The Impact of Age and Health on Vehicle Choices among Elders

J Fam Econ Iss (2006) 27:437–457 DOI 10.1007/s10834-006-9023-x ORIGINAL PAPER The Impact of Age and Health on Vehicle Choices among Elders Hyungsoo Kim Æ Jinkook Lee Æ Doh-Khul Kim Published online: 14 June 2006 Ó Springer Science+Business Media, Inc. 2006 Abstract We examine older Americans’ choice of vehicle types and the impact of age and health status on this choice. Using the 1998 and 2000 Health and Retirement Study (HRS) and the 2001 HRS Consumption and Activities Mail Survey (CAMS), we estimate a multinomial logit model of older Americans’ choice of vehicle types. We find that both age and health status influence the type of vehicle purchased or leased. Compared to Americans aged 50–59, those aged 70 or older prefer passenger cars to trucks and sport utility vehicles (SUVs). We also find that elders with health problems are more likely to prefer SUVs to passenger cars than those without health problems. Keywords Age Æ Driving Æ Family burden Æ Health Æ Vehicle types Driving is an activity that enables elders to maintain their freedom, independence, and quality of life (American Association of Retired Persons, 2002; Stutts, Wilkins, Reinfurt, Rodgman, & Van Heusen-Causey, 2001). Elders often cite loss of the ability to drive as their number one worry, ahead of their spouse’s health and their fear of cancer (U.S. Department of Transportation, 1999). As long as they can drive, they can remain active on a daily basis, with less reliance on others, which has direct implications not only for themselves but also for their family members. H. Kim (&) University of Kentucky, 315 Funkhouser Building, Lexington, KY 40506, USA e-mail: hkim3@uky.edu J. Lee Ohio State University, 1787 Neil Avenue, Columbus, OH 43210, USA J. Lee e-mail: lee.42@osu.edu D.-K. Kim Mississippi State University, Meridan, MS 39307, USA D.-K. Kim e-mail: dkim@meridian.msstate.edu 123 438 J Fam Econ Iss (2006) 27:437–457 Elders who stop driving experience undesirable consequences, such as loss of independence, social isolation, reduced or even eliminated access to essential services, or time and money costs (Burkhardt, Berger, Creedon, & McGavock, 1998). Elders who can no longer drive may need to be assisted in such daily activities as grocery shopping, medical check-ups, and leisure, by other members of the family, which results in an increased burden on the family (Elliott, 2003; Kolodinsky & Shirey, 2000). Elders may also need to change the pattern of their economic activities. As non-drivers they tend to make fewer trips and pursue fewer economic activities, which may cause them to feel diminished control and even to be isolated from the rest of the world. Many elders even report emotional trauma and depression as their dependence on others increases (Marottoli et al., 1997, 2000). Family members, relatives, and friends may be forced to provide transportation for elders who stop driving. In fact, it has been reported that the most frequent form of assistance given to older parents is help with shopping and transportation. For example, members of the sandwich generation, who take care of children and parents, assist their aging parents most frequently with transportation (71%), followed by shopping (62%) (Nichols & Junk, 1997). Among the 9.3 million family members or friends who provide care (averaging about 29 h per week per caregiver), 80.6% help their elderly parents or relatives with trips outside the home (McNeil, 1999). Almost half (48%) of these caregivers are full-time employees (National Alliance for Caregiving and American Association of Retired Persons, 2004), which may result in reduced working hours and incomes (Burkhardt et al., 1998). Communities and society also share the burden of transportation. The annual cost of Medicare ambulance transportation is over $2.5 billion, and more than 50% of rural Medicare ambulance trips are non-emergency in nature and could be replaced by less expensive forms of transportation (U.S. Senate Special Committee on Aging, 2003). In 2004, the Federal Transit Administration’s Section 5310 Program for the Elderly and People with Disabilities received estimated funding of $400 million for equipment and services, such as replacement vehicles or new vehicles for expanded capacity and new services. Altogether, the total national estimates of unmet or uncompensated transportation needs for seniors exceed $1 billion per year, including funds devoted to various door-to-door transit expansions, voucher programs, and transportation provided by family caregivers (U.S. Senate Special Committee on Aging, 2003). Despite the negative consequences, elders eventually reach a point where they stop driving. Each year more than 600,000 persons aged 70 and older in the U.S. stop driving and become dependent on others for their transportation needs (Foley, Heimovitz, Guralnik, & Brock, 2002). In 2002, seven million people age 65 and older were not able to drive due to impaired cognitive and physical functions (American Association of Retired Persons, 2002). According to Stutts et al. (2001), 51.7% stop driving due to declining health, 28.9% due to age, and 9.4% due to income. To date, federal, state, and local actions to improve elders’ mobility, such as modifying the road system to assist older drivers through signage design, have been limited (Coughlin, 2001). Without improvements in vehicles, highways, and user programs, the nation will have difficulty in providing safe transportation for the older population. According to data from the Fatality Analysis Reporting System, fatality rates from crashes for those aged 65 or older have not declined much during the past two decades, while a significant decline in fatality rates has been observed in the 1–64 age group (U.S. Department of Transportation, 2003). 123 J Fam Econ Iss (2006) 27:437–457 439 Transportation and technological innovations take time, and changes to the transportation infrastructure and construction of new public transportation facilities are long-term solutions. Furthermore, innovative vehicle technologies intended to help older drivers drive more safely may be hindered by adaptation problems. For example, built-in systems, such as night vision and collision warning, have been proven to require a longer adjustment period, and are often associated with more accidents (Meyer & Coughlin, 2001). Others have proposed modifying the design of vehicles to minimize driving constraints for the elderly (Shaheen & Niemeier, 2001). Such modifications include modest adaptations to the design of such components as seats and doorways. Similarly, we propose that specific types of vehicles may help elders or those with functional limitations drive more easily by providing more convenience and comfort. In particular, (mini) vans and sport utility vehicles (SUVs) provide facilities and features, such as high visibility and easy entry and exit capabilities that can help elders drive more comfortably and safely by mitigating any decreased functional ability stemming from age and health problems. While aging and health problems may influence consumer preference toward different types of vehicles that possess more convenience features, limited research has been done on the influence of age on consumers’ choice of vehicle, and no study has examined the effect of health on vehicle choice. In this study, we investigate older consumers’ choice of vehicle type and the potential impact of age and health status on their choices, using the 1998 and 2000 Health and Retirement Study (HRS) and the 2001 HRS Consumption and Activities Mail Survey (CAMS) (University of Michigan, 2005). This study will enhance our understanding of elders’ behaviors with respect to coping with aging and health problems. In addition, our study’s findings will provide important baseline information for families, policy makers, and industry. Literature Review Determinants of Vehicle Choice Previous