Even “clean and green” New Zealand has its share of environmental problems. This is especially true in areas exposed to intensive agricultural production such as the Waikato region which accounts for around 30% of New Zealand's dairy production. Policy makers are torn between supporting the country's leading export industry and ensuring sustainably high environmental quality for the 400,000 people who live in the region. Water pollution from agricultural activities is considered to be one of the most important environmental issues facing New Zealand and is the most frequently mentioned environmental concern for the region's residents [1]. These concerns are well founded since levels of nitrogen and phosphorus in many streams, rivers and lakes have increased over the last two decades leading to a progressive decline in water quality and increased incidence of algal blooms in freshwater bodies [2].

Technical and regulatory mechanisms to reduce this non-point source pollution from agriculture are now the focus of an intensive research effort. Policy makers are showing increasing interest in non-market valuation and the use of market based tools to try and attain environmental improvement. It was in this context that a research program was started in 2008, to assess the potential tradeoffs between cost, water quality improvements and job losses, using choice experiments. It is intended that the findings will inform the policy process by allowing decision makers to consider both the costs and the benefits of different levels of water quality improvement for long term sustainability of the freshwater system in the catchment.

In this paper we describe a choice experiment on a community's willingness to pay for water quality improvements in streams. We investigate the preferences of residents of the Karapiro catchment which stretches over 155,000 hectares of the Waikato region from Lake Arapuni to the Karapiro dam. Land use is predominantly for dairy (34%), pastoral (13%) and forestry (48%) production. The amount of nitrogen and phosphorus reaching waterways in the catchment has generally been increasing and is expected to continue to rise because of intensification and conversion of land from forestry to dairy. Even with widespread adoption of “best management practices” [3] it is expected that the streams and rivers in the catchment will support more algae, water clarity will fall and the water system's ecological health may decline. Levels of E. coli may also increase. These changes may endanger the overall environmental sustainability of the current agricultural system.

Discrete choice experiments have gained widespread recognition since their early application by Louviere and Hensher [4] and Louviere and Woodworth [5] and their earliest application to environmental valuation by Boxall et al. [6]. Choice analysis is an attribute-based technique in which respondents are presented with different alternatives defined in terms of environmental attributes and cost. They are then asked to select their preferred one. The tradeoffs that they reveal during this exercise between the cost of the proposed options and their environmental attributes are used to derive implicit estimate of monetary value, under a set of well qualified assumptions.

In environmental valuation studies using choice experiments, researchers often need to provide respondents with descriptions of status quo conditions. Such descriptions are typically derived from environmental baseline studies and may differ from those perceived by respondents. Such discrepancy may lead to problem in benefit estimation because ignoring differences in utility baselines may affect the magnitude of utility changes and hence bias the implied estimates of benefits from the proposed environmental policies. We investigate this issue, taking the case of respondent perception of the quality of local streams.

In order to study the preferences of respondents with respect to departures from the current environmental conditions, the so-called status quo (SQ), analysts often place this as an alternative in all choice sets. However, recent studies have shown that description of the status quo, or its mere presence in the choice context is not neutral to the choice outcome [7-12]. Later in this paper we review the literature on current research results involving status quo in choice experiments, but we will focus on one area of relatively poor investigation, namely that of identifying the specific effect that respondent's perception of status quo conditions has on implied welfare estimates. In particular, respondents may or may not have a clear perception of how the status quo conditions they experience relate to the attributes and levels considered in the choice exercise. In short, some respondents may not be able to map into the descriptors of environmental status used by the researcher. In this case, it is necessary for the purpose of the choice exercise to provide respondents with a description of the SQ conditions using the specific metric selected for the experimental design. So, one can distinguish two types of respondents. A first type, whose perceptions of the SQ can be mapped into the choice experiment, and a second group, to whom a mapping needs to be supplied during the course of the interview on the basis of some previous, possibly technical, knowledge. Our contribution to the literature is that of investigating whether the effects of such an asymmetry of treatment systematically results in different welfare estimates from an endogenous split sample design.

We proceed by first reviewing the different formats for the SQ alternative in choice experiments. Hess and Rose [13,14] categorized the SQ alternatives into three formats as follows:

“Firstly, … the presence of a status quo alternative which is represented as a null alternative with the attributes and attribute levels of the alternative not shown as part of the experiment. A second form of these experiments involves respondents being shown alternatives with attribute levels based on their own experiences but not the exact levels as described. A final form of these experiments involves the inclusion of one or more alternatives in the choice task being described with exact levels representing each respondent's recent experiences.” (p. 299).

