V//särtryck
143 1989
Car Demand Modelling And Forecasting A New Approach
Jan Owen Jansson
Reprint from Journal of Transport Economics and Policy, May 1989, pp 125 - 140
dfv
Väg-06]! Trafik- Statens väg- och trafikinstitut (VTI) * 581 01 Linköping
, [IISt,t 18t Swedish Road and Traffic Research Institute * S-581 01 Linköping Sweden -s --.
CAR DEMAND MODELLING AND
FORECASTING
A New Approach
By Jan Owen Jansson*
INTRODUCTION
A forecasting model of car ownership has been developed at the Swedish Road and Traffic Research Institute for the Swedish National Road Administration. A new approach was chosen: its chief characteristics are that car ownership entry propensity (with its counterpart, exit propensity ) is introduced as the primary dependent variable, and that cross-section analysis and longitudinal cohort analysis have been combined to make maximum use of empirical evidence. The construction of the model, the estimation of its parameters, and the fore-cast are presented in Swedish in Jansson et al. (1986). This paper is an English summary of the points of most general interest.
RELATION TO PREVIOUS STUDIES
The work started with a separately published literature survey in English (Jansson and Shneerson, 1983). This identified three main types of models:
(1) Models for predicting the long-term development of car ownership, based on the notion of epidemic diffusion . The most refined version of these growth curve models has been worked out at the UK Transport and Road Research Laboratory by John Tanner (1974, 1977, 1978, 1979). (2) The so-called stock-adjustment models, which set out to explain the short- to medium-run fluctuations in purchases of new cars. These uctu-ations have a substantial impact on the level of activity of a motorised society. American pioneers in this field were Chow (1957 and 1961) and Suits (1958 and 1961); later prominent British works are those of Smith (1975) and Mogridge (1983).
(3) Cross-sectional models, based on comparisons of car ownership levels in different regions, cities and parts of cities. Early works in this mould are Tanner (1963), Beesley and Kain (1964) and Kain and Beesley (1965): later came the comprehensive, comparative study of car ownership in US metro-politan areas by Kain, Fauth and Zax (1977). The modelling technique of
* Swedish Road and Traffic Research Institute (VTI), Linköping.
the cross-sectional approach was in the 19703 inspired by discrete choice analysis. A number of car ownership models were estimated on disaggregate (household) data, often as a natural extension of the modelling of modal choice. Some of the best known contributions are: Lerman and Ben-Akiva (1976), Burns, Golob and Nicolaidis (1976), Bates, Gunn and Roberts (1978a), Train (1980), and Ben-Akiva, Manski and Sherman (1980).
This disaggregate approach has given us a deeper knowledge of the socio-economic factors of importance for household car ownership. It has also made more tractable the influence of local conditions such as the supply of public transport and parking. The in uential Leitch report took a firm stand in the debate on car ownership forecasting by recommending that The Department should as soon as practicable move away from the extrapolatory form of model currently used [growth curve models of type (1)] towards basing its forecasts on causal models (Adivsory Committee on Trunk Road Assessment, 1978, Paragraph 19.20). By causal models we are to understand models estimated on disaggregate household data.
The preference expressed in the Leitch report has also been that of the research community in the 1980s. However, so long as we rely only on cross-sectional data it is impossible to identify the dynamics of car demand, including the diffusion process and some important factors such as price elasticity which are important for long-term development. It was therefore decided that for the Swedish model none of the existing approaches was completely satisfactory; the model was instead calibrated on longitudinal cohort data, each cohort consisting of all males or all females born in a particular year. By studying the car ownership develop-ment path over the life-cycle of the individuals of a particular generation, on the one hand, and by comparing the development paths of different generations on the other, we hoped to come to grips with the true dynamics of long-term growth in car ownership. In addition we had the idea that, rather than focusing on the net result of entries into and exits from car ownership of each cohort, it would be rewarding to build separate models for entries and exits, since these are two completely independent occurrences. The basic philosophy was set out in Jansson (1982). In the next two sections the choice of approach will be explained more specifically.
