Time use, mobility and expenditure: an innovative survey design for understanding individuals’ trade-off processes

The value of travel time has always been subject to extensive debate in both academia and politics. As savings in travel time are often the major justification for infrastructure investments (BMVBW 2003), numerous studies deal with travel demand models which estimate the willingness to pay for a reduction of travel time, the subjective value of travel time savings (SVTTS). In most cases travelling is not a pleasure in itself, but a necessity to reach a location to engage in more pleasurable and useful activities.

However, not only reductions in travel time, but also improvements of travelling comfort are crucial transport policy measures. In recent years the comfort of public transport vehicles has increased significantly. In-vehicle time has become more entertaining and convenient due to the use of mobile devices which allow us to use our time more productively. These developments let expect that the travel activity itself might be perceived more positively for public transport compared to the car (EC 2015).

These two aspects of travel time, the time loss due to the duration of travel and the valuation of travel as an activity itself are reflected in the two components of the SVTTS: (1) the willingness to substitute travel time for other activities (or value of time as a resource VOR), which is the marginal utility of an additional unit of leisure in the DeSerpa (1973) model and (2) the direct valuation of time assigned to travel (VoTAT), also called the value of time as a commodity (VOC). These two components are expected to vary in different ways, so they should both be known in order to fully understand the effect of transport projects (Jiang and Morikawa 2007). In order to shed light on the individual components, it is necessary to integrate travel decisions into the larger framework of time assignment and consumer’s home production (Munizaga et al. 2008).

A main challenge of a joint time assignment and mode choice model with a microeconomic foundation is the merging of the two different types of models: Travel demand models are typically discrete choice models, e.g. mode choice, route choice or destination choice models, which model the indirect utility of travel decisions (Jara-Díaz 1998) based on random utility theory. Time and budget assignment models are continuous models, which model the direct utility of time use and budget allocation according to a Cobb–Douglas or an additive logarithmic utility function. The combination of discrete and continuous choice modelling approaches is a field of ongoing research (Jara-Diaz and Guevara 2003; Bhat 2008; Habib 2013).

Another challenge is the large amount of information required to model the complex trade-off processes between travel activities and non-travel activities. A prominent example of jointly estimating time assignment and travel decisions is the model developed by Jara-Diaz et al. (2008). Using a Lagrange optimization, they derive four equations (first order conditions) to be modelled in relation to each other:where U denotes the utility; X_j and P_j the amount and price of consumed good j; T_l, T_w and \(T_{t}^{MIN}\) the amount of time assigned to leisure, work, and travel; w the wage rate; λ, μ and κ_t the Lagrange multipliers representing the marginal utility of increasing available money, increasing available time, and reducing the minimum time constraint of travel, respectively. Besides the usual socio-demographic and socio-economic variables, the following information is required for estimating such a model:

Given that the objective is to model detailed trade-off processes between travel decisions, time use and budget assignment, it seems to be important to gain all necessary information from the same individual simultaneously. Moreover, data is required to be collected for an observation period of sufficient length to consider the allocation pattern as a representation of the individual’s long term equilibrium. To the best of our knowledge, no dataset exists which meets all these requirements and there is no survey procedure available to collect these data at sufficient quality and quantity in a diary format. Some travel surveys cover long periods (e.g. Chalasani and Axhausen 2004), but information about non-travel-activities can only be roughly inferred from ‘trip purposes’ and no information about budget assignment is included. A possibility to retrieve all required information is the matching of data from independent time use and expenditure surveys (Jara-Diaz and Rosales-Salas 2015; Konduri et al. 2011), but this procedure yields only probabilistic rather than direct relationships between time and budget assignment. Castro et al. (2012) mentioned that the merits and appropriateness of such a synthetic data generation are debatable and further efforts on obtaining combined data on time-use and expenditure are desirable. In 2009, a new module on time use and consumption has been added to the Longitudinal Internet Studies for the Social Sciences (LISS) panel, administered by the CentERdata (Cherchye et al. 2012). Retrospective information on time use during the past 7 days preceding the interview and household consumption expenditure within the previous month (within the last year for large durables) is collected. Yet this data collection method does not yield information on travel modes of on the level of individual activity episodes and can lead to systematically biased mean values of time use (Browning and Gørtz 2006; Juster and Stafford 1991). Dharmowijoyo et al. (2015) deployed a panel time use and activity diary throughout a 3-week period to capture day-to-day variability and repetition patterns. However, the activity diary was sampled at 15-min intervals, hence short duration activities and trips are not recorded and information on goods consumption is not included.

