Hedonic House Price ModelingBased on Multilevel Structured Additive Regression - Computational Approaches for Urban Environments - page 111

a b

Effect of the share of academics Effect of the age index

Mean regression
GAMLSS

Mean regression
GAMLSS

2400

1900

2200

1800

2000

1700

1800

1600

1600 1500

5 10 20 40 34 36 38 40 42 44 46

share of academics (%) age index

c d

Effect of the population density Effect of the WKO house price index

2400

Mean regression
GAMLSS

Mean regression
GAMLSS

2100

2200 2000

1900

2000

1800

1800

1700

1600

1600

1500

0.1 1 3 10 30 100 50 100 150 200

population density WKO house price index

Fig. 5.2 Effects of the neighborhood covariates. First row : effect of the share of academics (educ)
( a ) and the age index (age_ind) ( b ). Second row : effect of the log of population density (ln_dens)
( c ) and the house price index (wko_ind) ( d ). Shown are the posterior mean estimates of the mean
regression and the GAMLSS regression based on the gamma distribution

5.7.2 Quantiles

Figure 5.3 shows the effects for the structural covariates for the 20-, 50-, and 80 %-
quantiles. Beside the results of the mean regression and the GAMLSS regression,
we now also compare the effects of the quantile regression. Again, we hold all other
covariates constant at mean level of attributes and - if necessary - transform the
functions to natural units.
In general, the effects for the different quantiles are similar to those for the
expected house prices per sq. m., displayed in Fig. 5.1 . However, we can find some
interesting differences between the three models: Quantile regression estimates