Stage 5: Aggregation and/or Disaggregation
2.5 COMBINING REVEALED AND STATED PREFERENCE TECHNIQUES
There has been increasing interest in combining revealed preference (RP) and stated preference (SP) data in valuation of demand for environmental quality. The advantages of combining these two data sets include an increase in the amount of information available; the possibility of modelling goods with attribute levels outside the range of current levels; and reduction in the collinearity offered by the SP statistical designs (Adamowicz et al., 1997).
There are two main approaches of combining SP and RP data. The approaches are Random Utility Models combining SP and RP data, and the Contingent Behaviour Approach relating to either price or environmental quality changes. Adamowicz et al.
(1997) used RP and SP data based on recreational choices, where choice alternatives are described in terms of site attributes. This pooled Random Utility Method (RUM) approach is probably most suitable when the analyst wishes to focus on the value of
different attributes of recreational goods; and where changes in environmental quality produce site substitution effects across a group of sites (eg a group of fishing rivers when water quality alters).
Contingent Behaviour models are somewhat different. The word “contingent” implies that what is being measured is intended behaviour in some contingent market, rather than actual behaviour. Observations from contingent behaviour can be combined with observations of actual behaviour from the same individuals, using either pooled or panel data models. In Englin and Cameron (1996), four price-quantity estimates were made for each respondent, one real and three hypothetical. They conclude that the RP data gave lower estimates per angler than the hypothetical data; and that combining the real and hypothetical data improved the precision of the model estimates. The main feature of the Englin and Cameron paper is that the contingent behaviour relates to changes in trip frequency as prices changes. A natural extension is then to look at contingent behaviour when environmental quality changes. Such an approach was followed by Hanley et al. (2003), who look at the benefits of improved water quality standards on Scottish beaches. A more recent study, Kragt et al. (2006) also uses contingent behaviour approach to estimate the effect of quality of Great Barrier Reefs to the demand for recreational trips.
2.6 CONCLUSIONS
There are two types of valuation that can be used in valuing non-market goods such as environmental goods. These are the revealed preference approach and the stated
preference approach. For the revealed preference approach there are several methods, namely Hedonic Price Method, Averting Behaviour Method, and Travel Cost Method.
From all these methods, the TCM is the most appropriate method to be adopted in this study since it is a method that uses the cost of travelling to a non-priced recreation site as a means of inferring the recreational benefits which that site provides.
There are two types of the TCM normally used in past studies; Individual Travel Cost Method, and Zonal Travel Cost Method. Both methods have their strength and their weaknesses. For instance, the ZTCM is best suited to estimating consumer surplus for recreation at sites where visitor origins are relatively evenly distributed (Garrod and Willis, 1999). Problems arise when visitor origins distributed asymmetrically or where there are a few important points of origin to a single site. Another issue with the ZTCM raised by some authors is about the zonal definition. For example, Smith and Kopp (1980) demonstrated that the assumptions underlying the definition of zones can seriously impact on the resulting estimates of consumer surplus. Another limitation of the ZTCM is that it assumes that the estimated demand is generated by a “representative consumer” whose behaviour reflects the average behaviour in the population of a zone.
Another method is the ITCM that has an advantage over the ZTCM in that it takes more account of the inherent variation in the data, rather than relying on zonal aggregate data.
The procedure undertaken in the ITCM requires researchers to undertake an on-site questionnaire survey of visitors aimed at eliciting the estimates of household or individual visit frequencies over a given time period, plus information on the cost of travel to the site, recreational preferences, use of substitute sites, and socio-economic
characteristics. With the information, we can derive a demand curve from which consumer surplus may be estimated. Even though the ITCM is generally more flexible and applicable at a wider range of sites than the ZTCM, the former requires more information about individual visitors and it relies on surveys to elicit visitor characteristics, preferences and behaviour.
When both methods were carried out using the same data sets, considerable differences have been observed in estimated consumer surplus such as found by Willis and Garrod (1991), and Hanley (1989). Taking all the above into consideration, this study used both methods to elicit consumer surplus and at the same time to investigate whether we will agree with the abovementioned author or not.
