category: literaturenote citekey: rosselloroadaccidentstourismcase2011 title: "Road accidents and tourism: The case of the Balearic Islands (Spain)" authors: "Rosselló, Jaume; Saenz-de-Miera, Oscar" year: 2011 date: 2011-00-00 2011 doi: 10.1016/j.aap.2010.10.011 publication: "Accident Analysis & Prevention" url: "https://doi.org/10.1016/j.aap.2010.10.011" zotero_key: W88IFV9V zotero_storage: DTME455X collections: transpordi_planeerimine folder: 001_artiklid/Road Safety Indicators firstAuthor: "Rosselló, Jaume"

status: converted

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Road accidents and tourism: The case of the Balearic Islands (Spain)

Jaume Rosselló∗, Oscar Saenz-de-Miera

Centre de Recerca Econòmica, University of the Balearic Islands, Carretera de Valldemossa km 7.5, 07122 Palma de Mallorca, Spain

article info

Article history: Received 4 November 2009 Received in revised form 29 September 2010 Accepted 13 October 2010

Keywords: Externalities Road accidents Count data models Daily data Tourism

abstract

The increase in the number of tourists for many destinations and their increased mobility within host countries or regions has implied a rise in tourism-associated externalities, with vehicle crashes as the most common cause of injury for tourists. Within the transport literature, the number and variation in the amount of accidents has been related to a large set of determining variables, including weather conditions, socio-economic characteristics, exposure, physical characteristics of the road and a variety of dummies that try to capture effects such as safety laws and seasonal variations. However, the presence of tourism has been neglected. Using the case study of the Balearic Islands, the present study estimates the role of tourism in determining the number of accidents in a daily context, using the set of variables suggested by the literature and incorporating a daily measure for the stock of tourists at a host destination. Results show how tourism can be associated with a significant amount of the accidents that take place in the Balearics.

1. Introduction

Transport activities are essential for the movement of passengers with holiday or leisure purposes. However, in recent years, the increase in tourist numbers for many destinations and their increased mobility within host countries or regions has implied a rise in externalities, which can be associated with means of road transport (Palmer-Tous et al., 2007). This is consistent with an increasing body of literature in which travel has been recognized as a contributor to various problems (e.g., Dickinson and Robbins, 2008; Dickinson et al., 2009; Gursoy et al., 2002; Hall, 1999), among which road traffic accidents have stood out because they are often the main component of external costs from road transport activities (Lindberg et al., 1999; Lindberg, 2001).

Within the tourism literature, the issue of road traffic accidents has been addressed by various studies, like those of Bentley et al. (2001), Howard (2009), Page and Meyer (1996) and Page (2009). Tourists injured or killed in motor vehicle crashes have also become a key area of concern in travel medicine (Carey and Aitken, 1996; Leggat and Leggat, 2003; Wilks et al., 1999). This follows from the consideration that motor vehicle crashes have been identified, but not quantified, as the most common cause of injury or death for tourists (Wilks, 1999). Moreover, there are studies such as those of Page and Meyer (1996), Page (2009) and Wilks et al. (1999) that raise awareness of the problems that accidents may pose for the tourism industry, which has the image of a business selling positive holiday experiences (Clift and Page, 1996) and of a catalyst for change in both individuals and communities (Ryan, 1997). This concern is especially relevant in certain regions that have witnessed exponential growth in the number of tourists in relatively short periods of time and that should expect further growth. The number of international tourist arrivals was 806 million in 2005 and is expected to reach an estimated 1.6 billion in 2020 (UNWTO, 2008, 2009).

In spite of this, the nonexistence of a single economic sector that can be associated with tourism has made it difficult to account for the external costs of tourism activity. While the literature has been able to develop a set of techniques aimed at valuing the economic impacts of tourism (Tourism Satellite Accounts and General Equilibrium Models), there still exists a void when trying to assess the non-economic impacts of tourism.

