Part 3: Results and Analysis

Age Structure

The age structure of RI was checked for the presence of an uneven age structure needed to drive an age structure effect. Population profiles for the State of RI for the years 1970, 1980 and 1990 are shown in Figure 10. These illustrate the statewide age structure of the population as a whole and show the large amount of deviation between individual age groups over time. Notice that the groupings move forward in time every ten years by 2 age groups. Notice also the 10-14 age group in both 1970 and 1990. Despite a net increase in total population of 63,677[8] or a 7% increase from 1970 to 1990, the 10-14 year old age group was smaller in 1990 by 28,220 or 33% of the 1970 value.

Another more pronounced example of the extreme difference the age structure has over time is the 30 -34 year old age group. In 1970 there were 47,276 and in 1990 the number had shot up to 87,772, which is almost double its 1970 value. As stated earlier, during this same period of time the population as a whole had only increased by 7%. This later cohort (30-34 in 1990) is part of the generation also known as the baby boom generation. That phenomenon is part of the cause of this significant shift in the population. Another significant cause for this variability is the "baby bust" (born in the 1930's) that preceded the baby boom. As far as a comparison with the US as a whole, Rhode Island has a similar age structure except there is a larger elderly population. This made Rhode Island, in 1990, the state with the third oldest population (percent over 65), with Florida and Arizona 1st and 2nd respectively. All told, the presence of an uneven age structure shows that there is a possible contribution to sprawl from an age structure effect.

Cohort Analysis

A cohort analysis was done to determine if the cohorts of Rhode Island's population had changed to any great extent over the study period. Figure 11 is a cohort analysis of Rhode Island over the study period that illustrates the fact that each cohort, or group born at the same time, has not changed significantly over the twenty year study period, with two notable exceptions. The first of these two exceptions is the effect of mortality on those born before 1920 over the study period. The second illustrates two things; the period effect alluded to above and the impact of RI's large college population. The period effect is a net out migration during the 1970's for all cohorts born before 1970, which was a combination of military base closings and economic factors which led to a decline in parts of the population. Based on Figure 11, this out migration appears to have affected the cohort of those born between 1945 and 1955 the most, or those most likely in college or part of the military. The differences between the 1980 and 1990 census shows the impact of the college population predominately.

Components of Change

The historical population was modeled using birth and death rates for the state while holding migration at zero in order to determine the sources of change in the demographic profiles. Figure 12, Figure 13, Figure 14, Figure 15 are results from the model runs (Appendix 2 is a table of results) breaking down the overall change in each age group over time into the proportional impact from migration and aging (natural increase is the total amount of the aging column). Keep in mind that out migration is a number that is subtracted from natural increase in order to determine total change. Notice the variability of contributions to change from aging and that in the relevant age groups of 30+ aging is a greater contributor to change than migration is in most of those groups. This is due to the differences in the age structure of the population. Also, notice that although migration has played a significant role in the increase of population in the non-urban areas, natural increase accounts for two-thirds of all increase in that region during the 1970's as seen in Figure 16 and Figure 17. However, in the 1980's the contribution is only about a third. This shows that natural increase was a factor in population increases in the non-urban areas but that migration also played a role but over the two decades that role changed significantly. In other words migration from urban to non-urban areas has increased.

Internal Migration and Age Structure Deviation Analysis

At first glance, there does appear to have been an increase in migration into the non-urban areas. It appears that migration has doubled into the non-urban region. Assuming that the migration rate to and from the non-urban areas is equal for all people, this is where the analysis would stop. However, as seen in Figure 18 (Large), there appears to be a residential preference of those under 18 for certain areas of the state. At first glance these areas seemed to correlate with the location of the non-urbanized areas. Assuming that each age group had a preference for one area over another, there was a way to determine if that preference had changed over time. This is where Age Structure Deviation Analysis was used to ascertain if age-specific residential preference, and therefore internal life-cycle migration [9], did in fact exist and had it changed over time.

Age Structure Deviation Analysis relies upon comparing the age structure of two regions (assuming there are differences between them) against a baseline (or average) age structure. In this case the baseline is the state age structure and the two regions are subsets of that overall age structure. It is assumed that the differences seen between the age structures constitute a deviation in residential preference. Figure 19 represent the results of the deviation analysis method for the years 1970, 1980 and 1990 where positive numbers represent the residential preference for that region. Note that the representations here are simply snapshots of the population at a given time. Therefore this is a period analysis and not a time-series analysis. Notice that there is a significant preference towards non-urban areas for those under the age of 40 in 1970. It is not until 1980 that a slightly different pattern begins to emerge. There is a preference of those under the age of 20 towards non-urban residence that is then seen again in the 30 to 49 year age groups. What is also significant is this latter group happens to be, statistically speaking, the parents of the former group, the sub 20 year old age groups. Those in the 20 to 30 year range prefer the urban areas as well as the 55+ age ranges. This pattern continues, with some variation, in 1990. Most notably there is a shift in the observed residential preference from non-urban areas to urban in the age groups spanning 55-65.

