BACKGROUND
The practice of making the value of travel time savings (VTTS) for a single trip an estimate based on a percentage of the average wage rate is still alive and well. Some figures circulating, including those being used in evaluation guidelines, are typically around 40% of the average wage rate for private travel. This amounts to $13.76/person hour (2011–2012), in Table 1 of Appendix 4 of the Transport for NSW March 2016 Principles and Guidelines for Economic Appraisal of Transport Investment and Initiatives (“Principles and Guidelines for Economic Appraisal of Transport Investment and Initiatives Transport Economic Appraisal Guidelines” 2016). Footnote 23 of The Australian Transport Assessment and Planning Guidelines (ATAP 2018) report (authored by Neil Douglas) states that the Australian value is therefore close to the 40% wage rate assumption as recommended by the Austroads working group in 1997 in referring to evidence in text: “The overall average of $12.80/hr was 43% of AHE for Australia.” In many countries, when obtaining a base VTTS as a percentage of the average wage rate, it is typically assumed that on average individuals work and are paid for 2,000 hours annually. This is a simplistic assumption, as it does not inform us how the VTTS was initially obtained. For example, was it based on a model that distinguishes the components of travel time for a trip, such as free-flow, slowed-down, and stop–start time, and does it allow for the variability in total time and/or components over repeated trips (the latter being the true measure of trip time variability)? We are not focusing on the value of travel time variability (or reliability); however, we are including it within a model to estimate the single trip value of time to ensure that we control for this influence and avoid confoundment with the travel time components of a single trip.
In the late 1960s and early 1970s, when the focus was on a single VTTS, simplistic studies (often a regression equation and not a discrete choice model) found that the VTTS as a percentage of the average wage rate (noting most people surveyed were commuters and worked fixed hours each week) was close to 25% (Quarmby 1967; Beesley 1965). In the US, the magic number often cited was 40% (Lave 1969; Lisco 1967; Thomas 1967) These estimates, often quoted, have become a rigid benchmark to obtain “acceptable” values.
The earlier studies that were used to establish such benchmarks have treated travel time as homogeneous across a sample (i.e., fixed parameters), whereas the state of practice now allows for preference heterogeneity across a sample through random parameters. A fixed average wage rate becomes almost impossible to support, given growing flexibility in working hours and discretionary income. As methods have evolved and more data are available, there is no theoretical rationale for such a fixed percentage.
RESEARCH QUESTION AND HYPOTHESIS
How appropriate is it to attribute the acceptability of the mean VTTS to a specific percentage of the average wage rate, when heterogeneity of travel time and work practices no longer align with a dominating homogeneity assumption on working hours? To illustrate the behavioral implications of such an “imposition,” we consider how the mean VTTS decomposed into appealing heterogeneous components (i.e., free-flow and congested time) deviate from this strong condition and posit that using the 40% rule is very likely to detract from the gains we have made in understanding the heterogeneity of the behavioral VTTS.
METHODS AND DATA
We have progressed a long way in enriching the dimensionality of travel time that should be considered when evaluating transport initiatives (projects, programs, and policies), especially in behavioral interpretations, econometric methods, and the design of choice experiments (Hensher et al. 2015) Decomposing a specific trip’s time into free-flow, slowed-down, and stop–start time recognizes preference heterogeneity of travel time within a trip; by contrast, we also have trip time variability or reliability (the basis of value of reliability), which relates to variability across repeated trips. Overlayed with these distinctions is a random parameter treatment within a sample. If one can identify the role of heterogeneous free-flow, slowed-down, and stop–start time (and travel time variability across repeated trips), then one can truly identify the contribution of each to overall trip disutility (or utility) of travel. Our focus is on the within-trip estimates, since this is where the link to the average wage rate is made (see previous section).
To illustrate the points being made, we draw on a study we undertook in 2018 in Australia as a sizeable stated preference pilot study of 750 observations where all willingness-to-pay (WTP) estimates associated with car travel were derived from a single, and hence behaviorally consistent and contextually comparable, model. A mixed logit model was estimated with two unlabeled route alternatives where all travel time attributes (free flow, congested, travel time variability) had estimated random parameters normally distributed, and the cost attributes (running and toll cost) were estimated as fixed parameters. WTP estimates were obtained using the delta test to obtain standard errors, and hence z values (Hensher et al. 2015), all of which are statistically significant at the 1% level. This pilot study was undertaken by ITLS and Deloitte Access Economics for Austroads, the National Association of Road Authorities in Australia. It is yet to be published. This evidence is indicative of what we observe in relation to the average wage rate in many WTP studies.
FINDINGS
The results of interest are summarized in Table 1. The findings suggest that the mean value of travel time savings experienced in different conditions varies and is not well aligned to the 40% rule. It should be noted that the distribution of VTTS results in a distribution of the values as a percentage of the individual’s wage rate. Researchers (particularly consultants) who develop separate models for trading subcomponents of travel time with cost are effectively confounding components of travel time that should be separated out within a single model (and by implication a single choice experiment). Some researchers have used choice experiments on subsets of attributes and then pooled them into a single model with scaling. This we deem inappropriate unless it can be proven that the findings are significantly similar to those from a single choice experiment. The outdated view that individuals cannot process more than three attributes and two alternatives has failed to appreciate the need for relevance, as well as comprehensive and comprehensible preference designs.
Any consistently estimated model is better than assuming a fraction of the average wage rate, and it produces different results, so existing procedures distort investments. Updating by a constant will ensure that all WTP estimates of interest move over time in identical relativity, which we believe is of questionable merit.
We recommend not using the average wage rate measure as a defining benchmark for many reasons, particularly because there are so many new developments in data, models, and interpretation of relevant attributes, which means that the average wage rate relationship is highly volatile and hence unreliable as a single measure of relevance.
Acknowledgment: I thank David Levinson and a referee for comments on an earlier version.