In this context, while a broad range of studies (Brijs et al., 2008; Eisenberg, 2004; Fridstrom, 1999; García-Ferrer et al., 2006; Hermans et al., 2006; Hutchings et al., 2003; Keay and Simmonds, 2006; Levine et al., 1995; Lord and Persaud, 2000; Quddus, 2008; Van den Bossche et al., 2005a,b; Wang and Abdel-Aty, 2006) have addressed the underlying causes of road traffic accidents, only few have linked them to tourism. Studies that do consider the influence of tourism on the externality that road accidents represent do so through the use of dummy variables for labor/school holidays and vacation periods. The only exception found in the literature is Levine et al. (1995), who consider monthly tourist arrivals but do not include them in the definitive results due to the nonsignificance of the variable.

Thus, the present study tries to fill this gap by explaining the number of accidents in a daily context, using most of the vari-

^∗ Corresponding author. Tel.: +34 971 172 459; fax: +34 971 172 389. E-mail address: jrossello@uib.es (J. Rosselló).

ables suggested by the literature and including a daily measure for the amount of tourists at a host destination. The Balearic Islands are taken as case study because of (i) the relative importance of the tourism activity in the region (with a population of 1 million inhabitants, it accounts for 13 million tourists per year); (ii) the geographical nature of the islands, which makes it possible to thoroughly estimate the daily tourist stock from the ports and airports and (iii) the availability of the rest of the variables required to carry out a study of this nature.

The remainder of this paper is structured as follows. Section 2 contains a literature review on the variables commonly included in studies of this nature. Section 3 reviews the various methods used to approach the modeling of road accidents. Section 4 provides a description of the data. Section 5 contains the estimation results. Finally, concluding remarks and some limitations can be found in Section 6.

2. Literature review

Previous research suggests a number of well-identified variables that have proved to be significant determinants of the number of crashes. They can be classified into broad categories, and a literature review reveals that the weather conditions constitute some of the most commonly used variables in empirical studies (Eisenberg, 2004; Andreescu and Frost, 1998; Brijs et al., 2008; Fridstrom et al., 1995; Hermans et al., 2006). Common examples of these variables are precipitation, hours of sun, temperature, air pressure, wind speed, relative humidity or snow for places with more extreme conditions.

Besides the weather, most studies (Van den Bossche et al., 2005b; Fridstrom, 1999; Fridstrom et al., 1995; Brijs et al., 2008; Quddus, 2008) have agreed that the volume of exposure (the amount of units exposed to accident risk) is a key variable to be included in any road accident model. The most commonly used measure to account for exposure is the traffic volume in terms of the number of vehicle-kilometers driven on the road network. However, there are various cases in which no valid measure of exposure is available (Hermans et al., 2006; Van den Bossche et al., 2005a). This issue has been discussed by Brijs et al. (2008) and Van den Bossche et al. (2005a,b), who wanted to test how well the safety model could perform without any measure of exposure and concluded that even without a variable such as exposure, valid models can be constructed.

Dummy variables can be considered as a valid alternative to address the consideration of exposure. Thus, for instance, if information on daily traffic exposure is not available for some reason, day-of-the-week dummies may account quite well for the dayof-the-week variability in exposure and still produce consistent results for the weather effects (Brijs et al., 2008).

Additional special situations have been frequently approached through the use of dummy variables. Hermans et al. (2006), Quddus (2008) and García-Ferrer et al. (2006) employed dummies in order to capture the effect of certain policies and countermeasures for the occurrence of traffic accidents, like the compulsory use of seat belts. Keay and Simmonds (2006) used dummies to capture the effect of school and public holidays, and Fridstrom (1999) presents a brief discussion on the use of the calendar effects and their relationship with fuel sales, a variable commonly used to proxy traffic volume. Of particular note for this study, Eisenberg (2004) included spatial dummies to account for tourism in states in which this activity is relatively higher than in others.

Nevertheless, in spite of the importance of evaluating tourism activity in understanding crash counts for an increasing set of regions, it seems clear that the need to take such influence into account through a quantitative measure has been neglected. The exception to this finding is Levine et al. (1995), who incorporate tourist arrivals within the set of explanatory variables. However, the variable turns out to be non-significant in the main model. It is important to stress that in that study, arrivals were introduced on a monthly basis, even though the model was set out in terms of daily data and thus unable to accurately capture daily variations in tourism.

