The Scottish Government

« Previous | Contents | Next »

Listen

CHAPTER SIX: QUANTITATIVE SURVEY - CHOICE EXPERIMENT

6.1 Introduction

The results from the case study workshop exercises involving choice sets were used in the design of the choice experiment survey. The second part of the survey questionnaire consisted of a choice experiment in which respondents were asked to make choices between different hypothetical bundles of service attributes. Each attribute represented a different element of service provision and in each option took one of either two or three levels. Each option also had an associated price in terms of an increase in annual council taxes per household. The inclusion of a price allowed the economic value of the service attributes to be estimated.

6.2 Methods

The choice experiment consisted of four local service attributes relating to the location, quality and accessibility of services, two of which had three levels and two levels. The price attribute had six levels (see Table 6). The attributes were developed from workshop discussions with members of the public and service providers in each of the case study areas, and the levels were chosen to provide a contrast across the range of service provision within each attribute. Although the workshop discussions raised the issue that it would be easier if choices were based around specific services it was necessary to maintain a degree of abstraction in the choice experiment. As was found in the Applecross workshop there may have been a tendency to anchor on services that already exist rather than expressing any willingness to pay more for new services. We were also interested in exploring preferences for the common attributes of services rather specific services.

In the service type and location attribute the co-location level was expressed in terms of services being co-located "in your nearest village". This potentially creates problems in that some respondents will live within villages where services are available, whilst other will live outside villages and have some distance to travel. This in turn will affect the relative distances of the other levels within this attribute. Two approaches to analysis were used to alleviate this problem. Firstly, preferences for co-located services locally were analysed offering a more general assessment of preferences for local co-located services relative to single services at a distance. Secondly, respondents were also asked how far they lived from their local village, these values were entered into the first level of the attribute and analysed against services being either 10 or 20 miles away to estimate the preferences for marginal changes in distance to services, in effect giving an estimate of the disutility for each additional mile travelled to access services.

The remaining service attributes relate to the quantity and quality of services, in terms of opening hours, who staffed the service, and the level of service. Staffing was raised as an important issue in the Stornoway workshops where participants considered it very important that they were familiar with the staff. Opening hours and level of service reflect the possible trade-off between limited local services and fuller services provided at a distance.

Table 6: Choice experiment attributes and levels

The price attribute was presented in terms of an increase in annual household council tax payments. This was considered to be a realistic payment mechanism for the delivery of local services, one which respondents could easily relate to, and consequently would encourage respondents to carefully consider their choices. Respondents were asked to complete six pairwise choices between different combinations of service attributes, an example of which is illustrated in Figure 15.

The choice experiment survey was piloted with samples of 30 respondents in each of the three areas. The results of the pilot survey indicated that respondents were making choices based on the attributes and that the price attribute adequately covered the range of willingness to pay. Consequently the choice experiment was not amended for the main survey, although minor changes were made to other aspects of the questionnaire concerning attitudinal and socio-economic questions. The main survey was administered to 170 respondents in both the Eastriggs area and on Lewis, and 176 respondents in the Applecross area.

The choice experiment responses from the main survey were combined with those from the pilot survey, as there had been no changes in the design between the two surveys. This yielded total samples of 200 for both Eastriggs and Lewis and 206 for Applecross. With six choice pairs per respondent, which are treated as independent observations, this would yield 1200 choice sets from Eastriggs and Lewis, and 1236 from Applecross. However, a number of respondents were unable or unwilling to make a choice between options A, B or neither, their responses being recorded as "don't know". These respondents were subsequently removed from further analysis.

The non-price attributes are treated as qualitative variables and enter the model as dummy variables in which the "worst" level for each attribute takes the value zero and is treated as a reference level. These 'worst-case' levels for each of the attributes are:

Implicit prices represent the willingness to pay to move from the reference level of an attribute to an improved level and are calculated as:

where is the estimated coefficient for a change between levels of attribute i representing the utility of that change and is the estimated coefficient of the price attribute, representing the marginal utility of income. This ration is multiplied by -1 due to the expected negative sign of the estimate price coefficient.

Figure 15: Example choice set

6.3 Results

Table 7 presents the results for multinomial logit models estimated for each of the three samples and a combined model of all three samples.

