Empirical Results of Panel Regression Model of Urban Sprawl

Table 3 presents a fixed-effects model of panel regression analysis for the 36 metropolitan areas in Taiwan for three time periods—1966-80, 1980-90, and 1990-2000; this model shows the variables that cause a metropolitan to become less compact or more sprawling during the three observation periods. This panel model is a dynamic model which shows what factors cause the change of urban form, rather than a stable model showing the characteristics of metropolitan areas with different degrees of compactness/sprawl. This fixed-effects model is selected over fixed effects model since the results of Hausman test suggests its superiority where the

probability is not significant (i.e., .009<.05). In addition, Robust standard errors are adopted to control for heteroskedasticity. The model has fairly good predictive powers with goodness of fit of 61.2%.

First of all, the panel model results show that generally the pulling-outwards forces cause less compact or more sprawling development though some of the results are not statistically significant (Table 3). Controlling for two metropolitan variables—population and household density— the more rapidly the population grows over a some-one-decade-long period in a

metropolitan area, indicating more residential spatial needs, the less compact or more sprawling it becomes, without considering how compact or sprawling a metropolitan area is at the beginning of an observation time period; the coefficient and P-value of population growth rate reveals that the change of Sprawl Index is statistically positively affected by population growth. However, even though household size has a negative coefficient, indicating that shrinking household also contributes more residential needs and hence less compact development or urban sprawl, it is not statistically significant. Similarly, emigration from central cities to outskirts has a positive coefficient, meaning it will cause less compact or more sprawling development, but it is not statistically significant. All the findings as a whole suggest that when the need for residential space a metropolitan area in Taiwan grows, the metropolitan appears to grow un-smartly from the population distribution point of view. Besides, this analysis again supports that population is an ideal variable to gauge the degree of urban sprawl, but the number of households is superior than population in measuring needs for residential space.

Those who moved outward: (1) Zoned out, (2) Intended out Expected

POP growth Real (3) Met residential demand POP growth

POP t2: From statistics POP t1 POP t1

13

Secondly, the panel model results show that the higher the proportion of all available housing (existing unoccupied housing stock plus newly built housing) in the outskirts, the less compact or more sprawling the metropolitan area becomes (Table 3). However, the impact is only borderlinely significant. This result may imply that, given others equal, the housing available in the outskirts pulls households outwards to outskirts, and hence cause less compact development or urban sprawl, or vice versa that the housing available in central cities may lead to more compact development. Thirdly, the model also reveals that improved accessibility due to higher automobile ownership, lead to a less compact or more sprawling development spatial pattern. In addition, the number of highway exits does not statistically affect urban form (Table 3).

Finally, the model shows that the degree to which a metropolitan will become more sprawling or compact is also affected by its population size and household density. Given others equal, the larger the metropolitan population, the less compact or more sprawling its urban form will become (Table 3). This result is consistent with the impact of population growth presented above. Household density of a metropolitan area affects USI differently; the higher the household density, the more compact or less sprawling the metropolitan area will become.

Unlike population size, which is largely caused by the nature of the society, and metropolitan area, household density to certain degree can be affected by land use policy and consequently can be a significant policy tool affecting urban form in terms of urban sprawl.

Further, the relative significance of each factor that leads to less compact/more sprawling urban form is revealed by the magnitude of standardized coefficients (Table 3). This

significance analysis of factors will be analyzed separately for policy-related and for natural characteristics of a metropolitan area, respectively, even though the line sometimes is hard to draw when policies are mostly made to cope with development trend. This arrange of analysis framework make it easier to understand what natural characteristics cause a metropolitan area to grow in a more compact or more sprawling way, and what policies can shape future urban form more significantly.

Table 3 Dynamic Panel Model of Urban “Sprawl,” Taiwan’s Metropolitan Areas, 1966-2000

Variables Coefficient (Sig.) Robust S.E.

Standardized Coefficient Pushing outward factors:

Overall pushing outward:

Population growth rate between previous time point and

current time point (△Pop%(t-1)~t0) 14.881(.029) 6.644 .119

Percentage change in household size (△HH-Size% (t-1)~t0) -39.602(.381) 44.900 -.052 Pushing outward from central cities:

Ratio of migration from central cities to outskirts between previous time point and current time point to metropolitan households (Ratio-Migrationcc (t-1)~t0)

16.282(.805) 65.724 .026

Pulling outward factors:

Ratio of all available housing in the outskirts to metropolitan

households(Ratio-Outskirt-All-Housing(t-1)~t0) 34.580(.060) 18.082 .142 Accessibility variables:

Public policy related:

Number of highway exits at mid-time-period

(Highway-Exit(t-0.5)) .362(.612) .711 .041

Socioeconomiccharacteristics:

Household automobile ownership at mid-time-period

(Automobile-Ownership(t-0.5)) .342(.000) .026 1.222

Control variables:

Population (t-1) 4.56e-05 (.000) 9.63e-06 1.449

Household density (t-1) (HH/KM²) -.253 (.000) -.046 -2.056

Constant 2.582(.778) 9.119 0.262

Summary Statistics:

Total number of observations 108 (3 time periods, 36 metropolitan areas)

R² 0.612

Hausman test (Fixed/Random): Prob. .009

In the arena of natural characteristics, population size of a metropolitan area is the most significant factor, and followed by household automobile ownership, and population growth rate (Table 3). Across all metropolitan areas, the larger population size, household automobile ownership, and population growth rate, the less clustered/more dispersed the metropolitan area becomes. Among these non-policy variables, household automobile ownership, in fact, can be affected by such policy as transportation infrastructure, taxing and pricing policies, and density of land use policy.

The policy-related variables affecting the change of degree of urban sprawl, are household density, and the availability of housing in the outskirts (Table 6), all of which are land-use related factors. Household density has a lot higher impact than the availability of housing in the

outskirts, and in fact is the most crucial factors of all. The higher the overall household density, and less supply of housing in the outskirt, the more compact or less sprawling a metropolitan area will become.

15