To conclude this account of the relation of our approach to previous studies and other work in progress, we mention a number of articles and reports which we have been made aware of after the completion of our own work. Firstly, a somewhat similar attitude to ours towards the dynamics of car ownership was expressed in Goodwin and Mogridge (1979). Secondly, cohort analysis of car ownership and travel behaviour based on panel data has apparently become a main interest of travel demand analysis in the eighties, as witnessed by Goodwin (1986, 1988), Golob (1987), Hensher et al. (1987), Kitamura (1987a, 1987b), and Madre (1988). Thirdly, the use of panel data is, of course, not a necessary condition for cohort analysis. Van den Broecke (1987) and Kawashima (1988) have conducted demographic cohort analysis of car registrations and driving licenses on the basis of total population forecasting. The Swedish model is based on statistics of total population combined with car register and inland revenue
CAR DEMAND MODELLING AND FORECASTING J. O. Jansson
records. Under the Swedish system of personal number identification, it has been possible to trace both the income and the car ownership history of all the birth-year cohorts in the ten birth-year period 1975 84.
Before we develop the model we must explain the problems of data on car registrations in Sweden.
PROBLEMS OF DATA ON SWEDISH CAR REGISTRATIONS
The chart in Figure 1 looks like a typical three-scenario forecast. It is not. The highest and the lowest curves correspond to historical figures from the Swedish car register, representing two different measurements of the total car eet. In recent years the small gap between these two alternative measurements total number of cars on register, and total number of cars registered as being in active use has suddenly widened, and the two curves have taken rather different courses.
The work reported in this paper started in 1983. At that time Sweden had experienced a relatively long period of a seemingly stagnant rate of car owner-ship, after an actual decrease in 1977. This is seen in Figure 1, and in greater detail for recent years in Table ]. It was generally believed that it could be explained by the slight fall in household disposable income and rise in fuel prices that had happened in the same period. It was not as simple as that, as we soon fOund. To be sure, official statistics as conventionally interpreted at that time showed stagnation in Swedish car ownership from 1977 to 1982. But a new phenomenon had appeared in the car register, which cast some doubt on the official figures, or was at least a disturbing factor. The stock of cars registered as temporarily not in use (passive cars) made a sudden leap in 1977, and from then on developed at an unprecedented rate of growth. In 1976 the proportion of passive cars (in the Swedish car register) to cars registered as being in active use was 7: 100, and ten years later it was 21 : 100.
Our first task was to examine what the registered stock of passive cars really consisted of (Cardebring and Jansson, 1986). In briefest summary, the findings were that roughly two thirds of the passive cars were, in reality, scrapped, and that this stock of scrapped cars was steadily accumulating. For our purpose, however, the main problem was that a substantial proportion of the one third of the total passive stock, which could be expected to be put into use again sooner or later, would previously have been registered as being in active use. The main categories involved were (1) used cars in the course of being traded, and (2) cars out of order, currently undergoing repair or intended to be put in order again. In reality the proportions of cars in these categories remained unchanged during the 1970s. But in the car register they have changed from active to passive status, mainly because in the middle of the 1970s it became possible to obtain a vehicle tax and/or insurance refund without a fee when the status of the car was changed in the register.
In order to compensate for this irrelevant irregulariy in the time-series of car ownership, we adopted a modified definition of the total car fleet: the true eet includes all cars in active use (according to the car register) and also those 127
Car ) Total Numbers + ,'I I q I A", [I ,! ,. True I v . II, A Actlve I
I" '~A--A-A-A H_A.A
2.700-1
2.200 - Key:
Total = Total cars on register
Active = Total cars registered as in active use
__ True = True number of cars
A,
'646 '68 '70 '72 '74 '76 '78 '80 '82 '84 year
FIGURE 1
Total Car Fleet Development in Sweden
passive cars that can be expected to be put into active use some time in the future. As is shown in Figure 1 and Table 1, this gives a fairly smooth develop-ment a continued increase of Swedish car ownership throughout the period of observation.
This somewhat tedious discussion of a specifically Swedish problem is justified by the result. There seems to be a widespread belief abroad that Sweden is one of the few countries to have experienced a fall in the rate of car ownership during the lengthy period of economic stagnation after the oil crisis. It is important to rectify this misinterpretation. The sharp trend-break shown in official statistics is an unfortunate result of certain changes in the car register, which have no relation to reality.
CAR DEMAND MODELLING AND FORECASTING J . O. Jansson
TABLE l Cars/I 000 Persons
(1)
(2)
(3)
(4)
_
Year Registered as Registered as Estimated number True car ownership in active use temporarily of (2) which are = (1 ) + (3)
not in use expected to
become active
1976
350
26
8
358
1977
346
41
16
362
1978
345
49
19
364
1979
345
54
20
365
1980
347
_
59
22
.369
1981
348
63
23
371
1982
353
68
25
378
1983
361
70
24
385
1984
369
76
23
392
1985
377
80
25
402
1986
388
82
26
414
WHY A NEW APPROACH?
