* This research was conducted while being an intern at the Netherlands Authority for Consumers and Markets
The Impact of a Change in the Scope
of the Dutch Universal Service
Obligation on Consumer Welfare
Maaike Vermeer*
M.Vermeer.4@student.rug.nl
University of Groningen, the Netherlands Faculty of Economics and Business June 2016
Supervisor ACM: dr. P.T. Dijkstra Supervisor RUG: prof. dr. M. Mulder
Abstract:
This paper estimates the change in consumer welfare for different changes in the scope of the Dutch Universal Service Obligation. The willingness to pay (willingness to accept) for a positive (negative) change is estimated by means of a choice-based conjoint analysis. These estimated welfare measures should be compared with the change in producer welfare to estimate the effect on social welfare. This paper suggests that authorities should be careful with changing the price or the delivery time of a letter. An alternative is to decrease the delivery frequency by one day or increase the distance to the closest mailbox in areas with a (very) strong degree of urbanization.
JEL Classification Codes: C25, C83, D11, D61, H41
I. Introduction
The optimal regulation of the Universal Service Obligation (USO) is a subject that has often been discussed by the Dutch government. In recent years, postal volumes have dropped significantly and expectations are that postal markets will continue to decline in the future (Jaag and Trinkner, 2011). The liberalization of the postal market and online substitution seem to be the main causes of this decline. The financial crisis has strengthened the decline in postal volumes. In response to the crisis, organizations have searched for cost savings and have switched to less expensive alternatives (WIK-Consult, 2011). Authorities are searching for options to keep the USO affordable for the Universal Service Provider (USP) and for consumers.
Most studies and literature focus only on the costs of the USO and how to reduce those costs in order to maintain the USO. For example, Panzar (2003) develops a methodology for calculating the costs of the USO, taking into account the competitive environment of the postal market. Furthermore, Oxera (2007) describes three different ways to fund the increasing net costs of the USO. They both leave out a fundamental part of the USO: the users. The USO is designed to fulfill the needs of its users. A retrenchment of the postal service offered by the USP may decrease the costs of the USP, however it may also decrease the utility individuals obtain from using the postal service. It is therefore important to weight the change in producer welfare of a certain policy up against the change in consumer welfare. By doing this, authorities can determine whether a policy increases social welfare.
The aim of this paper is to determine how changes in several attributes of the Dutch USO influence the obtained consumer welfare. This is estimated by means of the willingness to pay (WTP) or willingness to accept (WTA) for the selected attributes. A choice-based conjoint analysis is used to obtain data about consumer preferences and a conditional logit model employs this data to estimate the WTP or WTA for the different attributes. Authorities can determine policy implications by combining this information on changes of consumer welfare with information on changes of producer welfare.
determine the change in social welfare of a certain change in the USO. The last Section provides a discussion and a conclusion.
II. Government intervention in a market economy
Gruber (2012) explains the role of government intervention by sketching the economy as a series of trades between on the one hand producers and on the other hand consumers. A trade can be seen as a Pareto improvement when it makes at least one party better off without making another party worse off. An allocation is Pareto efficient or Pareto optimal, when the efficiency of the economy is maximized in such a way that no improvement can be made for one party, without making another party worse off (Rosen and Gayer, 2010). In most cases, the competitive market outcome can be seen as the most efficient market outcome for the society. Based on the assumptions that (1) a market exists for each commodity and (2) all consumers and producers are price takers, the First Fundamental Theorem of Welfare Economics states that any competitive equilibrium allocation is Pareto efficient (Rosen and Gayer, 2010). If the competitive market outcome can be seen as the most efficient market outcome for the society, then why does the government intervene in some markets? This can be motivated by the existence of market failures and redistribution.
II.A. Market Failures
The first explanation for government intervention is the existence of market failures, problems that induce a market economy to deliver a market outcome that is not the most efficient market outcome for the society (Gruber, 2012). Market failures arise when firms have market power or when markets do not exist.
The First Fundamental Theorem of Welfare Economics only holds when all consumers and producers are price takers. However, some producers may have the power to affect prices. This market power leads generally to an inefficient allocation of resources. A firm that can set prices may be able to set the price above the marginal costs. This higher price results in a lower quantity consumed than would be the case with perfect competition (Rosen and Gayer, 2010).
offers both types a different contract and both types choose to fully insure (Gruber, 2012). However, individuals know more about their health status than does the insurer. This may cause adverse selection; only high-risk (low health status) individuals select insurance, which leads to an inefficient market outcome.
An externality is a cost or a benefit of an economic transaction that affects directly the welfare of an unrelated third party in a way that is outside the market mechanism and is not reflected in the market price (Rosen and Gayer, 2010). An example of a negative externality, which decreases the welfare of the unrelated parties, is air pollution due to automobiles. Since externalities are not reflected in the market price, individuals are not stimulated to increase (decrease) the consumption of goods with a positive (negative) externality.
A public good is a good that is non-excludable and non-rival. Non-excludable implies that individuals cannot be excluded from the using the good. Non-rival implies that the use by one individual does not reduce availability of the good to others (Rosen and Gayer, 2010). The problem with a public good is that individuals might have incentives to hide their true valuation of the public good. Individuals might claim that the public good is not beneficial for them and that they do not want to pay for it. This is because they know that when the public good is there, they will benefit from it anyway, and they will get a ‘free-ride’ after other individuals pay for it.
II.B. Redistribution
II.C. How does the government intervene?
The government can use taxes and subsidies to intervene in the economy. Taxes can raise the price for overproduced goods, while subsidies can lower the price for underproduced goods (Gruber, 2012). Furthermore, the government can regulate the behaviour of firms and individuals. The difference between regulation and taxation is the restriction on decisions of economic agents. With regulation, the government can restrict the choices of economics agents, while taxation, although it influences agent’s welfare, does not restrict the agent’s choices (Viscusie et al., 2005). The government can affect the behaviour of firms and individuals by, for example, setting a minimum wage, a maximum price, control the amount of pollution emitted in the firm’s production process, or restrict (mandate) private sale of goods that are overproduced (underproduced). In addition, the government can provide the good directly, in order to reach desired level of consumption. Or when the government does not want to be involved directly, the government can finance private firms to provide the level of consumption that maximizes social welfare (Gruber, 2012).
II.D. Government intervention in the postal market
In 1997, the regulatory framework introduced the USO for all European Member Countries, based on the European Commission’s Green Paper from 1992 (Crew and Brennan, 2016). The USO is a set of binding restrictions on the USP, which ensures that all inhabitants of a country have access to a specified quality of postal services at an affordable price. The European Parliament has set minimum requirements for the USO. National governments can determine their own implication on condition that it satisfies the minimum requirements. Roughly speaking, the USO covers single-piece letters and small packages sent and received by consumers and small or medium-sized businesses. This covers only a small fraction of the total postal market, which implies that the great majority of the postal market is liberalized. A detailed description of the Dutch implementation of the USO is given in Section III.
price for sending a letter in high and low cost areas. Therefore, the USO can be seen as a cross subsidy from low to high cost areas (Crew and Kleindorfer, 2001).
