Hedonic House Price ModelingBased on Multilevel Structured Additive Regression - Computational Approaches for Urban Environments - page 109

a b

Effect of the floor area Effect of the plot area

1900

Mean regression
GAMLSS

2500

1800

1700

2000

1600

1500

1500

1400

Mean regression
GAMLSS

1300

1000

50 100 150 200 250 300 200 600 1000 1400

floor area plot area

c d Effect of the age Effect of the time index

2100

Mean regression
GAMLSS

1900

2000

1900

1800

1800

1700 1700

1600

1600

Mean regression
GAMLSS

1500

1400

0 20 40 60 80 1997 2000 2003 2006 2009

age time index

Fig. 5.1 Effects of the continuous structural covariates of level-1. ( a ) Effect of the floor area
(variable area); ( b ) effect of the plot area (area_plot). ( c ) Effect of the age of the building (age);
( d ) effect of the time index (time_index). Shown are the posterior mean estimates of the mean
regression and the GAMLSS regression based on the gamma distribution

categorical variables at their mode level and evaluate all neighborhood covariates
and spatial effects at the mode of the municipals (which we call the average effect )
and transform the functions from the mean regression to natural units (prices in Euro
per sq. m.). Since the effects are quite different in magnitude, we do not show them
on the same scale.
In panel (a), the effect of the floor area (variable area ) is shown. We find a
monotonically decreasing and very pronounced effect of additional floor area on
prices per sq. m., accounting for a variation of more than 1,650 Euro. However, the
decreasing effect weakens as the floor area becomes larger. The results of the mean
regression and the GAMLSS regression are virtually the same.
Additional plot area ( area_plot , panel (b)) yields higher prices per sq. m. of floor
area, although this effect becomes weaker as plot area increases and levels off at