The modification of the statistics of Swedish car ownership in recent years had the effect of falsifying the simple and straightforward economic model of car ownership. Household disposable incomes had actually been decreasing for the first time since the war, and the long downward trend in fuel prices had abruptly turned upwards, and yet the rate of car ownership had continued to grow. Some other powerful in uence must have been at work to offset the negative influences of income and price of fuel.
The real price of cars had stayed roughly constant in the period in question (1977 82). Better roads could hardly be the explanation; road building in Sweden had reached something of a saturation level. The bulk of expenditure on roads was for maintenance and repair. Prices of substitute modes of transport had gone down rather spectacularly at the beginning of this period. In 1978 the domestic airlines introduced substantial off-peak rebates, and in 1979 the Swedish Railways followed suit.
Almost the only remaining possible explanation seemed to be that the so called diffusion process was still at work. One should interpret diffusion in this context as a change in the order of preferences in favour of cars, or simply a rise in the taste for cars. Traditional car ownership forecasting models were often completely focused on this factor, as time was made the sole explanatory variable in the fitting of logistic growth curves to car. ownership. This was quite in line with the general idea of an introduction phase for every new product, in which potential
buyers learn about the novelty and acquire a taste for it. In the present case one would have thought that the introduction phase for cars was well over in the mid-19705 in Sweden. However, other empirical evidence also indicated that the diffusion process was still going on in the 1970s. The Engel curves of Figure 2 show that at a given real income level the precentage of car-owning households was substantially higher in 1978 than in 1970 (let alone in 1954). This should be a sign of a positive diffusion effect if we bear in mind that, as stated above, most other possible influences were counteracting. For our forecasting purpose it was imperative to determine whether this force the diffusion effect could be expected to remain potent for very long in the future, or whether it could be expected to peter out fairly soon. To that end we needed to take a closer look at the phenomenon of diffusion.
THE DIFFUSION PROCESS
A well-established notion about the diffusion process is that it consists of two stages vertical diffusion and horizontal diffusion. (See, for example, Bonus, 1973.) If we add to the three observed Engel curves in Figure 2 a fourth, hypo-thetical, one, representing the assumed position at some time in the very beginning of motorisation, the meaning of vertical and horizontal diffusion is made clear. When a new, expensive product appears on the market, only the rich trend-setters venture to acquire it: the right-hand portion of the Engel curve shifts upwards. Sooner or later lower income groups follow the example. With an analogy from mechanics, horizontal diffusion is represented by a force pulling the hypothetical Engel curve in Figure 2 horizontally to the left. If the product in question becomes as popular as the car, the end result will be an Engel curve like the one for 1978, taking the diametrically opposite shape to the first one: the product has changed character from an extreme luxury to a pronounced necessity.
This is a neat way of looking at the diffusion process. Nevertheless we found it wanting, at least for the recent past years for which data are available. It focuses too much on income, whereas we found age to be the strategic factor. Income and age are strongly correlated up to the age of retirement; this confuses matters, or at least complicates them. '
Let us change the focus from household car ownership to entries to and exits from car ownership by individuals. To consider car demand on an individual basis (rather than treating the car as a household good) was both a necessity, given the personal nature of data on car registrations, and a virtue, in view of our findings from a number of Swedish cross-section data sets that household characteristics do not contribute significantly to explaining the number of cars per adult in different households. Age, sex, urban/suburban/rural location and income per adult turn out to be the only significant explanatory variables.
Who are first-time buyers of cars (rather than buyers of cars for replacement), and who are getting rid of their cars? In Figure 3 the continuous line gives the total number of persons of every age from 18 to 88 years en tering car ownership in a particular year, according to the car registers. This shows that the very
CAR DEMAND MODELLING AND FORECASTING J. O. Jansson
% A 100 90 " 1978 80 Hr
70 '"
1970
60 Hypothetical pmwm 50 curve 40 -... 30 __ 1954 20 _-10 __ Household income 1000 Sw/Cr I l 1 | l | | l x T l l l l | ' T I 7 10 20 30 40 50 60 70 80 90 100 FIGURE 2Percentage of Car Owning Households at Different Incomes {1975 Price Level) An Illustration of the Working of Diffusion in Bonus s Sense
youngest legally entitled motorists are strongly dominating. The number of persons of different ages leaving car ownership in each year is given by the broken line. It should be noted that neither the exits nor the entries are necessarily of a permanent nature. Most likely many of the younger persons leaving car ownership in a particular year will re-enter in a later year. We have not been able to check that. Our data thus contain an unknown mixture of short-term and temporary movements in and out of the car register with more permanent entries and exits.
