Empirical Article

Tourism Economics
2022, Vol. 0(0) 1–24
Modeling censored tourism © The Author(s) 2022
Article reuse guidelines:
expenditures in Turkey with sagepub.com/journals-permissions
DOI: 10.1177/13548166211067191
journals.sagepub.com/home/teu
non-normal and heteroscedastic
errors: An application of the inverse
hyperbolic sine double-hurdle model

Faruk Urak
The Turkish Radio and Television Corporation, Erzurum Directorate, Erzurum, Turkey

Nihat Küçük
Harran University, Şanlıurfa, Turkey

Abdulbaki Bilgiç 
Bilecik Şeyh Edebali University, Bilecik, Turkey

Steven T Yen
National Taiwan University, Taipei, Taiwan

Abstract
Censoring, extreme values, and non-normal distribution are common features in microdata. These
data features can compromise the statistical distributions of the estimators when an appropriate
model is not used. We use an inverse hyperbolic sine (IHS) transformation in the double-hurdle
(DH) model to accommodate extreme values and skewness in censored accommodation spending
among Turkish households. The full-parameterized model nests many restricted specifications
which do not accommodate dependence, heteroscedasticity, and non-normality in the error terms.
Statistical test results support the use of the fully parameterized dependent IHS-DH specification.
The statistically significant correlation between the binary and level decisions of accommodation
precludes the use of a model with a two-step structure. Some of the findings in our study have a
determining or driving force in expressing causality relationship in monthly accommodation
probability and level decisions. The study made also prudent policy recommendations about what
the results might mean to policymakers and stakeholders.

Corresponding author:
Abdulbaki Bilgiç, Bilecik Seyh Edebali University, College of Economics and Administrative Sciences, Department of
Management Information Systems, A Block, 3rd Floor, Gülümbe Campus, 11230-Bilecik, Turkey.
Email: [email protected]
2 Tourism Economics 0(0)

Keywords
Accommodation, double-hurdle model, households, inverse hyperbolic sine, Turkey

Introduction
Aggregate data such as time series often contain little detail about individual expenditures or
characteristics of consumer units such as individuals or households, for which microdata offer a
better alternative. As Alegre and Pou (2004) and Disegna and Osti (2016) stated, the most desirable
feature of microdata at the individual or household level is its compatibility with consumer choice
theory, which is about whether the goods or services in question, such as tourism, are preferred by
the consumer rather than being completely ignored as in aggregate data. Microdata are worth
collecting as they reveal both the complex interactions between individual characteristics, the
nonlinear relationships, and the causality of price and income with the quantity demanded (Blundell,
1988). They also allow uncovering of the determinants of demand such as the number of travels and
level of spending and have the advantage of capturing heterogeneity among consumer units. The
unique features of microdata, are often compromised or neglected in their use. As pointed out by
many (e.g., Pellegrini and Scagnolari, 2019; Rashidi and Koo, 2016; Zhang et al., 2012), individual
or family decisions regarding tourism are multifaceted as they are highly discretionary involving
sequential or simultaneous decisions on participation, choice of destinations, spending level, and
length of stay. Quantitative analysis of tourism requires attention to such innate characteristics.
One prominent issue in the use of microdata relates to the zero observations in consumer choice,
as a result of a binding non-negativity constraint(s) leading to zero value(s) in one or more activities
in the utility function or other non-economic sources of zero activity. A rational consumer may not
consume that good due to budget, time, or non-economic constraints. Consequently, zero obser-
vations result, in the forms of a truncated number of trips, length of stays, or zero spending.
The most widely used estimation methods in investigating tourism expenditures or length of stay
with zero observations include Tobin’s (1958) Tobit (censored regression) model (e.g., Disegna and
Osti, 2016; Kwiatkowski and Konecke, 2017; Park et al., 2019, 2020), Heckman’s (1979) sample
selection model (Alegre et al., 2013; Jang and Ham, 2009; Nicolau and Más, 2005; Rashidi and
Koo, 2016), and other forms of sample selection models such as the single-hurdle (Cragg, 1971,
equations (7) and (9)) and double-hurdle (Cragg, 1971, equation (5) and (6); Blundell and Meghir,
1987) models (Bernini and Fang, 2020; Brida et al., 2012; Brida et al., 2013a, 2013b; Jang and Ham,
2009; Martinez-Garcia et al., 2012; Sun et al., 2015). In applications of the aforementioned selection
models, including a recent study by Bernini and Fang (2020), independence is typically imposed
between the two equations (stochastic processes or error terms), which often facilitates estimation.
Additional statistical procedures include the infrequency-of-purchase and scobit discrete-
continuous models (Sun et al., 2015; Wu et al., 2013). Also proposed was Heckman (1979)
sample selection models with a logit link function for the binary selection outcome (Lin et al., 2020;
Rashidi and Koo, 2016; Wu et al., 2013; Zhang et al., 2012) and specifications with binary logistics
and multinomial logistic regression, in which tourism expenditure was transformed into a binary
variable to take advantage of some features of the discrete utility model to focus specifically on the
selection process between tourism and other goods (Alegre et al., 2011; Saayman and Saayman,
2006). Also in the spirit of endogenous sample selection, Alegre and Cladera (2010) fitted the
switching regression model with two sample regimes, viz, first time and repeat visitors, using the
same set of variables, including mixed metric and categorical regressors, to characterize spending
behavior among individuals. Additional innovative models, such as the random forest model, have
Urak et al. 3

recently gained ground in the literature (Brida et al., 2018), as was an approach to tourist spending at
different points in the underlying distribution (Marrocu et al., 2015; Thrane, 2014). Based on this
myriad of frameworks, researchers have succeeded in formulating empirical models that come
closer to consumer theories by combining both analyses of statistical distributional properties and
economic approaches. Along the lines of nonparametric methods, the conditional median was
considered rather than the mean level, in conditional quantile regression (Gomez-Deniz and Perez-
Rodriguez, 2019; Marrocu et al., 2015; Thrane, 2014). More recently, unconditional quantile
regression was also used to analyze tourist spending that can be generalized to the population
(Perez-Rodriguez and Ledesma-Rodriguez, 2019). As an alternative, the logarithmic transformation
replaces the censoring mechanism that complicates model specification. Jones and Yen (2000)
propose the Box-Cox transformation of the dependent variable for the dependent double-hurdle
model (Blundell and Meghir, 1987) but, as widely discussed in the literature (Carboni, 2012; Lee
and Maddala, 1985) there are inexplicable problems in the model, hindering its use in empirical
applications. Of all these approaches, the extent to which the dependent variable is compatible with
the underlying distribution is neglected. Among the most notable problems in these, forcing an
otherwise non-normal random variable to the normal distribution can cause statistical distortions,
compromise onerous modeling issues, and confound policy implications. The germane issue lies in
the nature of tourism expenditure which is largely skewed by zero values and contains pile-ups of
observations on both tails of the distribution, invalidating normality and likely homoscedasticity of
error terms. To avoid such eminent problems, we consider an alternative specification with the
inverse hyperbolic sine (IHS) transformation (Burbidge et al., 1988) to the censored dependent
variable,1 which is better suited than the logarithmic and Box-Cox transformation in terms of
statistical properties.
Although many tourism studies with microdata have featured a comprehensive model selection
process, the issues of non-normality and unequal variance (heteroscedasticity) have remained
unsolved. As Brida and Scuderi (2013) pointed out, the effort to adopt new econometric models is
commendable for the diverse spectrum of literature that would otherwise emerge in a kind of
“methodological stagnation” where research would produce empirical findings rather than in-
strumental tools for knowledge. This current study can be, therefore, considered as a methodological
contribution since the IHS double-hurdle (DH) model is applied for the first time to the field of
tourism as a proposed solution to the problems that have dominated the investigation of tourism
expenditures, beyond the empirical issue of uncovering the driving forces behind expenditures.
Further, the framework is not only limited to lodging but also applicable in many key non-normal
continuous dependent variables for tourism activities such as transportation, food and beverage,
gifts and souvenirs, entertainment and recreation. By preserving the bivariate relationship between
the binary spending decision and the level of expenditure, the IHS-DH model constitutes a novelty
to the empirical literature in that both non-normality and heteroscedasticity are accommodated.

