The move to housing ownership in temporal and regional contexts

Transcription

1 Environment and Planning A 1994, volume 26, pages The move to housing ownership in temporal and regional contexts M C Deurloo Department of Human Geography, University of Amsterdam, Nieuwe Prinsengracht 130, 1018 VZ Amsterdam, The Netherlands WAV Clark Department of Geography, University of California, Los Angeles, CA 90024, USA F M Dieleman Faculty of Geographical Sciences, Utrecht University, PO Box 80115, 3508 TC Utrecht, The Netherlands Received 16 April 1993; in revised form 28 June 1993 Abstract. A previous longitudinal study of households who make the change from renting to owning demonstrated the close connections between the tenure change and family composition. Specifically, there is a short period in which decisions with respect both to family changes and to house purchase occur. In this paper the authors extend that work and elaborate the findings by directly incorporating a measure of family composition change and analyzing its 'triggering effect' on the tenure change, and by enlarging the temporal and regional context analysis. Shifts from couples to families and increases in income trigger moves to ownership. Also, there are interaction effects between the regional contexts and time periods. A notable finding is that the economic climate affects some groups of households more than others. From the 1980s on, low-income households and one-earner families have been seriously affected in their ability to enter the homeowner housing market. 1 Introduction The move from renting to homeownership is inextricably bound up with the characteristics of the household; with age, race, income, and family size. The entry into the homeownership market does not occur in a vacuum. The move is bound up with the economic context at the time of the move. In addition, circumstances in the housing market vary regionally and this also impacts the probability of entering the homeowner market. In this sense the move from renting to owning can be treated as an event that is closely linked to other events in the life course of a person, such as marriage and income change, but also to circumstances and events in the housing market, such as the price inflation of owner-occupied housing, increasing mortgage rates, and the relative costs of owning versus renting. In the research presented in this paper we use this conceptual structure to examine the move from renting to owning in the United States in the period based on data from the Panel Study on Income Dynamics. Most of the literature on housing choice and tenure transition employs various forms of conventional logit models to link housing and tenure choice to other explanatory variables. These models are not truly time dynamic and do not relate the timing of the transfer from renting to owning to the timing of other events in the household or in the housing market. In this paper we use a proportional hazards model to examine the relationship between the timing of the rent-to-own move and changes in household composition. Further we embed this analysis in a regional and time-period context. Because our other research has clearly shown that the majority of moves into the homeownership market occur for households in the age-group, and indeed primarily in the latter part of this period, we focus on these households in

2 1660 M C Deurloo, WAV Clark, F M Dieleman this analysis. We also limited our research to couples and families, because these are the only types of households where the move from renting to owning occurs in significant proportions. Previous research has been focused on the tenure behavior of couples and families as separate groups (Clark et al, 1994). We were able to show that the influences on the timing of the move for both household groups are remarkably similar. Therefore they are treated here as one group and we measure the actual trigger effect of a change from couple to family status on the rent-to-own shift. Moreover, we illustrate that temporal and regional effects on the move from renting to owning vary across different household types.' 2 Background and previous research The research literature on tenure change is part of a larger literature on the demand and supply of housing services (Olsen, 1987), and specific studies of tenure changes have documented that income has a fundamental influence on the housing consumption decision (Boehm, 198.1; Henderson and Ioannides, 1983; 1985; 1987; 1989; Rosenthal, 1989), that the move to ownership is age related (Clark, 1986; Morrow- Jones, 1988), that the number of earners influences the likelihood of movement to ownership (Myers, 1985), and that there are links between ownership, inflation, and rising interest rates (Mayer, 1981; Rudel, 1987). The series of papers by Henderson and Ioannides (1987; 1989) incorporate family characteristics, wealth, price, and expenditure measures as independent variables to explain the tenure decision. The 1989 paper is one of the first papers with an attempt at a longitudinal (dynamic) analysis of the tenure decision process. Previously we introduced a proportional hazards model as a method of estimating and explaining the temporal nature of moves to ownership for groups of couples and families separately (Clark et al, 1994). In that work we examined couples and families as separate groups and computed the timing of the shift from renting to owning for each group. In addition we considered the timing of the shift to ownership for couples in relation to the transition from a couple household to a family household. For couple households who move to ownership, about 80% become owners within three years after the start of the household. For families who move to owning this is about 72% in three years. Proportional hazards models, where the dependent variable is the time from the start of the episode (as a renter) to the eventual move to ownership, showed that household characteristics including race, income, and the number of earners all play an important role in the shift to ownership, and, significantly, that both economic and regional variables were important additional explanatory variables. It is not unexpected, but we were able to document the positive role of new construction in facilitating the shift to ownership. Inflation also has a positive effect on the shift to ownership. When inflation is high, a house purchase is a hedge against continuing price rises. The results for couples and families are remarkably similar. In the previous research we introduced a specific dynamic model and integrated both household and time-varying housing, and household and contextual variables. The setting is now ready for the integration of the same variables outlined by us (Clark etal, 1994) with specific measures of change in household composition and analyses of the way in which specific types of households are affected by economic and local circumstances in their decision to rent or own a house. 3 Data, units, and methods of analysis The data for the analysis were abstracted from the Panel Study of Income Dynamics (PSID) which is a large panel study initiated in the United States in 1968 (Hill, 1992). The twenty waves (or successive panels) have several hundred variables on a