literature suggests the following as key determinants of vehicle purchasing behaviors: household characteristics (such as age, education, gender, household income, race, household size, and work status), vehicle attributes (such as price, operating cost, etc.), and existing vehicle stocks (Goldberg, 1995; Lave & Train, 1979; Mannering & Winston, 1985). A major focus of the existing literature has been on examining the determinants of vehicle ownership (Bunch, 2000). Higher income increases the ownership of new vehicles and multiple vehicles. Age is negatively associated with vehicle ownership, and as household size increases, the likelihood of having larger, older, and multiple vehicles increases (Bunch, 2000; Manski & Sherman, 1980). Pickrell and Schimek (1999) investigated the determinants of a household vehicle’s utilization patterns, and found that higher income is associated with increased distances traveled. They also found a negative relationship between age and vehicle use, but the use of vans did not rapidly decline with age. Kockelman and Zhao (2000) also examined the differences in ownership and use patterns among different types of vehicles, including passenger cars, minivans, pickup trucks, and SUVs. They found that if a household owns multiple cars, minivans and pickups are favored as additions, but the addition of an SUV is not affected by existing vehicle stock. 123 440 J Fam Econ Iss (2006) 27:437–457 The age of the primary driver in the household is a significant predictor of the choice of vehicle type (Kavalec, 1999; Lave & Train, 1979). Investigating relationships between household characteristics and vehicle types, Berry, Levinsohn, and Pakes (1998) found that older households tend to purchase larger cars with both more safety features and more accessories, while they are less likely to choose SUVs and pickups than cars. Kavalec (1999) investigated the hypothetical vehicle choices of different age cohorts, using a 1993 survey of Californian residents. Kavalec found that the demand for vans is the greatest at 40 and the demand for pickup trucks is most valuable for those in their late twenties, when Californians are given hypothetical choices of different types of vehicles. Age, Health, and Ability to Drive Even though previous studies have shed some light on choice of vehicle type with regard to household characteristics, they do not provide any indication of the impact health status has on this choice. With physiological aging, older drivers experience changes in basic functions, specifically lessened sensory, cognitive, and physical functions, which all affect driving (Shaheen & Niemeier, 2001). In driving activities, 85–95% of the sensory cues are visual, so weakened visual performance affects the driving ability of elders (Malfetti & Winter, 1986). As people age, peripheral vision, dynamic visual acuity, contrast sensitivity, and night vision gradually decrease, and this weakened vision creates difficulties in detecting approaching vehicles (Wist, Schrauf, & Ehrenstein, 2000), identifying changes on the road surface (Owsley, Sekuler, & Siernsen, 1983), and seeing objects in dim light (Kline et al., 1992). Cognitive ability declines with aging, although the decline is not dramatic (Rodgers, Ofstedal, & Herzog, 2003), and influences only a small fraction of elders; among noninstitutionalized elders, only 1.1% of those aged between 65 and 74, and 17% of those aged 85 or older are cognitively impaired (Kelsey, O’Brien, Grisso, & Hoffman, 1989). However, short-term or long-term memory loss is common among elders, which leads to difficulties in remembering driving instructions for route choice and using new invehicle technologies (Meyer, 2004). The speed of information processing also tends to slow down with aging, which results in slower responses to unexpected shocks. Stelmach and Nahom (1992) found that older adults (aged 60–65) are disproportionately slower than middle-aged or younger adults in brake-pedal reaction time. Changes in physical function, such as decreased range of motion, strength, and flexibility, make driving more difficult for elders (Ehrenman, 2003; Meyer, 2004). For example, muscular strength decreases about 12–15% between the ages of 30 and 70 (Blocker, 1992), and those aged 60 and older experience a decrease in range of motion and strength of up to 25% (Ehrenman, 2003). Quite a few elders have chronic conditions such as arthritis, heart disease, vision problems, and fractures, and these chronic health problems affect their daily activities, including driving and shopping (Guccione et al., 1994). These changes in physical function create, or are the result of, pain and often lead to limitations in ordinary driving-related functions, such as getting in and out of a vehicle, bending over to adjust seat controls, and twisting around to look over their shoulders when parking or changing lanes (Coughlin, 2001; Ehrenman, 2003; Wallace & Herzog, 1995). In coping with such changes, older drivers often take some self-regulated actions. They tend to drive fewer miles and do not drive as often at night, during rush hour, or during bad weather (Straight & Jackson, 1999). Similarly, older drivers may also 123 J Fam Econ Iss (2006) 27:437–457 441 choose specific types of vehicles that are more comfortable and convenient in coping with aging and functional limitations. Recent trends in the demand for specific types of vehicles are noteworthy. Among the vehicles commonly used for passenger transportation, SUVs have continued to grow in terms of both the market share for new sales and the total number of registered vehicles. From 1992 to 1997, the market share of SUVs increased from 8.2% to 16.9% (as of 2004, this share had increased to 26.1%), and the number of SUVs registered increased 93% over the same period. Even though the market share for new minivan sales decreased from 10% in 1992 to 8.8% in 1997 and to 7.0% in 2004, the registered number has still increased 61% over the same period. The market share for passenger cars decreased from 66.6% to 57.7% over the same period (as of 2004, the share was 51.7%) (U.S. Census Bureau, 1997; U.S. Department of Energy, 2003). Baby boomers in particular have purchased minivans and SUVs in record numbers (Krebs, 2000). This trend indicates that minivans and SUVs have been substituted for passenger cars (Pickrell & Schimek, 1999). Some argue that this trend simply reflects temporal popularity, while others argue for a cohort effect (Kavalec, 1999). In this study, we offer an alternative explanation that this trend reflects the changing preferences of elders due to aging and declining functional health. For example, SUVs offer features like higher seat position and increased passenger room (Glover, 2000; Willoughby, 2003), which appeal to those with weakened vision or decreased flexibility and strength in the neck or back. Minivans also have more interior room and sliding doors, which allow for easy entry and exit and additional comfort, along with higher seat position (Petrin, 2002). The Conceptual Framework The conceptual framework of this study is based on the choice model for vehicle type derived from the utility maximization theory (Bunch, 2000; Train, 1990). This theory posits that a consumer chooses a vehicle that provides the greatest utility among different types of vehicles. The utility derived from a particular type of vehicle is a function of vehicle attributes, such as price, performance, comfort, etc., and household characteristics, such as income, age, and health status (Goldberg, 1995; Kavalec, 1999). Consider an individual (i) who chooses among four different types of vehicles (j): passenger cars (c), pick-up trucks (t), minivans (v), or SUVs (uv), from which he/she gets some utility (Uij). This utility function is assumed to consist of an observable part (uij), which is the average utility an individual gets from choosing a vehicle type, and an unobservable stochastic part (ij): Uij ¼ uij þ eij ; ð1Þ where i = 1, 2, ..., n, and j = c, t, v, and uv. The uij term is further assumed to consist of a vector of individual characteristics and vehicle attributes (Xi). Uij ¼ uij ðXi Þ þ eij : ð2Þ This model can be estimated using vehicle type choice probabilities (Pr) defined by: 123 442 J Fam Econ Iss (2006) 27:437–457 Prik ¼ Pr ðUik[Uij forallj 6¼ kÞ ¼ Pr ðuik ðXi Þ þ eik[uij ðXi Þ þ eij Þ : ð3Þ ¼ Pr ðeik