The estimated coefficients therefore represent preferences for changing from the reference level to each of the other levels in each attribute. For the Applecross, Lewis and combined models all of the estimated coefficients are significant at the 5% level and have the expected signs ( i.e. positive preference to move away from the "worst" level and negative preferences for higher council tax), with the exception "all day opening (9 to 5)" versus "all day opening and some evenings". This suggests that respondents are indifferent between having all day opening and having evening opening. The issue of opening hours did not arise as a major discussion point at any of the workshops. In the Eastriggs model the coefficients for the opening hours and staffing attributes are insignificant indicating that respondents in this sample were indifferent to these attributes. All other coefficients in the Eastriggs model are significant and have the expected sign.

Likelihood ratio tests of model equivalence between the different samples indicated that although no significant differences exist between particular attribute levels, particularly in the case of Applecross and Lewis, none of the overall models were found to be equivalent. However, the degree to which the test of equivalence between models was rejected was much smaller when comparing Applecross and Lewis, than when comparing either of these to Eastriggs. This result means that direct applications of the models to other areas would not be defensible.

A possible reason for the differences in preferences is the different degrees of accessibility of the sample areas as indicated by their Scottish Executive Urban-Rural classification. Eastriggs is classified as being an accessible rural settlement, whilst nearby Annan (approximately 3.5 miles away) is classified as an accessible small town. Applecross is classified as a very remote rural settlement, as is the Island of Lewis outside Stornoway, which itself is classed as a very remote small town The degree of remoteness and consequent expectations of service availability may be reflected in the lower preferences (and values, see below) for closer services for the Applecross and Lewis respondents relative to those for Eastriggs. In other words as remoteness increases, the expectation of service availability decreases. This does not mean that locally available services are not important in more remote areas.

Table 7: Choice experiment results for the three areas (t statistics in brackets).

* Significant at the 5% level

6.4 Relative preference orderings

The different sizes of the coefficients reveal the relative preference orderings amongst the attributes and levels of the three areas. Figure 16 presents the estimated coefficients in graphical form with the 95% confidence intervals for the estimated coefficients. This allows visual comparisons to be made both across the attributes within each sample and across the samples. Overlapping confidence intervals indicate that there are no statistically significant differences between the coefficient estimates.

It can be seen from Figure 16 that for the attribute levels for which the estimated coefficients were significant, the preferences expressed by respondents in Eastriggs were higher and significantly different from the preferences of both the Applecross and Lewis samples. The large overlaps in the confidence intervals for the Applecross and Lewis coefficients (with the exception of "all day versus limited opening") indicates that there are no significant differences between preferences for these attributes in these two areas.

6.5 Discussion of preference orderings

Co-located services rather than single services up to 20 miles away were the most preferred level in both the Eastriggs and Applecross samples, and the second (out of eight) most preferred in the Lewis sample.

All day opening was the most preferred attribute level in Lewis when compared to limited opening. This was the sixth most preferred attribute level in Applecross, but was insignificant in Eastriggs.

Having single services up to 10 miles away compared to up to 20 miles away was the second and third most preferred option in Eastriggs and Applecross respectively. This was sixth most preferred in Lewis.

Having co-located services rather than single services up to 10 miles away was second most preferred in Eastriggs, and fourth and fifth most preferred in Applecross and Lewis. This may reflect the lesser degree of remoteness in Eastriggs, where services are generally closer.

Services that were open all day and some evenings compared to limited opening was the second and third most preferred level in Applecross and Lewis respectively, but again this was insignificant in Eastriggs. There were no significant preferences for having services open all day and some evenings compared to having services open all day.

The different preference orderings between service location and opening hours indicates that respondents on Lewis are more willing to travel to access services, and instead prefer more convenient opening hours. This may reflect a better quality of roads on Lewis compared to Applecross, with resulting lower journey times. Other factors such as place of employment relative to home may also be important as this may allow more convenient access to services at a distance. However, we did not directly observe these factors.

The preference for having staff who know you rather than staff you do not know was insignificant in Eastriggs, but was the second and third most preferred level in Applecross and Lewis respectively. This may reflect the fact that community cohesion is of greater importance to residents in these areas, and that the sense of community is still strong. This difference was demonstrated by comments made at the evening workshop in Eastriggs where many of the (more elderly) participants expressed concern about the loss of sense of community.

The preference for having a full range of information or services rather than a limited range was significant, although the least important attribute, in each sample. The significance of this attribute confirms comments made in the Applecross workshop, where some people had to travel to Kyle of Lochalsh (41 miles) or Inverness (80 miles) to use Post Office services that were not available locally. Similarly, some Applecross residents travel to Dingwall (69 miles) for some council services.

The price coefficient (marginal disutility of price) is twice as large in Eastriggs (-0.0323) than in Applecross (-0.0157) or Lewis (-0.0123) indicating greater resistance to increases in council tax in Eastriggs.