An example of the use of the first format is provided in the study by Campbell et al. [15] on rural environmental landscape improvements in the Republic of Ireland, in which the SQ alternative was labeled “No Action” without specifying the attribute levels. In this case it is quite obvious that the respondent is left to her own devices as to what conjecture to make about the SQ. Furthermore, the analyst does not collect any information on such conjecture. In this study we are particularly interested in the second and third formats above. The attributes described to respondents might either represent some average population measure of the good being valued—and as such be described quantitatively to respondents (as in the second case above)—or might be tailored to suit each individual's specific experiences (as in the third case above and Rose et al. [16]). The use of the second approach is the most prevalent in the existing literature on environmental valuation, to which our study contributes. Typically, this approach involves the use of the SQ alternative described in terms of the average population measures of the prevailing environmental quality (e.g., [17,18]).

Such average population measures are obtained through a consultative process involving the recording of expert assessments and public opinions, usually through focus groups. Additionally, other information obtained from a literature search may also be incorporated [19]. In as much as the latter approach is the most commonly used in environmental valuation the following issues are worth addressing. First, what if the predicted average levels of environmental quality deviate from the attribute levels perceived by respondents? Second, in the face of a discrepancy between the perceived attribute levels and predicted average attribute levels for the SQ alternative, how will respondents perceive the choice tasks presented to them? Third, what are the implications for the implied welfare measures of using SQ scenarios that directly account for individual specific perceived knowledge of environmental quality?

Exploratory and pioneering work on the differences between perceived and objective attribute measures was published as early as 1997 [20]. The first and second questions above were more recently addressed by Barton et al. [21] and Kataria et al. [22]. The former analyzed respondents' understanding of water quality in different lakes compared to objective measures. The latter, asked respondents whether they believed in the description provided for the status quo and whether they found the overall scenarios presented to them credible. They found that not accounting for respondents' beliefs in the proposed scenarios could lead to biased welfare estimates.

To date, we are aware of only one other study [23] in environmental valuation that has attempted to address the third question presented above. It is against this backdrop that this study endeavors to contribute to the environmental valuation literature by assessing the implications on welfare estimates of using a SQ alternative based upon each respondent's specific perceptions of water quality vs the use of a fixed SQ based upon average measures of water quality for the overall population.

We use choice experiment data on streams in the Karapiro Catchment to investigate whether respondents' perceptions agree with our chosen description of the SQ alternative (an average measure of stream quality in the catchment), which we provided to them. Instead of simply asking respondents whether or not they believed in the described SQ scenario—as was the case in a study by Kataria et al. [22]—respondents in our study were asked to state their perceived water quality attribute levels at the SQ. Only those respondents who were unable to give their own assessment were given “the average assessment of the current condition of streams in the catchment”. Such treatment is labeled henceforth as SQ provided. Respondents who were able to assess current water quality used their own SQ in the choice experiments, or SQ perceived. We investigate the nature of the SQ effect emanating from the use of these two alternative formats for the SQ alternative and the implications for the implied welfare estimates.

The remainder of the paper is organized as follows. The next section briefly reviews the nature of status quo effects in choice experiments. Section 3 covers methods and the empirical model used in this study. An outline of the survey and experimental design are presented in Section 4. Results and discussions are presented in Section 5, and finally, conclusions and implications of the study are presented in Section 6.

Initially the use of SQ alternatives in choice experiments was supported mainly on the basis of making choice tasks more realistic. It was shown that individuals making decisions tend to refer to past experiences. Therefore, relating experimentally designed alternatives to a previously experienced reference point makes stated choice tasks more realistic to respondents and informative to analysts [24,25]. This is consistent with psychological and behavioral theories, for example, prospect theory by Kahneman and Tversky [26] and case-based decision theory by Gilboa et al. [27]. In later studies the inclusion of the SQ alternatives in choice experiments was justified on other grounds, including avoidance of forced choices [11,28], improvement in model fit, ensuring unbiased estimates [7] and increase in design efficiency [29].

More recently, studies have shown that the status quo description and even its mere presence in the choice context is not neutral to the choice outcome. In particular, it has been found that respondents presented with both SQ and experimentally designed alternatives have a bias towards sticking with the SQ alternatives, generally referred to as the status quo bias effect, even though Scarpa et al. [12] discuss how SQ effect can be due to either a predilection for the SQ or a reluctance to stick with it, depending on the definition of the attributes of alternatives. This asymmetry in preferences between the SQ alternative and non-experienced alternative is consistent with reference-dependent utility theories [26,30-32]. The main explanations that have been put forward for this SQ effect include loss aversion [33] cognitive misperceptions and regret avoidance [31], protesting [34,35] and choice task complexity [36]. It has also been argued that respondents tend to avoid the cognitive burden associated with evaluating choice task alternatives that have not been experienced [10,11] and that respondents presented with unattractive alternatives are likely to choose the SQ [8].