eij[uij ðXi Þ
uik ðXi ÞÞ Equation 3 indicates that an individual chooses (k) type vehicle if, and only if, the utility obtained from (k) type of vehicle is higher than that of all other types of vehicles. For estimation, a functional form of uij (Xi) and the distribution of ij need to be specified. The term uij (Xi) is generally assumed to be a linear function of individual characteristics and vehicle attributes. The error term ij is assumed to be independent and identically distributed across i and j and follows the Type I extreme-value distribution (f(ij) = exp[–exp(–ij)]). This assumption makes this model lead to the well-known multinomial logit model derived from utility maximization. Based on the assumptions associated with functional form and distribution, the vehicle type choice probability can be expressed as follows: expðX 0 ibk Þ : Prik ¼ P expðX 0 ibj Þ ð4Þ j The impact (bj) of individual characteristics on utility across groups who choose different types of vehicles can be estimated with this model (4), using the maximum likelihood method. However, it is structurally impossible for this model to estimate the value of each parameter, such as bj, where j = c, t, v, u (Long, 1997). Instead, only the difference of each parameter, such as (bk – bj), can be estimated. That is, Log Prk ¼ Xi ðbk
bj Þ; k 6¼ j: Prj ð5Þ Practically, one of the parameters bj needs to be held constant (for example, bc = 0). In this study, we employ the probability of choosing passenger cars as our base, comparing it with the probability of choosing other types of vehicles. This yields the following three general logits (log of the odds) estimated. Prc ,Prt, Prv, and Prsuv are the probabilities of choosing passenger cars (cars), pickup trucks, minivans, and SUVs, respectively. Log (Prt/Prc) = the logit of the probability of choosing pickup trucks over choosing cars Log (Prv/Prc) = the logit of the probability of choosing minivans over choosing cars Log (Prsuv/Prc) = the logit of the probability of choosing SUVs over choosing cars Three additional logits are also used for comparisons among pickup trucks, minivans and SUVs. Log (Prv/Prt) = the logit of the probability of choosing minivans over choosing pickup trucks Log (Prsuv/Prt) = the logit of the probability of choosing SUVs over choosing pickup trucks 123 J Fam Econ Iss (2006) 27:437–457 443 Log (Prsuv/Prv) = the logit of the probability of choosing SUVs over choosing minivans Estimating Eq. 5 in terms of individual characteristics (Xi) provides us with the impact of Xi on relative probabilities of vehicle type choice. For example, if the sign of the coefficient of an independent variable is negative for Log (Prt/Prc), we can interpret this to mean that the independent variable decreases the probability of buying a pickup truck relative to buying a car. Methods Data The data set used for this study is the 1998 and 2000 HRS and the 2001 HRS CAMS (University of Michigan, 2005). The 1998 HRS sample is representative of the US population aged 50 and older. This sample was re-interviewed in 2000 and 2002 and is scheduled for follow-up interviews on a biennial basis. The HRS provides data on individuals’ health status and households’ wealth status, permitting investigation of how health and wealth change in later life. In the fall of 2001, a questionnaire, the 2001 CAMS, was sent to 5000 of the households interviewed in the 2000 HRS (13,214 households). This questionnaire included questions on household consumption patterns such as vehicle purchases. One member was randomly selected if a household had two members. The 2001 CAMS collected data from 3813 respondents among the 5000 households. This represents a 76% response rate. We combined the 2001 CAMS data with the 1998 and 2000 HRS data on health and wealth status. Among the 3813 CAMS respondents, 771 had purchased or leased vehicles in the previous 12 months.1 Fifty-one of these respondents did not answer the question on vehicle type and were eliminated from further analysis. The remaining 720 respondents make up the sample for this study. We used a chi-square test to check where there is a significant difference between the remaining sample members (720) and the 51 eliminated respondents. Among all independent variables used in this study, only the race variable (minority) showed a statistically significant difference between the two samples (v2 (1) = 15.23; p < .001). Eliminating 51 respondents does not seriously affect our results. 1 Our sample consists only of those who purchased or leased a vehicle. Thus, sample selection bias may influence our results. We use the Hausman-type test of the independence of irrelevant alternatives (IIA) to test whether there is sample selection bias by adding an alternative choice, ‘‘Did not buy or lease vehicles,’’ to the existing four vehicle types. Table 1 shows the result in the case of using self-rated health (SRH bad) as a measure of health status. bf represents the coefficient of SRH bad in the unrestricted model with five alternatives including ‘‘Did not buy or lease vehicles,’’ and br is the coefficient of SRH bad in the restricted model, with four alternatives excluding ‘‘Did not buy or lease vehicles.’’ The results of estimation showed that SRH bad is statistically significant in utility vehicle in both models. We conducted tests to establish whether or not there are differences in the coefficients of SRH bad (H0: bf = br) between the unrestricted and restricted models. None of the test results rejects the null hypothesis that each pair of coefficients is the same, indicating that our result is not sensitive to sample selection bias. Table 2 shows the result in the case of using AIDL as a measure of health status, indicating no sample selection bias. 123 444 J Fam Econ Iss (2006) 27:437–457 Table 1 Result of sample selection bias test: SRH Chi statistics H0: bf = bar SRH Bad Truckc Vanc Utility vehiclec Unrestricted model (bf) Restricted model (br) –.164 (.305) .015 (.454) .674* (.330) –.177 (.307) .016 (.466) .713* (.333) v2(1) = .35 (Pr > v2 = .55)b v2(1) = .01 (Pr > v2 = .93)b v2(1) = 1.27 (Pr > v2 = .25)b Note. *p < .05 a Null hypothesis that each coefficient pair from the restricted and unrestricted models is the same b p value c Passenger car as a reference Table 2 Result of sample selection bias test: AIDL Chi statistics H0: bf = bar AIDL Truckc Vanc Utility vehiclec Unrestricted model (bf) Restricted model (br) –.717 (.417) –.496 (.477) .272 (.396) –.724 (.417) –.481 (.475) .263 (.399) v2(1) = .10 (Pr > v2 = .75)b v2(1) = .53 (Pr > v2 = .46)b v2(1) = .11 (Pr > v2 = .73)b Note. *p < .05 a Null hypothesis that each coefficient pair from the restricted and unrestricted models is the same b p value c Passenger car as a reference Measures The dependent variable in this study is the type of vehicle that respondents purchased or leased. The 2001 CAMS asked whether or not respondents had purchased or leased a vehicle in the previous 12 months. If they had, respondents reported the vehicle type chosen. Thus, the dependent variable is a categorical variable consisting of passenger cars (cars hereafter), pickup trucks, minivans, and SUVs. Independent variables include age and health status, which are the focal variables of this study, and other control variables that may influence consumers’ choice of vehicle type. The control variables include the following: purchase price of the vehicle, existing stock of vehicle(s), household income, household composition, and the respondent’s gender, race/ethnicity, education, and labor force