This example demonstrates the relevance of using disaggregated temporal data. Often considered the most common context when exploring crash counts (Quddus, 2008; Lord and Persaud, 2000; Brijs et al., 2008), time series data can be classified into aggregated and disaggregated data (either at a temporal or a spatial level). This distinction is important because, while changes in crash counts on a highly aggregated level can be explained by structural changes, they cannot easily pick up patterns of seasonality or weather effects. In contrast, the lower the level of aggregation, the more feasible it is to study the effects of weather conditions, traffic volume, and holidays on changes in crash counts (Brijs et al., 2008). Several authors have therefore issued warnings regarding biases that are introduced by modeling crash counts at high levels of aggregation (Golob et al., 1990; Jovanis and Chang, 1989).

Evidence on how the use of daily (disaggregated) versus monthly (aggregated) data improves the results can be found in Eisenberg (2004), where both monthly and daily data are used to study the effects of precipitation on the number of accidents. With monthly data, a rather surprising negative relationship between such variables is found, while further disaggregation (daily) helps to find the expected positive relationship. The source of the contrasting, as the author suggests, appears to be a substantial negative lagged effect of precipitation across days within a state-month.

3. Methods

Within the regression framework, the choice of an adequate estimation technique for road traffic accidents depends on specific characteristics of the data. If the distribution followed by the number of accidents tends towards the normal, Ordinary Least Squares (OLS) can be an appropriate estimation technique (Levine et al., 1995; Keay and Simmonds, 2006). However, while the normality assumption is reasonable for (roughly) continuous dependent variables that can take on a large range of values, a count variable such as accidents cannot have a normal distribution (since the normal distribution is for continuous variables that can take on all values). If it takes on very few values, the distribution can be very different from the normal (Wooldridge, 2000). Furthermore, the assumptions for OLS methods are violated by the discrete, non-negative nature of accident count data and the reality that the variance in the number of accidents increases as the traffic flow increases (Lord and Persaud, 2000). For these reasons, there are alternative methods that are better suited to this particular type of data, whose starting point is the Poisson, a probability distribution that is specifically used for count data (Cameron and Trivedi, 2001; Gujarati, 2004; Wooldridge, 2000).

The Poisson for the number of occurrences of the event has a probability density function given by

$$Pr(Y = y) = \frac{e^{-\mu} \mu^y}{y!}, \quad y = 0, 1, 2, \dots,$$ (1)

where is the intensity or rate parameter (Cameron and Trivedi, 2001; Gujarati, 2004). The expression can also be referred to as P[and its two first moments are E[Y] = and V[Y] = -. The Poisson regression model can then be derived from the Poisson distribution by parametrizing the relation between the mean parameter and covariates (regressors) x (Wooldridge, 2000). The standard assumption is to use the exponential mean parametrization,

$$\mu = \exp(x_i'\beta), \quad i = 1, \dots, n, \tag{2}$$

where by assumption there are k linearly independent covariates, usually including a constant.

If we are interested in the effects of the xj on the mean response, there is little reason to go beyond the Poisson regression (Wooldridge, 2000). In fact, the purpose of using a statistical model in the present study lies basically in the effects of the regressors on the dependent variable. However, the Poisson's first two moments evidence the well-known equality of the mean and variance property of the distribution (Cameron and Trivedi, 2001; Gujarati, 2004), a feature better known as equi-dispersion. Since count data are often characterized by exhibiting over-dispersion (i.e., the variance is greater than the mean), the Negative Binomial has gained further popularity (Dionne et al., 1995; Fridstrom et al., 1995; Eisenberg, 2004; Lord, 2006; Martin, 2002; Noland and Oh, 2004) and has been considered more suitable, due to its ability to deal with this issue (Lord, 2006; Eisenberg, 2004; Fridstrom et al., 1995; Dionne et al., 1995; Quddus, 2008).