Literature review
While tourism was seen as a mass phenomenon in the years before the eighties, changes in
motivations, travel patterns, and technologies resulted in a complex structure of complementary
services highlighting a highly segregated market. Therefore, empirical analysts would have to use
sophisticated tools to characterize tourist demand or level of spending, in which new econometric
models compatible with the data features have been successful in providing techniques and an-
alytical approaches to explain tourist spending and better understand its counterfactual situations
(Brida and Scuderi, 2013; Jurdana and Frleta, 2015). Unlike macro data, collecting microdata
4 Tourism Economics 0(0)

requires initial fieldwork and is time-consuming and relatively expensive, creating difficulties in
accessing individual or family spending behavior and socioeconomic factors. Following up on
econometric models suitable for the structure of the data discussed in the previous section, instead, a
mini-portrait of the literature on the existence of silent variants of the causal relationship with
tourism expenditures and classification of these factors are presented in this section.
The literature has grouped the determinants of tourist expenditures into four main categories:
economic, socio-demographic, travel-related, and psychological factors (Bernini et al., 2017; Lin
et al., 2020; Mudarra-Fernández et al., 2019; Thrane, 2015a, 2015b). The relationship of these
factors with tourism expenditure has been reviewed by Sainaghi (2012) and Brida and Scuderi
(2013), as well as numerous analyses with cross-sectional data.
Income is the leading economic factor, followed by other factors such as occupation, property
size (or assets), financial difficulty, duty-free import limits, various licenses, and loyalty cards
(Alegre et al., 2010; Bilgic et al., 2008; Saayman and Saayman, 2009). Even the general health
variable of the individual or household, proxied by health expenditures, has been identified as a
factor in tourism economic constraints (Hong et al., 2005; Hung et al., 2012). The literature
(Marrocu et al., 2015; Park et al., 2020), which defines income as a personal budget constraint that
forms the purchasing capacity of individuals or households, has agreed on a positive causal re-
lationship with tourist expenditures. The more income, the more spending, and vice versa (Brida
et al., 2013a; Lee and Lee, 2017; Middaugh et al., 2013; Parola et al., 2014). On the other hand,
studies that chose the occupation variable as a ruling factor in tourism expenditures concluded that
tourists with a job tend to spend more (Brida et al., 2013b; Di Vaio et al., 2018; Parola et al., 2014).
While socio-demographic factors such as age, education, gender, household size, marital status,
nationality, type of residence, ethnicity, and family history are characterized as the characteristics
that shape tourist expenditures, the causal relationships of age and gender with tourism expenditures
have not settled (Amir et al., 2017; Marksel et al., 2017; Wu et al., 2013). In a U-shaped relationship
between age and expenditure, an additional year would have a lower marginal effect among young
people who reported lower spending than the middle-aged and older people (Brida et al., 2013a; Lee
and Lee, 2017). On the other hand, there are also studies showing that tourism expenditures decrease
with age (Di Vaio et al., 2018; Parola et al., 2014). Such age-related non-overlapping causation is
difficult to generalize because the effect is dependent on distinct travel spending patterns for the
social and political environments that people experience over temporal generations (Jang and Ham,
2009) and on how the variable(s) are classified within themselves by the researcher(s) (Marrocu
et al., 2015). On the other hand, although a significant relationship between gender and tourism
expenditures is rarely reported in the literature, there is no consensus on the direction of the re-
lationship (Gargano and Grasso, 2016; Saayman and Saayman, 2012). Similarly, the empirical
literature has provided imprecise information about the causal relationship between tourism ex-
penditures and other socio-demographic factors such as education, marital status, number of
children, and family type (Brida and Scuderi, 2013; Di Vaio et al., 2018; Hung et al., 2012; Parola
et al., 2014).
The roles of compulsive drivers in tourism expenditures such as length of stay, size, and
composition of the traveler group, travel experience, travel purpose, accommodation, type of travel
and activities in the destination, and the destination itself were identified (Brida and Scuderi, 2013;
Lew and Ng, 2012; Wu et al., 2013; among others). Contrary to findings by Alegre et al. (2011) and
Marrocu et al. (2015), length of stay, which is one of the travel-related traits, has a strong stimulus in
tourism expenditures (Brida et al., 2013a, 2013b, 2014, 2015; Di Vaio et al., 2018; Marksel et al.,
2017; Wang et al., 2014). Similarly, travel party size is among the most frequently researched
determinants of tourist spending. This variable has been found significant in most studies, albeit
Urak et al. 5

with varying signs such as positive, negative, or even a U-shape beyond certain numbers, indicating
a declining marginal effect of an additional person (Alegre et al., 2011; Disegna et al., 2012; Thrane
and Farstad, 2011; Wu et al., 2013). In the literature, causality between tourist expenditures and
other travel-related factors such as travel type, business or pleasure, and travel purpose, does not lag
behind the discussion of previous estimators, presenting a similar argument (Marksel et al., 2017;
Marrocu et al., 2015; Toudert and Bringas-Rabago, 2016). Also, the expectation that tourism
expenditures will naturally increase as travel distance increases has made distance an important
stimulus in tourism expenditure models (Wang et al., 2006; Wu et al., 2013). Such finding is
probably at least related to the selection effect, where the cost of traveling to distant places is
negligible for those with much to spend (Van Loon and Rouwendal, 2017).
Compared to the determinants in the other three groups, less salient factors such as motivation,
satisfaction, and taste are rarely included in the tourism analysis as psychological traits (Aguiló
et al., 2017; Brida et al., 2014; Di Vaio et al., 2018; Domènech et al., 2020) because these variables
pose problems of endogeneity with the residuals of dependent variables in econometric analyses,
presenting challenges in statistical analysis (Bernini and Galli, 2019). Although satisfaction with
different aspects of the destination has a lasting relationship with the amount of money spent (Brida
et al., 2014; Disegna and Osti, 2016; Gargano and Grasso, 2016; Jurdana and Frleta, 2015), Wang
and Davidson (2010) reported that satisfaction does not contribute much to tourist spending.
Analyses of other psychological traits have also produced very striking results; Di Vaio et al. (2018),
for instance, noted that cruise passengers traveling on a super-sized vessel tend to spend signif-
icantly less while onshore compared to other passengers (Doménech et al., 2020).
In addition to all the above triggers, family habits, unfortunately, did not receive the attention
they deserve in tourism expenditure analysis. As is well documented, the social life of the family is
one of the most important factors shaping the consumption decision and the amount spent in
modeling consumer preferences in microeconomics. In this context, this current study strives to
bring novelty to the literature, in addition to the methodological contribution, by including such
factors in the analysis.

Material and method

Econometric model
Zero accommodation expenses occur in many different ways—in cases when families did not travel
throughout the year, in day trips that did not require accommodation, in trips with accommodation
provided by relatives or by their own summer home, and in business travel with accommodation
covered by the employer. Heckman’s sample selection model may not be appropriate for this last
source of zeros, as the model is generally regarded as the “first hurdle dominance model,” which
assumes that only accommodators who make the trip report expenditures (Bernini and Cracolici,
2015, 2016). As noted by Bernini and Cracolici (2015, 2016) and Bernini et al. (2017), Cragg’s
(1971) double-hurdle (DH) model is the most promising alternative that accommodates all four
sources of zero spending. The model assumes that a family desires a positive amount of goods or
services under the “participation decision” in the first stage (Cragg, 1971, equation (5)), and in the
second stage, under the “consumption decision”, favorable conditions must be created for a positive
expenditure to occur (Cragg, 1971, equation (6)). A positive accommodation expense occurs if the
family or its members decide to stay and spend. The DH accommodates zero values at both decision
stages, characterized by the aforementioned two equations or stochastic processes. Zeros in the first
stage are due to not being accommodated, and zeros in the second stage arise from not consuming
6 Tourism Economics 0(0)

due to deliberate decisions or random circumstances such as the reasons mentioned above, es-
pecially in the fourth category. The issue of zero accommodation spending also permits the control
of selection bias and is addressed through hurdle models that can contemporaneously analyze
distinct determinants of spending decision and spending level. Models that do not address both of
these causes of zeros can produce biased, inconsistent, and inefficient parameter estimates (Pudney,
1989).
Econometric approaches to a censored dependent variable include (1) Tobit model (Tobin, 1958)
featuring one censoring mechanism to govern zero and positive outcomes, (2) sample selection
model (Heckman, 1979), or single-hurdle (SH) model (Cragg, 1971, equations (7) and (9)) in which
a binary selection rule dictates the zero-one outcome, with positive level determined by a continuous
stochastic process, and (3) double-hurdle (DH) model (Cragg, 1971, equations (5) and (6); Blundell
and Meghir, 1987) featuring both a binary selection rule to determine the zero-one outcome and an
additional censoring mechanism to govern level which also in itself admits zero values. In this
current study with accommodation expenditure, we are also interested in transforming a random
variable that allows a non-normal spending variable to have an explicit normal distribution. Such
transformation maps the original spending level and its corresponding binary spending decision to a
bivariate normal distribution which allows for contemporaneous interdependence. We also ac-
commodate heteroscedasticity of the two error terms. The joint treatment of both non-normality and
heteroscedasticity is important as failure to accommodate either can lead to biased and inconsistent
estimates (Wooldridge, 2010).
In the same spirit as the conventional DH model (Cragg (1971) and Blundell and Meghir (1987),
the DH model, with an inverse hyperbolic sine (IHS) transformation to the dependent variable,
features one stochastic process z0i α þ ui governing selection and an additional process x0i β þ vi
governing level (which can be censored) such that, for observation i
Iðyi ,θÞ ¼ x0i β þ vi if z0i α þ ui > 0 and x0i β þ vi > 0
(1)
¼0 otherwise

where zi and xi are vectors of explanatory variables explaining the binary decision to spend and level
of spending, respectively, α and β are conformable parameter vectors, and the error terms ðui ,vi Þ,
specific to each equation and occur outside the discretion of the researcher, are assumed to be
distributed as bivariate normal with means (0, 0), standard deviations ð1,σÞ and correlation ρ. The
value of x0i β þ vi amounts to the latent variable, say y∗i , and if this latent value exceeds the reservation
amount previously determined by the consumer unit at the time of booking, the decision will be to
pay and stay overnight(s) at the accommodation facility. Of course, the reservation amount pre-
viously determined by the family or its members will face family income constraints, which plays a
leading role in determining the accommodation price. In this context, two decision mechanisms
should be overcome simultaneously. In the first stage, it is necessary to decide on whether or not to
have a paid accommodation, and in the second stage, it should be decided to pay overnight fees
(including the amount and length of stay), the amount of which can be zero as well. The dependent
variable yi is transformed by the IHS transformation with scaling parameter θ such that (Burbidge
et al., 1988)
h  1=2 i.
Iðyi ,θÞ ¼ log θyi þ θ2 y2i þ 1 θ ¼ sinh1 ðθyi Þ=θ (2)

where, operationally, θ is restricted to positive because the transformation is symmetric about 0.