status. First, age is a categorical variable: age 50–59 (n = 249 respondents), age 60–69 (n = 276 respondents), and age 70+ (n = 195 respondents). We group the sample using this categorization because the older population is heterogeneous, and we want to evaluate the potential nonlinear effect of age. These age groups also separate the most affluent demographic group for car buying (50–59) and the least affluent and most inactive adults (age 70+). Health status is measured using the following two proxy variables: (1) self-rated overall health status (SRH); and (2) having functional difficulties in Activities of Daily Living (ADL) or Instrumental Activities of Daily Living (IADL). While 123 J Fam Econ Iss (2006) 27:437–457 445 self-reported overall health status measures one’s overall physical health condition, ADLs and IADLs measure one’s functional abilities, which are directly associated with driving. Because these two measures capture different aspects of one’s health status, we employ both measures to examine their impacts on vehicle type choice. Both variables are further investigated using baseline status in 1998 and changes in status reported in 2000. First, to identify an elder’s overall health status, the 1998 HRS asks the respondent the following question: ‘‘Would you say your health in general is excellent, very good, good, fair, or poor?’’ The answers to this question are coded on a 1–5 scale, with 1 reflecting excellent health and 5 poor health. A binary variable (SRH Bad in 1998) is created, taking a value of 1 if the respondent rates his or her health as fair or poor, and 0 otherwise. The 2000 HRS asks the following question: ‘‘Would you say that your health is better now, about the same, or worse compared with your health in 1998?’’ Based on the answers, we create a binary variable (Change of SRH: Worse), taking a value of 1 if the respondent rates his or her health as worse, and 0 otherwise. Even though the self-rated overall health measure may be affected by the potential bias of respondents, there has been evidence demonstrating the validity of a self-rated health measure, which is positively correlated with assessment by a health care professional (Ferraro, 1980; Hoeymans, Feskens, Kromhout, & Van Den Bos, 1997) and the incidence of serious health conditions (Hurd & McGarry, 1995).2 Functional disability refers to limitations in performing independent living tasks. ADL are the activities necessary for hygiene and personal care, including walking, dressing, bathing, eating, bedding, and toileting; IADL are the activities essential to residing in the community, including activities such as preparing a hot meal, shopping for groceries, making phone calls, taking medications, and managing money (Spector & Fleishman, 1998). The 1998 HRS asks whether or not the respondent has any difficulties in performing the above ADL or IADL activities due to a health or memory problem. For this study we create a binary variable, Existing ADL or IADL, taking a value of 1 if the respondents had difficulty with any of the ADL or IADL activities in 1998 and 0 otherwise. Another binary variable, NEW ADL or IADL, represents changes in this measure during the past two years. This variable takes a value of 1 if the respondent has developed any new impediments in ADL or IADL in 2000 and 0 otherwise. ADL and IADL are combined in this study, because this combined measure provides an improved measure of functional disability (Spector & Fleishman, 1998). Among vehicle attributes, the only available information from the data set is purchase price, which is a continuous variable. In order to capture existing vehicle stock, we include a binary variable with a value of 1 if an additional vehicle is available in the household and 0 otherwise. The descriptions of all other control variables, which include household and respondent characteristics, are presented in Table 3. 2 The correlation coefficients are .596 between self-rated health and disability, and .492 between self-rated health and number of illnesses (Ferraro, 1980). Elderly men with disabilities affecting mobility and basic activities of daily living had an odds ratio of poor self-rated health of 4.7 and 8.9, respectively (Hoeymans et al., 1997). Probabilities of living to the age of 85 are .53 for elders with excellent self-rated health and .16 for those with poor self-rated health (Hurd & McGarry, 1995). 123 446 123 Table 3 Variable description Variable Description Dependent variable Vehicle type A categorical variable: standard car, pick-up truck, van, or sport utility vehicle (SUV) Independent variables Age Age 50–59 Age 60–69 Age 70+ Health SRH Bad Change of SRH: Worse Existing ADL or IADL New ADL or IADL A binary variable, 1 if a respondent in the household rates his or her overall health as fair or poor in 1998, 0 otherwise A binary variable, 1 if a respondent rates his or her health as worse in 2000 compared to that in 1998, 0 otherwise A binary variable, whether or not a respondent had difficulties in ADL or IADL in 1998: 1 = had any difficulty in the 11 types of activities: walking, dressing, bathing, eating, bedding and toileting, preparing a hot meal, shopping for groceries, making phone calls, taking medications, and managing money in 1998, 0 = otherwise A binary variable, whether or not a respondent without difficulty in 1998 had difficulty in 2000: 1 = if new ADL or IADL, 0 = otherwise A continuous variable, purchase price /$1,000 A binary variable, 1 = have additional vehicles, 0 = otherwise A continuous variable, annual household income in 2000 /$1,000 A A A A A A A A binary binary binary binary binary binary binary binary variable, variable, variable, variable, variable, variable, variable, variable, 1 1 1 1 1 1 1 1 = = = = = = = = single only, 0 = otherwise (reference group) single with other members, 0 = otherwise couple only, 0 = otherwise couple with other members, 0 = otherwise female, 0 = male college or more education, 0 = less than college African American or Hispanic, 0 = non-Hispanic white working in 2000, 0 = otherwise J Fam Econ Iss (2006) 27:437–457 Control variables Price of vehicles purchased Have additional vehicles Household income Household composition Single only Single with others Couple only Couple with others Female College Minority Work A binary variable, 1 = age 50–59 in 2000, 0 = otherwise (reference group) A binary variable, 1 = age 60–69 in 2000, 0 = otherwise A binary variable, 1 = age 70 or older in 2000, 0 = otherwise J Fam Econ Iss (2006) 27:437–457 447 Analysis The HRS is a multistage probability sample of households with an oversample of African Americans and Hispanics (Heeringa & Connor, 1995). We use a weight variable below and take clustering and stratification into account for descriptive statistics. Specifically, the 2001 CAMS provides two household weight variables: one for adjustment for non-response and the other for normalizing the sample representation. As the HRS recommends, we multiply the two weight variables and create a single variable, correcting for (1) the initial selection probabilities, (2) the different attrition rates, and (3) the different participation rates in the CAMS study (University of Michigan, 2005). Vehicle type choice probabilities are estimated using Eq. 5 of the multinomial logit model. The multinomial logit model assumes the independence of irrelevant alternatives (IIA) (McFadden, 1973). This means that the relative probability between two choice alternatives is not affected by other alternatives; in other words, utilities from each alternative are not correlated. Thus, McFadden (1973) and Amemiya (1981) suggest that the multinomial logit model should be used only when the alternatives are clearly different from each other. However, some of the choice alternatives in this study tend to be similar. For example, minivans and SUVs may share some characteristics, indicating that there may be a correlation between the errors. If the IIA does