Revealed preference, or behaviour in the market place, cannot value all environmental goods. As an example, by using TCM, one can estimate the value of a national park by assessing the demand for a related market good, which is by calculating how much people are prepared to spend on travel to gain access to that particular park. However, one cannot estimate non-use values, since there is, by definition, no related market good for the mere existence, as distinct from use, of the park by using TCM. Thus, stated preference approaches is the most appropriate approach to use to value public goods such as wilderness and landscape preservation; or the value of preserving historical artefacts, monuments, or the character of old towns. This is because in stated preference approach, respondents are directly asked of the WTP they put on a good in a study through a survey question.
In the stated preference approach, currently there are two most prominent methods namely the Contingent Valuation Method and the Choice Experiment or Choice Modelling. One advantage that CM has over CVM is the ability to separately identify the value of individual attributes of a good or programme. However, the CM questionnaire needs respondents to answer complex choices or rankings between bundles with many attributes and levels.
So, for stated preference approach, we decided to use a single-bounded (or referendum) and a double-bounded CVM. This is because this format has become the pre-eminent approach to contingent valuation throughout the world since mid-1990s (Garrod and Willis, 1999). The CVM studies can actually be conducted using several elicitation formats. Until the mid-1980s, the elicitation formats are based on open-ended questions and iterative bidding games (with or without the use of payment cards). Problems associated with these techniques led a number of researchers to investigate alternative elicitation formats that did not require respondents to construct their maximum WTP for a particular environmental good but instead asked them to choose between discrete alternatives relating to the specification of that good and its cost. The discrete choice question format, often known as referendum CVM becomes the most popular format of WTP elicitation after the recommendation made by the US Department of Commerce’s National Oceanic and Atmospheric Administration’s (NOAA) Blue-Ribbon Panel (Arrow et al. (1993). This format only requires respondents to answer “yes” or “no” to a given amount. It is like a market situation, where for each good the price is given, and consumers choose whether to accept it or not. The advantage of this type of question format is that respondents are said to have no incentive to behave strategically (Arrow
et al., 1993). Thus, when a respondent is asked whether they are willing to pay some amount of money for a specific environmental improvement, and they said “yes”, it imply that their “true” WTP for that improvement is at least that amount of money.
Even though the referendum CVM is widely used not just for research purposes but also for policy decision making, it is still associated with some weaknesses. In order to improve the statistical efficiency of the referendum CVM, some researchers have proposed that a further round of bids follow the first round, with the level of the second bid dependent upon the response to the first. Thus, an affirmative response to the first bid amount would lead to the respondent being asked about a higher amount, while a rejection of the first bid would lead to the second bid amount being lower. The double-bounded model is asymptotically more efficient than the single-double-bounded model, as proved by Hanemann et al. (1991).
The assumption underlying the resulting analysis is that identical value distributions are elicited by both the initial and the follow-up questions (Cameron and Quiggin, 1994).
However, there are several studies which found that they yield WTP estimates that are substantially different from the estimates implied by the first responses alone such as Hanemann et al. (1991), McFadden and Leonard (1995), and Herriges and Shogren (1996). Herriges and Shogren investigate the existence of an anchoring effect caused by the first bid, and conclude that it affects, at least in part, the estimates. Some other researchers investigate a “yea-saying” effect. For example, Whittington et al. (1992) found that giving respondents time to think had a clear influence on their answers,
producing consistently lower estimates. There is concrete possibility that some respondents tend to say “yes” if an answer is needed on the spot.
Calia and Strazerra (2000) used a Monte Carlo analysis in search of the bias and efficiency of a single-bounded versus a double-bounded model. They confirmed that the double-bounded CVM is more statistically efficient compared to the single-bounded model. It produces more precise point estimates of parameters and central tendency measures of the WTP, as well as narrower confidence intervals around the mean or median WTP. On the contrary, no clear-cut results are obtained for the point estimates given by the two models, even for a small sample size, so they conclude that neither estimator can be said to be less biased than the other.
CHAPTER 3
PAST STUDIES USING THE TRAVEL COST METHOD AND CONTINGENT