Part 2: Research Design and Methods
The hypothesis is that sprawl (measured primarily through housing demand) or urban-rural migration is not just a result of a behavior change (changes to the age specific internal migration rates or household size) but also more significantly a result of changes in the age structure of the population (little or no change to the age specific migration rates). Given age-specific migratory patterns driven by residential preference, growth in sprawl indicators will occur cyclically if the age structure is uneven. Growth in sprawl indicators will also occur seemingly independently of population growth unless one corrects for the delay between birth and when the behavior, such as purchasing of a home, is most likely to occur. Therefore, one cause of suburban sprawl is the combined effect of internal life-cycle migration, age specific behavior patterns, age specific residential preferences and the age structure of the population.
At the core of the argument on sprawl is the belief that as time has progressed, people of similar demographic characteristics have had an ever-increasing affinity for the non-urban areas over urbanized areas, thereby increasing housing demand in non-urban areas over urban areas. To determine if that were the case, it would, in part, require knowing the age specific migration rates between non-urban and urban areas. If there existed a rate of people moving from urban to non-urban areas (age specific internal migration rate) and if it were in effect increasing, then this would constitute a behavioral change, or in other words a shift in residential preference. As these internal migration rates were not available for this dataset, an alternative analysis was undertaken to determine residential preference from differences in age structures of populations residing in various regions within the state at multiple points of time. Then an analysis of the age structure will be done to determine the impact of migration and natural increase on each region (urban and non-urban). Finally, an analysis of housing will be done using the results from the first two methods.
The census figures used for this study were collected into a database
manually from the US Census Bureau books for the years 1970 to 1990 and
consisted of counts of people living in the State of Rhode Island on the
Census date. Municipality of residence and 5 yr age group (starting 0-4)
grouped the data with no delineation between sex and racial composition.
Numbers for the state totals were collected and used as a baseline to
ensure accuracy. This was accomplished by totaling each municipality by
age group and comparing against the collected state totals. All municipalities
were classified two ways as described above by the GSRI definitions of
Urban and Non-Urban as well as Urban Core, Urban Ring, Sub-Urban and Rural.
These classifications were made based on the municipality in the 1990's
and were not changed throughout the entire time period covered by the
study. Note again that the US Census definitions of Urban and Rural were
Therefore, Middletown was removed from the study. Three other municipalities out of 39 were excluded from the survey due to lack of data. These were municipalities not represented by name in the state's US Census General Characteristics book so the tract data representing the municipality was found but ultimately not used. West Greenwich, Little Compton and New Shoreham (Block Island) were removed because the data was unavailable in 5 yr. age groups or in 10 yr. age groups where 0-10, etc; was the breakdown due to their small population sizes. These municipalities are very sparse rural areas.
Birth rates for the state were collected from the Rhode Island State Department of Health (RI-DOH) since the Census Bureau does not provide fertility or mortality data. The data was collected for the age groups of 15-44 by five year grouping in per 1000 numbers. The rates were only considered accurate for the Census years, therefore only the rates in census years were collected. Birth rates were determined by state only and not on an individual municipality basis as they were unavailable for most of the study period.
Death rates were created for the state using total deaths using data collected from the RI-DOH. This was accomplished by dividing the number of deaths in an age group by the number of people in an age group. Similar to births, this was done statewide and not by municipality, as numbers were unavailable on a place of residence basis for most of the study period and only for place of death.
The classifications of municipalities into urban, non-urban etc; were done in a way to continue an established set of conventions used in other studies regarding sprawl in Rhode Island. This is from the work on the subject by RI Statewide Planning Program (RI Statewide Planning Program 1999). This will promote compatible data and will hopefully allow those studies to re-evaluate their original conclusions from within the new framework the present research. The results of the classifications were shown above in Figure 1 (Large) and were based on 1990 data.
Municipalities were classified based on meeting the following criteria from Statewide Planning:
Urban (Urban Ring and Urban Core is delineated out of this larger group by age of city)
Note that Urban is separated into Urban Core and Urban Ring by age of the city, where older cities are pre WWII and new urban centers are post war era.
The first part of the analysis was to look at the state's overall age structure to illustrate the very uneven nature of the age structure in Rhode Island. It was then necessary to determine what, if any, impact migration out of the state had on the age structure of the population and how individual cohorts had changed over time. Performing a standard cohort analysis on the population using the census data from 1970-1990 accomplished this and showed that the change was minimal (results in the next section).
Next the components of change for the population were decomposed into migration and natural increase. This method uses a standard cohort component population model in use by demographers for modeling population projections. (See Appendix 1 for more information) Running the model yielded the level of natural increase each region would expect given zero migration. The resulting numbers for natural increase were then used to compute net migration for the study period. This showed the impact of each demographic component comprising the total change in the population.
The next part of the analysis used a method referred to as Age Structure Deviation Analysis and is detailed further in Appendix 1. This was used to compute the difference in the age structure between a particular region at a particular time and the state age structure as a whole at the same point in time (a period analysis). Any deviation between the two is assumed to represent an age specific residential preference. If a similar pattern of age specific preference occurs year after year, then there must be a pattern of migration occurring internally in the state in order for this to occur. Therefore, in the absence of direct data on age specific migration rates between the two regions, evidence of migration (in net terms) can still be found based on the effect it has on the age structure of the population, providing there is an age specific preference for housing.
Decomposing the sources of housing growth in the state and each region by using the demographic components of change and age-specific householder rates will complete the analysis. This will quantify the actual sources of housing change and show to what extent, if any, behavioral changes have contributed to the phenomenon of sprawl.
[Thesis HTML] [Thesis PDF*] [Soup Sem] [Defense] [Other Works]
*Requires Adobe Acrobat Reader