NUMBER OF PERSONS PER YEAR
25000-zzsooa
El20000-
\
m..- g
15000-
l
l
12500-
l
100004
\
7500-
\
5000
2500-AGE
FIGURE 3Number of Entries and Exits to/from Car Ownership for Different Ages in 1984
Nevertheless, the difference between the total number of entries and exits in any year gives accurately the overall change in that year in the total car eet owned by physical persons.
Looking at the two curves in Figure 3 together, we see clearly which age groups make positive and negative contributions, respectively, to the change in the car eet: only pe0ple up to about 40 years of age contribute positively. In middle life (40 65) the total net effect is virtually zero; in old age, of course, the net effect is negative. As long as the positive net effect of young people is greater than the
CAR DEMAND MODELLING AND FORECASTING J. O. Jansson
negative net effect of old people, total car ownership will be on the increase. At saturation the two effects are exactly offsetting.
Car ownership entry and exit propensities
The main reason why young people constitute a majority of total new entrants (into car ownership) is trivial enough: they do not have cars to the same extent as middle-aged people. A more interesting relationship is that of new entrants to the total number of have-nots in each age group. The ratio of the number of
persons entering car ownership during a particular year to the number of carless
persons at the end of the pervious year is called the entry propensity; and its counterpart, the ratio of the number of persons leaving car ownership during a particular year to the number of car owners at the end of the previous year, is called the exit propensity. The most rewarding use of these concepts is to relate them to age. Figure 4 gives entry and exit propensities for men and women separately, for every age from 18 and upwards. We have calculated these pro-pensities, relating them to age, for ten different years.
The curves in Figure 4 show ten different relationships between a particular propensity and age, observed for each of the years in the ten-year period 1975 84. The shapes of both relationships entry propensity with respect to age, and exit propensity with respect to age are very characteristic and comparatively stable; some shifts occurred in the period of observation, but the general shapes remained the same.
As expected, the entry propensity for men is considerably higher than for women at all ages. -It was more surprising that the female exit propensity would be so consistently higher than the male. For both sexes it is quite clear that young people have the greatest propensity to change status with regard to car owner-ship in both directions. In middle life consumption habits become increasingly steady. In old age involuntary exit from car ownership is, needless to say, a salient feature.
We regard the entry and exit propensities as the primary dependent variables instead of the car ownership level, because the latter is derived from these pro-pensities and the continuous turnover of population. This means that the car ownership level can go on growing for a very long time after the entry and/or exit propensities have ceased to change. It is here that cohort analysis is a helpful tool. To illustrate its use, a simulation has been made. We took the observed situation of income, prices, car ownership, the entry and exit propensities with respect to age, and the birth and death rates of the year 1975 as the point of departure. On the assumption that the propensities and the birth and death rates stay constant for ever, we can trace the development of car ownership with respect to age from 1975 to the final equilibrium, which in this example would occur in the middle of the next century. As is shown in Figure 5, car ownership of middle aged and old persons would rise substantially in this rather long transitory stage. Because the general level of car ownership was already relatively high in Sweden in 1975, most of the rise occurs before 1990. The increase occurring in the last 50 years (2000 2050) is quite small. If a corresponding simulation had been made with 1950 or
1960 as the point of departure, the population turnover effect causing horizontal diffusion would have appeared even greater.
O_35_
ENTRY PROPENSITY 1975 84
0.25- 0.20-PE RC ENT 0.15 0.10-0.05 0-00 I I I l l IO 20 30 40 50 50 70 80 90 FIGURE 4aCar Ownership Entry Propensities by Age for Men and Women
Vertical and horizontal diffusion
A more relevant division of the diffusion process for cars than the one illustrated in Figure 2 is to let vertical diffusion be represented by an upward shift in the entry propensity (and/or downward shift in the exit propensity) unspurred by any change in income, price, etc., while horizontal diffusion is represented by the subsequent increase in car ownership due to the population turnover effect, which continues long after the entry and exit propensities have ceased to change. Vertical diffusion then corresponds to a genuine change in taste: individuals of a given age have a higher propensity to enter car ownership than individuals of preceding generations had at the same age. When the entry and exit propensities cease to change and (in the hypothetical absence of economic forces to the contrary) stay constant, it does not follow that individuals of different gener-ations from then on will show the same level of car ownership at a particular age.
