Why and how would you segment tourism markets?

Answer 1

Market Segmentation in Tourism

Every tourist is different. Every tourist feels attracted by different tourist destinations, likes to engage in different activities while on vacation, makes use of different entertainment facilities and complains about different aspects of their vacation. While all tourists are different, some are more similar to each other than others: many people enjoy culture tourism, many tourists like to ski during their

winter holiday and many tourists require entertainment facilities for children at the

destination. Acknowledging that every tourist is different and that tourism industry

cannot possibly cater for each individual separately forms the basis of market segmentation.

Smith (1956) introduces the concept of market segmentation as a strategy. He states that "Market segmentation […] consists of viewing a heterogeneous market (one characterized by divergent demand) as a number of smaller homogeneous markets". When segmenting a market, groups of individuals are developed which are similar with respect to some personal characteristic. The particular personal characteristic with respect to which similarity is explored is the segmentation criterion or segmentation base. Segmentation criteria / bases can be socio-Demographics (for instance, old versus young tourists), behavioral variables (skiers versus sightseers) or psychographic variables (tourists motivated by rest and relation versus those motivated by action and challenges).

Market segmentation can be applied by any unit operating in tourism industry: hotels, travel agencies, tourist attractions, restaurants, and local charities. A tourism destination is the entity for which market segmentation is conducted.

The benefit of market segmentation lies in a tourist destination being able to specialize on the needs of a particular group and become the best in catering for this group. In doing so the destination gains a competitive advantage because (1) competition can be reduced from the global market to tourism destinations

specializing on the same segment (e.g., all ecotourism destinations), (2) efforts can be

focused on improving the product in a specific way rather than trying to provide all things to people at high cost (e.g., a family destination is unlikely to need extensive nightlife options), (3) marketing efforts can be focused by developing the most effective message for the segment targeted (e.g., a sun and fun message for young tourists traveling with friends) and by communicating the message through the most effective communication channel for the segment (e.g., in national geographic or other nature magazines for ecotourists), and finally, (4) tourist experiencing a vacation at a destination that suits their special needs are likely to be more satisfied with their stay and, consequently, revisit and advertise the destination among like-minded friends. Or, as Smith stated in his seminal paper (1956): "market segmentation

tends to produce depth of market position in the segments that are effectively defined

and penetrated. The [organization that] employs market segmentation strived to secure one or more wedge-shaped pieced [of the market cake]."

The examples above demonstrate that the expected outcome from market segmentation is competitive advantage. Consequently, the aim of the actual segmentation task is to Group tourists in the way that is of most managerial value. In order for a segment to be managerially useful a number of requirements should be fulfilled:

1. The segment should be distinct meaning that members of one segment should be

as similar as possible to each other and as different as possible from other

segments.

2. The segment should match the strengths of the tourism destination.

3. The segment should be identifiable. While female travelers can be identified very easily, identification of those visitors who are motivated by rest and relaxation may not be as simple.

4. The segment should be reachable in order to enable destination management to communicate effectively. For instance, surf tourists are likely to read surf magazines which could be used to advertise the destination.

5. A segment should be suitable in size. This does not necessarily imply that a bigger

segment is better. A tourism destination may choose to target a small niche segment that represents a large enough market for the particular destination and has the advantage of having very distinct requirements.

The above criteria for the usefulness of segments have to be considered when one or more of many possible segments are chosen for active targeting.

Market segments can be derived in many different ways. All segmentation approaches can be classified as being either a priori (commonsense) segmentation

approaches (Dolnicar 2004a ; Mazanec 2000) or a posteriori (post hoc, data-driven)

segmentation approaches (Dolnicar 2004a; Mazanec 2000; Myers and Tauber 1977).

The names are indicative of the nature of these two approaches. In the first case

destination management is aware of the segmentation criterion that will produce a

potentially useful grouping (commonsense) in advance, before the analysis is

undertaken (a priori). In the second case destination management relies on the

analysis of the data (data-driven) to gain insight into the market structure and decides

after the analysis (a posteriori, post hoc) which segmentation base or grouping is the

most suitable one.

COMMONSENSE SEGMENTATION

In the case of commonsense segmentation destination management informs

the data analyst about the personal characteristics believed to be most relevant for

splitting tourists into segments. The choice of personal characteristics can be driven

by experience with the local market or practical considerations. Most tourism

destinations, for instance, use country of origin as a

segmentation criterion. They

profile tourists from different countries of origin and develop customized marketing

strategies for each country. Even if this method is not the most sophisticated, country

of origin segmentation offers major practical advantages of taking such an approach:

most countries of origins speak a different language which requires customized

messages to be developed anyway, each country of origin has different media

channels.