3 The move to housing ownership 1661 variety of socioeconomic characteristics of individuals and families. The file has standard socioeconomic data and detailed data on the movement behavior of individuals and families for each year. For our analysis we used the combined 1987 family-individual respondent and nonrespondent PSID files. With this combination we have a record of each person who was ever in the PSID sample between 1968 and Because we are also using decennial census data we started the analysis in For the waves we created episodes for households as the basic unit for the analysis. For this analysis we focused on family households, which includes married or long-term cohabiting heads with or without people up to 17 years of age. The subgroup of these households without anyone under 17 years of age is referred to as couple households. An episode is the time period that the household (represented by the head) exists either as a couple or as a family. An episode was initiated when the individual was a head of the current couple or family at the time of the interview, and was a renter. Subsequent years were added to the episode as long as this remained the same, except for the renter status (the household could still be renting or may have made a move to ownership). The interval from the start of the episode as a renter until the eventual move to owning is the length of the spell as a renter household. This is the dependent variable in the proportional hazards models. We ended the episode as soon as the household type changed to anything other than couple or family status. Such an episode is treated as right censored, because the household has left the population at risk before the spell was completed. Obviously the episode was also ended and treated as right censored if information about the household was missing for the next interview year. As our research is directed at residential mobility we restricted the data to all intracounty moves and moves outside the county where the respondents indicated that the move was for a consumption or mixed reason (PSED definition). Figure 1 gives the total number of episodes we analysed over the whole period and the number and proportion of episodes that have a move to ownership. There were 124 episodes that were observed in the first year (1970) and still existed in the last year (1987). The start year and end year of these episodes is unknown (left and right censored). For 2423 episodes the beginning and end of the episode were observed in the period (uncensored). The other episodes were Number of Number of episodes episodes with a move Figure 1. Episodes started in a rental dwelling derived from the PSID and with the age of the head at the start less than 45 years old.

4 1662 M C Deurloo, WAV Clark, F M Dieleman either right or left censored. For the episodes we recorded the race and age of head, whether couple or family status, whether the head was employed, the number of earners in the household, and the household income (adjusted by the purchasing power of the US dollar) at each year of observation, variables which have been identified as critical in the tenure-change literature. These are the household characteristics which will be linked to the rent-to-own move. We also looked at the timing of the tenure change in relation to the timing of changes in the household type between couple and family, of changes in the number of earners, and changes in income. The influence of the rental housing situation was measured by the rent (adjusted by shelter consumer price index) and size (number of rooms). The proportional hazards model was fitted to only the uncensored and right-censored observations. Left-censored observations cannot be used adequately, because one cannot equate the beginning of the observation period with the beginning of the risk period. fi t y South 4 A $**^\ * North 1/7 \\ /**\ Central n 8H (c) Figure 2. struction, rates and inflation. T i i i i i i i i i i i i i i i Year Economic context variables, (a) The price of new single family home, (b) New conthe number of private units started as a percentage of the total stock, (c) US mortgage