Analytically, it is hypothesized that the distribution of a random count y is Poisson, conditional on the parameter , so that f(y|) = exp(−)y/y! and = -, where is a deterministic function of x, a matrix of N × K explanatory variables; for example, exp(x ˇ), and > 0 is iid distributed with density g(|˛). Different observations may have different (heterogeneity), but part of this difference is due to a random (unobserved) component . If f(y|) is the Poisson density and g(), > 0, is the gamma density with E[] = 1 and V[] =˛, we obtain the negative binomial density,

$$h(y|\mu,\alpha) = \frac{\Gamma(\alpha^{-1} + y)}{\Gamma(\alpha^{-1})\Gamma(y+1)} \left(\frac{\alpha^{-1}}{\alpha^{-1} + \mu}\right)^{\alpha^{-1}} \times \left(\frac{\mu}{\mu + \alpha^{-1}}\right)^{y}, \quad \alpha > 0,$$ (3)

where (·) denotes the gamma integral that becomes a factorial for an integer argument (Cameron and Trivedi, 2001).

Since exp(·) is a nonlinear function, we can neither use linear regression methods nor nonlinear squares because all standard count data distributions exhibit heteroskedasticity. Instead, we will use maximum likelihood estimation methods (Wooldridge, 2000).

While it is common to estimate the parameters of the model using the maximum likelihood (ML) of a negative binomial specification, the ML estimation can also be performed under a number of alternative distributional assumptions. Moreover, when doubt exists about the form of the variance function, the use of the quasi-ML (QML), also referred to as pseudo-ML (PML), is recommended. Computationally, it is the same as Poisson ML, with the qualification that the variance matrix must be recomputed (Cameron and Trivedi, 2001).

The quasi-maximum likelihood estimators are robust in the sense that they produce consistent estimates of the parameters of a correctly specified conditional mean, even if the distribution is incorrectly specified. More specifically, this is done through a quasi-generalized pseudo-maximum likelihood estimator (QGPMLE) (Gourieroux et al., 1984a,b). If the variance of the Negative Binomial is ωi = ⁱ + ˛ⁱ and

$$\tilde{\alpha} = \frac{1}{n-k} \sum_{i=1}^{n} \frac{\left{ \left( y_i - \hat{\mu}_i \right)^2 - \hat{\mu}_i \right}}{\hat{\mu}_i^2}$$ (4)

is a consistent estimator of ˛, then the QGPMLE ˇˆ QGPML maximizes

$$\ln L{\text{LEFN}} = \sum{i=1}^{n} \left{ -\tilde{\alpha}^{-1} \ln \left( 1 + \tilde{\alpha} \mu{i} \right) + y{i} \ln \left( \frac{\tilde{\alpha} \mu{i}}{1 + \tilde{\alpha} \mu{i}} \right) + b \left( y_{i}, \tilde{\alpha} \right) \right},$$ (5)

where LEFN is a "linear exponential family with nuisance parameter" density (Cameron and Trivedi, 2005).

If the count data used for an application constitute a time series, an additional issue has to be considered. In particular, the independence of event occurrences in successive time intervals can be a reasonable assumption when the underlying stochastic process for such events, conditional on covariates, has no memory. But if the series presents serial correlation, there are various models to choose from. In the present study, we will focus on an autoregressive, or Markov model, which is a simple adjustment to the cross-section model detailed above that enters lagged values of y into the formula for the conditional mean of current y (Wooldridge, 2000).

4. Data description

This study benefits from different databases collected and/or elaborated by the Centre de Recerca Econòmica¹ (Center for Economic Research) that come from different sources. Thus, daily counts of road traffic accidents (where accidents involving pedestrians are not included) were provided by the Dirección General de Tráfico (Traffic Department) and consist of daily vehicle crashes that take place in each one of the Balearic Islands: Mallorca, Menorca and the Pitiuses (Eivissa & Formentera). The time series data range from January 1, 2000, to December 31, 2006, and can be seen in Fig. 1.

Meteorological variables are based on data provided by the Instituto Nacional de Meteorologia (National Institute of Meteorology) of the Balearic Islands and are taken from the airport stations. The daily stock of the population was elaborated directly from the Centre de Recerca Econòmica. Due to the importance of this variable for the present study, details about its estimation are described in the upcoming section.

4.1. The daily stock of residents and non-residents in the Balearic Islands

The variable used to determine the effect of tourism on the level of accidents comes from the estimation of the daily population stock of each one of the Balearic Islands and of the archipelago as a whole and was calculated for the first time by Riera and Mateu (2007) as a "Daily Indicator of Human Pressure" (DIHP) and has been also analyzed by Haldrup et al. (2007). The DIHP estimates the demographic burden that the Balearic Islands support on a given day, quantifying the people on the whole archipelago and each one of the islands.