Commonsense segmentation has a long history in tourism research with many

authors referring to it as profiling. As early as 1970 tourism researchers did

investigate systematic differences between commonsense segments with a publication

titled "Study Shows Older People Travel More and Go Farther" (author unknown)

appearing in the Journal of Travel Research. A vast amount of commonsense

segmentation studies have been published since and are continuing to be published.

4

Dolnicar (2004a) concludes that commonsense segmentation remains the most

common form of segmentation study conducted in academic (and most likely also

industry) tourism research: 53 percent of all segmentation studies published in the last

15 years in the main outlet for tourism segmentation research (the Journal of Travel

Research) were commonsense segmentation studies. Recent examples include

Kashyap and Bojanic (2000), who split respondents into business and leisure tourists

and investigates differences in value, quality and price perceptions, Israeli (2002),

who compares destination images of disabled and not disabled tourists, Klemm

(2002), who profiles in detail one particular ethnic minority in the UK with respect to

their vacation preferences, and McKercher (2002), who compares tourists who spend

their main vacation at a destination with those who only stop on their way through.

Other commonsense studies are discussed in Dolnicar (2005).

Typical examples of areas in which commonsense segmentation approaches

are regularly used include profiling respondents based on their country of origin,

profiling certain kinds of tourists (e.g., culture tourists, ecotourists) and profiling

tourists who spend a large amount of money at the destination (big spenders). In fact,

geographical segmentation such as grouping tourists by the country of origin were

among the first segmentation schemes to be used (Haley 1968).

A step by step outline of commonsense segmentation is given in Figure 1.

Commonsense segmentation consists of four distinct steps: first, a segmentation

criterion has to be chosen. For example, destination management may want to attract

tourists from Australia. Country of origin represents the segmentation criterion in this

case. In Step 2 all Australian tourist become members of segment 1 and all other

tourists (or a more specific subset of other countries of origin) become segment 2

members.

Figure 1: Steps in commonsense segmentation

Analyses of variance, t-tests, Chi-square tests or binary logistic regressions

represent suitable techniques to test whether Australian tourists are significantly

different from other tourists in Step 3. Note that the kind of test used depends on the

number of characteristics that are tested and the scale of the variables. If many

Step 1: Selection of the segmentation criterion

(e.g. age, gender, $ spent, country of origin)

Step 2: Grouping respondents into segments by assigning each

respondent to the respective segment

Step 3: Profiling of segments by identifying in which personal

characteristics segments differ significantly

Step 4: Managerial assessment of the usefulness of the market

segments (and formulation of targeted marketing activities).

5

characteristics are available in the data set the computation of independent tests for

each characteristic overestimates the significance. Therefore, a Bonferroni correction

is necessary on each p-value to account for this systematic overestimation, or

researchers must choose methods, such as binary logistic regression, which

automatically account for potential interaction effects between variables. The test

chosen in Step 3 also needs to be appropriate for the scale of the data. If the profile

regarding nominal (e.g., gender, type of vacation), binary (e.g., prior experience with

the destination on a yes - no scale) or ordinal (e.g., income groups, level of expressed

satisfaction) characteristics is tested, analysis of variance and t-tests are not the

appropriate tests as they assume metric, normally distributed data. For some ordinal

data this can be shown, but should be demonstrated before a test for metric data is

applied.

Finally, in Step 4 destination management has to evaluate whether or not the

commonsense segment of interest (e.g., Australian tourists) does represent an

attractive market segment. This evaluation is made using the criteria outlined above.

If the segment is attractive, destination management can proceed to customize the

service to best suit the segment needs and develop targeted marketing activities which

will enable most effective communication with the segment.

DATA-DRIVEN SEGMENTATION

Data-driven segmentation studies do not have as long a history as

commonsense segmentation studies do. Haley (1968) introduces data-driven market

segmentation to the field of marketing. While acknowledging the value of geographic

and socio-demographic information about consumers, Haley criticizes commonsense

approaches as being merely descriptive rather than being based on the actual cause of

difference between individuals and instead proposed to use information about benefits

consumers seek to form market segments. This approach requires groups of

consumers to be formed on the basis of more than one characteristic and,

consequently requiring different statistical techniques to be used. As Haley (p. 32)

states,"All of these methods relate the ratings of each respondent to those of every

other respondent and then seek clusters of individuals with similar rating patterns."

