support |
Ma et al. (2018) [24] | Geographically and temporally weighted regression (GTWR) | — | Explanatory variables are eliminated with the index of Pearson correlation lager than 0.6; however, the multicollinearity may exist among the variables and should be tested with variance inflation factor (VIF) index. |
He et al. (2019) [25] | GWR | GWR4 | The explanatory variables are selected from 14 factors; the coefficients of population and official land use that resulted from GWR have negative sign in some areas around stations |
Zhu et al. (2019) [26] | Bayesian negative binomial regression model/GWR | SPSS/GIS | Station ridership determinants are obtained from the method of Bayesian negative binomial regression; the multicollinearity could be addressed as well |
Tang et al. (2019) [27] | GWR/generalized linear model (GLM) | — | The multicollinearity is tested with a Pearson correlation coefficient (PCC) greater than 0.7 and a VIF greater than 7.5 |
|