5 The move to housing ownership 1663 To measure the impact of the housing-market context on the timing of the move from renting to owning a dwelling, we collected time-series data for each of the four regions (Northeast, Midwest, West, and South) used by the US Bureau of the Census to subdivide the United States. The data included the price of new single-family homes (adjusted), the general rate of inflation, mortgage rates, and the amount of new construction, all variables which have been identified as having significant impacts on tenure change (figure 2). Rather than enter these variables as separate independent variables we used the variations in these measures over time to identify four distinct economic periods: , , , and A major reason for using these time periods, analogous to the use of regions, is that the independent variables are highly intercorrelated, generating periods in which we can say either the indicators are 'green', very favorable for the move to ownership, or the economic incentives do not favor the move to ownership (Clark et al, 1994). In the conditions were moderately favorable for tenure changes, not in the least for households with modest incomes. Prices and mortgage rates were low and the level of new construction was high. In comparison, the period , with rapid inflation, house-price increases, and (relatively) low mortgage rates, was more favorable for households with a higher income to make the move from renting to owning. The periods and were much less favorable time periods for the move to ownership, especially for those with a lower income. New construction slumped and mortgage rates were high, especially in the first period in the 1980s, even though the inflation rates had returned to 1970s levels. And in the West and Northeast, house prices soared between 1984 and 1987 and consequently made the decision to purchase more difficult for an increasing number of households. 4 The propensity to move to owning, triggers, economic circumstances, and regions The powerful corollary effect of age (of the head of household) is present in this analysis as in our study of couples and families separately (table 1). Effectively, the great majority of the movers are less than 35 years old. Income and the number of earners also have predicted relationships. 62% of the movers have incomes in the top two quartiles, and almost 65% of those households moving from renting to owning have two earners. These relationships of the household type, stage in the Table 1. Characteristics of the episodes with a move from renting to owning. Age category in the year of the move <25 25-<35 35-< Total number % Income quartile in the year of the move Total number % Number of earners in the year of the move Total number %

6 1664 M C Deurloo, WAV Clark, F M Dieleman family life-cycle, income, and job participation with the decision to rent or own a house seem to be present at all times and in all circumstances in Western societies. However, what is the nature of the linkages amongst these variables over time? Through the notions of the life course, the process through life can be viewed as a sequence of related events (Mayer and Tuma, 1990). In this approach we can visualize a household proceeding through a series of changes related to occupational careers, family-composition changes, income trajectories, and other events such as the decision to rent or own a dwelling. Thus the move from renting to owning in this perspective relates to three other changes that might trigger the move: (a) a change from a couple to a family, (b) a change from a one-earner to a twoearner household, and (c) a significant positive income change in the year preceding the move. Figure 3 shows the propensity to move from renting to owning by year broken down by these three changes. It is clear from these simple graphs that the move from renting to owning is consistently related to other changes over the period First, more of the households who changed from a couple household to a family moved to owning in the next year than those who did not change family status. Second, a larger number of those who had a positive income change in a year moved to owning in the next year. However, the third change, from a oneearner to a two-earner household in a year, seems less clearly related to the eventual shift to owning in the next year. An appropriate way to evaluate the moves from renting to owning in relation to explanatory variables such as family composition, income changes, and variations in economic and regional context variables is the use of event-history models. These models are focused on the time or the duration in one state, in this case renting, before the change to a new state, in this case owning. Event-history models overcome the conceptual difficulties that arise in a normal regression context in handling uncompleted spells and in measuring explanatory variables whose values change during the time that the household is at risk of making the move. We used the relatively straightforward Cox's proportional hazards model. This model can be written as: loghm = h 0 (t)+pz t, where h t (t) is the hazard function of the survival time of each household, h 0 (f) is any function of time (baseline hazard), Z t fi is a vector of measured explanatory variables for the ith household, and is the vector of unknown regression parameters associated with the explanatory variables (Z). The principle advantage of the proportional hazards model over parametric eventhistory models is that the /? parameters can be interpreted as constant proportional effects (independent from duration!) of the explanatory variables on the conditional probability of completing a spell. This is the analog in a hazard-function framework of the interpretation of an unstandardized regression coefficient in a linear regression model. Moreover, in Cox's partial-likelihood approach to the proportional hazards model, the fts can be estimated without specifying the form of the baseline hazard function. The advantage of this is that there is no risk of misspecification of h 0 (t) and consequent misspecification of ft. A thorough discussion of this model can be found in Kiefer (1988) and Yamaguchi (1991). The actual calculation of the parameters was performed with the SAS procedure PHREG (SAS, 1991). Although there are many statistical computer packages that can run proportional hazards models, in our experience this relatively new SAS