Jaag and Tinker (2012) argue that information asymmetries might exist in postal markets. Senders of normal letters cannot follow the collection, sorting and delivering process. Therefore, they cannot observe the level of quality the postal service provide. The USO enables regulatory authorities to monitor the postal service provided.
Moreover, the uniform and universal postal network presents characteristics of a public good (Cremer et al., 2008). This is supported with the idea that the postal network is valuable itself. It non-excludable since all individuals can make use of the same postal network. Due to this network, all individuals (with a known address) can receive mail. It is non-rival because using the postal network by one individual does not reduce availability of the network to others. All individuals benefit from a high quality of the universal postal network. However, they might hide their true valuations, because they once the network is established they get a ‘free-ride’ after other individuals pay for it.
III. The influence of developments in the postal market on the USO
The USP, which may be one or more postal operators, has to provide a minimum quality of services at uniform and affordable prices to all inhabitants of a country. The USO can be seen as a set of binding constraints on the key variables of the USP: the price and the quality of services. It is the combination of uniform pricing and providing a uniform quality that makes it hard to incur no extra financial cost with providing the USO. When a postal provider wants to meet its obligation of providing a uniform quality without increasing financial costs, it can do so by differentiating in prices for different users. And vice versa, keep prices for users uniform without increasing financial costs can be done by providing different levels of qualities to users (Rodriquez et al., 1999). Since USPs are restricted to the binding obligation, it is fair to compensate them for their role as provider of this service.
service and reduces the quality compared to the situation of perfect competition (Hermann, 2011). In order to increase competition and efficiency, the European Union decided to fully liberalize the postal sectors in European countries from 2012 on (Quirós, 2011). The liberalization of the postal market is subject to the condition that the USO remains guaranteed. Competitors are also allowed to enter the regulated fraction of the market, however they have to obey the same regulations as the USP. In general, competitors are only entering in the unregulated market. Inter alia, the large costs to establish a nationwide sorting and delivery network, which is mandatory in the regulated fraction of the postal market, discourage competitors from entering that fraction of the postal market. Nevertheless, abolishing the monopoly right of the USP has serious consequences for its ability to meet the USO. The entering of new postal operators in the unregulated market decreases the market share of the UPS, and thus decreases its economies of scale and increases its costs. Since the UPS uses the same postal network for the provision of both not-USO and USO products, also the costs for USO products increases. To compensate for the higher USO costs, prices of the USO products have to increase or the USP should receive another financial benefit from the government. An increase in prices will reduce the quantity demanded from consumers, which put again more pressure on the costs of the USO.
IV. The Universal Service Obligation in the Netherlands
IV.A. The Dutch implementation of the Universal Service Obligation
The Dutch implementation of the USO is defined in the Postwet 2009. Table 1 summarizes the main features of the Dutch USO. PostNL provides the USO in the Netherlands and is the only postal operator that has the obligation to provide this specified service. PostNL operates on both the USO and the non-USO market. Since it uses the same network for both markets, it enjoys economies of scale. Other postal operators are not serving the USO market. This is because no other operator has such a large network as PostNL and hence it is not efficient for other operators to enter this market.
Table 1: Overview of the most important features of the Dutch USO (Rijksoverheid, 2016)
Dimension USO The Dutch implementation
Products - Letters up to 2 kilogram
- Packages up to 10 kilogram - Registered mail
- Mail with a specified value (insured mail)
Network and access points Amount of service points: 1,000
Amount of mail boxes: 8,700
Collection and delivery - Collecting mail at least 5 times a week
- Delivering mail at least 5 times a week - Collecting medical and bereavement mail
at least 6 times a week
- Delivering medical and bereavement mail at least 6 times a week
At least 95% of the mail has to be delivered within 24-hours. (With 24-hours delivery delivering on the next delivering day is meant. Hence the mail that is sent on Saturday and delivered on Tuesday, the next delivery day, also counts as 24-hours delivery.)
Table 2: Tariffs of USO products
Product Price in euros
Letters < 20 gram – 0.73
< 50 gram – 1.46 < 100 gram – 2.19 < 250 gram – 2.92 < 2 kilogram – 3.65
Package (up to 10 kilogram) 6.95
Registered mail 8.15
Insured mail 14.35
IV.B. Previous studies on developments in the Dutch Universal Service Obligation
ECORYS (2011)
on the actual costs and a determined norm return on sales of 10 percent. ECORYS expected the prices of USO products to increase with approximately 15 percent in the period from 2011 to 2015. The return on sales of the USO is expected to decrease further and becomes negative in 2011 and 2012. Due to the reorganization of PostNL the return on sales was expected to increase again from 2013 on. Since the return on sales is expected to be lower than the predetermined 10 percent norm, it is likely that this results in an extra price increase of 6.5 percent in 2016 (excluding inflation) on USO products. When consumers are sensitive for price changes, this can lead to a sharp decline in demand, which results in lower revenues for the USP and pressures the USO even further. It is very important to determine the scope of the USO conditional on the preferences of users. Intomart Gfk (2011) studies the preferences of consumers with respect to the USO and find that users believe that the price is the most important feature of the USO, followed by the delivery time and the delivery frequency. The distance to post offices and mailboxes are relatively unimportant to users. ECORYS believes that when the users value a low price higher than all other features, changing the relatively unimportant features, in order to keep the price increase low, can increase welfare. ECORYS discusses some concrete cost-saving measures. They discuss five scenarios: (1) an unchanged USO; (2) a change in the amount of collection and delivery days; (3) a change in the amount and dispersion of the mailboxes; (4) a change in the amount and dispersion of postal offices; (5) scenario 2, 3 and 4 together. They find that excluding the Monday delivery to households reduces the extra price increase in 2016 to 4 percent (exclusive price inflation). The fifth scenario could lead to a reduction in the extra price increase in 2016 to 2 percent. Therefore, ECORYS advises to ease the regulation of the USO by canceling delivery on Monday and reduce the amount of mailboxes and postal offices. These adjustments have already been implemented: PostNL does not deliver mail on Mondays anymore and the amount of mailboxes are reduced to 8,700 (Rijksoverheid, 2016).