About one decade after Haley has proposed data-driven market segmentation,

tourism researchers adopted the method and published the first data-driven

segmentation studies in tourism (Calantone, Schewe and Allen 1980; Goodrich 1980;

Crask 1981; Mazanec 1984). A large number for data-driven segmentation studies has

been published since with recent examples including work by Bieger and Lässer

(2002), who construct data-driven segments among Swiss population on the basis of

8

travel motivations. This study represents data-driven segmentation in its pure form

because no pre-selection of respondents takes place before the segmentation study is

conducted. Contrarily Hsu and Lee (2002) use a subset of the tourist population as a

starting point: only motor coach travelers. Among motor coach travelers they further

segment tourists in a data-driven manner by exploring systematic differences in 55

motor coach selection attributes. Further examples are discussed in Dolnicar (2005).

The large number of data-driven segmentation studies published in the past

two decades has led to a number of reviews of segmentation studies in tourism, some

of which focus more on content, some on methodology.

Frochot and Morrison (2000) review benefit segmentation studies in tourism.

They conclude that benefit segmentation leads to valuable insights in tourism research

in the past, but recommend the following improvements: careful development of the

benefit statements used as the segmentation base (some benefits are generic, but many

are specific to the destination under study), informed choice of the timing (asking

tourists before their vacation is less biased by the actual vacation experience), conduct

benefit segmentation studies regularly to account for market dynamics and conduct

them separately for different seasons.

Dolnicar (2002), based on a subset of studies reviewer by Baumann (2000),

analyzes methodological aspects of data-driven segmentation studies in tourism

concluding that only a small number of the available algorithms is used by tourism

researchers who prefer either the hierarchical Ward's algorithm or the k-means

partitioning algorithm. Dolnicar also identifies a number of problematic

methodological standards that have developed in data-driven segmentation in tourism.

To avoid data-driven segmentation studies that are of limited scientific and practical

value it is important for data analysts and users to be aware of a number of basic

principles upon which data-driven segmentation is based. These foundations are

described in detail in the following section.

Foundations of data-driven market segmentation

Foundation 1: Market segmentation is an exploratory process. Many statistical

techniques enable researchers to conduct test that provide one single correct answer

for a research question. For instance, if an analysis of variance is conducted on

destination brand image data, the test results inform the researcher whether or not

there is a significant difference in the way respondents from different countries of

origin perceive a destination. This test result is exactly the same, no matter how often

the analysis is repeated. This method is not the case in data-driven market

segmentation. Market segmentation is a process of discovery, an exploratory process.

Aldenderfer and Blashfield (1984) refer to clustering, the algorithm typically used in

data-driven market segmentation in tourism, as "little more than plausible algorithms

that can be used to create clusters of cases." Each algorithm produces a different

grouping and even repeated computations of one algorithm will not lead to the same

segments. This point is very important to both researchers conducting data-driven

market segmentation and managers using segmentation results. As a consequence, the

choice of the segmentation algorithm and the parameters of the algorithm can and do

have a major impact on the results. Data analysts must be aware of the fact that their

selection of a data-driven segmentation procedure is "structure-imposing"

(Aldenderfer and Blashfiled 1984) and that segmentation results from one algorithm

9

are unlike to have revealed the one and only true segmentation solution for any given

data set.

Foundation 2: Market segments rarely occur naturally. The exploratory nature

of market segmentation leads to a question which has rarely been discussed in

marketing or tourism research: are market segments real and is the data analyst's aim

to identify such naturally occurring segment or are market segments an artificial

construction of groups for a particular purpose. Different authors take distinctly

different positions on the matter. The seminal market structure analysis and market

segmentation studies (Frank, Massy, and Wind 1972; Myers and Tauber 1977) imply

that the aim of market segmentation is to find natural groupings. More recently,

Mazanec (1997) and Wedel and Kamakura (1998) state explicitly that market

segmentation typically means that artificial groupings of individuals are constructed.

Empirically both cases can occur and represent to extremes on the continuum

of highly structured to not structured data sets. These two extreme options have been

referred to as "true clustering" and "constructive clustering" by Dolnicar and Leisch

(2001).