Figure 16: Comparison of estimated coefficients (error bars represent 95% confidence intervals)

6.6 Implicit prices

Table 8 presents the implicit prices for the three areas. These should be interpreted as willingness to pay additional council tax per household per year. Because the estimated models were specified as a linear function of the attributes the implicit prices are additive. Consequently, comparison between levels not explicitly estimated (for example, co-located services versus single services up to 10 miles away) can be made by calculating the difference between the estimated implicit prices.

The estimated implicit prices reiterate the results of the choice experiment models in terms of the relative preferences within each of the samples for the attribute levels. However, the differences in price sensitivity, indicated by the sizes of the price attribute coefficients, slightly alter the relative preferences between samples as we are dividing the estimated coefficients by slightly different price coefficient. Where price sensitivity is higher, as indicated by a higher price coefficient, the resulting willingness to pay will be lower. The price coefficient for Eastriggs was more than twice as large as those for both Applecross and Lewis, indicating a greater disutility from council tax increases. Consequently, the larger preferences for attribute levels in Eastriggs as illustrated in Figure 16 are reduced by approximately a half in relative terms. Figure 17 illustrates this by presenting the implicit prices in the same format as for the estimated coefficients in Figure 16.

Table 8: Implicit prices (£/household/annum).

* Indicates that the underlying coefficient estimates were significant at the 5% level.

Figure 17: Implicit prices (error bars represent 95% confidence intervals)

Marginal value of increasing distance to access services

The attributes used in the choice experiment were primarily qualitative and as such the estimated coefficients and implicit prices reveal preferences and values for moving from one state of the world to another. They cannot therefore be used to estimate the value of incremental changes between levels. However, it is possible to interpret the service location attribute as being quantitative if we make the assumption that there is no particular difference between local co-located services and single services at a distance. By entering the distance that respondents stated they lived from their nearest village where services can be accessed (see Annex 24) in place of the level "Co-located services" we can directly compare this to the other levels of this attribute, 10 miles and 20 miles. This will allow us to estimate the loss of utility for each additional mile travelled to access services.

The multinomial logit models were re-estimated for each sample to estimate the marginal value of each mile travelled to access services. The resulting implicit prices can be interpreted as the annual value of the loss in welfare per household for each additional mile travelled to access services. These were £3.76 for Eastriggs, £3.40 for Applecross and £2.97 for Lewis, indicating that the loss in welfare for having services at a distance is lower for the two more remote areas. Care must be taken in the interpretation of these figures: they do not indicate the loss in welfare for each mile travelled on each occasion a service is accessed. Rather, if a household in Applecross had to travel 15 miles to access a typical range of services the annual loss in welfare would be 15*£3.40 = £51.

Table 9 presents these welfare losses for each sample area over a range of distances. Note that the estimates differ slightly from the implicit prices for the co-located services versus single services up to either 10 or 20 miles away presented in table 8, this is because we have explicitly considered the distance to be travelled to the nearest village in the marginal cost estimates. These welfare losses are over and above direct costs such as transport and the opportunity cost of time.

Table 9: Welfare losses of increasing distances to access services

6.7 Conclusions

The results of the quantitative survey and choice experiment have implications for the provision of co-located services and local services more generally. Co-location was recognised by survey respondents as offering a way of ensuring provision of services locally. Indeed co-location of services such as post offices and grocery shops has existed in more remote areas such as Applecross and Lewis for some time. The non-market benefits of having services provided locally rather than at a distance were demonstrated in all three sample areas. When determining the optimum level of service provision these non-market benefits should be considered in addition to any commercial benefits and they offer a guide to the level of public financial support that could be justified.

Local services are also seen as having an important social role in remote areas in terms of providing a focus for the community. This was backed up by the strong preferences for having staff known to local people delivering services in Applecross and Lewis. This was not important in Eastriggs. Therefore, models of service provision in more remote areas should be cognisant of these factors, for example by encouraging service provision from within local communities. This might be achieved through encouraging volunteerism and assisting local individuals or community trusts to set up or maintain service outlets.

Although not important in the more accessible Eastriggs sample area, all day opening of services was important in both Applecross and Lewis when compared to limited hours. Again this suggests that there may be a case for public financial support for longer opening hours or the encouragement of volunteerism for services where purely commercial considerations would justify only limited opening hours.

In all three samples there was a preference for having a full range of services over a limited range. This issue is exemplified by the experience of some participants in the Applecross workshop who had travelled either 41 miles to Kyle of Lochalsh or 80 miles to Inverness for Post Office services that were not available locally.

« Previous | Contents | Next »