The literature suggests heteroscedasticity is often inevitable with unit-level data (Maddala, 1983;
Urak et al. 7

Yen and Jones, 1997). When this subtlety is ignored in the prediction of nonlinear models (by
maximum-likelihood techniques), biased and inconsistent parameter estimators can result, by
maximizing an incorrect probability density function (Brown et al., 2015). We accommodate
heteroscedasticity of the error term by parameterizing its standard deviation ðσÞ as an exponential
function of a vector of variables wi (without a constant) with conformable parameter γ such that
 
σ i ¼ σexp w0i γ (3)

The specification (3) is equivalent to one with a constant in w, viz., σ i ¼ expðγ0 þ w0i γÞ, another
popular specification in which γ0 ¼ logðσÞ. An obvious advantage of this exponential specification
is that the transformation guarantees that the standard deviation is both identified and positive with
γ ≠ 0 (Brown et al., 2015).
In practice, breaking down the two decision steps is crucial in modeling accommodation because
the decision to stay in a destination resort is mainly related to social factors, whereas the decision on
the amount of spending reflects budget constraints. For maximum-likelihood estimation, the sample
likelihood function for the heteroscedastic IHS-DH model is
  
x0 β
L ¼ ∏ 1  Φ2 z0i α, i ,ρ
yi ¼0 σi
(  " #) (4)
 1=2 Iðy ,θÞ  x 0
β z 0
α þ ρ Iðy i ,θÞ  x 0
β σi
σ1
i
× ∏ 1 þ θ2 y2i i φ
i
Φ i  1=2
i

yi > 0 σi 1  ρ2

where Φ and φ are the univariate standard normal cumulative and probability distribution functions,
1=2
respectively, Φ2 is the standard bivariate cumulative distribution function, and ð1 þ θ2 y2i Þ is the
Jacobian of the transformation which maps Iðyi ,θÞ to yi . Among the prominent advantages of the
IHS transformation is that it allows both positive and zero values in the dependent variable, ac-
commodates negative values while damping outliers, and is known to handle extreme values more
robustly than the Box-Cox transformation (Burbidge et al., 1988). Another advantage of the IHS
transformation is that it is scale-invariant, and accommodates both zero and negative values in the
latent dependent variable. The transformation also exhibits linearity around the origin and con-
verges to the logarithm on the right tail (Brown et al., 2015; Pence, 2006). Thus, the scaling
parameter, restricted to positive as noted above, governs the ratio of the domain range at which the
function is essentially linear and the ratio at which it approaches logarithm. Another important issue
in endogenous selection models relates to parameter identification, which requires exclusion re-
strictions, viz, variables that can generate non-trivial variation in the selection process but do not
have direct impacts on the outcome variable. In this context, based on the random utility theory
(Pudney, 1989), some of our variables were included in the selection model but excluded from the
level equations as dictated by the demand theory (for example, household factors such as stove and
property category variables were included in the selection equation). Although not required for
identification, household income group variables are used only in the level equation. Drawing on the
SH and DH literature with heteroscedastic errors (e.g., Su and Yen, 1996; Yen and Jones, 1997),
household scale variables available to us are selected to enter the heteroscedasticity equation, such
as the number of children and adults, number of cars, and household income groups.
Imposing parametric restriction ρ = 0 to the model leads to an independent, heteroscedastic IHS-
DH (Su and Yen, 1996), while restriction θ ¼ 0 further reduces the model to the independent DH
model (Cragg, 1971, equations (5) and (6)), and the homoscedastic specifications amount to γ ¼ 0.
8 Tourism Economics 0(0)

All restricted specifications are nested and tests can be done by conventional means for nested
specifications, viz., by likelihood-ratio, Lagrange multiplier, or Wald test (Engle, 1984). Also, the
hypotheses of weak instruments, characterized by zero coefficients of unique instruments in the
selection equation, are nested as well and can be tested accordingly.
To explore the effects of explanatory variables, the probability of a positive observation and
unconditional mean of the accommodation expenditure are, respectively
 
Prðyi > 0Þ ¼ Φ2 z0i α,x0i β σ,ρ (5)

Z∞
 1=2
Eðyi Þ ¼ yi 1 þ θ2 y2i
0
(6)
(  " #)
0 0
Iðy ,θÞ  x 0
β z α þ ρ Iðy i ,θÞ  x β σ i
×σ1
i
φ i
Φ i
 1=2
i
dyi
i
σi 1  ρ2
and the conditional mean is
Eðyi jyi > 0Þ ¼ Eðyi Þ=Prðyi > 0Þ (7)

using (5) and (6). Marginal effects of continuous (binary) explanatory variables are obtained by
differentiating (differencing) equations (5)–(7). Marginal effects are calculated for each observation
and averaged over the sample. For statistical inference, standard errors of average marginal effects
are derived by a mathematical approximation procedure known as the delta method (Wooldridge,
2010).

Data
Data used in this analysis were obtained from household budget surveys conducted between 1
January and 31 December 2018 in Turkey by the Turkish Statistical Institute (TSI). The TSI limited
seasonality effects on spending by replacing nearly a thousand families with similar characteristics
each month. In the 2018 Household Budget Survey, a stratified two-stage cluster sampling method
was used, and the main sampling frame used in the selection of blocks, which are the first stage
sampling unit, is the National Address Database. In the second stage, blocks were created using this
framework, while blocks with probability proportional to the size of the settlement were determined
from urban, and rural areas and villages with a municipal organization. Households from each block
were systematically selected. On the other hand, if the survey could not be conducted with the main
sample households selected according to the sampling techniques in the survey, the “non-answer
form” was filled and the weight coefficients were calculated considering such situation. The re-
sponse rate in the 2018 Household Budget Survey was 76.05% across Turkey (11,828 households
responded to the survey out of 15,552).
The study sample included 11,636 households after removing observations with outliers and
missing data on important variables. Expenditures were collected under three categories: (1) hotels,
motels, and hostels, (2) vacation centers, camping areas, and youth hostels, and (3) other businesses,
including facilities offered by the state. Since expenditures in the last two categories were relatively
rare, we combined them with the first category and sum up expenditures in all three categories to
obtain monthly accommodation expenditure.
Urak et al. 9

Household accommodation expenses may be incurred by an individual(s) in the family for

compelling and non-compelling reasons. Individual(s) in the family may be obliged to travel away
from home every month for work or non-compelling reasons such as health and love for travel,
though we do not have information on such reasons for spending. Socio-demographic and economic
characteristics of the household and household head are presented in Table 1. Mean spending is
20.98 Turkish Lira (TL) per month for the whole sample and 206.74 among the 10.1% of
“consuming” households, which ranges from 20 TL to 1394 TL per month. The low 10% value of
accommodation spending is indeed due to monthly traveling behavior. In this context, it can include
family members or family who travel frequently each month, often business people, business
obligations of civil personnel, or people who routinely check for their medical treatment in
metropolitan cities such as Istanbul, Ankara, and Izmir. We might also note the active role of the
state in shaping accommodation prices in the tourism sector. The state provides accommodation to
its employees and their families for a fee below the market price, within the framework of the social
state concept without an aim to profit from such services. During 2018, there were approximately
12,650 state-owned facilities across the country (2057 in Istanbul, 247 in Ankara, 398 in İzmir, and
262 in Bursa), including coastal provinces such as Antalya (1826 facilities) and Muğla (1307)
(Ministry of Culture and Tourism, 2020). With this mechanism, the state has played an active role in
accommodation services and, as such, intervened in the tourism market by regulating accom-
modation prices.
Explanatory variables were chosen from empirical studies reviewed above. They include income
and wealth dictated by consumer theory, as well as socio-demographic characteristics such as age,
gender, education level, job status, family size, family type, and social habits. There is a long
tradition of including socio-demographic characteristics in empirical demand analysis (Prais and
Houthakker, 1955). Among the key variables, nuclear families, families with a tendency to save, and
families receiving in-kind and cash aid from the state and private institutions are expected to travel
less and also spend less, while families with a credit card(s) and a passenger car(s) are more likely to
travel and also spend more. Household habits such as the presence of persons in the house who
smoke tobacco or cigars, consume alcoholic beverages, eat lunch or dinner outside the house, buy
daily newspapers/magazines, and go to the cinema, theater, sports competitions, and entertainments
were also included in the analysis. Such habits cover a myriad of activities and expand the social
circle of the household and represent a parsimonious alternative in explaining accommodation
spending.
Definitions and sample statistics of all variables are presented in Table 1. About 7.1% were young
households headed by an individual at age 18–30, 44.3% at age 31–50, and 48.5% over age 50. The
proportion of male-headed households was 84.6%, 81.7% of household heads were married, 17.2%
graduated from high school, 14.4% from university, 84.5% had compulsory health insurance, 9.5%
had a green card (by which the government paid health expenses for low-income households). In a
family, salaried, employer, and pensioner-headed have a share of 44.3%, 22.9%, and 21.7%, re-
spectively. About 24.9% of households were tenants, and 49.0% resided in houses heated by a stove.
While 25% of the families live with a maximum monthly income of 2477 TL, those living in the
last quarter of income earn a monthly income of at least 5595 TL. The difference of at least 2.26
times between the lower and upper-income groups points to widespread income inequality among
households in the country. On the other hand, 38.2% were saving,2 18.9% were spouses only,
49.7% were nuclear families with kids, and 27.2% received in-kind or monetary aid from the state.
Almost half of the families in the country use credit cards, while 43% own a passenger car. About
52% of households smoke, while about 6% have a habit of consuming alcohol. While the habit of
eating out was 52%, the habit of buying and reading newspapers and/or magazines remains scant,
10 Tourism Economics 0(0)

Table 1. Variable definitions and sample statistics.