Similarly, methodologies for accounting for the SQ effect on utility have been developed. The common approach has been to include the alternative specific constant (ASC) to capture the SQ effect on the systematic component of utility. The conditional logit model is usually applied to measure such effects. On the other hand, the SQ effect on the stochastic component of utility which represents the correlation of the error structure between alternatives, is commonly modeled through the nested logit framework; see for example [37,38].

Currently, studies have demonstrated that such specifications are limited in that they fail to simultaneously account for the SQ effect on the systematic component of utility and the variance differences in utilities between experienced SQ and conjectured utility from experimentally designed alternatives. To overcome such limitations, Scarpa et al. [12,39] proposed the use of error components (MXL-EC) in which the additional variance of utility of alternatives different from the SQ can be identified. Since their application, numerous other studies have found the MXL-EC to be better suited in capturing the SQ effects than the conditional logit and nested logit frameworks, and even MXL models without error components [13,15,39-43]. Within the MXL-EC framework, the SQ effect on the systematic component of utility can be measured by the ASC, while the effect on the stochastic component of utility can be captured by introducing a common error component shared by the utilities associated with alternatives different from the SQ, which takes account of the correlation patterns and increased error variance due to the conjectural nature of the experimentally designed alternatives.

It has already been argued that when the SQ alternative is included in the utility specification, the utility from experimentally designed alternatives tends to be more correlated amongst these, than with the SQ alternative. This correlation pattern can be attributed to the fact that the utility associated with the SQ alternative is experienced by the respondents while that of experimentally designed alternatives is not and can only be conjectured, giving rise to higher variance. Additionally, the attribute levels pertaining to the SQ alternative are fixed while those of experimentally designed alternatives are variable across choice occasions. This implies that respondents face a higher cognitive burden in evaluating experimentally designed alternatives than the SQ alternative and therefore, extra errors in addition to the usual Gumbel Type I error are expected to be made. These extra errors would induce a common correlation structure across the experimentally designed alternatives and can be captured within the MXL-EC framework through the introduction of a dummy variable [12,15,39,40,42]. For this reason we adopt this modeling approach in our estimation.

We employ a mixed logit specification that combines both the random parameter and error component interpretation, following the approach detailed in Scarpa et al. [44]. Train [45] has shown how the mixed logit model can give rise to two different interpretations, the random coefficient and the error component interpretations. The random coefficient interpretation accounts for taste variations over the sampled individuals and has been widely applied in many studies, e.g., [46-48]. On the other hand, the error component interpretation refers to the decomposition of the error term and accounts for different correlations patterns among utilities for different alternatives [45,49-51].

In the case of this study, the choice tasks consisted of two experimentally designed alternatives and the SQ alternative. We therefore define the following utility structure: (1) { U ( a ) = β ∼ x a + ɛ + μ a U ( b ) = β ∼ x b + ɛ + μ b U ( s q ) = β ∼ x s q + μ s qwhere β̃ denotes the random preference parameters for different water quality attributes used in this study; β_sq is a fixed SQ specific constant which in our case takes a value of 1 for the SQ and 0 for the other alternatives; x is a vector of attributes describing the alternatives as well as selected respondents' characteristics; μ_a, μ_b and μ_sq depict the unobserved component of utility and are assumed to be i.i.d. Gumbel-distributed. Instead, the error component ɛ is distributed N(0,σ²). The σ² adds to the Gumbel variance of μ_a and μ_b.

Assuming a balanced panel of discrete choices, with T choices made by each individual n, the joint probability of a sequence of T choices 〈y₁,y₂,y₃,....y_T,〉 made by an individual is given by: (2) P ( y 1 , y 2 , …. , y T ) = ∫ β ∫ ɛ ∏ t = 1 T exp ( β ∼ x t i + ɛ i ) ∑ j = a , b , s q exp ( β ∼ x t j + ɛ j ) φ ( ɛ | σ 2 ) f ( β | θ ) d ɛ d βwhere ε_j is equal to zero when j = sq.

Since the integral in Equation (4) has no closed-form, it is approximated in the log-likelihood function by numerical simulation, in our case by using quasi-random Halton draws [48,52]. We first illustrate the methods for the estimation of the random utility model and then the specific tests used to evaluate the difference between simulated distributions from models with different SQ data.