The goal of this study is to develop a survey design that meets all requirements formulated above and at the same time ensures acceptable response burden and high data quality. Such a survey procedure should have an observation period of at least 1 week, since this period captures the rhythms of most activity types sufficiently and is a suitable compromise between response burden and data requirements (Jara-Diaz and Rosales-Salas 2015; Minnen et al. 2015; Senbil and Kitamura 2009; Zerubavel 1985). A combined data collection approach is meant to take advantage of three currently separate survey techniques:

We merge these three survey traditions into the innovative “Mobility–Activity–Expenditure-Diary” (MAED). This paper reports about the lessons learned from two pilot studies and the results of the following main survey, which uses the final MAED-design. With this research we hope to contribute to the advancement of methods for collecting data on travel behaviour in the context of individual’s overall activity and consumption patterns.

The remainder of this paper is organised as follows: The state of practice for travel surveys, time use surveys and consumer expenditure surveys is presented in “Current state of survey practice” section. Findings from pilot studies which marked the path towards the final questionnaire design are discussed in “The Mobility–Activity–Expenditure-Diary (MAED) design” section. We first develop the general concept of the MAED based on the state of art and the goals of this study in “Approach to integrating the three survey traditions” section. We then explain the questionnaire designs for the pilot studies in “Findings from pilot studies” section and the final MAED design in “The final MAED design” section. “Survey procedure, response rates and incentives” section describes the survey procedure and the tested incentive schemes. Quantitative results of time use and expenditure patterns are presented and compared with data from national Austrian surveys in “Quantitative results of main survey” section. Conclusions and an outlook on further research are provided in the final section.

A major challenge of the MAED survey is the high response burden caused by the large amount of information, the complexity of information, and the long observation period. In order to achieve high response rates and data quality it is necessary (1) to motivate the respondents at the beginning and again at crucial stages of the survey, (2) to provide individual support during the reporting period—written instructions are important but not sufficient, and (3) to offer an incentive in return for the high effort. In terms of information channels we used only self-administered mail-back questionnaires with telephone announcement and support. We decided against online questionnaires for several reasons:

The MAED survey procedure is based on the NKD (Socialdata 2009). In answer to the higher complexity of information and the higher response burden we integrated additional phone calls. Figure 6 gives an overview of the survey procedure used in the main MAED survey.

The addresses of survey participants were gained from a random selection of Austrian households according to 18 pre-defined strata, which were arranged by region and settlement structure. For around 50% of sampled households a telephone number could be found. The availability of a telephone number makes a difference for the recruitment process, because announcement calls yield much higher participation rates than announcement letters:

Households with telephone number were called for motivation 3 days after the announcement postcard was sent. They were informed about the study and the incentives, and were asked to participate in the survey. Households which agreed to participate were sent a questionnaire package with a pre-defined reporting period. On the first diary day households were called again. This call served two purposes: (1) as second motivation and reminder to start with the diary—if they had not started already; (2) to give the participants some key information how to fill out the diary correctly. Interviewees were trained to explain those parts of the questionnaires that were most important or tended to cause difficulties, e.g., how to code frequently visited locations or how detailed activities should be reported. Respondents could also ask questions. Households that returned the questionnaire in due time received no further call. Otherwise they received a reminder call every week until the questionnaire had been sent back or the household refused to participate.

Households without an available telephone number could only take part if they actively replied to the announcement letter. The answer (folding card, SMS or e-mail) should include the number of required diaries and a telephone number. From that point on the household was treated as described above: sending the questionnaire, phone call on the first diary day etc.

The returned questionnaires were first checked for completeness of documents and then entered into a database. This step included a validation with checks for plausibility and missing values on an automatic and manual basis. In case of missing or implausible answers respondents were called again; open questions were discussed and solved step by step together with the respondent to ensure high data quality.

Table 5 shows the sampling of the main survey. The response rate of telephone households (V1) is at a similar level as in the pilot studies, which results from two balancing effects: On the one hand we had an additional selection criterion ‘non-employed households’, what causes a lower rate; on the other hand we had by definition no ‘unavailable households’ in this group, because households of group V1 were shifted to group V2 (households without telephone number) if the phone number was not valid or if contact attempts failed. A comparison of response rates of both groups shows that households motivated by telephone (V1) responded almost three times more often than those without known telephone number (V2) which had to reply actively to the announcement letter.