Variable Definition Mean (SD)

Dependent variables
Expenditure Accommodation expenditure per month (TL, Turkish lira) 20.983 (91.611)
Among the consuming (10.1% of sample) 197.724 (187.051)
Household head characteristics: Binary variables (yes = 1, no = 0)
Age ≤30 Age 18–30 (reference) 0.071
Age 31–50 Age 31–50 0.443
Age 51–65 Age 51–65 0.314
Age >65 Age >65 0.171
Male Gender is male 0.846
Married Married (reference) 0.817
Unmarried Never married 0.040
Divorced Divorced 0.044
Widow Widow 0.099
No School Not holding a diploma (reference) 0.113
Elementary Had elementary school education 0.422
Secondary Had secondary school education 0.146
High school Had high school education 0.172
College Had college and higher (master or PhD) education 0.144
Salaried Salaried (paid or full time) or part-time employee 0.443
Employer Employer or full and part-time self-employed 0.229
Retired Retired 0.217
Other job status Other job status (e.g., disabled and/or unable to work due to 0.087
permanent health problems, housework, and compulsory
military service, and other conditions) (reference)
Agricultural job Worked in an agricultural job 0.187
Household characteristics: Binary variables (yes = 1, no = 0)
Compulsory insurance Had compulsory health insurance 0.849
Greene card Health expenses covered by the state 0.095
State aids Received cash or in-kind aids from the government 0.272
Private aids Received cash or in-kind aids from private person and or intuitions 0.176
Apartment Resided in an apartment 0.544
Renter Resided in rental house 0.249
Stove heating House with traditional stove heating 0.490
Spouses only Family consisted of only spouses 0.189
Spouses with kids Family consisted of spouses and kids 0.497
Other family types At least one nuclear family and other members or more than 0.147
one member without a nuclear family (reference)
Savings Made monthly savings 0.382
First quartile income Family income less than 2477.17 TL per month (reference) 0.250
Second quartile income Family income between 2477.17 – 3712.50 TL per month 0.250
Third quartile income Family income between 3712.50– 5595.47 TL per month 0.250
Fourth quartile income Family income greater than 5595.47 TL per month 0.250
Car Owned a passenger car 0.430

(continued)
Urak et al. 11

Table 1. (continued)

Variable Definition Mean (SD)

No real estate No real estate like a detached house, apartment, summer 0.329
house, shop, and hotel (reference)
One real estate Owned only one of the properties described above 0.468
Two real estates Owned only two of the properties described above 0.128
≥ Three real estates Owned three or more of the properties described above 0.075
Credit card(s) habit Had habit of using credit cards 0.498
Tobacco habit Had habit of using tobacco products 0.523
Alcohol habit Had habit of using alcohol products 0.055
Eating out habit Had habit of having food away from home 0.523
Cinema habit Had habit of going to the cinema 0.083
Newspaper habit Had habit of occasional reading of newspapers 0.053
Game habit Had habit of playing game 0.043
Coffee house habit Had habit of frequently going to coffee house 0.278
Bazaar habit Had habit of going to bazaar frequently 0.634
Internet habit Had internet connection at home 0.100
Wage earners Households more than two working individuals 0.070
Residence area Resided in house with living area of >160 sq m 0.056
Household characteristics: Continuous variables
Kids Number of children in family 1.027 (1.295)
Adults Number of adults in family 2.418 (1.025)
Sample size 11,636

Note: Standard deviations, in parentheses, are reported for continuous variables only.

at 5%. This low habit is followed by the cinema habit while shopping at Bazaar is at a high rate of
63%. Meanwhile, Internet use at home has remained low, perhaps due to the prevalence of mobile
phones.

Results and discussion

Variance inflation factors (VIFs) are calculated for regressors in both the selection and level
equations, and all VIFs are well below the rule-of-thumb threshold of 10, suggesting that mul-
ticollinearity is not a problem.3 We first report some specification test results. Vuong’s (1989) non-
nested specification tests were carried out for selection among the IHS single-hurdle (IHS-SH),
double-hurdle (IHS-DH), and Heckman sampling selection (IHS-SS) models, all with inverse IHS
transformation and heteroscedastic errors (see online appendix for the IHS-SH and IHS-SS models).
The IHS-DH and IHS-SH are found to perform equally well by Vuong’s test. The IHS-DH has a
lower Akaike Information Criterion (AIC) (log-likelihood = 10,317.886, AIC = 20,819.77) than
the IHS-SH (log-likelihood = 10,319.316, AIC=20,822.63) model, suggesting the preferability of
the former, which is also more consistent with the two-stage process, following consumer choice
theory (Brida and Scuderi, 2013; Disegna and Osti, 2016; Pudney, 1989).4 Further, based on the
IHS-DH, Wald test results suggest rejection of θ ¼ 0 ðχ 2 ¼ 46:02, df ¼ 1Þ, ρ ¼ 0
ðχ 2 ¼ 4:37, df ¼ 1Þ, and ðρ ¼ 0,θ ¼ 0Þ ðχ 2 ¼ 53:85, df ¼ 2Þ, all with a p-value < .0001, favoring
the fully parameterized dependent and heteroscedastic IHS-DH over the dependent heteroscedastic
12 Tourism Economics 0(0)

DH, independent heteroscedastic IHS-DH, and independent heteroscedastic DH models, respec-

tively. Wald tests also reject homoscedasticity ðvector γ ¼ 0Þ in the full model ðχ 2 ¼ 23:83Þ and all
nested models above, each with df = 5 and a p-value < .001. Likelihood-ratio and Lagrange
multiplier tests produce similar results. In sum, the above tests support dependent errors, IHS
transformation, and heteroscedasticity of error term of the level equation, favoring the fully
specified IHS-DH model over all nested specifications.
Although the IHS-DH model does nest the two-part model, the IHS-SH model does, however,
evolve into a two-part model where the first step involves the probit model and the second step
involves the heteroscedastic truncated OLS regression. Wald test rejects the two-part model ðρ ¼ 0Þ
ðχ 2 ¼ 13:19, df ¼ 1, p  value < :0001Þ when the error-dependent IHS-SH model is considered,
implying the impossibility of making a separate decision between the spending decision and the
amount spent on accommodation for an average household in Turkey.
The instruments in the selection equation (viz., area of residence, heating system (stove), and real
estate dummies) are jointly significant by the Wald test ðχ 2 ¼ 29:14, df ¼ 5, p  value < :0001Þ,
rejecting the hypothesis of weak instruments and suggesting adequate instrumentation for parameter
identification. Numbers of both kids and adults and the three unique income variables in the level
equation are also jointly significant ðχ 2 ¼ 66:52, df ¼ 5, p  value < :0001Þ, justifying their
inclusion. Finally, Wald test results suggest the explanatory variables are jointly significant in both
the selection ðχ 2 ¼ 173:19, df ¼ 41Þ and level ðχ 2 ¼ 142:14, df ¼ 41Þ equations, both with a p-
value < .0001 (Table 2).
The error correlation coefficient is significant and negative, suggesting unobserved character-
istics affect the binary and level decisions in opposite directions. Also, although the signs of the
parameter estimates are generally compatible with economic theory, signs of most of the variables,
such as age categories, divorced head, education and employment categories, state aids, savings,
credit cards, some of the habit variables, etc. are opposite between the selection and level equations,
highlighting an advantage of the double-hurdle parameterization versus the Tobit model. The Tobit
model is known to be restrictive in this context, in that all variables are forced to affect probability
and level in the same direction.
The regressors generally play a more important role, in terms of their statistical significance, in
the spending level than the likelihood of accommodation (Table 2). Although the signs of in-
dependent variables overlap with economic theory these values do not account for the marginal
impacts of regressors on the probability and levels of spending. To explore the roles of ex-
planatory variables more fully, we calculate average marginal effects of explanatory variables on
the likelihood to spend, conditional level (viz., among accommodators), and unconditional level
(viz., overall population) of spending, along with their standard errors; results are reported in
Table 3.
Among the household head characteristics, there exists an inverted U-shaped association be-
tween age and spending levels. Compared to households headed by an individual aged ≤30, those
headed by an individual aged 30–50 years are more likely to spend at a paid facility and spend more
where they stay. Families headed by an individual aged 51–65 and >65, on the other hand, are less
likely to spend and spend less overall, unconditional on spending. Our results stand in contrast to
findings by Brida and Scuderi (2013), Bernini and Cracolici (2015), Bernini and Fang (2020), Lin
et al. (2015), and Thrane (2014) that households closer to retirement spend more on tourism. Bernini
and Cracolici’s (2016) however do also find inverted U-shape findings for Italian tourists, and Lin
et al. (2015, 2020) also find accommodation expenditures vary over the course of the life cycle,
supporting the notion of households’ income discretionary spending pattern. Factors such as
deterioration in health with aging, meeting their daily treatment needs if they are sick, and duties
Urak et al. 13

Table 2. Maximum-likelihood estimates of IHS-DH model.