Conducting data-driven market segmentation

A data-driven segmentation study contains all the components of a

commonsense segmentation study. The way in which respondents are grouped is the only difference between the commonsense and the data-driven approach: in commonsense segmentation one criterion is selected which usually is one single variable such as age or gender or high versus low levels of tourism spending. In data driven segmentation a number of variables which ask respondents about different aspects of the same construct (e.g., a list of travel motives, a list of vacation activities) form the basis of segmentation and a procedure - in tourism research typically a

clustering algorithm - is used to assign respondents to segments based on the

similarity relationships between respondents. Figure 3 illustrates the additional steps

needed for data-driven segmentation as steps 2a-2c.

Figure 3: Steps in data-driven segmentation

In step 2a the data analyst selects one or more segmentation algorithms. The

predominant algorithms used in tourism research are k-means clustering and Ward's

clustering. Ward's clustering is one form of hierarchical clustering procedures.

Hierarchical - more precisely agglomerative hierarchical - clustering procedures

determine the similarity between each pair of two respondents and then choose which

two respondents are most similar and places them into a group. This process is

repeated until all respondents are in one single group. The disadvantage of

hierarchical algorithms is that they require computations of all pair-wise distances at

each step which can be a limiting factor when working with very large data sets. The

second most frequently used data-driven segmentation algorithm in tourism research

is k-means clustering. K-means clustering is an algorithm from the family of

partitioning techniques. This technique does not require the computation of all pair

wise distances. Instead the number of segments to be derived has to be stated in

advance. Random points drawn from the data set represent these segments. In each

Step 1: Selection of the segmentation base

(e.g. travel motivations, vacation activities)

Step 2: Grouping of respondents

Step 3: Profiling (external validation) of segments by identifying

in which personal characteristics segments differ significantly

Step 3: Managerial assessment of the usefulness of the market

segments (and formulation of targeted marketing activities).

Step 2a: Selection of segmentation algorithm(s)

Step 2b: Stability analysis

Step 2c: Computation of final segmentation solution

13

step of the iterative procedure the distance between each of the respondents and the

"segment representatives" is computed and the respondent is assigned to the segment

that best represents his or her responses. For example, if a five segment solution is

computed, only five distance computations have to be calculated using partitioning

techniques as opposed to as many distance computations as there are respondents in

the sample when using hierarchical techniques.

Although k-means and Ward's clustering dominate data-driven segmentation

studies in tourism, a large number of other algorithms is available to the data analyst:

a wide range of alternative clustering algorithms (Everitt, Landau, and Leese 2001),

neural networks (e.g., Mazanec 1992; Dolnicar 2002), bagged clustering (e.g.,

Dolnicar and Leisch 2003), latent class analysis (e.g., Van der Ark and Richards

2006), and finite mixture models (Wedel and Kamakura 1998).

When selecting an algorithm the data analyst should be aware of the

advantages and disadvantages of the alternative methods and in particular the way in

which they are known to impose structure on data. Most clustering algorithms allow

the data analyst to define which distance measure should be used. Again, a large

number of alternative distance measures are available. The data analyst has the

responsibility to select a distance measure suitable for the data scale. For instance,

metric and binary data can be analyzed using Euclidean distance. This choice is not

necessarily the case for ordinal data. For a detailed discussion of alternative distance

measures see Everitt, Landau, and Leese (2001).

Another point that should be noted while discussing the selection of a suitable

clustering algorithm is the term "factor-cluster segmentation" which appears to have

developed in tourism research. Researchers using this approach typically select a large

number of items, conduct factor analysis to reduce a large number of items to a

smaller number of factors and subsequently use factor scores as the basis for

segmentation. This approach has two effects: (1) the original items are actually not

used to segment. Consequently, resulting segments cannot be interpreted using the

original items, because they emerged from a heavily transformed data space. (3)

Factor analyses typically explain between 50 and 60 percent of the information

contained in the original items. Conducting factor analysis before clustering essentially means that 40 to 50 percent of information is lost. Direct clustering of original items is therefore preferable if the aim of the segmentation study is to develop segments based on the questions asked in the survey (benefits, motivations, and behavior). Sheppard (1996) compares cluster analysis with factor-cluster analysis methods and concludes that factor-cluster analysis is not suitable if the study's aim is to examine heterogeneity among tourists; factor analysis may be a valuable approach for the development of instruments for the entire population assuming homogeneity.

Arabie and Hubert (1994) are less diplomatic by stating that "`tandem´ clustering is an

outmoded and statistically insupportable practice" because the nature of the data is

changed dramatically through a factor analytic transformation before segments are explored.

Data analysts also should keep in mind that the number of variables that can be analyzed with a sample of a certain size is limited. Although there are no specific rules for non-parametric procedures, a rule of thumb proposed by Formann (1984) provides some helpful guidance: for the case of binary data (yes no questions) the minimal sample size should include no less than 2k cases (k = number of variables), preferably 5*2k of respondents.