not hold, the results lead to bias. In order to test this assumption, Hausman and McFadden (1984) propose a test of IIA, which compares two estimators of the same parameter: one is consistent and efficient under the null hypothesis, while the other is consistent but inefficient. In practice, the unrestricted model with all alternatives is estimated first. Second, the restricted model is estimated, with one of the alternatives having been eliminated. If these two estimators are not different, IIA holds, suggesting that the result of the multinomial logit model is not biased. The result of the IIA assumption test indicates that there is no difference in the parameter estimates between the unrestricted and the restricted models. Therefore, we can conclude that the specification in this study does not yield biased results. Results Descriptive Analysis Table 4 profiles vehicle buyers. One in five elders bought a vehicle during the past 12 months. Those who have the following characteristics are more likely to purchase a vehicle than their counterparts: people who are younger, healthier, more affluent, married, living with others, male, non-Hispanic whites, more educated, and working. Table 5 profiles buyers by vehicle type. Approximately 60% of vehicle buyers chose cars, followed by trucks, SUVs, and minivans. The preference for vehicle type differs significantly by age group but is not associated with health status. The purchase price, existing stock of vehicles, household composition, and labor market status are also found to differ across vehicle types. For example, the most favored vehicle type for those aged 70 or older is cars (72.8%), followed by minivans (12.8%). Even though the majority of those between 50–59 years old also prefer cars (51.6%), higher proportions of people in this group chose trucks (21.4%) and SUVs 123 448 J Fam Econ Iss (2006) 27:437–457 Table 4 Profile of sample Buy % Age 50–59 60–69 70+ Total Health status Self-rated health in 1998 Fair or poor Above fair Total Self-rated health in 2000 Worse Not worse Total ADL or IADL in 1998 Yes No Total New ADL or IADL in 2000 Yes No Total Income Median Mean Standard deviation Marital status Married couple Single Total Family size One Two Three or more Total Gender Female Male Total Race/ethnicity Minority Non-Hispanic whites Total Education College graduate or more Less than college graduate Total Work (in 2000) Yes No Total Totala Not buy n % n % n 28.3 20.6 16.0 271 290 210 771 71.7 79.4 84.0 693 1119 1101 2913 100 100 100 954 1409 1311 3684 50.7*** 17.7 24.5 151 600 751 82.3 75.5 783 2050 2833 100 100 934 2650 3584 17.8*** 21.5 23.0 153 644 797 78.5 77.0 617 2259 2876 100 100 770 2903 3673 .8 17.6 23.7 81 690 771 82.4 76.3 405 2498 2903 100 100 486 3188 3674 10.1** 18.3 23.4 87 684 771 81.7 76.6 459 2444 2903 100 100 546 3128 3674 6.55* 2886 30,696 49,730 80,061 3651 49.4*** 45,640 69,820 106,111 765 27,692 45,298 72,199 26.6 13.9 540 231 771 73.4 86.1 1495 1408 2903 100 100 2035 1639 3674 71.3*** 12.9 25.0 26.2 105 70 596 771 87.1 75.0 73.8 334 255 2314 2903 100 100 100 439 325 2910 3674 55.4*** 20.1 26.9 435 336 771 79.9 73.1 1965 938 2903 100 100 2400 1274 3674 23.3*** 17.2 24.2 113 658 771 82.8 75.8 599 2304 2903 100 100 712 2962 3674 17.5*** 29.8 20.9 203 568 771 70.2 79.1 512 2391 2903 100 100 715 2959 3674 26.8*** 28.2 20.1 297 474 771 71.8 79.9 806 2097 2903 100 100 1103 2571 3674 29.8*** Note. *p < .05; ** p < .01; ***p < .001 a Percentages and means are weighted b The statistics are t statistics for continuous variables and v2 for binary variables 123 Statisticsb J Fam Econ Iss (2006) 27:437–457 449 Table 5 Profile of vehicle buyers by types Car % Total Age 50–59 60–69 70+ Health status SRH Bad Yes No Change of SRH: Worse Yes No Existing ADL or IADL Yes No New ADL or IADL Yes No Truck n % Van n 59.9 431 17.5 126 % Totala SUV n 9.4 68 51.6 129 21.4 58.0 160 19.9 72.8 142 9.2 53 7.3 18 55 9.1 25 18 12.8 25 57.8 82 16.9 60.6 339 17.4 24 97 61.5 88 20.3 59.4 342 16.8 29 9.1 13 97 19.6 55 9.8 14 9.5 53 % n % Statisticsb n 13.2 95 100 720 19.7 49 100 249 40.45*** 13.0 36 100 276 5.1 10 100 195 701 15.5 22 100 142 12.5 70 100 559 719 9.1 13 100 143 14.2 82 100 576 .94 3.18 66.3 47 11.2 9 59.0 384 18.3 117 7.9 7 9.7 61 14.6 12 100 75 13.0 83 100 645 3.40 63.8 51 16.2 13 59.4 380 17.7 113 8.8 7 9.5 61 11.2 9 100 80 13.4 86 100 640 .61 Price of vehicle (1000$) Mean Std dev 17.9 408 17.4 118 19.3 64 11.1 9.9 9.2 22.4 86 100 676 10.5 4.77** Have additional vehicle(s) Yes No 58.8 387 18.1 119 9.1 60 71.0 44 11.3 7 12.9 8 14.0 92 100 658 4.8 3 100 62 7.34 41.7 427 44.7 125 48.4 65 67.0 94 100 711 69.4 64.1 58.7 92.2 119.4 91.6 45.3 110.6 1.99 Household income (1000$) Median Mean Std dev Household composition Single only Single with others Couple only Couple with others 73.9 105 13.4 56.3 40 26.8 57.4 206 16.4 54.1 80 19.6 19 7.1 10 5.6 8 100 142 24.4** 19 9.9 7 7.0 5 100 71 59 9.8 35 16.4 59 100 359 29 10.8 16 15.5 23 100 148 Gender Female Male 60.6 245 14.8 58.9 186 20.9 60 10.2 41 66 8.5 27 Race/ethnicity Minority Non-Hispanic whites 59.4 57 17.7 17 59.9 374 17.5 109 8.3 8 9.6 60 14.6 14 100 96 13.0 81 100 624 .31 Education College graduate or more Less than college 63.8 122 13.1 25 10.0 19 58.4 309 19.1 101 9.3 49 13.1 25 100 191 13.2 70 100 529 3.66 Work (in 2000) Yes No 58.4 160 20.1 60.7 271 15.9 15.7 43 100 274 10.01* 11.7 52 100 446 55 5.8 16 71 11.7 52 14.4 58 100 404 11.7 37 100 316 5.2 Note. *p < .05; **p < .01; ***p < .001 a Percentages and means are weighted b The statistics are t statistics for continuous variables and v2 for binary variables 123 450 J Fam Econ Iss (2006) 27:437–457 (19.7%) than those in the 70 or older age group, and few of them purchased minivans (8.0%). Multinomial Logit Analysis The estimation results of Eq. 5 are presented in Table 6a, b. Table 6a shows the results of the model with the self-rated health status measures, while Table 6b presents the results of the model with the ADL/IADL measures. Comparing the results in Table 6a, b, we find consistency in terms of both signs and magnitude of parameter estimates. As expected, we find that both age and health status significantly influence the choice of vehicle type. First, those aged 70 or older are less likely to buy trucks and SUVs than cars, compared to those aged 50–59, after controlling for other variables in the model. This effect of age is not observed, however, for the choices between minivans and cars. That is, when we compare the two age groups (those in their 50s and those aged 70 or older), their probability of purchasing minivans instead of cars is not significantly different. Comparing the elders in their 60s with the elders in their 50s, we did not find any significant difference in their choice of vehicle type. Thus, for these two age groups, age is not an influencing factor when it comes to choice of vehicle type. Therefore, we can conclude that elders aged 70 and over have a significantly different preference toward vehicle type than those in the 50–59 age group, and such a difference in preference is observed in cars versus SUVs and pickup trucks. We find that existing health problems at the baseline significantly influence older consumers’ choice of vehicle type after controlling for other variables in the model. Self-rated health status of fair or poor (SRH Bad) increases the probability of buying SUVs instead of buying either cars or trucks (Table 6a). Existing ADL or IADL difficulties also increase the probability of purchasing SUVs instead of buying trucks (Table 6b). However, negative changes in health status, either declining self-rated health status (Change of SRH: Worse) or new ADL or IADL problems during the 2year period, are found to be insignificant. This finding can be explained as a lag between the decline in functional health and its impact on choice of vehicle type. That is, an elder’s health status, measured in both self-rated overall health and functional health, influences his/her choice of vehicle type. However, such an impact takes a bit of lead time to influence the purchasing decision. We also find that the following control variables influence older consumers’ choice of vehicle type: price3, stock of vehicles, and household composition. On the other hand, household income and other characteristics of respondents, such as gender, race/ethnicity, and education, are found to be insignificant. To further illustrate the impacts of age and health status, we compute predicted probabilities using Eq. 5, along with the estimated coefficients and observed values of variables in the sample. The estimated probability changes are presented in Table 7a, b, highlighting the different probability distributions of choosing each vehicle type across age and health status. 