Selection equation Level equation Heteroscedasticity

Variable Estimate SE Estimate SE Estimate SE

Constant 1.005 0.722 253.276 183.678 – –

Age 31–50 0.191 0.262 75.497* 40.920 – –
Age 51–65 0.527** 0.267 80.661* 45.465 – –
Age >65 0.921*** 0.312 114.517** 56.870 – –
Male 0.289 0.264 3.703 39.615 – –
Unmarried 0.305 0.335 76.630 51.252 – –
Divorced 0.723* 0.424 129.062** 55.919 – –
Widow 0.100 0.342 64.036 64.229 – –
Elementary 0.370 0.231 119.315** 57.236 – –
Secondary 0.186 0.263 77.057 63.298 – –
High school 0.515** 0.232 110.815** 54.510 – –
College 0.776*** 0.278 107.281*** 54.748 – –
Compulsory insurance 0.441* 0.250 59.457 55.900 – –
Greene card 0.091 0.338 11.153 85.579 – –
Salaried 0.490 0.328 121.192** 60.225 – –
Employer 0.288 0.352 109.113* 65.305 – –
Retired 0.572* 0.340 75.448 56.716 – –
Agricultural job 0.054 0.218 22.701 54.294 – –
State aids 0.228 0.176 58.364* 34.164 – –
Private aids 0.182 0.163 33.327 28.281 – –
Apartment 0.381 0.250 102.696* 53.809 – –
Renter 0.260* 0.153 34.205 26.083 – –
Spouses only 0.113 0.217 40.387 44.447 – –
Spouses with kids 0.152 0.187 41.575 38.054 – –
Savings 0.074 0.146 21.375 27.225 – –
Credit card(s) habit 0.112 0.185 73.937** 36.288 – –
Internet habit 0.187 0.138 46.253* 25.049 – –
Eating out habit 0.268 0.177 15.572 39.609 – –
Cinema habit 0.528*** 0.192 21.703 26.277 – –
Game habit 0.211 0.184 38.411 34.951 – –
Coffee house habit 0.074 0.122 33.152 22.729 – –
Bazaar habit 0.103 0.104 3.299 19.860 – –
Tobacco habit 0.008 0.106 14.342 20.045 – –
Alcohol habit 0.170 0.159 64.635** 30.237 – –
Newspaper habit 0.163 0.179 10.521 26.666 – –
Car 0.266** 0.111 1.902 23.415 – –
Wage earners 0.032 0.254 37.842 58.644 – –
Stove heating 0.196** 0.080 – – – –
Residence area 0.140 0.101 – – – –
One real estate 0.033 0.075 – – – –
Two real estates 0.312*** 0.107 – – – –
≥ Three real estates 0.517*** 0.126 – – – –

(continued)
14 Tourism Economics 0(0)

Table 2. (continued)

Selection equation Level equation Heteroscedasticity

Variable Estimate SE Estimate SE Estimate SE

Kids – – 31.341*** 8.883 0.014 0.025

Adults – – 33.008*** 9.942 0.038 0.030
Second quartile income – – 176.507*** 69.115 0.190 0.148
Third quartile income – – 289.317*** 73.197 0.462*** 0.140
Fourth quartile income – – 384.959*** 75.458 0.538*** 0.148
Error std. dev. (σ) – – – – 314.811*** 46.622
Error correlation (ρ) 0.331** 0.158 – – – –
Scaling (θ) – – 0.004*** 0.001 – –
Log-likelihood 10,317.886 – – – – –
Wald (df)[p-value] 173.191 (41) 142.140 (41) – – – –
[< .001] [< .001]
Note: ***p < .01, **p < .05, *p < .10. Wald tests are for joint significance of all variables in equation. SE stands for standard error.

taking care of the elderly at home even if they are healthy limit mobility of the family to the outside
world. As younger and middle-aged households tend to spend more, additional promotions, such as
additional discounts on the cost of accommodation after a certain period of stay, or additional
services such as airport or bus station pick-up and drop-off can generate additional revenues for
accommodation resorts.
Male-headed families are 2.8 percentage points less likely to spend and spend 5.05 TL less
overall per month. Although gender has not been found to play a role (Amir et al., 2017; Brida
and Scuderi, 2013; Marksel et al., 2017; Tchetchik et al., 2009; Wang et al., 2006; Wu et al.,
2013), with one exception by Zhang and Feng (2018), who corroborate our finding. Es-
sentially, in Turkey’s social environment, male-headed households travel more independently
than female-headed households, while women are always accompanied in travel by someone
who may prefer facilities located in the city center, perhaps in commercial places such as
hotels rather than government-run facilities, as both the rate of women in business life and
their rate of being civil personnel are low in Turkey. Male-headed households can increase the
demand for accommodation to the extent that gender discrimination is a factor. Divorced
households, on the other hand, spend approximately 22 TL less per month conditional
spending compared to households with married heads. Although divorced families may be
more independent, they may be weaker financially, causing them to spend less on accom-
modation than the married.
Consumer preference models often stipulate education as a contributing factor in tourism
spending decisions and spending levels. Although the parameter estimates of high school and
college are positive and significant in both the selection and expenditure equations, marginal effects
of elementary and high school education variables remained statistically important in terms of
conditional expenditure level. Interestingly, as the education level of the household head increases,
families spend less on accommodation. This finding contradicts those reported in the literature
(Alegre and Pou, 2004; Alegre et al., 2013; Bernini and Cracolici, 2015; Lin et al., 2015; Nicolau
and Más, 2005; Park et al., 2020), perhaps because families with educated heads are likely to opt for
state-run accommodation centers that are decent, clean, and relatively inexpensive. Better educated
Urak et al. 15

Table 3. Marginal effects of explanatory variables on probability and levels of spending.

Probability Conditional level Unconditional level

Variable Estimate SE Estimate SE Estimate SE

Age 31–50 0.010 0.012 19.217** 9.711 2.684 1.997

Age 51–65 0.018 0.012 13.540 12.250 2.146 2.139
Age >65 0.038*** 0.012 15.285 17.507 5.575** 2.352
Male 0.028* 0.015 7.439 8.832 5.045* 2.754
Unmarried 0.004 0.015 15.689 10.455 1.439 2.309
Divorced 0.004 0.016 21.988* 11.972 0.766 2.665
Widow 0.014 0.019 18.780 16.160 3.490 3.570
Elementary 0.011 0.010 28.026** 14.339 3.247 2.125
Secondary 0.012 0.011 18.290 13.333 2.768 2.101
High school 0.000 0.012 21.354* 12.597 1.242 2.218
College 0.020 0.015 15.757 12.201 2.009 2.922
Compulsory insurance 0.018 0.011 8.803 14.098 2.411 2.044
Greene card 0.012 0.019 5.480 21.561 2.322 4.139
Salaried 0.000 0.014 26.996** 13.160 1.452 2.350
Employer 0.011 0.018 29.395* 16.726 3.657 3.091
Retired 0.019 0.015 10.683 12.660 2.290 2.709
Agricultural job 0.003 0.008 5.881 12.919 0.862 1.703
State aids 0.002 0.007 12.591 7.609 0.999 1.293
Private aids 0.003 0.007 6.174 5.997 0.151 1.213
Apartment 0.003 0.007 23.498** 11.001 1.790 1.294
Renter 0.010 0.007 4.851 5.255 1.227 1.240
Spouses only 0.026** 0.013 15.494 10.884 5.277** 2.666
Spouses with kids 0.027*** 0.009 16.013* 8.472 5.309*** 1.729
Savings 0.001 0.005 4.974 5.505 0.465 0.911
Credit card(s) habit 0.016*** 0.006 20.255*** 7.226 3.744*** 1.082
Internet habit 0.037*** 0.009 19.551*** 5.803 7.696*** 2.082
Eating out habit 0.028*** 0.006 10.521 8.788 5.048*** 1.341
Cinema habit 0.065*** 0.014 18.931*** 7.032 12.457*** 3.387
Game habit 0.005 0.010 7.414 8.599 0.464 1.757
Coffee house habit 0.005 0.005 8.726* 4.936 1.350 0.969
Bazaar habit 0.008 0.005 1.211 4.488 1.288 0.953
Tobacco habit 0.004 0.005 4.224 4.387 0.933 0.844
Alcohol habit 0.045*** 0.012 26.083*** 7.980 9.746*** 2.807
Newspaper habit 0.010 0.010 0.249 5.913 1.655 1.771
Car 0.023*** 0.006 6.325 5.511 4.138*** 1.248
Wage earners 0.010 0.008 10.384 12.301 2.105 1.550
Stove heating 0.016** 0.007 0.572 9.643 2.550* 1.387
Residence area 0.014 0.009 12.955 8.290 2.953* 1.537
One real estate 0.007 0.011 48.858 40.597 3.962 3.802
Two real estates 0.005 0.020 86.631*** 28.594 5.445* 3.052
≥ Three real estate 0.001 0.027 91.304*** 26.253 5.099 3.218
Kids 0.012*** 0.004 10.102*** 2.836 2.544*** 0.800

(continued)
16 Tourism Economics 0(0)

Table 3. (continued)

Probability Conditional level Unconditional level

Variable Estimate SE Estimate SE Estimate SE

Adults 0.014*** 0.006 12.372*** 4.217 3.569*** 1.349

Second quartile income 0.070** 0.029 63.163** 28.900 17.586** 8.555
Third quartile income 0.123*** 0.035 116.980*** 37.994 35.726*** 12.871
Fourth quartile income 0.167*** 0.041 175.221*** 50.231 57.180*** 17.385
Note: ***p < .01, **p < .05, *p < .10. SE stands for standard error.