CAR DEMAND MODELLING AND FORECASTING J. O. Jansson 0.30q
EXIT PROPENSITY 1975 84
0.25 0.20 |... Z LJ O 0.151 0: DJ l 0.10 0.05 0-00 l 14 l T I 1 r I 10 20 30 40 50 50 70 80 90 ACE FIGURE 4bCar Ownership Exit Propensities by Age for Men and Women
History matters. Because before the vertical diffusion process had petered out older generations had a lower entry propensity at any given age than younger generations, car ownership levels (at a given age) will differ between generations even after the entry and exit propensity curves (like those of Figure 4) have ceased to shift. The horizontal diffusion process is not over till all generations older than 17 at the time of the hypothetical state of constancy of the entry and exit propensities have died off. That is why it takes about 75 years, in the simul-ation illustrated in Figure 5, before the rate of car ownership stops growing after the entry and exit propensities have stopped moving.
The effect of horizontal diffusion can also be called the population turnover effect as a pointer to what is basically involved. It is a quantitively important effect, and what is nice for forecasters about the population turnover effect is, of course, that it is predictable by cohort analysis.
The main lesson
The paradox presented at the outset continuously rising car ownership in the period of economic stagnation after the oil crisis was solved when entry pro-pensity was substituted for car ownership level as the dependent variable in the econometric analysis. We found that the entry propensity had actually fallen in the period 1975 82 in response, we concluded, to slightly decreasing disposable incomes and rising fuel prices. The fall in entry propensity was not great enough to offset the positive effect on the car ownership level of horizontal diffusion. This experience was an important lesson to us. We conclude that econometric work on car ownership forecasting runs the risk of seriously biased estimates unless we have a true understanding of the car diffusion process. Perhaps this may not apply in the USA, but it certainly does in other countries where motorisation is largely a phenomenon of the period since the second world war.
MALE AND FEMALE CAR OWNERSHIP
The use of entry and exit propensities (rather than household car ownership) as the primary dependent variables is easier if we view the car as an individual rather than a household good. Household entry and exit would be considerably more cumbersome concepts, in view of the increasing instability of the household itself. An essential ingredient in our approach is to relate the entry and exit pro-pensities to age, and to study how these relationships have behaved for different generations. It is thus the average behaviour of all individuals of particular birth-year cohorts that is the object of study. Such far-reaching averaging blunts the analysis. We made the most obvious further division of cohorts by separating men from women. Still further disaggregation would be theoretically desirable, but practically very expensive in data collection.
The male/female division turned out to be very rewarding, but also quite problematic. Who owns the car in a single-car household including two adults? Note here in passing that this is a different problem from the aforementioned issue of whether household characteristics contribute significantly to explaining the number of cars per adult in various households. The question raised here does not require an answer when we are examining the former issue, but it does, of course, when separate rates of car ownership are to be determined for male and female birth-year cohorts. We defined car owners strictly by reference to the car register, in spite of the fact that we know that the registered owner is not always the main user of the car. We took a sample from the car register of couples (with or without children) owning just one car, and used a questionnaire to determine to what extent the registered owner and the main driver of the car differed. We found that the man was the registered owner in 80 per cent of the cases but the main driver in just 72 per cent. For longitudinal male and female cohort analysis this discrepancy creates a real problem only if it changes in magnitude over time - for example, because new rules for car insurance or taxation may make it profitable to switch ownership of the car in the household. To guard against this source of bias, we checked whether there was any correlation between female entries and male exits, and vice versa, in age groups where cohabitants are to be
CAR DEMAND MODELLING AND FORECASTING J. O. Jansson Car Ownership
(Cars per person at each age)
0.6 '" __
/ N \
0.4 "
Legend: \
I Car ownership rate 1975 0.2 "'
D Car ownership rate 1982
. Car ownership rate 1990
0 Car ownership rate 2000
A Car ownership rate 2050
0-0
|
|
1
l
l
0 20 40 AGE 60 80 100
FIGURE 5
Simulation of the Population Turnover Effect Causing Horizontal Diffusion
found. No correlation came to light, so we decided finally that separate male and female cohorts should be formed in the analysis.