3 There may be a potential multicollinearity between purchase price and household income. The correlation coefficient between the two variables is .32. To avoid the potential impact, we exclude each variable in turn and examine the results. However, the results are robust in that price is significant regardless of household income, while household income is not significant, with or without the price variable. That is, the potential multicollinearity between purchase price and household income does not influence the result substantially. 123 (1) log (prt/prc) (2) log (prv/prc) (a) Multinomial logit models for vehicle type choice: SRH Age Age 60–69 –.344 (.274) –.076 (.384) Age 70+ –1.130** (.370) –.024 (.435) Health SRH bad –.177 (.307) .016 (.466) Change of SRH: Worse .237 (.278) .134 (.429) Price .009 (.011) .014 (.014) Have additional vehicle(s) 1.291* (.631) –.076 (494) Household income –.001 (.001) .002 (.002) Household composition Couple and others .373 (.371) 1.098* (.503) Couple only .104 (.316) .708 (.428) Single and others .886* (.429) .695 (.595) Female –.409 (.229) .082 (.289) Minority –.405 (.373) –.079 (.440) College –.528 (.281) .140 (.329) Work –.125 (.266) –.712* (.377) Constant –2.005** (.783) –2.338** (.804) Log likelihood –664.20 Pseudo R2 .0705 (4) log (prv/prt) –.475 (.305) –2.128*** (.544) .267 (.431) 1.105* (.523) –.131 (.360) –.998 (.614) .713* (.333) –.502 (.369) .041** (.012) 1.231 (.765) –.0004 (.001) .150 (.450) –.549 (.428) .004 (.016) –1.368 (.757) –.0009 (.002) .890* (.398) –.739 (.415) .032* (.014) –.060 (.950) .0003 (.001) .740 –.190 .027 1.308 .001 (.466) (.491) (.017) (.870) (.002) .733 (.584) 1.004* (.526) –.313 (.755) .396 (.312) .395 (.473) .373 (.372) .020 (.356) 1.872 (1.182) .008 .401 –.121 –.095 .069 –.295 .608 –1.538 (.683) (.608) (.876) (.363) (.534) (.414) (.444) (1.203) 1.107* (.518) 1.109* (.468) .573 (.702) –.013 (.268) –.010 (.381) –.154 (.305) –.104 (.302) –3.877*** (.985) .724 .603 –.191 .491 .325 .669 –.587 –.333 –.542 (.302) –2.220*** (.542) .241 (.426) 1.052* (.514) .263 (.399) –.178 (.436) .039** (.012) 1.250 (.763) –.0004 (.001) .243 –.406 .010 –1.332 –.001 (.580) (.501) (.678) (.337) (.530) (.398) (.424) (1.050) (.593) (.585) (.016) (.754) (.002) (5) log (prsuv/prt) –.101 (.356) –1.034 (.610) .988* (.522) –.086 (.512) .033* (.014) –.039 (.947) .0003 (.001) (6) log (prsuv/prv) –.399 (.448) –2.103** (.654) –.342 (.444) –2.087** (.650) .744 .319 .023 1.293 .001 (.578) (.628) (.017) (.867) (.002) 451 123 (b) Multinomial logit models for vehicle type choice: ADL or IADL Age Age 60–69 –.441 (.272) –.199 (.382) Age 70+ –1.186** (.365) –.133 (.430) Health Existing ADL or IADL –.724 (.417) –.481 (.475) New ADL or IADL in 2000 –.091 (.367) –.498 (.507) Price .005 (.011) .015 (.014) Have additional vehicles 1.290* (.630) –.042 (490) Household income –.0007 (.001) –.002 (.002) (3) log (prsuv/prc) J Fam Econ Iss (2006) 27:437–457 Table 6 (a) Multinomial logit models for vehicle type choice: SRH, (b) Multinomial logit models for vehicle type choice: ADL or IADL 452 123 Table 6 Continued (1) log (prt/prc) Household composition Couple and others .346 (.364) Couple only .150 (.308) Single and others .883* (.426) Female –.400 (.229) Minority –.405 (.225) College –.526* (.274) Work –.216 (.264) Constant –1.756* (.771) Log likelihood –684.40 Pseudo R2 .0706 (2) log (prv/prc) (3) log (prsuv/prc) (4) log (prv/prt) (5) log (prsuv/prt) (6) log (prsuv/prv) .926* (.486) .581 (.410) .615 (.583) .104 (.288) –.085 (.435) .059 (.325) –.733* (.371) –2.131** (.781) 1.109* (.514) 1.155* (.465) .663 (.699) .011 (.264) .003 (.368) –.296 (.299) –.092 (.297) –3.766*** (.974) .579 .430 –.268 .504 .247 .585 –.517 –.375 .762 (.576) 1.004* (.519) –.220 (.749) .401 (.307) .336 (.456) .230 (.363) .124 (.350) 2.010 (1.168) .182 .573 .048 –.103 .088 –.355 .641 –1.634 (.560) (.479) (.662) (.333) (.517) (.390) (.416) (1.024) (.668) (.594) (.864) (.360) (.524) (.408) (.437) (1.184) Note. *p < .05; ** p < .01; ***p < .001 (b) n = 671. The 49 respondents who did not answer questions on health status or vehicle price were eliminated. Health status is measured by ADL or IADL. Standard errors are shown in parentheses J Fam Econ Iss (2006) 27:437–457 (a) n = 653. The 67 respondents who did not answer the self-rated health question, vehicle price or income were eliminated. Health status is measured by selfrated health (SRH). Standard errors are shown in parentheses J Fam Econ Iss (2006) 27:437–457 453 First, Table 7a reports the impact of age on the type of vehicle chosen by using predicted probability distribution (hereafter probability distribution). Column (1) of the SRH section shows the probability distribution for those aged 50–59 of choosing each vehicle type in the model with self-rated health as the measure of health status. Among this group, 51.8% chose cars, followed by trucks (21.8%), SUVs (18.0%), and minivans (8.3%). Among those age 70 or older (column (2)), the probability distribution for choosing each vehicle type differs substantially: cars increase from 51.8% to 78.3%; trucks decrease from 21.8% to 7.8%; minivans shift from 8.3% to 11.7%; and SUVs decrease from 18.0% to 2.2%. That is, for the age group 70 or older compared to the 50–59 age group, the probability of choosing cars increases by 26.5% and minivans by 3.4%, while the probability of purchasing trucks decreases by 14.0% and SUVs by 15.8% (column (3)). Similar results are found using the model with the ADL/IADL health measure; these are also presented in Table 7a. Second, the impact of health status on the type of vehicle chosen is presented in Table 7b. Column (1) shows the probability distribution of choosing each vehicle type for those whose self-rated health status is good, very good, or excellent. Among those in this group, 60.9% chose cars followed by 18.0% who chose trucks, 11.3% who opted for SUVs, and 9.7% who preferred minivans. Those with fair or poor selfrated health (column (2)) have a different probability distribution of vehicle type Table 7 (a) Impact of age on vehicle type choice: predicted probability (%), (b) Impact of health status on vehicle type choice: predicted probability (%) (a) Impact of age on vehicle type choice: predicted probability (%) Age Age 50–59 (1) Age 70 or older (2) Difference (3) = ((2) – (1)) Health measure SRH ADL/IADL SRH ADL/IADL SRH ADL/IADL Car Truck Van SUV 51.8 21.8 8.3 18.0 50.4 22.6 8.7 18.3 78.3 7.8 11.7 2.2 80.3 7.4 10.3 2.0 26.5 –14.0 3.4 –15.8 29.9 –15.2 1.6 –16.3 (b) Impact of health status on vehicle type choice: predicted probability (%) Health problem No (1) Yes (2) Difference (3) = ((2) – (1)) Health measure SRH Bad ADL/IADL SRH Bad ADL/IADL Car Truck Van SUV 60.9 18.0 9.7 11.3 59.2 18.5 9.9 12.3 57.0 14.0 8.8 20.2 65.7 10.1 6.8 17.5 SRH Bad –3.9 –4.0 –0.9 8.9 ADL/IADL 6.5 –8.4 –3.1 5.2 Note. (a) Predicted probabilities were computed using Eq. 6 in the text by using estimated coefficients and the observed values of variables in the sample. Probability differences in column (3) signify Pr (y = j | Xb, x = 1) – Pr (y = j | Xb, x = 0), where y = j (j = c, t, v, uv). X is the independent variable vector, x represents the binary variable of age group, and