families also have more travel opportunities and may prefer to stay in such low-cost places to the
extent that they may have too much tourism information and high search costs. Private accom-
modation businesses should take this fact into account to attract such segregated groups by creating
a competitive environment with various price-cutting mechanisms.
Regarding occupational variables, salaried and employer-headed households spend approxi-
mately 27 TL and 29 TL more per month, respectively, conditional on spending, compared to
households with another job status. These families are more likely on constant travels for job duties
and may spend more in overnight hotels. They may also be more likely to discover new places by
being exposed to intense short-term travel for business needs, and to visit these discovered places
alone or with the family (Brida and Scuderi, 2013; Lin et al., 2015). The literature also suggests that
general managers, self-employees, and business owners are more likely to travel than the un-
employed, fishermen, and handicraft (Bernini and Cracolici, 2015; Eugenio-Martin and Campos-
Soria, 2011). Exceptions include finding by Bernini and Fang (2020) in China, where both em-
ployed and self-employed households are less likely to participate in the tourism market, probably
because of a greater incentive to devote time to work.
Families living in apartments spend about 23.50 TL less per month than those residing in
detached houses, perhaps due to economic barriers. Compared to other family types, childless
couples and couples with children are 2.6–2.7 percentage points more likely to spend and spend
15.49 to 16.01 TL more per month conditional on spending and 5.28 to 5.31 TL unconditionally.
These positive effects are expected as childless couples are more likely to travel together and
families with children are more likely to travel with their children. Rashidi and Koo (2016) find that,
interestingly, traveling with a partner increases the likelihood of being in a married relationship.
Compared to large households with more economic and physical constraints and a more complex
family structure, families with and without children are found more likely to travel and also spend
more (Bernini and Cracolici, 2015; Nicolau and Más, 2005), corroborating our results. Interestingly,
Alegre and Pou (2004) find that couples without children, with children, and with other adults are
less likely to spend in tourism compared to one-person households. In this context, additional
revenues can be generated by amenities such as swimming pools, PlayStation centers, and
playgrounds for children.
On habit traits, families with credit cards, the Internet, eating out, testing, and alcohol use are
more likely to spend and also spend more conditionally and unconditionally on spending. Bernini
and Fang (2020) reported that Chinese homeowners who feel happy 5–7 days a week are more likely
to spend, corroborating our results, considering the high probability of their social interaction
networks. Since the Internet is a gateway to the world, it provides access to information about
complex tourism products before deciding to participate in tourism (Bernini and Fang, 2020).
Urak et al. 17

Again, as this literature suggests, families using the internet in China are 13.33 percentage points
more likely to participate in tourism than families without the internet. Also, families adopting
cinema culture as part of their lifestyle are likely to spend more. As expected, credit card(s), which
allow the use of future income in current spending, results in more accommodation expenses. In
addition, these results show that accommodation resorts that include bars, mini coffee houses, movie
theaters, and open buffet dining halls for social and relaxation activities can have an advantage that
can easily attract such families. Owning a passenger car plays a role, increasing the probability by
2.3% points and unconditional level of spending by 4.14 TL per month. We expect a positive role of
vehicle ownership as it reflects the wealth and a means to facilitate travel, especially on short trips
(Lin et al., 2020). Our findings are consistent with findings for Europe (Mergoupis and Steuer,
2003), Spain (Alegre et al., 2013), and China (Lin et al., 2020). Family or family members often tend
to travel in their car on short-distance trips (Lin et al., 2020). Travelers, in particular, are likely to
look for other accommodation facilities with a parking lot on their next trip when they experience
parking problems at a resort, so facilities with ample parking spaces and valet services are likely to
fair better than those without.
Families who follow the old tradition and heat homes with stoves are 1.6% points less likely to
spend and spend 2.55 TL less unconditionally each month. These families generally have limited
connections with the outside world as income is a major barrier to travel. Contrarily, households
residing in a large house (with an area >160 sq m) spend 12.96 TL more conditionally and 2.95 TL
more unconditionally per month, compared to households with a smaller living space. This factor
can be primarily correlated with the income effect, and such results are expected since those living in
large houses are typically wealthier families, often dense families with many and likely highly
educated individuals. Interestingly, the number of real estate lowers the accommodation expenses.
Families with two and three or more real estimates spend as much as 86.63 to 91.30 TL more
conditional on spending and 5.45 to 5.10 TL unconditionally, compared to families without real
estate. The reason for this negative role of real estate is probably that such well-to-do families have
close relatives or houses where they can spend the night, especially in metropolitan areas like
Istanbul, Ankara, and Izmir. Meanwhile, such behavior reflects the existence of more frugal
consumption after income control, particularly concerning the fact that such households generally
face greater life pressures (Lin et al., 2020). Also, our results support the life cycle hypothesis, viz.,
that families consider not only current conditions but also future living conditions when choosing
financial assets (Lin et al., 2020; Modigliani and Brumberg, 1954). Thus, this cult of properties can
be exacerbated by investment in the future (such as children’s education, their weddings, first time
home buying, and/or car purchases), which limits current travel frequency, time, and accommo-
dation expenses, especially in families with parents over 50 in Turkey.
The numbers of children and adults in the family decrease the probability and level of spending,
collaborating findings by Bernini and Cracilici’s (2015) for Italy but contradicting findings by
Nicolau and Más (2005) for Spain and Eugenio-Martin and Campos-Soria (2011) for 15 European
countries. With more children in the household, both the expenditure items and the time allocated by
the parents increase, limiting the holiday and accommodation expenses of the households. Although
parents are decision-makers of the family, they may not be able to dictate travel by children (Alegre
and Pou, 2004; Mergoupis and Steuer, 2003).
Compared to lower-income families in the first quartile, families in the second, third, and fourth
quarter groups are 7.0–16.7 percentage points more likely to spend; these households spend as much
as 63.16 to 175.22 TL more conditional on spending and 17.59 to 57.18 TL more overall per
month. As income increases, so do the probability and level of expenditure. The impact on the
highest income group is worth mentioning. They spend about 2 times more than families in the
18 Tourism Economics 0(0)

second income quartile and 1.5 times more than families in the third income quartile, and this
effect is even more pronounced in their spending amount. Considering that families in the high-
income group are more likely to stay in more expensive hotels, these results are consistent with
theory, reflecting an income factor as the literature suggests (Brida and Scuderi, 2013; Hung
et al., 2012; Lin et al., 2015, 2020; Rashidi and Koo, 2016). Park et al. (2019) also find that
household income exerted a positive relation with lodging expenditure. As Hung et al. (2012)
pointed out, it is generally accepted that tourism product is an inferior good for light spenders
and a normal good for heavy spenders, when focusing on the lowest and highest income
quartiles. Lin et al. (2020) note that economic capacity is one of the most important factors in
increasing tourism consumption.

Conclusion and recommendations

This study aims to explore the driving forces of household accommodation expenditures using a
large sample from a household budget survey in Turkey. Uncovering the factors behind tourism
spending can inform policy deliberation towards restructuring and planning of tourism services
which would enhance the competitiveness of the tourism industry at the country level. On the
methodological front, the study demonstrates the importance of correcting for both hetero-
scedasticity and non-normality of the error terms in a sample where extreme values and a large
number of zeros are present in the expenditure level. The IHS transformation is applied to the
expenditure variable and heteroscedasticity of error terms accommodated, both of which are
supported by statistical tests. Governed by just one additional (scale) parameter, the IHS trans-
formation offers a powerful alternative to traditional transformation methods such as the logarithmic
transformation in regression analysis. The transformation has an advantage in handling extreme
values in the dependent variable—both small values due to cheap accommodation at state social
facilities and large values at luxury hotels. When only accommodators are taken into account,
adopting a logarithmic transformation to amounts where 65% of the observations are less than
average can confound empirical estimates; the difference between large and small values in the
logarithmic transformation would be unpalatable at the modeling phase. We, therefore, use the IHS
transformation to account for extreme values and skewness in the distribution of accommodation
spending. In the IHS-DH model, the negative correlation between the binary decision and level
decision indicates that unobserved household characteristics, not controllable by the researcher,
affect the family’s travel decision and expenditure level in opposite directions. The fact that many
regressors have opposite signs in the selection and level equations highlights an important ad-
vantage of the double-hurdle parameterization over the Tobit model; in the latter specification, all
explanatory variables are forced to affect the probability and level of the dependent variable in the
same direction.
Results of the study suggest socio-demographic and economic characteristics of the household
and the household head have profound effects on the probability and levels of spending on ac-
commodation. Among the life cycle patterns of the family’s accommodation expenditure decisions
and expenditure amounts, important driving forces such as gender, age, marital status, education,
occupational groups can be listed from the characteristics of the household head. In addition, there is
a profound association between family spending decisions and spending amounts and triggering
factors such as lifestyle, family type, family habits, car ownership, real estate assets, numbers of
children and adults, and, of course, income. Similarly, destination marketing organizations
(DMOs) should take such segregations into account and include amenities such as minibar, coffee
house, buffet place, movie theater, swimming pool, physical play equipment for children, large
Urak et al. 19