The interesting result of this admittedly somewhat problematic separation was that female car ownership appeared as the strategic factor for the future develop-ment of motorisation.
In the main economic scenario constant incomes for men and 2 per cent growth in incomes for women the separate forecasts for male and female car ownership, when combined, predict an increase in the total car eet for the period 1985 2000 of 16 per cent, and for the 25 years 1985 2010 the equi-valent figure is 23 per cent; but, as is shown in Table 2, male car ownership is forecast to grow only by 3 per cent to the year 2010, while female car owner-ship is forecast to grow by 70 per cent.
TABLE 2
Car Ownership Forecast for Sweden
Year Total Cars per Cars per Cars per
number of 100 persons ] 00 men 100 women
cars (aged > 1 7) (aged > 1 7)
millions
1984 3.32 40 71 27
2000 3.84 46 73 41
2010 4.10 48 73 46
CONCLUDING REMARKS
The approach to car ownership forecasting presented in this paper was chosen with the specific purpose of coming to grips with the diffusion process. The birth-year cohort analysis, in which entries and exits over the life-cycle of different generations are modelled separately, has turned out to be well suited for this purpose. The approach also seems to be well in line with the apparent redirection of general travel demand analysis towards a more dynamic view of car ownership and use than is afforded by established cross-sectional methods of analysis. However, no claims are made that the Swedish car ownership forecasting model is a complete, dynamic model. One perhaps the most important of the dynamic aspects has been captured, but others remain to be tackled. The modelling of car ownership entries and exits was made in a conventional static framework. We are developing these parts of the model. Some preliminary results are reported in Jansson (1988), but (for example) the problem of how the pent-up demand for cars in a period of economic stagnation will express itself in later periods has not been solved.
Date of receipt offinal typescript: April 1988
REFERENCES
Advisory Committee on Trunk Road Assessment (1978): Report to the Ministry of Transport (The Leitch Report). HMSO, London.
Bates, J. J., H. F. Gunn and M. Roberts (l978a): A Disaggregate Model of Household Car Ownership. Research Report 20, Departments of the Environment and Transport. HMSO, London.
Bates, J. J., H. F. Gunn and M. Roberts(1978b): A Model of Household Car Ownership . Part 1. Traf c Engineering and Control, vol. 19 (November).
CAR DEMAND MODELLING AND FORECASTING J. O. Jansson Bates, J. J., H. F. Gunn and M. Roberts (197 8c): A Model of Household Car Ownership . Part
2. Traffic Engineering and Control, vol. 19 (December).
Beesley, M. E. and J. F. Kain (1964): Urban Form, Car Ownership and Public Policy: An Appraisal of Traffic in Towns . Urban Studies, vol. 1, no. 2 (Nov.), 174 203.
Ben-Akiva, M. E., C. F. Manski and L. A. Sherman (1980): A Behavioural Approach to Modelling Household Motor Vehicle Ownership and Applications to Aggregate Policy Analysis. Paper to the World Conference on Transport Research, London.
Bonus, H. (1973): Quasi-Engel Curves Diffusion and the Ownership of Major Consumer Durables . Journal ofPolitical Economy, vol. 81, No. 3 (May/June), 655 677.
van den Broecke, A. (1987): De mogelijke groei van het personenautobezit tot 2010. Ministerie van Verkeer en Waterstaat, The Hague.
Burns, L. D., T. F. Golob and G. C. Nicolaidis (1976): A Theory of Urban Households Auto-mobile Ownership Decisions . Transportation Research Record, 569.
Button, K. J., A. D. Pearman and A. S. Fowkes (1982): Car Ownership Modelling and Fore-casting. Gower, Aldershot.
Cardebring, P., and J. O. Jansson (1986): Avstallda bilar och bilstatistiken VTI-meddelande 445. Swedish Road and Traffic Research Institute, Linköping.
Chow, G. C. (1957): Demand for Automobiles in the United States. North-Holland, Amsterdam. Chow, G. C. (1961): Statistical Demand Functions for Automobiles and Their Use for Fore-casting . In: A. C. Harberger (ed): The Demand for Durable Goods. University of Chicago Press.