This appears to confirm that migration had in fact changed over the study period and increased migration was most likely having an effect on sprawl.

Age-Structure Effects and Residential Preference

Referring back to the original hypothesis, a changing age structure, age specific behavior patterns and a residential location preference that is driving internal migration between regions are working in conjunction to drive sprawl rates much higher than overall population increases. Until now all of these issues have been treated separately but when combined some significant findings are elucidated.

Looking back at the deviation analysis graph Figure 19, it appears that the age specific residential preference over time has changed significantly for different cohorts from 1970 to 1990 denoting a possible fundamental change in the preference for housing location.

Notice also that these graphs are in numbers and not in percentages (or relative proportions). Taking into account an uneven age structure, comparing these charts over time may prove misleading since at one point an age group may be larger or smaller than at another point. This is because of the amount of variation in a short period of time that was shown to be possible in the example above in Figure 10 when the 30 -34 year group doubled over the study period. Therefore changes over time may appear to be of a different magnitude but in reality, they may be relatively similar. This adjustment can be seen in the graph Figure 20 where, contrary to the trend occurring above, it appears that the age-specific residential preference over one's lifetime has not changed significantly between generations and therefore migratory patterns have not changed greatly over the study period. This calculation was accomplished by taking the size of the deviation and dividing it by the number of people expected in that age group, assuming the age structure of the region matched the age structure of the state. This made the numbers relative percentages to the size of the age group.

Age Shifts in the Thirty-Something and Middle-Aged Transitions

Looking back on Figure 20 there is a shift in the residential preference for those in the middle-aged and thirty-something groups while no shift in preference is occurring for the under twenty set (statistically the children of those in their thirties and forties) and the over seventy groups. If the shift were due to aging-in-place there would likely be shifts in all of the age groups towards the right on the age axis and that is not being observed. However, these shifts may be the result of a much more complex interaction between the age-specific fertility rate (and possible deviations of that rate between the various regions of urban and non-urban) and other factors. The most likely of these explanations is that a delay in childbearing has pushed the 30+ age groups to the right as seen in the shifts in age-specific rates for fertility in Figure 21. As you can see the peak in the fertility rates has shifted somewhere between five and ten years. This entire issue is worthy of more study by modeling these interactions in a hope to better understand what it is people are valuing the most when then they choose a new home and choose to leave that home for another.
Another interesting result of Figure 20 is that there is a change in the magnitude of people in the 20-39 age groups where they are increasingly looking to the urban areas for residential housing. The reasons for this are not clear but this behavior is actually beneficial as it means at least some age groups in the population are beginning to favor the urban areas more than they had historically.

Housing Analysis

Household Size

Since housing is a big component of sprawl an analysis of these demographic components effects on housing will show what effect demographics has on sprawl. Growth rates of sprawl indicators have been higher than the growth rate of the population as a whole for Rhode Island over the study period. As noted earlier, one demographic explanation for this deviation of rates is that the average household size has changed over time. The implication of a shift in this metric is that individual behavior has changed and people are settling in different patterns than before. With regards to Rhode Island, in 1970 there were 3.24 people per household whereas in 1980 that number had fallen to 2.8. In 1990 the number had fallen a bit more to 2.65. The impact of this shift in average household size led to 46,493 more housing units, assuming all households count as one housing unit [10], needed during the 1970's. Over the 1980's, had the average household size stayed at 1970 levels, there would have been 22,021 less housing units demanded, ceteris paribus. This raises the grand total of additional housing units needed due to the shift in average household size, again assuming all households count as one housing unit, to 68,514 for 1970 - 1990. Over that same 20-year period the number of housing units built was 106,172. The shift in average number of people per housing unit seemed to account for more than half of all construction during the 1970's and 1980's.

Interestingly enough, even this explanation is not entirely sufficient. For starters, the population of Rhode Island has an age structure that is rather uneven. An analysis of average household size does not take this into account nor does it explain the increases in the other indicators of sprawl in the state such as demand for motor vehicles and average VMT per person (used as indicators of sprawl) as each of these would be affected by variation in the age structure because they impact people in different age groups differently. Therefore, shifts seen in the averages of a specific behavior (such as home buying) per person do not necessarily constitute proof that a shift in that behavior has occurred. Since it is this change in behavior that most people cite for causing sprawl, the research from here out will seek to quantify what portion of the change in the number of people per household is due to behavioral changes and what are simply changes due to the composition of the population.