and safe car parking lots, and valet services. These amenities can create an advantage in the
cutthroat competition in the tourism market. As such, tourist traits, such as using segregation
analysis, can offer tourism businesses the opportunity to develop and optimize product and
service understanding to increase tourism demand and provide guests with more satisfying and
rewarding experiences. Such arrangements are long-term and can have strategic implications. For
example, highly educated and high-income guests may prefer shorter trips because they face
higher opportunity costs, despite the inclination to visit more places and consume more goods.
Such families can choose high-quality, new, educational, and at the same time relaxing actions. By
keeping these features on their agenda, hospitality operators or DMOs can achieve success and
sustainability by designing shorter and more poignant tourism products rather than traditional
mass-market products.
Choosing the most appropriate statistical model has been the main theme of this study, in
addressing departures from the assumptions of normality and homoscedasticity in tourism ex-
penditures. We demonstrated the applicability of the IHS-DH model by applying it to the 2018
cross-section data of the TSI. In this context, it may not be appropriate to generalize the marginal
effects derived from the IHS-DH model with a one-year dataset, which will change as the time-
specific family profile changes. Such impacts could be valid only for the short-term but not for the
long-term decisions. A panel dataset may be needed to ascertain long-term effects on tourism
expenditures, but unfortunately, there is no dataset compiled by TSI for the time dimension. Instead,
a mean-based pseudo-panel dataset can be constructed from repeated cross-sections which, with its
pros and cons, can perform an alternative functional task, as birth cohorts, birth-education cohorts,
and birth-gender cohorts may be formed to elicit causality between accommodation expenditures
and some characteristics of the family. By tracking cohort means, mean-based pseudo-panels are
subject to fewer measurement errors than individual-level data. However, such pseudo-panels are
not without problems. These mean-based pseudo-panels, for instance, do not provide information
about intra-group mobility and cohort-level estimates are a potential source of bias if events such as
migration or death affect the size and composition of cohorts. Further, the construction of a pseudo-
panel should involve a balance between the number of cohorts and the number of observations in
each cohort. If the size of a cohort derived from the data is not large enough, the average char-
acteristics per cohort will reflect an error-based measurement of the actual cohort population values.
Finally, the structures of qualitative variables in cross-sectional data turn into a quantitative form
that eliminates the opportunity of comparison between qualitative variables. Despite the above
problems and in the absence of a genuine panel dataset, the causality relationship between ac-
commodation expenditures or spending types and some characteristics of the household may be
among the topics worth studying in future studies within the scope of totally different statistical
models (e.g., with fixed and random effects)5 featuring multiple categories of expenditures and in
addition to existing factors, integrating additional new travel-related factors such as business or
pleasure, and travel purpose within a system to arrive at satisfactory information about the richer
family segregation.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or
publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
20 Tourism Economics 0(0)

Supplemental material
Supplemental material for this article is available online.

ORCID iD
Abdulbaki Bilgiç  https://orcid.org/0000-0001-5946-0915

Notes
1. Superiority of the model is discussed later in the econometrics model sub-section. Also, since the IHS-
transformed DH model is the preferred model as a result of model specification test (see Results section),
we present only the IHS-DH model. Likelihood function and marginal effects calculation for the in-
dependent IHS-DH model can be obtained by restricting the error correlation ρ ¼ 0 to those of the IHS-
DH model; also see Su and Yen (1996). Model characterizations, likelihood functions, and marginal
effects (conditional means and probabilities) for the IHS-SH and IHS-SS are presented in the
Supplemental Online Appendix.
2. The saving variable may potentially be endogenous. We investigated endogeneity of saving in both the
selection and level equation. Following Terza et al. (2008), this was done by first estimating the savings
equation to obtain the least-squares residuals. The residual term was then included in each of the selection
and level equations as an additional regressor. Statistical significance of the residual term would amount to
enogeneity of the saving variable in the corresponding equation. We found no evidence of the saving
variable in the selection or level equation.
3. We also calculated Cramer’s V correlation coefficients matrix (available upon request). Results of analysis
by VIFs are corroborated.
4. The AIC is calculated as AIC ¼ 2logL þ 2k, where k is the number of parameters estimated.
5. We are grateful to an anonymous reviewer who directed our attention to such issues in future studies.

References
Aguiló E, Rossello J and Vila M (2017) Length of stay and daily tourist expenditure: a joint analysis. Tourism
Management Perspectives 21: 10–17.
Alegre J and Cladera M (2010) Tourist expenditure and quality: why repeat tourists can spend less than first-
timers. Tourism Economics 16(3): 517–533.
Alegre J, Cladera M and Sard M (2011) Analysing the influence of tourist motivations on tourist expenditure at
a sun-and-sand destination. Tourism Economics 17(4): 813–832.
Alegre J, Mateo S and Pou L (2010) An analysis of households’ appraisal of their budget constraints for
potential participation in tourism. Tourism Management 31(1): 45–56.
Alegre J, Mateo S and Pou L (2013) Tourism participation and expenditure by Spanish households: the effects
of the economic crisis and unemployment. Tourism Management 39: 37–49.
Alegre J and Pou L (2004) Micro-economic determinants of the probability of tourism consumption. Tourism
Economics 10(2): 125–144.
Amir S, Osman MM, Bachok S, et al. (2017) Socio-demographic variation on tourism expenditure in Melaka
UNESCO world heritage area. Advanced Science Letters 23(4): 2958–2961.
Bernini C and Cracolici MF (2015) Demographic change, tourism expenditure and life cycle behaviour.
Tourism Management 47(C): 191–205.
Bernini C and Cracolici MF (2016) Is participation in the tourism market an opportunity for everyone? Some
evidence from Italy. Tourism Economics 22(1): 57–79.
Urak et al. 21

Bernini C, Cracolici MF and Nijkamp P (2017) Place-based attributes and spatial expenditure behavior in
tourism. Journal of Regional Science 57(2): 218–244.
Bernini C and Fang R (2020) Living standard and Chinese tourism participation. International Journal of
Tourism Research: 1–14.
Bernini C and Galli F (2019) Italian otbound tourists, tourists’ expenditure and satisfaction. SSRN. Pre-print
DOI: 10.13140/RG.2.2.26618.95685.
Bilgic A, Florkowski WJ, Yoder J, et al. (2008) Estimating fishing and hunting leisure spending shares in the
United States. Tourism Management 29(4): 771–782.
Blundell R (1988) Consumer behaviour: theory and empirical evidence–a survey. The Economic Journal
98(389): 16–65.
Blundell RW and Meghir C (1987) Bivariate alternatives to the Tobit model. Journal of Econometrics 34(1–2):
179–200.
Brida JG, Bukstein D and Tealde E (2015) Exploring cruise ship passenger spending patterns in two Uruguayan
ports of call. Current Issues in Tourism 18(7): 684–700.
Brida JG, Chiappa GD, Meleddu M, et al. (2012) The perceptions of an island community towards cruise
tourism: a factor analysis. Tourism: An International Interdisciplinary Journal 60(1): 29–42.
Brida JG, Disegna M and Osti L (2013a) Visitors’ expenditure behaviour at cultural events: the case of
Christmas markets. Tourism Economics 19(5): 1173–1196.
Brida JG, Disegna M and Scuderi R (2013b) Visitors to two types of museums: do expenditure patterns differ?
Tourism Economics 19(5): 1027–1047.
Brida JG, Lanzilotta B, Moreno L, et al. (2018) A non-linear approximation to the distribution of total ex-
penditure distribution of cruise tourists in Uruguay. Tourism Management 69: 62–68.
Brida JG, Pereyra JS and Scuderi R (2014) Repeat tourism in Uruguay: modelling truncated distributions of
count data. Quality & Quantity 48(1): 475–491.
Brida JG and Scuderi R (2013) Determinants of tourist expenditure: a review of micro econometric models.
Tourism Management Perspectives 6: 28–40.
Brown S, Greene WH, Harris MN, et al. (2015) An inverse hyperbolic sine heteroskedastic latent class panel
Tobit model: an application to modelling charitable donations. Economic Modelling 50(C): 228–236.
Burbidge JB, Magee L and Robb AL (1988) Alternative transformations to handle extreme values of the
dependent variable. Journal of the American Statistical Association 83: 123–127.
Carboni OA (2012) An empirical investigation of the determinants of R&D cooperation: an application of the
inverse hyperbolic sine transformation. Research in Economics 66(2): 131–141.
Cragg JG (1971) Some statistical models for limited dependent variables with application to the demand for
durable goods. Econometrica 39(5): 829–844.
Disegna M and Osti L (2016) Tourists’ expenditure behaviour: the influence of satisfaction and the dependence
of spending categories. Tourism Economics 22(1): 5–30.
Disegna M, Scuderi R and Brida JG (2012) Visitors of two types of museums: do expenditure patterns differ?
Tourism Economic 19(5): 1027–1047.
Di Vaio A, Lepore L and Varriale L (2018) Self-organised cruiser’s expenditures in a port of call: the interaction
effect between city interface satisfaction and super-sized ships. International Journal of Culture, Tourism
and Hospitality Research 12(4): 385–406.
Doménech A, Gutiérrez A and Anton Clav´e S (2020) Cruise passengers’ spatial behaviour and expenditure
levels at destination. Tourism Planning & Development 17(1): 17–36.
Engle RF (1984) Chapter 13 Wald, likelihood ratio, and lagrange multiplier tests in econometrics. In: In-
triligator MD, Griliches Z and Griliches Z (eds) Handbook of Econometrics. Amsterdam, The Neth-
erlands: North Holland Publishing Company, 775–826.
22 Tourism Economics 0(0)