Furthermore, to make the USO affordable in the long run, ECORYS advises to look at the possibility to introduce 48-hours delivery service within the USO and the possibility to introduce unstaffed pickup points. Besides this, ECORYS discusses the idea of a stand-alone USO company. However, they find that this idea will not increase the efficiency of the USO. A company that provides both USO and non-USO services can make use of economies of scale. These economies of scale are not exploited when a company only provides the USO. Cost will be approximately 20 percent lower due to the economies of scale. Another noteworthy point is that a stand-alone USP will not be established overnight. This means that large investments have to be made before a new company can provide this service.
declines with the same pace in 2015. This implies that it is more likely that the volume of the USO decreases with 25 percent in the period from 2011 to 2015 instead of the 30 percent ECORYS estimated. Furthermore, ECORYS expected the prices of USO products to increase with 15 percent. The price of a stamp increased with 50 percent to 69 cents in 2015, and currently the stamp price is 73 cents. Given that the volume of the USO is expected to decline less than ECORYS estimated while the price has increased more than ECORYS expected, mail seems to be less price elastic than ECORYS estimated. In other words, people do not completely want to switch to alternative forms of sending mail.
Intomart Gfk (2011)
Intomart Gfk (2011) investigates the needs and behavior of the USO users. The aim of their study is to create an insight into consumers’ perceptions, needs and expectations of the universal postal service. Intomart answers what the current sociological, economical and psychological meaning of mail is. They look at aspects of the postal service were users attach value to, which price and quality requirements users have, and the substitution behavior of users. Moreover, Intomart studied which groups are most affected by changes in the postal service.
Their study consists of three parts: a qualitative phase, a quantitative phase and a conjoint analysis. The qualitative phase makes sure that the right questions are asked and tests whether the expected outcomes of a comparable study in 2004 are obtained. The quantitative phase should give a representative overview of consumers’ ideas about the universal postal obligation. For this research they take an online survey among 2000 participants and a written survey among 150 participants who are not active online. The conjoint phase aims to conclude the extent to which different characteristics of the USO influence the valuation of the universal postal service.
them are known with the fact that the postal market was fully liberated in 2009. However, since PostNL is still the only provider of the USO, it has not affected consumers directly.
Intomart compares the study about the behavior of USO users of 2004 with the behavior users have in 2011. They find that users send and receive less mail than in 2004. An explanation could be the increase in Internet usage. Decreasing the delivery frequency has become more acceptable in 2011. In 2004, 41 percent found it unacceptable to move away from six-times a week delivery. In 2011, only 17 percent found this unacceptable. The usage of mailboxes has also declined in 2011. In 2004, users used on average 7 times a month a mailbox, compared to 5 times a month in 2011. Again the increased use of the Internet seems to be the major cause of this decline. Almost one-third of the users would send fewer letters after an increase in the stamp price. Intomart finds that the maximum acceptable amount of hours to deliver a letter is 41 hours, which is pretty close to the 48-hours delivery. The substitution of written letters by email or other mobile communication is more observable by young people. On average young people visit postal offices significantly less than older people, 3.6 times a year and 7.6 times a year respectively. It seems that the younger generation is adjusted more to the increasing online lifestyle.
Table 3: Attributes and levels conjoint analysis Intomart
delivery days. In 2004, less than 5 delivery days was not acceptable. In 2011, less than 4 delivery days are not acceptable.
Intomart claims that users are not willing to pay more than 49 cents for a stamp. The current stamp price is 73 cents, which is almost 50 percent higher than the maximum amount users are willing to pay according to Intomart. Clearly, the demand for mail is not that elastic as Intomart expected or users preferences have changed dramatically in the previous five years. It is therefore interesting to study the current valuation of the different characteristics of the USO and determine how several changes in the USO influence the welfare obtained from this service by consumers. This can be done by means of the willingness to pay (willingness to accept) for a certain positive (negative) change.
V. Methodology
V.A. The willingness to pay and willingness to accept defined
Kalish and Nelson (1991) define the willingness to pay (WTP) as the maximum amount an individual is willing to pay for a given quantity of a good. Likewise, Jedidi and Zhang (2002) define the WTP of a consumer as the price at which the consumer is indifferent between buying and not buying the good.
Attribute Levels
Stamp price 29 cents
34 cents 39 cents 44 cents 49 cents 54 cents 59 cents 64 cents
Delivery time Within 24 hours
Within 48 hours Within 72 hours
Delivery frequency 3 times a week
4 times a week 5 times a week 6 times a week
Distance to closest mailbox Halved
Unchanged Doubled
Distance to closest post office Halved
This definition can be described with the following equation:
𝑈𝑖(𝑥, 𝑝) − 𝑈𝑖(0, 𝑝) = 0 (1)
Here, 𝑈𝑖(𝑥, 𝑝) is the utility consumer i gets from buying good x at price p, and 𝑈𝑖(0, 𝑝) is the utility consumer i gets from not buying good x at price p. The other way around, the willingness to accept (WTA) can be defined as the minimum amount required for an individual to forgo a good (Martín-Fernández et al., 2010)
Hanemann (1991) uses the concept of changes to describe the WTP and WTA. He defines the WTP as the maximum amount an individual is willing to pay to undergo a positive change and the WTA as the minimum amount an individual is be willing to accept to forgo a positive change. In the same way, the WTP can be identified as the maximum amount an individual would be willing to pay to forgo a negative change and the WTA as the minimum amount an individual would be willing to accept to undergo a negative change. Because the WTP or WTA can give a monetary value to non-monetary attributes, the willingness to pay (accept) is relevant for improvements (deteriorations) of non-market goods or non-monetary attributes (Erbert, 2008). One might think that the WTP for a positive change of a certain good and the WTA for a similar negative change in that good are equivalent. However, there exist a difference between the two from a theoretical point of view. Both the WTP and WTA can be represented by means of two Hicksian welfare measures: the Compensating Variation (CV) and the Equivalent Variation (EV) (Hanemann, 1991).
Graph 1: Panel (a) displays the CV after a price increase of good 1 (X1). Panel (b) shows the EV after a price increase of good 1(X1)1
In a similar way, the EV can be explained by means of panel (b) of graph 1. The EV is the income change that is needed to end up on the new indifference curve at old prices. Assume again that the initial utility-maximizing point is A and the price of good 1 increases. Due to the income effect and the substitution effect, the new utility-maximizing point is point B. The point on the new indifference curve that has similar prices to the initial point A is point C. The distance between N and M is the amount has to be taken away from the individual to end up at the new indifference curve at original prices.
Thus the difference between CV and EV is that the CV reflects new prices and the old utility level, while the EV reflects the new utility level and old prices. In other words, the CV is the compensation needed to maintain the old utility level, while the EV is the compensation needed to maintain the new utility level (Amiran and Hagen, 2003). Generally, for a change that negatively affects individual’s welfare, the CV is identified as the minimum amount an individual is willing to accept to undergo the negative change and the EV is identified as the maximum amount an individual is willing to pay to forgo the negative change. In this case the CV equals the WTA and the EV equals the WTP. For a change that positively affects an individual’s welfare, the CV is identified as the maximum amount an individual is willing to pay to undergo the positive change and the EV is identified as the minimum amount an individual is willing to accept to forgo the positive change. In this case the CV equals the WTP and the EV equals the WTA (Jakobsson and Dragun, 1996).