7 The move to housing ownership 1665 program is one of the most powerful and flexible. The hazard h f.(/) is the rate at which the spell as a renter household will be completed at time t by a move to owning, given that the renter situation lasted until /. An addition of one unit in an independent variable reduces (if it is a negative coefficient) or increases (if it is a positive coefficient) the logarithm of the hazard by the value of the regression parameter while controlling for the effects of other variables and time. We also present the (conditional) risk ratio of a variable, defined as the exponent of the regression. coefficient, which is the ratio of the two hazard functions according to whether or not the variable is increased by one unit. To indicate which factors are 1 to 2 earners no change positive income change no change Year Figure 3. Propensity to move to ownership for those who changed from couples to families, by changes from one-earner to two-earner status, and for those with a positive change in income.

8 1666 M C Deurloo, WAV Clark, F M Dieleman more important than others, we could have added, for example, the Wald % 2 statistic of each factor. We do not present that detail, because the values are not directly comparable between dichotomous and nondichotomous variables. Instead we give the significance of the Wald statistic for one level (0.01 for the total group of episodes and 0.05 for subgroups of episodes) which indicates important and lessimportant variables. The hazards model related five groups of variables, the characteristics of households, changes in the characteristics of households, the characteristics of the previous rental dwelling, the'regions "arid the time periods, to the propensity of, the hazard of, a move from renting to owning (table 2). With respect to the characteristics of the household, the results show that the presence of white households, highincome households, employment, and additional earners all significantly increase the hazard of buying a house. For example, the risk ratio for employment status is 1.96, implying that the hazard for employed heads is 1.96 times the hazard for unemployed heads. Age has the correct sign; that is, increasing age decreases the likelihood of buying a house. Couples have a higher (not significant) propensity to move to ownership than do families. This may be caused by the interaction of this variable with the number of earners. Table 2. Hazards model for the total group of episodes with a start age less than 45 years. Variable Parameter Risk ratio Age (t) Race Income (t) Employment status of head (t) Number of earners (t) Couple or family status (t) Change couple to family in previous year (t) Change 1 to 2 earners in previous year (t) Positive income change in previous year (t) Rent (t) Size of rental house (/) Region midwest versus northeast Region west versus northeast Region south versus northeast Time period versus Time period versus Time period versus * 0.67* 0.24* * * 0.46* * (t) Time varying. * Significant at the 0.01 level. -21ogL: without covariates ; with covariates Model x 2 is (p = ) Two changes in the household characteristics have trigger effects on the move to ownership. Changing from a couple to a family in the year before the move has a significant and strong effect and having an increased income shortly before the move is also positively related to moving to ownership, although the coefficient is not significant. The change from one to two earners in the previous year is not clearly related to moving to ownership, as was already evident in figure 3. The influences of the characteristics of the rental unit are as expected.