As stated above, the age structure effect has a large effect on the significance of this metric. Using an example to illustrate this point, assume there is a hypothetical population of 10 people. The population consists of two couples, two children to each couple and living in two housing units. There are two other people in the population who are single and each has their own housing unit [11]. That makes 4 housing units to 10 people or an average household size of 2.5 people per household. Assuming no one else enters the population (births or migration) or leaves (death), 10 years go by but the population now looks somewhat different. First the children have all grown up to the age where they move out of the house. None of the four are married and all have their own housing unit. The two people who were single and living apart 10 years earlier have gotten married and are living in one housing unit. The two couples are still married and each couple occupies one housing unit. The average household size is now 1.43 people per household as there are 7 housing units occupied by 10 people. The important aspect of this is there has been no shift in behavior as there has been no change in the age specific behavior patterns. Those who recently left their parents homes are younger than the age at which people get married and cohabitate. The two newly weds, although being married, are not at the age where people normally have children. The two original couples are past the age of giving birth but are still married and therefore continue to cohabitate but have no children. Therefore, the age specific rates for these behaviors have not changed but the proportion of people engaging in those behaviors has changed.

Age-Structure and the Components of Change Effect on Housing Demand

One way to adjust for age structure effects is to look at age specific headship rates [12] for the population in question. Change over time in these rates constitutes a shift in behavior pattern. When there is no shift then all change is attributable to the age structure. However, there may be a combination of both and to adjust for this headship rates are held constant at 1990 levels. Doing this shows what number of housing units would have been needed for any period of time given a particular set of rates. The difference between that number and the observed number is considered the impact of the shift in behavior patterns.

Past Historic Housing Activity

During the 1970's the number of housing units in net demand, attributed to migration, was 3,692 while in the 1980's the net demand attributed to this same in migration was 7,464. Table 3 shows the total number of estimated housing demand (determined by using 1990's headship rates) and the observed numbers of housing demand (See Table 4 for Age Specific numbers). Note that using 1990's headship rate yielded rather accurate housing demand numbers for 1980 but was a little high for 1970 showing that there was likely no significant change in headship rates during the 1980's but some change in the 1970's. Also, note that there is a distinction between actual units and occupied units and that the headship rates yield only the number of occupied units. As noted earlier, the number of occupied units is being used as a proxy for housing demand.

In Table 3 it is clear that the change in housing stock and the change in current total population size was uncorrelated, as most studies have suggested it would be. However, demand based on headship rates shows that as the age structure changes through time, the estimated demand correlates with the observed demand. This is because the age structure changes the proportion of people likely to head a household out of the total population. But what does not change is the likelihood that, at a given age, someone is more or less likely to be the head of a household. The later being an example of a shift in behavior.

Future Projections

Table 5 shows projected housing demand (AKA: occupied households) for the state of RI from 2000 through 2025. The population projections are based on US Census Bureau Estimate A for RI (graphed in Figure 22) and the housing projections use 1990 headship rates applied to all years (Seen in Table 6). Headship rates were unavailable in prior census years due to a lack of age specific household headship numbers from the Census Bureau for renters [13]. One issue with the population projections (2000-2025) is the totals are low (2000 Census totals for RI came out in time for this study and were higher than expected), most likely due to the higher than expected migration into the state during the 1990's.

As time goes on, the age structure will most likely begin to stabilize as Figure 22 illustrates with 2025's population profile. With this stabilization of the age structure will come a stabilization of the housing construction. This means only that there will be less variability in housing demand. One thing that will remain is the fact that housing will continue to outpace population growth as the population growth rate continues to decrease. However, the disparity will be far lower than had been in the past and given enough time (and no more period events such as that causing the baby boom), over which no population growth occurs and the disparity will likely be completely removed. This phenomenon of disparate growth rates is due to the delay between housing demand attributed to one person and the event of their birth. In other words, increases in housing demand occur 30-40 years after an increase in population (providing births is the driving force behind the increase). This displacement is anchored primarily against births because birth rates (and immigration) will be the primary component of change in the population where death rates are considered to be roughly constant into the future.

Table 7

Effect of Vacancy Rate and Other Factors on Housing Demand

As Table 8 demonstrates, the vacancy rate has fluctuated over the study period with a sharp increase over the 1970's and a slight rebound in the 1980's. This is probably a result of a period effect on the population where in the 1970's there was a large out migration of people from the state. It is important to note that although the number of vacation homes has increased, its proportion to the total housing stock has decreased over the study period. The 2000 Census will probably show a further decrease in the vacancy rate.

Summary of Age Structure Effect on Housing Demand

Housing demand across the whole state was much higher than population growth during the period of 1970-1990 due to the long-term effects of the "baby bust" of the 1930's (the cohort aged 50-60 in 1990) and the subsequent "baby boom" of the 50's and 60's. Appendix 3a and Appendix 3b contain a full breakdown of the housing demand per age group, region and component of change. Using these numbers, housing demand over the study period in non-urban areas, absent migration, had a net gain of 34,563 housing units attributed to aging in place. In the urban areas net gain of housing units attributed simply to aging in place would have been 36,579. Adjusted for net in migration, the total demand on the non-urban areas rose to 45,917 while net out migration from the urban areas dropped the demand there to 25,847 housing units. Notice that aging in place accounted for over two thirds of the increased housing demand in the non-urban areas. This leaves a third of the increase in housing demand as a result of migration.

Next: Discussion...