1 The original graphs from Tutor Globe (2016) are adapted in order to make it easier to define the CV and
Graph 2 can be used to illustrate how the WTP or WTA can be derived to give a monetary value to a change in a non-market good or public good. In this example the quality of the USO is used as public good. The horizontal axis represents the quality of the USO and the vertical axis the consumption of market good 2. The individual obtains utility from the quality of the USO and the consumption of good 2. The quality of the USO can be expressed in the size of the USO and the extent to which authorities are monitoring regulatory compliance. Assume that this quality of the USO is supplied to individuals free of charge. This could be a valid assumption given that individuals can freely receive mail, and thus will benefit from a better quality USO without bearing a cost. In this case, the budget line is a horizontal line and the individual spends all his money on good 2 (Johannson, 1993). The initial utility-maximizing point is A. After a decrease in the quality of the USO, for example an increase in the maximum time it takes to deliver a letter, the utility of the individual decreases and the individual’s new utility-maximizing point is B. At this point the individual still spends all his money on good 2, however the obtained utility is lower than in point A due to the decrease in the quality of the USO. The CV, which is the distance between N and M in the graph, is the compensation needed to obtain the same utility as before the change in the quality of the USO. Therefore, the distance between N and M is the minimum amount an individual is willing to accept to undergo the negative change. Thus in this case the WTA gives a monetary value to a change in a public good.
Randall and Store (1980) argue that for goods that are perfectly divisible and exchanged at zero transaction cost in large markets the CV and EV, and thus the WTP for a positive change of a certain good and the WTA for a similar negative change of that good, are equivalent.
Graph 2: The WTA for a negative change in a public good (X1)
However, due to market failures it might be that individuals do not value benefits and costs equally. For example, due to information asymmetry a gap between the WTP and WTA for a good can arise (Makokha et al, 2007).
According to Cranfield and Magnusson (2003), a consumer’s WTP or WTA is influenced by their individual preferences and tastes, attitudes towards and perception of the different goods, income and demographic characteristics. WTP or WTA measures can be estimated in several ways. The most used methods are revealed preference methods and stated preference methods. Revealed preference methods are based on choices made already by consumers and use this observed behavior to approximate consumer preferences and utility functions (Snyder and Nicholson, 2012). By stated preference methods, individuals make choices under experimental conditions to estimate their preferences (Abley, 2000). According to Tietenberg and Lewis (2016), distinction can be made between direct and indirect stated preference methods. A direct form is contingent valuation. With contingent valuation, individuals explicitly state their WTP or WTA for a specific good. They do so by either directly stating their WTP or WTA for a good (open-ended contingent valuation), or by indicating whether they would buy a good at a specified price (closed-ended or discrete-choice contingent valuation) (Jedidi and Jagpal, 2009). A disadvantage of this method is that direct questions about WTP or WTA are cognitively difficult to answer. Moreover, direct questions may trigger respondents to answer strategically. Alternatively, attribute based methods, such as conjoint models, can be used to identify the WTP or WTA more indirectly (Tietenberg and Lewis, 2016). Instead of asking individuals to explicitly state their WTP or WTA, individuals are asked to choose among different profiles. Each profile consists of a combination of attributes and the levels of each attribute vary across the different profiles. To be able to estimate the WTP or WTA, one of the attributes in each profile has to be a price measure. In this indirect case, the ratio of the attribute coefficient to the negative of the price coefficient is defined as the WTP or WTA for that attribute (Train and Weeks, 2009). Thus the WTP or WTA can be expressed as follows:
𝑊𝑇𝑃 𝑜𝑟 𝑊𝑇𝐴 = 𝛽𝑖
−𝛽𝑝 (2)
A disadvantage of stated preference methods is that the ‘stated’ preferences might not match the actual preferences, which is not the case with revealed preferences (Wardman, 1988). According to Kroes and Sheldon (1998) however, stated preference methods also have advantages over revealed preference methods. It can be hard to obtain sufficient variation in the revealed preference data in order to determine all variables of interest. Moreover, with revealed preference methods there are often strong correlations between explanatory variables. When strong correlations occur, it becomes difficult to estimate parameters that reflect the proper trade-off ratios. Stated preference methods are most of the time more easy to control, since the researcher can choose the conditions that are going to be evaluated by the respondents, and they are more flexible, given stated preference methods’ ability to deal with a wider variety of variables. Given these advantages of stated preferences over revealed preferences, this paper uses a stated preference method to estimate the WTP or WTA for different attributes of the USO.
V.B. Conjoint analysis
This paper uses a choice-based conjoint analysis to study consumer’s valuation of the characteristics of the USO. A choice-based type of analysis is very suitable for this experiment, since it is closer to real-world decision making than a ratings-based analysis and aggregate levels can be taken (Rao, 2014). With this type of analysis respondents receive different choice sets and have to indicate which profile of attribute levels they find most attractive. First the relevant attributes of a product are defined. Next, the different levels those attributes can take are specified. Combining levels of the different attributes can make profiles with a unique combination of product characteristics. The profiles are similar enough to be close substitutes, however they will be different enough so that respondents will have a clear preference. Different profiles can be combined to generate a choice set, from which respondents have to choose the most preferred profile. The data obtained from these choice sets can be analyzed using a discrete choice model; usually a multinomial or conditional logit model is used. Using this discrete choice model, a function that relates the attribute levels to the probability of choice can be obtained.
With a limited number of both attributes and levels, it is possible to present respondents all possible profiles. When the respondents see all possible combinations, the design is called a ‘factorial design’. However, when the number of attributes and levels is large, a ‘fractional factor design,’ in which only a carefully chosen fraction of the profiles presented, will be sufficient. (Van Poll, 1997)
For the conjoint analysis that is designed for this research, five attributes of the USO are specified. Table 4 displays the different attributes and their levels.