b is the estimated coefficients vector of the variables in the model (b) Probability differences in column (3) signify Pr (y = j | Xb, x = 1) – Pr (y = j | Xb, x = 0), where y = j (j = c, t, v, uv). X is the independent variable vector, x represents the binary variables of health status measured by self-rated health (bad or not) and having ADL or IADL, respectively, and b is the estimated coefficients vector of the variables in the model 123 454 J Fam Econ Iss (2006) 27:437–457 from those with good self-rated health (column (1)). Among those with fair or poor health, 57.0% chose cars, 14.0% trucks, 8.8% minivans, and 20.2% SUVs, a decrease of 3.9% in the probability of choosing cars, a decrease of 4.0% for trucks, and a decrease of .9% for minivans, but an 8.9% increase in the probability of choosing SUVs (column (3)). Similarly, existing ADL or IADL problems increase the probabilities of choosing cars (6.5%) and SUVs (5.2%) but decrease the likelihood of selecting trucks (8.4%) and minivans (3.1%). Discussion, Conclusion, and Implications As they age, older consumers experience declining driving abilities caused by weakened vision, physical disability, or other manifestations of declining health. While coping with such unfavorable changes, they may develop particular preferences toward vehicle types. This study empirically examines consumers’ (those aged 50 or older) selection of vehicle type along with the impacts of age and health status. First, we find that the 70 and older age group prefers passenger cars to pickup trucks and SUVs compared to the 50–59 age group, indicating that age is an important determinant of consumer demand for different types of vehicles. This finding is consistent with Berry et al.’s findings (1998) that age is negatively associated with the purchase of SUVs and pickup trucks. On the other hand, studying the effect of age cohorts on vehicle preference, Kavalec (1999) found that the demand for minivans peaks at age 40. Given the fact that our sample does not include anyone younger than 50, we cannot make a direct comparison with Kavalec’s study. However, any comparison with his study should be done with the recognition of the differences in sample and preference decisions. Kavalec’s sample consisted of California residents aged 16 to 88 who lived in the major urban areas, and his study was based on the respondents’ preferences on hypothetical purchase decisions rather than on actual purchase data like ours. Second, we find that existing health problems significantly influence older consumers’ choice of vehicle-type. Those who self-rated their health as bad, or had ADL or IADL problems at baseline, were more likely to buy SUVs. One possible explanation for this result is that SUVs provide various favorable features for older consumers with health problems, such as increased visibility, ease of entry and exit, comfort, or larger space. The results indicate that an elder’s health status influences his/her preference in the selection of specific vehicle types. Unlike existing health problems, the impacts of declining self-rated health and new incidence of ADL or IADL difficulties on choice of vehicle type are not significant. These results can be explained by a reaction lag concerning a change in health conditions. Lingering existing health problems may make older consumers prefer vehicles more suited to their health conditions. However, in coping with the occurrence of new health problems, it may take time for them to realize their new needs and react accordingly. We draw the following implications from our findings. First, for elders who start experiencing difficulties in driving and their families, we recommend that they consider switching to a different type of vehicle that provides more comfort and convenience, such as an SUV or a minivan, before giving up driving. Given that these vehicles are more expensive than passenger cars, tax incentives could be considered for these elders, especially with their difficulties in functional health. As we discussed 123 J Fam Econ Iss (2006) 27:437–457 455 earlier, prolonging elders’ independence by enabling them to drive safely enhances their quality of life and reduces the burden of care on their families and society. Second, we further recommend continual improvements in vehicle design, which would help elders to drive. Our findings indicate that elders with health problems prefer specific types of vehicles, such as SUVs. Built-in new technologies for greater safety while driving is not effective for elders due to adaptation issues. Some auto manufacturers have tried developing a car for older drivers, but it has stayed at the conceptual stage. Although the designers considered introducing items such as swivel seats to facilitate ingress and egress and improved lighting, it is likely such a car would be a commercial failure. Thus, policy makers and industry should pay more attention to the development or modification of existing vehicle types of minivans or SUVs for the general population, as well as elderly people. This study contributes to a better understanding of older consumers’ demands for different types of vehicles. To the best of our knowledge, these findings are the first of their kind. However, the results of this study should be generalized within the following limitations. First, the data employed are cross-sectional vehicle purchase data, capturing purchases made during a 12-month period. Elders’ preferences for different types of vehicles may change as they age and the condition of their health changes over time. The temporary popularity of specific vehicle types may also influence choice of vehicle. Without longitudinal data, we could not study whether or how such change occurs over time. Second, this study focuses on the impact of age and health on choice of vehicle type, with limited attention to the specific attributes of the vehicles. Although we are able to control for the impact of purchase price and the availability of other vehicles, the data set is limited in the sense that no other attribute data are available. Using a hedonic price approach and data on specific features, future researchers could estimate willingness to pay for specific features, such as visibility or exterior style, and compare the willingness to pay of those with or without health problems. Finally, even though we investigate the effect of age, we are not able to examine the cohort effect, given the nature of the cross-sectional data set of the 2001 Consumption and Activities Mail Survey (CAMS). Because baby boomers show many differences from the previous generation, an investigation of potential cohort effect would also be worthwhile. References American Association of Retired Persons. (2002). Transportation: The older person’s interest (FS 44R). Washington, DC. Amemiya, T. (1981). Qualitative response models: A survey. Journal of Econometric Literature, 19, 1483–1536. Berry, S., Levinsohn, J., & Pakes, A. (1998). Differentiated products demand systems from a combination of micro and macro data: The new car market. (Working Paper 6481). Cambridge, MA: National Bureau of Economic Research. Blocker, W. (1992). Maintaining functional independence by mobilizing the aged. Geriatrics, 47, 42– 56. Bunch, D. S. (2000). Vehicle demand and type choice. In D. A. Hensher, & K. J. Button, (Eds.), Handbook of transport modeling (pp.463–479). Amsterdam, Netherlands: Pergamon Press. Burkhardt, J. E., Berger, A. M., Creedon, M., & McGavock, A. T. (1998). Mobility and Independence: Changes and challenges for older drivers. Bethesda, MD: Econometrics, Inc. 123 456 J Fam Econ Iss (2006) 27:437–457 Coughlin, J. (2001). Transportation and older persons: Perception and preferences, a report on focus groups. Washington, DC: American Association of Retired Persons. Ehrenman, G. (2003). New wheels for grandma. Mechanical Engineering Magazine Online. Retrieved April 4, 2004, from http://www.memagazine.org/backissues/apr03/features/grandma/ grandma.html Elliott, M. (2003). Work and family role strain among university employees. Journal of Family and Economic Issues, 24, 157–181. Ferraro, K. F. (1980). Self-ratings of health among the old and old–old. Journal of Health and Social Behavior, 21, 377–383. Foley, D. J., Heimovitz, H. K., Guralnik, J. M., & Brock, D. B. (2002). Driving life expectancy of persons aged 70 years and older in the United States. American Journal of Public Health, 92, 1284–1289. Glover, L. (2000). Driving under the influence: The nature of selling sport utility vehicles. Bulletin of Science, Technology and Society, 20, 360–365. Goldberg, P. K. (1995). Production differentiation and oligopoly in international markets: The case of the U.S. vehicle industry. Econometrica, 63, 891–951. Guccione, A., Felson, D., Anderson, J., Anthony, J., Zgang, Y., & Wilson, P., et al. (1994). The effects of specific medical conditions on the functional limitations of elders in the Framingham study. American Journal of Public Health, 84, 351–358. Hausman, J., & McFadden, D. (1984). Specification tests for the multinomial logit model. Econometrica, 52, 1219–1240. Heeringa, S. G., & Connor, J. H. (1995). Technical description of the health and retirement survey sample design. Ann Arbor, MI: The Institute for Social Research, University of Michigan. Hoeymans, N., Feskens, E., Kromhout, D., & Van Den Bos, G. (1997). Ageing and the relationship between functional status and self-rated health in elderly men. Social Science & Medicine, 45, 1527–1536. Hurd, M. D., & McGarry, K. (1995). Evaluation of the subjective probabilities of survival. Journal of Human Resources, 30, S268–S292. Kavalec, C. (1999). Vehicle choice in an aging population: Some insights from a stated preference survey for California. The Energy Journal, 20, 123–138. Kelsey, J., O’Brien, L., Grisso, J., & Hoffman, S. (1989). Issues in carrying out epidemiological research in the elderly. American Journal of Epidemiology, 130, 857–866. Kline, D. W., Kline, T. J. B., Fozard, J. L., Kosmick, W., Schieber, F., & Sekuler, R. (1992). Vision, aging, and driving: The problems of older drivers. Journal of Gerontology: Psychological Sciences, 47, 27–34. Kockelman, K. M., & Zhao, Y. (2000). Behavioral distinctions: The use of light-duty trucks and passenger cars. Journal of Transportation and Statistics, 3, 47–60. Kolodinsky, J., & Shirey, L. (2000). The impact of living with an elder parent on adult daughter’s labor supply and hours of work. Journal of Family and Economic Issues, 21, 149–175. Krebs, M. (2000). Shifting gears. American Demographics, January 52–55. Lave, C., & Train, K. (1979). A disaggregate model of vehicle type choice. Transportation Research, 13A, 1–9. Long, J. S. (1997). Regression models for categorical and limited dependent variables. Thousand Oaks, CA: SAGE Publications. Malfetti, J. L., & Winter, D. (1986). Drivers 55 plus: Test your own performance. Washington, DC: AAA Foundation for Traffic Safety. Mannering, F., & Winston, C. (1985). A dynamic empirical analysis of household vehicle ownership and utilization. Rand Journal of Economics, 16, 215–36. Manski, C., & Sherman, L. (1980). An empirical analysis of household choice among motor vehicles. Transportation Research A, 14A, 349–366. Marottoli, R. A., Mendes de Leon, C. F., Glass, T. A., Williams, C., Cooney Jr. L. M., & Berkman, L. F., et al. (1997). Driving cessation and increased depressive symptoms: Prospective evidence from the New Haven EPESE. Journal of the American Geriatric Society, 45, 202–206. Marottoli, R. A., Mendes de Leon, C. F., Glass, T. A., Williams, C., Cooney Jr. L. M., & Berkman, L. F. (2000). Consequences of driving cessation: Decreased out-of-home activity levels. Journal of Gerontology: Social Sciences, 55B, S334–S340. McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. In P. Zarembka, (Ed.), Frontier of econometrics (pp.105–145). New York: Academic Press. 123 J Fam Econ Iss (2006) 27:437–457 457 McNeil, J. M. (1999). Preliminary estimates on caregiving from wave 7 of the 1996 Survey of Income and Program Participation. Retrieved April 7, 2004, from http://www.census.gov/dusd/MAB/ wp231.pdf Meyer, J. (2004). Personal vehicle transportation. In R. L. Pew, & S. B. Van Hemel, (Eds.), Technology for adaptive aging. (pp. 253–281). Washington, DC: The National Academies Press. Meyer, J., & Coughlin, J. F. (2001). Older drivers and new in-vehicle devices. Public Policy and Aging Bulletin, 11, 11–14. National Alliance for Caregiving and American Association of Retired Persons. (2004). Caregiving in the U.S. Retrieved April 7, 2004, from http://research.aarp.org/il/us_caregiving.pdf Nichols, L. S., & Junk, V. W. (1997). The sandwich generation: Dependency, proximity and task assistance needs of parents. Journal of Family and Economic Issues, 18, 299–326. Owsley, C, Sekuler, R., & Siernsen, D. (1983). Contrast sensitivity throughout adulthood. Vision Research, 23, 689–699. Petrin, A. (2002). Quantifying the benefits of new products: The case of the minivan. Journal of Political Economy, 110, 705–729. Pickrell, D., & Schimek, P. (1999). Growth in motor vehicle ownership and use: Evidence from the Nationwide Personal Transportation Survey. Journal of Transportation and Statistics, 2, 1–17. Rodgers, W. L., Ofstedal, M. B., & Herzog, A. R. (2003). Trends in scores on tests of cognitive ability in the elderly U.S. population, 1993–2000. Journal of Gerontology Social Sciences, 58B, S338– S346. Shaheen, S., & Niemeier, D. (2001). Integrating vehicle design and human factors: Minimizing elderly driving constraints. Transportation Research Part C, 9, 155–174. Spector, W. D., & Fleishman, J. A. (1998). Combining activities of daily living with instrumental activities of daily living to measure functional disability. Journal of Gerontology Social Sciences, 53B, S46–S57. Stelmach, G. E., & Nahom, A. (1992). Cognitive-motor abilities of the elderly driver. Human Factors, 34, 53–65. Straight, A., & Jackson, A. M. (1999). Older drivers. Washington, DC: American Association of Retired Persons. Stutts, J. C., Wilkins, J. W., Reinfurt, D. W., Rodgman, E. A., & Van Heusen-Causey, S. (2001). The premature reduction and cessation of driving by older men and women. Chapel Hill, NC: University of North Carolina. Retrieved April 7, 2004, from http://www.hsrc.unc.edu/pdf/2001/ G7Report.pdf Train, K. (1990). Qualitative choice analysis: Theory, econometrics, and an application to vehicle demand. Cambridge, MA: MIT Press. University of Michigan. (2005). An overview of Health and Retirement Study components. Retrieved November 11, 2005, from http://hrsonline.isr.umich.edu/intro/sho_uinfo.php?hfyle=overview&xtyp=2 U.S. Census Bureau. (1997). The vehicle inventory and use survey. Retrieved April 7, 2004, from http://www.census.gov/prod/ec97/97tv-us.pdf U.S. Department of Energy. (2003). Light truck sales by type, 1976–2002. Retrieved April 4, 2004, from http://www.ott.doe.gov/facts/archives/fotw281supp.shtml U.S. Department of Transportation. (1999). Remarks as prepared for delivery keynote address by the Secretary of Transportation Rodney E. Slater. Retrieved April 4, 2004, from http://www.dot.gov/ affairs/1999/110799sp.htm U.S. Department of Transportation. (2003, November). Safe mobility for a maturing society: Challenges and opportunities. Washington, DC: Author. U.S. Senate Special Committee on Aging, (2003). Keeping America’s seniors moving: Examining ways to improve senior transportation (Serial No. 108-15). Washington, DC: U.S. Government Printing Office. Wallace, R. B., & Herzog, A. R. (1995). Overview of the health measures in the Health and Retirement Study. The Journal of Human Resources, 30, S84–S107. Willoughby, B. (2003). SUVs: A work in progress. Consumers’ Research Magazine, January, 26–29. Wist, E. R., Schrauf, M., & Ehrenstein, W. H. (2000). Dynamic vision based on motion contrast: Changes with age in adults. Experimental Brain Research, 134, 295–300. 123