Finally, the most unresolved question in market segmentation remains how to

select the number of segments that best represents the data or most suitably splits

respondents into managerially useful segments. A large number of heuristics exist to assess the optimal number of clusters but comparative studies show that no single one

of these indices is superior to the others. If the data is well structured, the correct number of clusters will be identified by most heuristic procedures. If the data is not well structured, which is typically the case in the Social Sciences, heuristics are not helpful to the data analyst. The approach the author finds most useful is based on the above mentioned concepts of segmentation (Figure 2) where data structure is the driving force and stability is the criterion. To determine the number of clusters using the stability criterion, a number of repeated computations are conducted and the agreement across alternative solutions is assessed. The number of clusters that leads to the most stable results over repeated computations wins.

OTHER APPROACHES TO CREATING MARKET SEGMENTS

Although the majority of market segmentation studies in tourism are typically

classified as being commonsense segmentation studies or data-driven segmentation

studies, combinations of both approaches are possible and may represent a useful

alternative for tourism managers to explore potentially attractive target segment for

their purposes. Dolnicar (2004a) gives an overview of such alternative segmentation

approaches. The classification of these approaches (left side of Figure 5) assumes that a two-stage process is taken where the data analyst first creates a commonsense or a

data-driven segmentation and then continues with an additional analysis afterwards.

For instance, destination management could first split tourists based on their country

of origin and then in the second step either (1) search for distinct groups differing in

their travel motivations (which would represent a Concept 5 segmentation) or (2) split respondents into first time and repeat visitors (Concept 3).

Figure 5: A systematics of market segmentation approaches (modified from Dolnicar, 2004a)

Which group is described first?

A subgroup of the total tourist

population determined by data-driven

segmentation on multivariate basis

A subgroup of the total tourist

population determined by data-driven

segmentation on multivariate basis

CONCEPT 1

= commonsense

= a priori segmentation

CONCEPT 2

= data-driven

= a posteriori

= post-hoc segmentation

Which groups are explored next?

A subgroup determined by an a priori

or common sense criterion

A subgroup determined by data-driven

segmentation on multivariate basis

CONCEPT 3

commonsense /

commonsense

segmentation

CONCEPT 4

data driven /

commonsense

segmentation

CONCEPT 5

commonsense /

data-driven

segmentation

CONCEPT 6

data-driven /

data-driven

segmentation

CONCEPT 7

Types of tourist

emerge as cells from a

cross-tabulation of two

independently

conducted

segmentation studies

which could be

commonsense or

data-driven.multaneous

Of course, managers may be interested in exploring combinations of simultaneously constructed market segments. Combination methods are done by conducting two independent segmentation studies based on different segmentation bases and then simply cross-tabulating the resulting groups. For instance, destination management could construct segments based on motives and segments based on vacation activities independently based on the same data set and then investigate whether these two segmentations are associated and result in interesting vacation

types. One example for such a simultaneous segmentation study is provided by

Dolnicar and Mazanec (2000).

Note that while such alternative segmentation approaches are useful in

exploring potentially interesting target segments they can also be used to externally

validate segments. For instance, if country of origin is used as an a priori segmentation criterion, researchers could investigate whether segments of tourists who differ with respect to their tourism motivations are associated with the country of origin grouping.

CONCLUSION

Market segmentation is a strategy any entity in the tourism industry can use to

strengthen their competitive advantage by selecting the most suitable subgroup of

tourists to specialize on and target.

A wide variety of alternative techniques can be used to identify or construct

segments. Approaches range from simple commonsense segmentations (where

tourists are split on the basis of a predefined personal characteristic) to

multidimensional data-driven approaches where a set of tourist characteristics is used

as the basis for grouping. Once tourists are grouped using the correct and most

suitable analytical techniques the resulting segmentation solution has to be assessed

by the users (tourism managers) who will not only evaluate the segmentation solution

per se but also the fit of potentially interesting segments with the strengths of the

tourism destination.

Tourism managers can benefit from market segmentation by using it actively

as a method of market structure analysis. In doing so, they can gain valuable insight

into the market and specific sections of the market and identify the most promising

strategy to gain competitive advantage. Typically such a strategy will not only require

market segmentation, but also product positioning. Both approaches will have to be

evaluated in view of competitors' segmentation and positioning choices to be

successful. Segmentation solutions should be computed regularly to ensure that

current market structure is captured.