Table 4: The attributes of the USO and their levels
Attribute Levels
Stamp price 58 cents
63 cents 68 cents 73 cents 78 cents 83 cents 88 cents
Delivery time Within 24 hours
Within 48 hours Within 72 hours
Delivery frequency 3 times a week
4 times a week 5 times a week 6 times a week
Distance to closest mailbox Halved
Unchanged Twice as far
Distance to closest service point Halved
The first attribute is the price of a letter. When a consumer has to pay less for sending a letter, he or she will have more money left to consume other goods or send more letters. Nowadays, the price of sending a letter is 73 cents. Both higher and lower prices are used as attribute levels. The second and the third attributes are the delivery time and delivery frequency per week respectively. A faster delivery time and a higher delivery frequency imply that consumers can receive and deliver information faster. There are three options for the delivery time: within 24 hours, within 48 hours and within 72 hours. It should be noted that, for example, within 24 hours implies within one delivery day. PostNL guarantees a 24-hours delivery, even though PostNL does not deliver on Sunday and Monday. Hence, the combination of a 24-hours delivery and a five times a week delivery is possible. For the delivery frequency four options are used: 3 times a week delivery, 4 times a week delivery, 5 times a week delivery and 6 times a week delivery. The fourth and fifth attributes are the distance to the closest mailbox and the distance to the closest service point respectively. A smaller distance to the closest mailbox or service point will make it more convenient to send a letter or receive help with sending a letter. The different profiles and the choice sets that are used for this analysis are created with Sawtooth Software, a program for designing and analysing conjoint analyses. Since in total 7 × 3 × 4 × 3 × 3 = 756 different profiles can be generated with the different attribute levels, this analysis uses a ‘fractional factor design,’ in which only a carefully chosen fraction of the profiles presented. Sawtooth Software is specialized in creating carefully a ‘fractional factor design’ and created 28 profiles and divided those profiles over 14 choice sets. During the survey, each respondent was asked to fill in the 14 choice sets. Every choice set consists of two profiles with different combinations of attribute levels and a ‘no-choice’ option. An example of a choice set that was shown to the respondents is displayed in Table 5. The full survey can be found in Appendix B.
Table 5: An example of a showed choice set
Stamp price 78 cents 68 cents ‘No-choice’ option
Delivery time Within 72 hours Within 48 hours Delivery frequency per week 6 times 5 times Distance to the closest mailbox Halved compared to the current situation
Halved compared to the current situation Distance to the
closest service point
Halved compared to the current situation
Unchanged compared to the current situation
Graph 3: The selection of the ‘no-choice’ option
When a respondent chooses the ‘no-choice’ alternative, it means that this respondent would not send any letters given the two profiles of that choice set. After choosing for the ‘no-choice’ alternative, a respondent was asked to state its second choice and choose between the two remaining profiles. This is done to obtain as much information about consumer preferences as possible. With this information a model with a ‘no-choice’ alternative and without a ‘no choice’ alternative can be compared, and the reason for choosing the ‘no-choice’ alternative can be identified. It might be that when respondents notice that they have to fill in an extra question after choosing the ‘no-choice’ alternative, they will not choose this option again as a matter of convenience. However, this does not seem to be the case in this research. Graph 3 displays the selection of the ‘no-choice’ alternative. From the trend line can be concluded that the selection of the no-alternative option slightly increases over time. It could be that respondents believe that some questions in the end of the survey are more unfavourable than questions in the beginning. However, the increase in the selection of the ‘no-choice’ option is very small. It is more interesting to see that respondents are not discouraged to choose the ‘no-choice’ option again after they found out that they have to fill in an extra question.
VI. Data
The survey is conducted among all employees of ACM. Since most of ACM’s employees are high educated, it should be taken in mind that the group is not representative for the overall Dutch population. A link to the survey was published on the home page of the ACM employees website. From the approximately 500 employees, 113 started with the survey. From these 113
respondents, 92 completely filled in the survey. This means that the response rate is approximately 18.4 percent. Xu and Yuan (2001) calculate the required sample size based on the number of parameter used in the conjoint analysis. The number of parameters is equal to the total number of levels of all attributes minus the total number of attributes plus one. The rule of thumb is that the ratio of the number of respondents to the number of parameters should be at least five. The conjoint analysis used for this survey has a total of 7 + 3 + 4 + 3 + 3 = 20 levels and five attributes. The number of parameters for this analysis is therefore 16 (20 – 5 + 1). The minimum number of respondents should therefore be 80. Most researchers agree that the minimum amount of respondents in a conjoint analysis should be at least between 70 and 100 (Xu and Yuan, 2001). Therefore, the sample size this paper uses is sufficient.
Respondents were asked to fill in several personal characteristics. Table 6 shows the distribution of the respondents per category. From this Table can be concluded that the share of male and female respondents is almost equal. Only one person who completed the survey is younger than 20. This is not surprisingly, since most workers have a high level of education and almost all students are older than 20 when they graduate. The other respondents are distributed almost equally over the other subcategories. Most respondents live in a region with a very Table 6: The distribution of the respondents per category
Category Subcategory % of respondents Gender Male Female 46 54
Age (in years) < 20
20 – 30 30 – 40 40 – 50 > 50 1 24 28 24 23 Degree of urbanization Very strong (> 2,500 addresses per km2)
Strong (1,500 – 2,500 addresses per km2)
Moderate (1,000 – 1,500 addresses per km2)
Weak (500 – 1,000 addresses per km2)
Rural (< 500 addresses per km2)
53 27 9 2 1
Size of the household 1 person
2 persons 3 persons 4 persons > 5 persons 28 38 10 11 5 Disposable income of the
strong or strong degree of urbanization (53% and 27% respectively). Also this is not surprisingly, since ACM is located in The Hague. Most respondents are duo or single households (38% and 28% respectively). That most of ACM’s employees are high educated is reflected in the disposable income of the households. Almost half of the households (47%) have a disposable income of more than 50,000 euros per year, and only few households (8%) have a disposable income lower than 20,000 euros per year.
VII. The model
This Section presents the model that is used for estimating the WTP or WTA for the different changes in the USO. Since a conjoint analysis is used for collecting the data, a discrete choice model is an appropriate model to analyse this data. A discrete choice model describes choices between two or more alternatives. In a discrete model the choice you make is all-or-nothing. For example, you choose A or B (Hofacker, 2007). Within discrete choice models distinction can be made between binary choice models and multinomial choice models. In a binomial choice model an individual choices between two discrete alternatives. In a multinomial choice model, an individual chooses between three or more available alternatives. There are also different multinomial choice models, e.g. a multinomial logit model and a conditional logit model. In a multinomial logit model, the explanatory variables are individual-specific characteristics, e.g. income, age and gender. In a conditional logit model the explanatory variables are alternative-specific characteristics, e.g. choosing between different brands (Verbeek, 2012).