The sample households for the survey were residents of the Karapiro catchment from Lake Arapuni to the Karapiro dam including contributing tributaries. Four focus groups were held to derive an understanding of people's views on water quality in the catchment and to identify attributes for inclusion in the choice experiment. These sessions were also used to test early versions of the questionnaire and to discuss the appropriate range of values for the payment variable. Best practice procedures for running the focus groups were developed drawing on Krueger [56] and on more specific New Zealand experience from Bell [57] and Kerr and Swaffield [58].

Focus group discussions highlighted the increasing number of fences on farms restricting livestock access to streams and creeks, and hence livestock pollution. This was recognized as an improvement and many participants thought that stream water quality was improving, especially when streams were protected by fenced areas of bush, which create a natural filter. Focus group participants from different areas had different perceptions of the quality of their local streams. For example, while some streams experienced by participants at the Karapiro focus group were perceived as with poor water quality, participants further upstream at the Waotu group reported high quality streams with trout, the water from which was used as a supply of domestic drinking water.

Questionnaire development and improvement took place over an extended period. Testing started using focus group participants and was followed by a pilot survey using two groups of six participants and a pre-test of 21 questionnaires. The water attributes identified by focus groups participants were supplemented by literature review and discussions with experts in the field. The attributes eventually selected for the final study were:

Suitability for swimming (percentage of E. coli readings that are satisfactory for swimming)

Suitability for swimming and ecological quality were defined by reference to criteria already defined by the Waikato Regional Council whereby water is assessed as being suitable for swimming (or not) and ecological health is assessed as being excellent, satisfactory or not satisfactory. The suitability for swimming attribute aligns with the proposed national policy statement for freshwater management that is designed to ensure that appropriate Freshwater Resources reach or exceed a swimmable standard. This attribute is also intended as a “catch all” that enables respondents to state their preference for water that is safe for all forms of contact recreation (swimming, paddling, fishing, eeling etc.).

The ecology attribute aligns with data collected by Waikato Regional Council (WRC) on the ecological health of waterways in the catchment. Based on 100 monitoring sites across the region, WRC reports that ecological health readings for undeveloped catchments range from 23% to 100% excellent, but for developed catchments the percentage of excellent readings is much worse, between 0 and 25%. The Karapiro catchment falls under the lower Waikato catchment zone where 68% of ecological health readings are reported to be unsatisfactory with only 2% excellent. Ecological health and “presence/absence of native fish and eels” vary together and so are both included in a single ecological health attribute, for example poor water quality results in “only small eels being found in most catchment streams” while high water quality leads to “large eels, bullies and smelt being found”.

The ecology of rivers and streams in the catchment has been adversely affected by clearance of forests and riverside vegetation, habitat loss and creation of barriers to fish passage (including dams). Aquatic plants and animals have also been affected by reduced water quality, changes to flow regimes, habitat loss (due to drainage and changes in land use) and introduced species that compete with or eat native fish [59].

Native fish populations in the Waikato Region are documented in Joy [60]. These species are highly affected by the Waikato dams which prevent fish migration. The population of eels depends on recruitment (which has been falling steadily in recent years) and the number of elvers transported over the hydro dams. Shortfin eels (Anguilla australis) are very tolerant of poor water quality and may even increase with rising levels of N and P. In poor conditions these eels would mainly be 30 to 40 cms in length. If water quality increases (and sufficient numbers are moved over the hydro dams), then the population of longfin eels (Anguilla dieffenbachia) should increase. This species is far less tolerant of poor water quality and can grow to 2 meters in length. Native bullies and smelt should be migratory but landlocked populations exist in Lake Taupo. Numbers of these species may be expected to increase with better water quality. Respondents were asked for their assessment of the condition of streams in the catchment based on the attributes and levels used for the choice cards. Respondents who indicated that they had ‘no idea’ of the quality of the streams in the catchment were presented with the status quo defined as ‘our assessment of the current overall condition of streams in the catchment’ (see Table 1).

During the survey, respondents who felt able to make their own assessment of stream quality in terms of the metrics used in the choice experiment scenario descriptions used their perceived quality assessment as the status quo. In this case attribute levels were entered onto a transparent overlay and placed on top of each page of choice cards to make it easy for respondents to compare their perceived status quo with the alternative levels offered in each choice card.

Attributes, attribute levels and labels used in the survey are defined in Table 1. Choice cards were based on an orthogonal design of 72 choice sets, with each respondent completing six choice tasks.