Golob, T. F. (1987): Effects of Income and Car Ownership on Trip Generation: a Structural Equations Model and its Temporal Stability. Institute of Transportation Studies, Report No. 877. California University, Irvine.
Goodwin, P. B. (1986): A Panel Analysis of Changes in Car Ownership and Bus Use. Transport Studies Unit, Oxford University. TSU Ref. 360.
Goodwin, P. B. (1988): Circumstances in which People Reduce Car Ownership: a Comparative Analysis of Three Panel Data Sets. Transport Studies Unit, Oxford University. TSU Ref:
412. '
Goodwin, P. B., and M. Mogridge (1979): Hypotheses for a Fully Dynamic Model of Car Ownership. Transport Studies Unit, Oxford University. TSU Ref. 105.
Hensher, D. A. et al. (1987): Modelling the Dynamics of Car Ownership and Use: a Method-ological and Empirical Synthesis. Fifth International Conference on Travel Behaviour. La Baume les Aix, October.
Jansson, J. O. (1982): Towards a Model for Explaining Changes in Car Ownership. VTI working paper. Swedish Road and Traffic Research Institue, Linköping.
Jansson, J. O. (1988): Car Ownership Entry and Exit Propensities of Different Generations a Key Factor for the Development of the Total Car Fleet. Oxford Conference on Travel
and Transportation, July.
Jansson, J. O., P. Cardebring and O. Junghard (1986): Personbilsinnehavet i Sverige 1950 2010. VTI report 301. Swedish Road and Traffic Research Institute.
Jansson, J. O., and D. Shneerson (1983): Some Approaches to Car Demand Modelling. VTI report 251A, Swedish Road and Traffic Research Institute.
Kain, J. F. and M. E. Beesley (1965): Forecasting Car Ownership and Use . Urban Studies, vol. 2, no. 2 (Nov.), 163 185.
Kain, J. F., G. R. Fauth and J. Zax (1977): Forecasting Auto Ownership and Mode Choice for
US Metropolitan Areas. Department of City and Regional Planning, Harvard University.
Kawashima, T. (1988): Driver s Licence Ownership in Japan: Towards Cohort Analysis for Car Ownership Forecasting. Oxford Conference on Travel and Transportation, July.
Kitamura, R. (1987 a): A Panel Analysis of Household Car Ownership and Mobility . Proceed-ings of the Japan Society of Civil Engineers. Mp/383/IV-7 : 13 27.
Kitamura, R. (1987 b): Determinants of Household Car Ownership and Utilization: a Dynamic Analysis. Fifth International Conference on Travel Behaviour, La Baume les Aix, October. Lerman, S. R., and M. E. Ben-Akiva (1976): Disaggregate Behavioural Model of Automobile
Ownership . Transportation Research Record, 569.
Madre, J. L. (1988): Long-Term Forecasting of Car Ownership and Car Use. Oxford Conference on Travel and Transportation, July.
Mogridge, M. (1983): The Car Market. Pion, London.
Smith, R. P. (1975): Consumer Demand for Cars in the USA. University of Cambridge Depart-ment of Applied Economics. Occasional Paper 44. Cambridge University Press.
Suits, D. (1958): The Demand for Automobiles in the USA, 1929 56 . Review of Economics and Statistics, vol. 40.
Suits, D. (1961): Exploring Alternative Formulations of Automobile Demand . Review of
Economics and Statistics, vol. 43, no. 1 (Feb.), 66 69.
Tanner, J. C. (1963): Car and Motorcycle Ownership in the Counties of Great Britain in 1960 . Journal of the Royal Statistical Society (Series A).
Tanner, J. C. (1974): Forecasts of Vehicles and Traffic in Great Britain. Report LR 650. Trans-port and Road Research Laboratory.
Tanner, J. C. (1977): Car Ownership Trends and Forecasts. Report LR 799. Transport and Road Research Laboratory.
Tanner, J. C. (1978): Long-term Forecasting of Vehicle Ownership and Road Traffic . Journal of the Royal Statistical Society, 141A, 14 63.
Tanner, J. C. ( 1979): Choice ofModel Structure for Car Ownership Forecasting. Supplementary Report SR 523. Transport and Road Research Laboratory.
Train, K. (1980): A Structured Logit Model of Auto Ownership and Mode Choice . Review of Economic Studies, vol. 47(2), 357 370.