VII.A. The conditional logit model
This paper uses the conditional logit model to determine the WTP or WTA for different changes in the USO. The respondents could choose from three options during the conjoint analysis, therefore a multinomial choice model should be used to evaluate the data. Since the explanatory variables have alternative-specific characteristics, the conditional logit model is preferred over the multinomial logit model. McFadden (1974) derives the conditional logit probability in a random utility model (RUM) framework. In RUMs the utility towards an alternative varies across individuals as a random variable. Moreover, decision makers act rational and thus to maximize utility (Hofacker, 2007). McFadden’s conditional logit model reflects the maximization of the utility U. Given that an individual has to choose between J alternatives, the utility, 𝑈𝑖𝑗, individual i derives from making choice j is given by:
Where 𝑉𝑖𝑗 is a stochastic observed component that represents the tastes of the population, and can be written as 𝑉𝑖𝑗= 𝛽′𝑋𝑖𝑗. Here, 𝑋𝑖𝑗 is a vector containing different attributes and 𝛽 is a vector that measurers the weights that the individual places on the attributes. 𝜀𝑖𝑗 is a random unobserved component, which is assumed to be an independent and identically distributed (IID) extreme value (Hole and Kolstad, 2012). Individual i will choose alternative j over alternative k when the utility he gets from alternative j is higher than the utility it get from alternative k: 𝑈𝑖𝑗 > 𝑈𝑖𝑘, given j ≠ k and j, k ∈ J.
McFadden (1974) describes the probability that individual i chooses alternative j over alternative k as follows:
𝑃𝑖𝑗 = Pr(𝛽′𝑋𝑖𝑗+ 𝜀𝑖𝑗 > 𝛽′𝑋𝑖𝑘+ 𝜀𝑖𝑘 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗 ≠ 𝑘)
= Pr( 𝜀𝑖𝑘 < 𝜀𝑖𝑗 + 𝛽′𝑋𝑖𝑗− 𝛽′𝑋𝑖𝑘 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗 ≠ 𝑘) (4) After some manipulations, the conditional logit model specifies the probability that individual i chooses alternative j as:
𝑝𝑖𝑗= exp(𝛽′𝑋𝑖𝑗) / ∑𝐽 exp(𝛽′𝑋𝑖𝑘) 𝑘=1
(5) Estimation of the unknown parameters is done by maximum likelihood (Hill et al., 2012). By maximum likelihood estimation, the probabilities of the observed outcomes enter the log likelihood function. Maximum likelihood then searches for the parameters that maximize the log likelihood function. A correctly specified model provides efficient, asymptotically normal and consistent estimators for the 𝛽 coefficients (Verbeek, 2012).
arises. This is because the ‘no-choice’ alternative is not a ‘real’ profile, and therefore it cannot be seen as just another alternative (Haaijer, 2001). When excluding the ‘no-choice’ alternative, only two ‘real’ profile alternatives remain. There is no alternative that can be omitted anymore, which implies that the IIA automatically holds.
The use of the conditional logit model is also valid in the case without the ‘no-choice’ option. In this case there are only two alternatives, which make the choice a binary choice. With only two alternatives, outcomes of the conditional logit model will be similar to the outcomes of the binary choice model (Verbeek, 2012).
VIII. Results
This Section discusses the estimation results. The conditional logit regression without a ‘no-choice’ alternative is used for interpreting the main results. This model is chosen because the underlying data contains more information than the model with the choice’ option. The ‘no-choice’ alternative does not include any of the attribute levels in the design, which implies that it can be modeled in the dataset by a series of zeros. However, in the model without a ‘no-choice’ alternative, those zeros are replaced by the second choice of the respondent, which increases the information about the choices of the attribute levels. The final Section of this chapter (VIII.G.) compares the model with a ‘no-choice’ alternative with the model without the alternative. However, first the (multi)collinearity problem is explained in Section VIII.A. The goodness-of-fit of the model is discussed in Section VIII.B. In Section VIII.C. the regression results are interpret. Section VIII.D. estimates the WTP or WTA of the different attributes and Section VIII.E. determines the relative importance of the attributes. Finally, Section VIII.F. looks at differences between demographic subgroups.
VIII.A. (Multi)collinearity
First, it is checked whether (multi)collinearity is a problem in the model. When (multi)collinearity exists, it implies that there exists an approximate linear relationship among the independent variables that leads to unreliable regression estimates (Verbeek, 2012). Small changes in the data can cause big changes in the estimated coefficients. To investigate whether (multi)collinearity exists, one can run an auxiliary regression for each independent variable. If the model has a high R2 it implies that the correlation between the chosen independent variable
concluded that the R2’s for halving the distance to the closest mailbox, doubling the distance to
the closest mailbox, doubling the distance to the closest service point and doubling the distance to the closest service point are very high (0.6117, 0.7481, 0.7239 and 0.7328 respectively). The R2 for the auxiliary regression for the delivery frequency is 0.1896, which is clearly lower (see
Table 17 in Appendix A). This suggests that there might be correlation between the four distance measures. To check which variables are correlated, a correlation matrix can be drawn. This correlation matrix can be found in Table 18 in Appendix A. The correlation coefficients between halving the distance to the closest mailbox and halving the distance to the closest post office, and doubling the distance to the closest mailbox and doubling the distance to the closest service point are very high (0.7143 and 0.8095 respectively). This indicates that there exists correlation between those independent variables, which may cause errors. It is not surprisingly that those variables are highly correlated, since they all are distance measurers. Because the existence of (multi)collinearity problems can result in highly inaccurately estimates, it is important to mitigate this (multi)collinearity. A solution could be to increase the sample size, or to omit one or more variables from the model when those correlated variables are assumed to measure approximately the same (Verbeek, 2012).
To mitigate (multi)collinearity in this model the variables for halving and doubling the distance to the closest service point are omitted. This can be done safely, since the variables for doubling and halving the distance to the closest mailbox variables and the variables for doubling and halving the distance to the closest service point are measuring approximately the same. Table 7 shows the regression results of the conditional logit model after mitigating the (multi)collinearity problem. This model is used for determination of the WTP or WTA of the different attributes. However, first the goodness-of-fit of the model is determined.
VIII.B. The goodness-of-fit of the model
The goodness-of-fit measure indicates the accuracy with which the model approximates the observed data (Verbeek, 2012). The goodness-of-fit measure for a conditional logit model can be calculated with the McFaddenR2:
McFaddenR2= 1 − log 𝐿1 log 𝐿0
(6)
Table 7: The conditional logit regression results
of 1 means that the estimated probability corresponds exactly to the observed values. However in practice, the McFaddenR2 is usually far below unity. Sarkar et al. (2015) argue that a
McFaddenR2 between 0.2 and 0.4 can be considered as highly satisfactory. Log likelihood values
cannot be compared between two different data sets, and neither can the McFaddenR2. However,
when comparing two models for the same data, the model with the greater likelihood will have a higher McFaddenR2. The McFaddenR2 for this model is 0.2832, which indicates a good model fit.
An alternative way to measure the goodness-of-fit is comparing the model’s predictions of a respondent choosing specific alternatives and the actual observations. The predicted probabilities of the model can be calculated with formula (5). This formula estimates the predicted probability of alternative j by taking the exponent of the expected utility an individual gets from choosing alternative j and divides this by the sum of the exponents of the expected utilities of from all possible alternatives. The actual observations are based on the respondents’ actual choices. Table 8 displays the predicted probabilities and the actual observations of choosing alternative 1 and 2 respectively in the first seven choice sets of the survey (see Appendix B). The table shows that the model as on average a good fit. The model miscalculates the probability on average with 6 percentage points.