The initial sample for this study was drawn by intersecting the Land Information New Zealand (LINZ) property title database with the catchment boundary layer in ArcGIS. In this way a list of all 7627 properties in the catchment was produced including physical location, territorial authority and other variables. The population was broken down into three geographical strata to reflect the markedly different socioeconomic characteristics of these areas; namely Tokoroa, Putaruru/Tirau and the remaining rural areas. Address lists were drawn up for each stratum and a pseudo-random number generator was used to draw up lists of addresses to be visited by each enumerator. Field work proved to be very time consuming with each enumerator only able to complete three to six surveys each day. Field work was carried out both during the day and at weekends to try to avoid bias towards people staying at home. In the later stages of the survey a quota system was used to try and reduce bias towards people over 60.

Comparison of socioeconomic and attitudinal characteristics for our sample, with data for the Waikato Region as a whole (Table 2) enables some conclusions to be drawn. Men appear to be under represented at 62%. This may be due to the fact that more males than females were at home during the time of the survey or in cases where a couple was at home then the male was more likely to participate. Differences between the sub-samples are also observed particularly in levels of education and income; for example 49% of the respondents in the perceived category achieved at least a diploma or a certificate compared to only 23% in the provided group. Similarly, 65% of respondents in the perceived category earn at least $50,000 compared to 39% in the provided category. Given random sampling, the differences in representation are mainly attributed to differences in propensity to take part in the survey, for example refusal rates were higher in lower socio-economic status urban areas and lower in rural areas.

Respondents in the SQ perceived subsample generally registered higher incomes and better education levels than their counterparts in the SQ provided subsample. So, we proceeded by comparing the two sub-samples before and after controlling for outliers in income and qualification. In Table 3 we report the models for these comparisons. Models 1 and 3 include all respondents and pertain to the subsamples SQ provided and SQ perceived, respectively. Models 2 and 4 are based on subsamples in which respondents with income levels of over NZ$50,000 and those with any tertiary qualification in education were excluded. We excluded these to try and ensure that differences in the estimated results can be attributed to differences in the SQ treatment alone, rather than to the effect of outliers in socio-economic covariates in one of the two sub-samples.

The broader purpose of this research was to assess a community's preferences for stream water quality improvements. A specific focus in this paper was placed on the effect of accounting for perceived vs described status quo levels. The study revealed that about 58% of respondents had their own perceived baseline condition of water quality and that they could map it into the framework of attributes and levels proposed in the survey. On the other hand 41% of respondents were provided a SQ description by researchers because these respondents either had little or no prior knowledge of the prevailing conditions of water quality in streams or they had this knowledge but could not map it into the proposed framework. We believe that such a dichotomy is common in many nonmarket valuation studies, and hence its consequences for policy prescription via value estimation are worth exploring.

The results of our investigation show marked differences in the marginal value that these two groups of respondents place on water quality improvements and this has implications for their willingness to pay values. The respondents who were provided with status quo descriptions expressed strong preference for water that is suitable for swimming, has good clarity and where trout can be found. Yet, this group displayed a reluctance to stay with the SQ scenario. We argued that this might be the case because of their comparative ignorance of baseline water quality conditions. The second group of respondents, who adopted their own perceived SQ scenario, expressed significantly stronger preference for improvements across all the attributes subject of this study, but this tendency was attenuated by a general reluctance to embrace policy options implying changes from the SQ, about which they had quite good knowledge. For this group, estimates of marginal willingness to pay values are higher across the entire distribution than for respondents to whom the SQ information was provided.

Economic theory suggests that marginal WTP should be proportional to the expected improvement and this in turn depends on individual perceptions in one group and the provided description in the other. In our individual perception data we observe that on average perceived quality of the SQ conditions was higher than the one that was provided. This might be the cause for the observed reluctance to abandon the SQ, as manifested by a positive and significant alternative specific constant for the SQ alternative. In principle for this group the expected improvement would be perceived as smaller, and so would the associated marginal WTP when compared to that held by the SQ provided group. However, this holds only for quality changes within evaluations by the same respondent. Unfortunately this cannot be tested here because of the lack of a counterfactual.

The present study demonstrates the effects of using a coding specification of the status quo directly built on respondents' perceptions. Our results are supportive of the findings by Kataria et al. [22] which showed that failure to take account of respondents' beliefs leads to biased welfare estimates and earlier similar findings by Adamowicz et al. [20] in the context of integrating revealed preference data, in which the status quo was based on respondent's subjective perceptions, and stated preferences, where it was objectively described to them.