**4. Analysis and Results**

Regarding the two variables—housing value and housing added value—two of the same sets of independent variables were put into two respective models: an OLS model based on mean estimation and a QR model based on 19 quantiles.

The parameter estimation of quantile regression parameters adopts the bootstrap method. Every quantile model conducts 500 samplings with replacements of initial samples. Clustered standard error is used with different cities as different clusters to estimate standard deviation.

Compared with the general robust standard error, a clustered standard error presumes that the random error terms in the regression equations are uncorrelated among cities but correlated within cities, and this helps to accurately estimate the vast difference among Chinese regions and strengthen the model; however, it has one disadvantage. The *p*-value of a clustered standard error is higher than that of a robust standard error and is less significant.