VIII.C. Interpreting the regression results
For the price, the delivery time, the delivery frequency and doubling the distance to the closest mailbox, the estimated coefficients take the expected sign. An increase in the price of a letter
Independent variable Coefficient
Price -8.497*** (0.734) Delivery time -0.035*** (0.003) Delivery Frequency 0.054 (0.102) The distance to the closest mailbox
Halving the distance Doubling the distance
-0.274 (0.174) -0.282 (0.145)
Alternative-specific constant for option 1 -0.253**
Table 8: Predicted and observed outcomes for the first seven choice sets
Choice set Alternative Prediction of choosing this alternative (%) Observation of choosing this alternative (%) 1 1 84 72 2 16 28 2 1 13 4 2 87 96 3 1 35 29 2 65 71 4 1 6 8 2 94 92 5 1 19 10 2 81 90 6 1 5 7 2 95 93 7 1 9 7 2 91 93
decreases the utility individuals get from sending a letter. Also increasing the time it takes to deliver a letter and doubling the distance to the closest mailbox decreases the utility individuals get from sending a letter. Moreover, increasing the delivery frequency increases the expected utility. The coefficient for halving the distance to the closest mailbox takes the opposite sign as expected: halving the distance to the closest mailbox decreases the utility individuals get from sending a letter. However, the coefficient for halving the distance to the closest mailbox is insignificantly different from zero, meaning one cannot be sure that the value of this coefficient is not really zero. The null hypotheses that halving the distance to the closest mailbox equals doubling the distance to the closest mailbox is highly insignificant at a the 0.05 level, which implies that it might be that there is no difference between both coefficients at all.
Table 9: The WTP/WTA in euros Delivery time (per extra 24
hours)
Delivery frequency (per extra day)
Halving distance mailbox Doubling distance mailbox WTP or WTA (in euros) -0.10 0.01 -0.03 -0.03
Table 7 shows that the alternative-specific constant for choosing alternative 1 equals -0.252. From this can be concluded that on average one obtains more utility from choosing alternative 2 relative to alternative 1.
VIII.D. The estimated willingness to pay or willingness to accept
Based on this conditional logit model, the WTP or WTA for the different attributes can be calculated using (2). Table 9 displays the WTP or WTA for the attributes. Positive outcomes imply a WTP and negative outcomes imply a WTA. The graph shows that individuals want 10 cents compensation per letter when the time to deliver a letter increases by one delivery day (24 hours). Individuals are willing to pay 1 cent extra per letter when the delivery frequency increases by one day. Both halving the distance to the mailbox and doubling the distance to the mailbox is accepted when individuals are compensated with 3 cents per letter.
The WTP or WTA for the attributes is relatively low compared to the current price of sending a letter. It might be that individuals attach a higher value to the price of sending a letter than to the other attributes, which will result in a lower WTP or WTA for the other attributes. To validate this argument, the relative importance of the attributes should be calculated.
VIII.E. The relative importance of the different attributes
When a consumer makes a choice, he attaches a particular value to every attribute. This is called the relative importance of the different attributes (Intomart Gfk, 2011). The relative importance is based on all choices respondents have made within the survey. Halbrendt et al. (1991) define the relative importance in percentages of attribute i with the following formula:
𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑖𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑐𝑒𝑖 = 𝑟𝑎𝑛𝑔𝑒𝑖 ∑𝐼 𝑟𝑎𝑛𝑔𝑒𝑖
𝑖=1
× 100% (7)
Graph 4: The relative importance of the attributes
meaningful to individuals. The delivery time (5%) and the distance to the closest mailbox (6%) seem to be less important for consumers. By designing the USO, the government should take into account which attirbutes consumers find most important for making their decision. By doing so, the USO might be designed in such a way that it increases welfare. Those findings are quite similar to the estimated relative importance by Intomart Gfk (2011). Intomart Gfk (2011) estimates that the price (40%) is most important to consumers, followed by the delivery time (25%). The big difference is the valuation of the delivery time. Where Intomart Gfk (2011) estimates a relative importance of 24%, this model can only give a weight of 5% to the importance of the delivery frequency. The fact that individuals indeed believe that the price is the most important attribute of the USO confirms the assumption that the WTP or WTA for the different attributes is low due to the high value individuals attach to the price of sending a letter.
VIII.F. The differences between demographic subcategories
In this Section, the model is estimated for different demographic subcategories. There are classifications made based on gender, age, degree of urbanization, the size of the household and disposable household income. The results of the conditional logit models as well as the relative importance for the subgroups can be found in Appendix A.
Gender
The models are estimated for both males and females. The price coefficient is higher (less negative) for females than for males. Hence, females get less utility from a price decrease than males, and thus attach less value to the price of a letter. Doubling the distance to the closest mailbox decreases the obtained utility more for females than for males. Females therefore believe that the distance to a mailbox is more important than males do. Since the marginal utility of income (the negative price coefficient) is lower for females, one would expect slightly higher WTP (WTA) for the attributes that positively (negatively) influence the utility an individual receives. Especially for doubling the distance to a mailbox one would expect a lower willingness to pay for females. Table 10 displays the WTP or WTA in euros by gender. Indeed, the
54%
35%
5% 6%
Table 10: The WTP/WTA by gender Subcategory Delivery time
(per extra 24 hours)
Delivery frequency (per extra day)
Halving distance mailbox Doubling distance mailbox Male -0.08 0.00 -0.02 0.01 Female -0.12 0.01 -0.05 -0.07
Table 11: The WTP/WTA by age Subcategory Delivery time
(per extra 24 hours)
Delivery frequency (per extra day)
Halving distance mailbox Doubling distance mailbox < 30 -0.10 0.00 -0.02 -0.04 30 – 50 -0.13 0.02 -0.04 -0.03 > 50 -0.04 0.00 -0.02 -0.04
compensation females want to accept a doubled distance to the closest mailbox is relatively large.
An explanation for this finding could be that females do value convenience higher than males do. This is in line with the study from Moschis et al. (2000), who estimate buying behavior of older persons by an online device and find that older females believe that convenience is more important and the price is less important than older males do.
Age
Within the survey an individual could choose between five subcategories for age. Since some of those categories were underrepresented, some subcategories are combined. The model is estimated for the following three subcategories: younger than 30, between 30 and 50, and older than 50.