The incentives were an integral part of the survey procedure. In the main study we offered 40 Euros for each completed diary. The incentive was paid after the diary is returned, checked and validated. This ensured the respondents’ interest to stay in contact with us until the data are finally validated and error corrected. In the pilot studies we tested different schemes:

An import question is whether our combined and condensed scheme measures the same values as the conventional surveys on time use, travel and consumer expenditure do. The combination of the three survey traditions is expected to affect the level of detail only, but not the distribution of main indicators.

In this section, we match our results with official figures of Statistics Austria as a benchmark. The latter had to be adjusted to fit the prerequisite of a population restricted to households with at least one employed person and the requirements in terms of model specifications described in “Approach to integrating the three survey traditions” section. Tables 6 and 7 present the values of some key socio-demographic characteristics of the MAED survey in comparison with the Austrian national travel, time use and consumer expenditure survey. These surveys are used in “Mobility”, “Time use” and “Consumer expenditures” sections for comparative analyses. In view of the MAED survey’s representativeness for the Austrian population of employed persons/households the Austrian national census (Registerzählung 2011) administered by the federal statistical agency of Austria (STAT, Bundesanstalt Statistik Österreich) is used as reference. Table 6 shows that women are slightly overrepresented in the MAED survey sample and the age distribution is left-skewed with younger employed persons being underrepresented, especially those aged 20–29. While the ratio of employed and self-employed persons corresponds well to the population the numbers on the highest educational degree attained indicate that more highly-educated people took part in the MAED survey. Graduates of universities, for example, are represented 2.5 times more often than in the Austrian census. As far as mobility surveys are concerned, this is a well-known phenomenon (Gerike et al. 2015). The overrepresentation of university graduates is also due to the higher average age of participants, so they are more likely to have already completed their education.

The group of single-person households is underrepresented in the MAED (14.5%), employed single-person households add up to over 30% of Austrian households (see Table 7). The group of households with 4 members, in contrast, is overrepresented (MAED: 27.1%, Austrian national census: 18.2%). Regarding the level of urbanisation response rates were higher in rural areas. This explains to some extent the low number of single-person households, because they are found more often in urban areas. In small municipalities only every fourth household is a single-person household, whereas in cities this applies for almost every second household (Statistik Austria und Österreichischer Städtebund 2014).

Our motivation for developing a novel survey instrument was to obtain a dataset, which includes all required components to model travel behaviour within the framework of consumers’ home production (Jara-Diaz et al. 2008). To this end we developed a questionnaire and survey design, which enables the collection of data about travel activities, non-travel activities and expenditures from the same individuals over a period of 1 week. The developed survey instrument is based on existing travel diaries, time use diaries, and consumer expenditure diaries. These stand-alone diaries were simplified and re-arranged in order to achieve an integrated MAED with the following features:

It has become evident that the integrated MAED survey performs well and delivers all queried information for travel and non-travel activities at acceptable response rates. Compared to conventional time use and expenditure surveys, it seems that the re-arrangements did not systematically affect the distribution of main activity categories; travel activities are reported more accurately than in conventional travel diaries. Individual telephone motivation and support of respondents as well as an incentive are required in return for the high response burden.

The quality of collected expenditure data is not fully satisfying. We were successful in including all required information in a condensed form into the MAED, however, conventional expenditure surveys put more effort into data collection by means of personal interviews and longer observation periods. This would also be desirable for the MAED to reduce unsystematic variation and to obtain better representations of individuals’ long term equilibrium. A technology-based version of the MAED could be a reasonable way to introduce automated data processing and balance checks in view of income and expenditures.

Despite this moderate weakness there is no doubt that the MAED survey yields a more accurate and consistent dataset for modelling travel behaviour within the framework of consumers’ home production than any existing survey (or the probabilistic merge of different surveys) may yield. We expect an increasing interest in this kind of integrated data, since both travel and non-travel activities have become more flexible and complex, as do the interdependencies between both. Many researchers work on a better understanding the multiple trade-off processes and travel decisions and how they will change in the future. Prominent fields of research are social interactions in multi-person households (Ho and Mulley 2015) and substitutive relationships between in-home and out-of-home activities (Srinivasan and Bhat 2005), which have the potential to heavily affect future travel demand. The research questions evolving in these fields inevitably rely on integrated data that can be obtained through a MAED survey.