Eugenio-Martin JL and Campos-Soria JA (2011) Income and the substitution pattern between domestic and
international tourism demand. Applied Economics 43(20): 2519–2531.
Gargano R and Grasso F (2016) Cruise passengers’ expenditure in the Messina port: a mixture regression
approach. Journal of International Studies 9(2): 158–169.
Gomez-Deniz E and Perez-Rodriguez JV (2019) Modelling distribution of aggregate expenditure on tourism.
Economic Modelling 78(C): 293–308.
Heckman JJ (1979) Sample selection bias as a specification error. Econometrica 47(1): 153–161.
Hong G-S, Fan JX, Palmer L, et al. (2005) Leisure travel expenditure patterns by family life cycle stages.
Journal of Travel & Tourism Marketing 18(2): 15–30.
Hung W-T, Shang J-K and Wang F-C (2012) Another look at the determinants of tourism expenditure. Annals
of Tourism Research 39(1): 495–498.
Jang S and Ham S (2009) A double-hurdle analysis of travel expenditure: baby boomer seniors versus older
seniors. Tourism Management 30(3): 372–380.
Jones AM and Yen ST (2000) A box-cox double-hurdle model. The Manchester School 68(2): 203–221.
Jurdana DS and Frleta DS (2015) Offseason tourist expenditure in the sun and sea destination. In:
Proceedings of the 9th International Conference on Business Administratio (ICBA’15): Advances
in Business and Economic Development, Dubai, United Arab Emirates. 22–24 February 2015,
pp. 193–200.
Kwiatkowski G and Könecke T (2017) Tourism and recurring sport events. Sport, Business and Management:
An International Journal 7(5): 464–482.
Lee G and Lee M-K (2017) Estimation of the shore excursion expenditure function during cruise tourism in
Korea. Maritime Policy & Management 44(4): 524–535.
Lee L-F and Maddala GS (1985) The common structure of tests for selectivity bias, serial correlation, het-
eroscedasticity and non-normality in the Tobit model. International Economic Review 26(1): 1–20.
Lew AA and Ng PT (2012) Using quantile regression to understand visitor spending. Journal of Travel
Research 51(3): 278–288.
Lin VS, Mao R and Song H (2015) Tourism expenditure patterns in China. Annals of Tourism Research 54:
100–117.
Lin VS, Qin Y, Li G, et al. (2020) Determinants of Chinese households’ tourism consumption: evidence from
China family panel studies. International Journal of Tourism Research 23(4): 542–554.
Maddala GS (1983) Methods of estimation for models of markets with bounded price variation. International
Economic Review 24(2): 361–378.
Marksel M, Tominc P and Božičnik S (2017) Cruise passengers’ expenditures. Tourism Economics 23(4):
890–897.
Marrocu E, Paci R and Zara A (2015) Micro-economic determinants of tourist expenditure: a quantile re-
gression approach. Tourism Management 50(C): 13–30.
Martinez-Garcia E, Raya JM and Majo J (2012) Differences in residents’ attitudes towards tourism among mass
destinations. International Journal of Tourism Research 19(5): 535–545.
Mergoupis T and Steuer M (2003) Holiday taking and income. Applied Economics 35(3): 269–284.
Middaugh J, Kwak L and Meng F (2013) How shopping satisfaction contributes to internationaltourism
satisfaction and tourism expenditures by gender: analysis of U.S. tourists in South Korea. International
Journal of Business Research 13(1): 5–12.
Ministry of Culture and Tourism (2020) 2019 Tourism certified facility statistics. Available at: https://yigm.ktb.
gov.tr/TR-201131/tesis-istatistikleri.html (accessed 20 August 2020).
Modigliani F and Brumberg R (1954) Utility analysis and the consumption function: an interpretation of cross-
section data. In: Kurihara K (ed) Post-Keynesian Economics. New Brunswick, NJ: Rutgers University
Press, 388–436.
Urak et al. 23

Mudarra-Fernández AB, Carrillo-Hidalgo I and Pulido-Fernández JI (2019) Factors influencing tourist ex-
penditure by tourism typologies: a systematic review. Anatolia 30(1): 18–34.
Nicolau JL and Más FJ (2005) Stochastic modeling. Annals of Tourism Research 32(1): 49–69.
Park K, Park J-Y and Back RM (2019) Determinants of Marathoners’ event expenditures: an in-depth ex-
ploration of past experience. Journal of Hospitality and Tourism Insights 2(2): 110–120.
Park S, Woo M and Nicolau JL (2020) Determinant factors of tourist expenses. Journal of Travel Research
59(2): 267–280.
Parola F, Satta G, Penco L, et al. (2014) Destination satisfaction and cruiser behaviour: the moderating effect of
excursion package. Research in Transportation Business & Management 13: 53–64.
Pellegrini A and Scagnolari S (2019) The relationship between length of stay and land transportation mode in
the tourism sector: a discrete-continuous framework applied to Swiss data. Tourism Economics 27(1):
243–259.
Pence KM (2006) The role of wealth transformations: an application to estimating the effect of tax incentives on
saving. The BE Journal of Economic Analysis & Policy 5(1): 1–26.
Perez-Rodriguez JV and Ledesma-Rodriguez F (2019) Unconditional quantile regression and tourism ex-
penditure: the case of the Canary Islands. Tourism Economics 25(8): 1–23.
Prais SJ and Houthakker HS (1955) The Analysis of Family Budgets, with an Application to Two British
Surveys Conducted in 1937-39 and their Detailed Results. New York, NY: Cambridge University
Press.
Pudney S (1989) Modelling Individual Choice: The Econometrics of Corners, Kinks and Holes. Oxford, UK:
Basil Blackwell.
Rashidi TH and Koo TTR (2016) An analysis on travel party composition and expenditure: a discrete-
continuous model. Annals of Tourism Research 56(C): 48–64.
Saayman A and Saayman M (2006) Sociodemographics and visiting patterns of arts festival in South Africa.
Event Management 9(4): 211–222.
Saayman M and Saayman A (2009) Why travel motivation and socio-demographics matter in managing a
national park. Koedoe 51(1): 1–9.
Saayman M and Saayman A (2012) Determinants of spending: an evaluation of three major sporting events.
International Journal of Tourism Research 14(2): 124–138.
Sainaghi R (2012) Tourist expenditures: the state of the art. Anatolia 23(2): 217–233.
Su S-JB and Yen ST (1996) Microeconometric models of infrequently purchased goods: an application to
household pork consumption. Empirical Economics 21(4): 513–533.
Sun M, Zhang X and Ryan C (2015) Perceiving tourist destination landscapes through Chinese eyes: the case of
South Island, New Zealand. Tourism Management 46(C): 582–595.
Tchetchik A, Fleischer A and Shoval N (2009) Segmentation of visitors to a heritage site using high-resolution
time-space data. Journal of Travel Research 48(2): 216–229.
Terza JV, Basu A and Rathouz PJ (2008) Two-stage residual inclusion estimation: addressing endogeneity in
health econometric modeling. Journal of Health Economics 27(3): 531–543.
Thrane C (2014) Modelling micro-level tourism expenditure: recommendations on the choice of independent
variables, functional form and estimation technique. Tourism Economics 20(1): 51–60.
Thrane C (2015a) On the relationship between length of stay and total trip expenditures: a case study of
instrumental variable (IV) regression analysis. Tourism Economics 21(2): 357–367.
Thrane C (2015b) Research note: does method of payment affect total trip expenditure? Tourism Economics
21(3): 669–676.
Thrane C and Farstad E (2011) Domestic tourism expenditures: the non-linear effects of length of stay and
travel party size. Tourism Management 32(1): 46–52.
Tobin J (1958) Estimation of relationships for limited dependent variables. Econometrica 26(1): 24–36.
24 Tourism Economics 0(0)

Toudert D and Bringas-Rábago NL (2016) Impact of the destination image on cruise repeater’s experience and
intention at the visited port of call. Ocean & Coastal Management 130: 239–249.
van Loon R and Rouwendal J (2017) Travel purpose and expenditure patterns in city tourism: evidence from the
Amsterdam metropolitan area. Journal of Cultural Economics 41(2): 109–127.
Vuong QH (1989) Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57(2):
307–333.
Wang Y and Davidson MCG (2010) A review of micro-analyses of tourist expenditure. Current Issues in
Tourism 13(6): 507–524.
Wang L, Law R, Hung K, et al. (2014) Consumer trust in tourism and hospitality: a review of the literature.
Journal of Hospitality and Tourism Management 21: 1–9.
Wang Y, Rompf P, Severt D, et al. (2006) Examining and identifying the determinants of travel expenditure
patterns. International Journal of Tourism Research 8(5): 333–346.
Wooldridge JM (2010) Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.
Wu L, Zhang J and Fujiwara A (2013) Tourism participation and expenditure behaviour: analysis using a scobit
based discrete-continuous choice model. Annals of Tourism Research 40(1): 1–17.
Yen ST and Jones AM (1997) Household consumption of cheese: an inverse hyperbolic sine double-hurdle
model with dependent errors. American Journal of Agricultural Economics 79(1): 246–251.
Zhang H, Zhang J and Kuwano M (2012) An integrated model of tourists’ time use and expenditure behaviour
with self-selection based on a fully nested Archimedean copula function. Tourism Management 33(6):
1562–1573.
Zhang C and Feng G (2018) More wealth, less leisure? Effect of housing wealth on tourism expenditure in
China. Tourism Economics 24(5): 526–540.

Author biographies
Faruk Urak who focuses on agricultural finance and price volatility transmission between
macroeconomic variables and agricultural products.
Nihat Küçük who specialized in consumer demand analysis and production economics.
Abdulbaki Bilgiç who specialized in applied econometrics including consumer demand analysis,
tourist behavior and health economics.
Steven T Yen who deals with econometrics of food demand and health.