Table 12: The WTP/WTA by degree of urbanization
Subcategory Delivery time
(per extra 24 hours) Delivery frequency (per extra day) Halving distance mailbox Doubling distance mailbox Very strong -0.10 0.00 -0.04 -0.03 Strong -0.08 0.00 -0.01 -0.03 Moderate - rural -0.15 0.03 -0.04 -0.08
Table 13: The WTP/WTA by household size
Subcategory Delivery time
(per extra 24 hours) Delivery frequency (per extra day) Halving distance mailbox Doubling distance mailbox 1 person -0.14 0.01 -0.08 -0.06 2 persons -0.07 0.00 -0.04 -0.05 > 3 persons -0.10 0.02 0.01 -0.02
increase in the delivery time differs notable. The minimum amount older individuals want to receive in order to accept an increase in the delivery time of one delivery day is lower for younger individuals. An explanation for this could be that the younger generation is more involved in the digital age, and is used to receive and deliver information faster.
Degree of urbanization
Table 12 displays the WTP or WTA by degree of urbanization. For the degree of urbanization the last three subcategories are combined, because those contained only a few respondents. The model is estimated for a very strong degree of urbanization, a strong degree of urbanization and a moderate – rural degree of urbanization. Individuals living in areas with a moderate – rural degree of urbanization get relatively less utility from a price decrease. Doubling the distance to the closest mailbox could have more consequences for individuals living in rural areas, since the dispersion of the mailboxes in rural areas is smaller than in the city. It is therefore not surprising that individuals living in an area with a moderate – rural degree of urbanization want a higher compensation for doubling the distance to the closest mailbox.
Size of the household
Table 14: The WTP by disposable income
Subcategory Delivery time
(per extra 24 hours) Delivery frequency (per extra day) Halving distance mailbox Doubling distance mailbox < 20,000 -0.16 -0.08 -0.13 -0.13 20,000 – 30,000 -0.10 0.03 -0.06 -0.08 30,000 – 40,000 -0.11 0.00 -0.08 -0.05 > 40,000 -0.09 0.01 -0.02 -0.03
delivery time increases by 24 hours (one delivery day). It is reasonable that larger households send more letters. They might heap mail so that they have to go less often to the mailbox, which means that they might care less about the delivery time. Moreover, households with children might send more time-independent letters, for example invitations for the birthday party of the child.
Disposable income
Finally, the model is estimated for different groups based on disposable income. The model is estimated for the following groups: a disposable income smaller than 20,000 euros, between 20,000 and 30,000 euros, between 30,000 and 40,000 euros, and larger than 40,000 euros. From Table 23 (Appendix A) can be concluded that the group with the lowest income does not receive the most disutility from a price increase. The group with the lowest income does approximately believe that all features are equally important. Due to the lower importance of the price and the relatively higher importance of other attributes, the WTP (WTA) for the lowest income group is higher for attributes that positively (negatively) influence the utility. Table 14 displays the WTP or WTA by disposable income. The finding that increasing the price does not give the most disutility to lower income households is not in line with the expectation. A price increase relative to income is higher for lower incomes, so that it is expected that lower income households are more price sensitive. However, since individuals on average only send a small amount of letters per year, which count only for a very small fraction of the individual’s budget, an increase in the stamp price does not restrict access to the USO for individuals with lower incomes.
VIII.G. The model with and without the ‘no-choice’ alternative
Graph 5: The second option based on price
Graph 6: The second option based on delivery time2
price and the delivery time is unchanged or the price is lowered by 5 cents delivery time has increased with at least one delivery day. The last point is in line with the calculated WTP or WTA in Section VIII.D.: individuals are only willing to accept an increase of the delivery time by one delivery day if they are compensated with a 10 cents price decrease. Moreover, all other options where the ‘no-choice’ alternative is chosen less often, contain at least one alternative that contains the current or a lower price. From this can be concluded that individuals indeed find the price very important and start to doubt whether to send mail or not when the price increases above the current level. After choosing the ‘no-choice’ alternative, the respondents got an extra question where they had to indicate which of the two alternatives they would chose if they did could not the ‘no-option’ alternative. From graph 5 and 6 can be concluded that individuals choose most of the time the alternative with the lowest price, even when the delivery time is the slowest of the two alternatives. The exception is choice set 13, where the respondents are almost indifferent between choosing alternative 1 and 2. The lowest price alternative in this choice set is combined with other attribute levels that are less favorable compared to the current situation. This indicates that respondents might be willing to pay more to prevent all other alternatives from changing in an unfavorable way.
Table 15 displays the model with and without the choice’ alternative. Since the ‘no-choice’ alternative does not include any of the attribute levels in the design, it can be modeled in the dataset by a series of zeros. However by doing this, the zeros become a fixed part of the utility (Haaijer et al., 2001). Using alternative-specific constants in the model can help to scale
2
The second option for the 10th choice set is excluded, because both alternative 1 and alternative 2
9% 11% 11% 19% 10% 52%
91% 89% 89% 81% 90%
48%
2 5 9 10 12 13
Alternative with the highest price Alternative with the lowest price
91% 11% 11% 90% 52% 9% 89% 89% 10% 48% 2 5 9 12 13
Table 15: The model with and without the ‘no-choice’ alternative
the utilities between the various choice sets. It does not matter which alternative is chosen as the base case. The model estimates do not change by choosing a different base case. Only the included alternative-specific constants change since these are relative to the base case. In the model with a ‘no-choice’ option, the ‘no-choice’ alternative is chosen as the base alternative. Therefore, an alternative-specific constant for both choosing alternative 1 and 2 respectively are added in the model. The constant for alternative 1 equals one when alternative 1 is chosen in a choice set, and zero otherwise. The constant for alternative 2 equals one when alternative 2 is chosen, and zero otherwise. When alternative 3, the ‘no-choice’ alternative, is chosen the constants for both alternative 1 and 2 equal zero. Since both the coefficient for alternative 1 as the coefficient for alternative 2 is higher than zero, choosing alternative 1 or 2 will increase an individual’s utility relative to the base case.
An increase in the price decreases individuals’ utility more heavily in the without a ‘no-choice’ alternative model. This is not what one expects, since the ‘no-‘no-choice’ alternative is always chosen in a choice-set where the price is too high. Since the ‘no-choice’ alternative is chosen regularly, including the ‘no-choice’ alternative implies that the number of choices for a ‘real’ profile decreases. It might be that due to this decrease not enough information is available to make a valid model.
Independent variable Without ‘no–choice’ Coefficient With ‘no-choice’ Coefficient Price -8.497*** (0.734) -6.879*** (0.528) Delivery time -0.035*** (0.003) -0.031*** (0.003) Delivery Frequency 0.054 (0.102) -0.033 (0.058) The distance to the closest mailbox
Halving the distance Doubling the distance
-0.274 (0.174) -0.282 (0.145) -0.522*** (0.115) -0.255 (0.136) Alternative-specific constant for option 1 -0.253**
(0.080)
7.554*** (0.448) Alternative-specific constant for option 2 --
