Palgrave Handbook of Econometrics: Applied Econometrics

(Grace) #1
Sergio J. Rey and Julie Le Gallo 1281

1.8

1.4

1.0

0.6

1.8

1.4

1.0

0.6

1.8

1.4

1.0

0.6

1.8

1.4

1.0

0.6

y1929

y2000

0.5 1.0 1.5 2.0
y1929

wy1929

0.5 1.0 1.5 2.0

y2000

wy2000

0.5 1.0 1.5 2.0
y1929

wy2000

0.5 1.0 1.5 2.0

Figure 27.7 Space-time kernels US
Note:y1929 is relative income in 1929 (per capita income for a state relative to national per capita income),
wy 1929 is the spatial lag of relative income for 1929 (see section 27.2.2).


The conventional HDR plot for US incomes from 1929 and 2000 is contained in
the upper left figure. Next to it is the Moran HDR plot, which extends the Moran
scatterplot to a conditional kernel density for the year 1929. The strong spatial
autocorrelation is clearly evident as the modes for the conditional distributions
for the spatial lag of high-income states in 1929 are all above the average, while
for lower-income states, the mode of the conditional distributions are all below
average.
The Moran HDR for 2000 in the lower left of the figure is radically different from
the HDR for 1929. First, all the conditional spatial lag distributions display much
less dispersion in 2000 than in 1929, while the range of the marginal distribution
for relative incomes in 2000 is also much narrower than in 1929. Second, the
strength of the spatial autocorrelation has weakened over the period, as reflected
in the number of conditional modes located in the HH or LL quadrants of the HDR
scatter dropping to 10 in 2000 from 15 in 1929.

Free download pdf