9 The move to housing ownership 1667 The regional effects are obvious. It is clearly easier to move to ownership in the lower-priced Midwest region and in the South in comparison with the Northeast. The level of new construction in the Northeast was relatively low and prices of owner-occupation were high, thus depressing the opportunities to move from renting to owning. None of the time periods have significant parameters, although the signs are correct. Based on the signs, we might suggest that the periods in the 1980s were less favorable for the move to ownership than the periods in the 1970s with more moderate mortgage levels, moderate price increases, and large amounts of new constructions. However, we agree that this is at best weak evidence for timeperiod effects. Household characteristics and changes in household characteristics seem to outweigh regional and, especially, time-period effects on the propensity to move from renting to owning. But it is quite possible that the contextual effects, especially the time-period effects, which indicate economic circumstances may have different influences on the decisions of different households. We have already suggested that economic circumstances, such as high rates of inflation, relatively high or low prices of owner-occupied housing, and so on, might affect low-income and high-income groups in a quite contrary way. Rapid price inflation might stimulate higher-income households to buy a house as a hedge against the inflation of assets, whereas for those on lower incomes the step from renting to owning might become out of reach in such circumstances. The effects in the aggregate may be to cancel out these individual effects. To examine this hypothesis we focused on specific groups of households to measure the effects of regional and time-period effects on subsets of the sample of household episodes. 5 Subgroups of movers Based on the income, the number of earners, or the family status we identified five simple and stable subgroups who might be quite differently affected by economic and regional circumstances. The subgroups are defined in table 3. When we graph Table 3. Hazards model coefficients for subgroups of household episodes. Variable Annual income below median Annual income above median Twoearner couple Oneearner family Twoearner family Age Race Income Employment status of head Number of earners Household type Rent Size of rental house Midwest region West region South region Period Period Period Total number * * * * -0.86* * 0.91* * 0.49* 0.36* 0.41* * 0.02* * * * Significant at the 0.05 level. Gaps mean the variable is not included * * * 0.54* * * * -1.20* 583

10 1668 M C Deurloo, WAV Clark, F M Dieleman the groups with above-median and below-median incomes, and the couples with two earners and families with one earner, clear distinctions emerge (figure 4). In essence, lower-income groups were increasingly excluded from the owner-occupation market from the early 1970s, with a slight improvement around 1977 when new construction was rapid and mortgage rates relatively stable and low. The high price levels and the high mortgage rates in the 1980s effectively limited these groups from entry into the ownership market. On the other hand, higher income groups previously renting a house moved into owner-occupation in relatively high proportions in the late 1970s, and were less affected in their access to owner-occupied housing by the circumstances in the 1980s Year Figure 4. Propensity to move to ownership for above-median and below-median income and for couples with two earners versus families with one earner. Not unrelated, families with one earner were also increasingly limited in their ability to enter the ownership sector. In the late 1970s in the inflationary spiral that affected much of the US housing market there was a rush to enter the market, and the increasing house prices buoyed up those who entered the market and substantially increased their equity. Renters were often willing to make sacrifices to enter the ownership market. With the recession of the early 1980s and increasing unemployment, and with house prices remaining high, there was a significant reduction in those willing and able to enter the ownership market. Two-earner couples seem to be the least influenced by economic circumstances in their decision to move to owning or to remain in rented accommodation. (1) C 1 ) We also looked at the pattern for families with two earners (a small group, not included in the graph). They seem to fall between the other groups we discussed.

11 The move to housing ownership 1669 The hazard model provides formal estimates which document the descriptive generalizations (table 3). For the subgroups which had difficulty entering the homeowner market in the 1980s, in particular lower-income households, but also increasingly one-earner families, the period effects are large, significant, and negative. For these subgroups, location is much less important than the economic context which decreases their access to owner occupation. Economic circumstances override the regional variations in price and the levels of new construction for these groups. For those household episodes with incomes above the median the model shows that period effects are less important, as already illustrated in figure 4. The period was a period of relatively high levels of movement from renting to owning as the positive, though not significant, coefficient shows. But for this group it is much more important where the household is located. Households in the Midwest, South, and even the expensive West are able to move to ownership more easily than those in the Northeast. These regions also had high levels of new housing construction. The two-earner couples are least affected by regional and time-period effects. They become owners whatever the time period or the region. They move to ownership in relatively large numbers and seem to delay that move only when circumstances are truly unfavorable, as in the very early 1980s. Finally, families with two earners again seem to fall in between the other groups. They are a small group but it is clear that if they are in the Midwest (with somewhat lower house prices) they move to ownership relatively easily. Indeed it is this group that seems to reflect both location and time-period effects in the decision to move to ownership. 6 Conclusions The analysis of the decision to remain a renter or to buy a dwelling in the period has reemphasized the usefulness of placing this decision in a longitudinal perspective. Renting or owning can be viewed as an event among other events in the life course of a household. The propensity to shift from renting to owning is not only dependent upon the household status, such as couple or family status, but also on two-earner status, or positive-income-gain status, as the parameters for these variables are consistently high and significant. The transition from couple to family and a significant positive income change clearly act as triggers for a move into owner-occupied housing as we demonstrated clearly. In addition, changes over time in economic circumstances and the conditions of the local housing market affect the propensity to remain a renter or to buy a house. But these contexts have different meaning for the various groups of households in our analysis. The high mortgage rates and high prices of owner occupation in the 1980s have clearly decreased access to ownership for the moderate income groups and families which have to rely on the income of one earner only. In the late 1970s the first group was also less able to enter owner occupation, which was then a good investment against the inflation of assets, than, for example, the group of couples with two earners. Our analysis in this paper indicates that, in particular, two-earner couples have easy access to the housing market as first-time buyers and are least affected by economic and regional variations in this decision. We also demonstrate that it is useful and relatively straightforward to apply a hazards-model approach to the tenure decision. The main difficulty of such an approach lies primarily in the handling of an enormous and complicated data set such as the PSID. It is not easy to define and identify units of analysis over time and to establish the values of time-varying and place-varying indicators of household circumstances and transitions in circumstances. But once that step is taken satisfactorily

12 1670 M C Deurloo, WAV Clark, F M Dieleman the application of a hazards model can be accomplished by existing computer packages. The models can account for uncompleted (right censored) spells and allow the introduction of time-dependent explanatory variables. Utilizing the concepts of the life course and the formal modelling of hazard functions has provided an enriched understanding of one part of the tenure-change process. An extension of this research will elaborate these findings for the shift from owning to renting (Dieleman et al, 1994). Acknowledgements. The authors acknowledge the support of the Priority Research Program of Population of the Netherlands National Science Foundation (NWO), and the support of the Netherlands Institute for Advanced Study in the Humanities and Social Sciences (NIAS), and the useful comments of an unknown referee. References Boehm T P, 1981, "Tenure choice and expected mobility: a synthesis" Journal of Urban Economics Clark WAV, 1986 Human Migration (Sage, Newbury Park, CA) Clark WAV, Deurloo M C, Dieleman F M, 1994, "Tenure changes in the context of micro-level family and macro-level economic shifts" Urban Studies Dieleman F M, Clark WAV, Deurloo M C, 1994, "Falling out of the home owner market" Housing Studies forthcoming Henderson J V, Ioannides Y M, 1983, "A model of housing tenure choice" The American Economic Review Henderson J V, Ioannides Y M, 1985, "Tenure choice and the demand for housing" Economica Henderson J V, Ioannides Y M, 1987, "Owner occupancy: investment vs consumption demand" Journal of Urban Economics Henderson J V, Ioannides Y M, 1989, "Dynamic aspects of consumer decisions in housing markets" Journal of Urban Economics Hill M S, 1992 The Panel Study of Income Dynamics: A Users Guide (Sage, Newbury Park, CA) Kiefer N M, 1988, "Economic duration data and hazard functions" Journal of Economic Literature XXVI Mayer N S, 1981, "Theory and estimation in the economics of housing demand" Journal of Urban Economics Mayer K, Tuma N, 1990 Event history analysis in Life Course Research (University of Wisconsin Press, Madison, WI) Morrow-Jones H A, 1988, "The housing life-cycle and the transition from renting to owning a home in the United States: a multi-state analysis" Environment and Planning A Myers D, 1985, "Wives' earnings and rising costs of home ownership" Social Science Quarterly Olsen E, 1987, "The demand and supply of housing service: a critical survey of the empirical literature", in Handbook of Regional and Urban Economics volume II Ed. E S Mills (North-Holland, Amsterdam) pp Rosenthal L, 1989, "Income and price elasticities of demand for owner-occupied housing in the UK: evidence from pooled cross-sectional and time-series data" Applied Economics Rudel T K, 1987, "Housing price inflation, family growth, and the move from rented to owner-occupied housing" Urban Studies SAS, 1991, "The PHREG procedure", TR P-217 SAS/STAT Software" version 6, SAS Institute Inc., Cary, NC, USA Yamaguchi K, 1991 Event History Analysis (Sage, London) p 1994 a Pion publication printed in Great Britain