The methodology for estimating the DIHP takes as a reference point the resident population on the first day of each year (RP0). Next, for each of the following days, the daily balance between arrivals (Ap) and departures (Dp) of people derived from interterritorial movements (passenger transit at ports and airports) is summed; an estimation of the population growth (Vp), which is a consequence of the "natural" evolution of the population for any given period (p), in this case a day, is summed as well. Vp is calculated under the assumption that the probability of either being born or dying is the same for every day of the year; hence it is derived

¹ www.cre.uib.es.

Fig. 1. Road accidents in the Balearic Islands.

from dividing the natural population growth by 365. Analytically:

$$DIHP_{RP} = RP0 + \sum{p=1}^{P} (A_p - Dp) + \sum{p=1}^{P} V_p$$ (6)

Although this equation summarizes the basic idea, there are a few methodological shortcomings. First, to the resident population on the first day of the year, it is necessary to add the non-residents (tourists) who are present in each of the islands on this initial day (namely PNRP0). In a similar way, the residents who are not present at this initial moment (PRNP0) should be subtracted. Considering the stock of people on each one of the islands as:

$$POP_0 = RP_0 + PNRP_0 - PRNP_0$$ (7)

the improvement of Eq. (6) takes the form:

DIHP = POP₀ + $$\sum_{p=1}^{P} (A_p - Dp) + \sum{p=1}^{P} V_p$$ (8)

The second point to be considered comes from the fact that airport arrivals are estimated by the authorities (while airport departures are recorded). This brings about the need to consider the stock of people, not only for the first period of the database, but also for January 1st of every year. Then the slight bias obtained from the two sources for January 1st of each year (own calculation through the methodology of Eq. (6), or re-estimating POP0 for each first day of the year) is weighted for each of the preceding days of the year. Then the net DIHP for each one of the islands can be calculated for the period January 1, 2000 to December 31, 2007.

Although through the variability of the DIHP one can assume that tourism is responsible for the strong seasonal component shown in the time series, the availability of Familitur data (the Spanish domestic tourism survey) allows for a decomposition into

Table 1 Regression based tests for overdispersion.

| | Test proposed by: | | | | | |------------------|------------------------------------------------|---------------------------------------|--|--|--| | | Cameron and Trivedi (1990)
Coeff. (p-value) | Wooldridge (1990)
Coeff. (p-value) | | | | | Balearic Islands | 0.0256 (0.0000) | 0.0237 (0.0000) | | | | | Mallorca | 0.0140 (0.0006) | 0.0179 (0.0002) | | | | | Menorca | 3.6759 (0.0000) | 0.0334 (0.7853) | | | | | Pitiuses | 0.1144 (0.0000) | 0.1099 (0.0122) | | | |

residents and tourists, both for the entire Balearic Archipelago and for each separate island.2 Because the Familitur survey collects monthly information on Spanish people (from the different regions) travelling to other regions (or abroad), the decomposition of the DIHP for the Balearic Islands into DIHP-Residents and DIHP-tourists can be computed and is shown in Fig. 2.

5. Empirical analysis

Following the methodological considerations explained in Section 3, Poisson and Negative Binomial models were initially estimated. However, given the presence of over-dispersion (the coefficients of the regression based tests3 in Table 1 are significant,

² For the cases of Mallorca, Menorca and the Pitiuses, an extrapolation was carried out, which is based on the daily population stock and monthly arrivals.

³ Cameron and Trivedi's (1990) test is based on an auxiliary regression of the squared difference between the dependent variable and the forecasted dependent variable, minus the dependent variable ((^y ⁻ ^y ^f)2 ⁻ ^y), on the square of the forecasted dependent variable (y f)2;Wooldridge's (1990)test is based on the estimation of the square of the standardized residuals minus one (sresid2 ⁻ 1) on the forecasted dependent variable (y f).

Fig. 2. DIHP for residents and tourists.

leading us to reject the Poisson restriction, v xi, ˇ ⁼ ^m xi, ˇ the results for the Poisson are not reported here. Estimation results from Negative Binomial models, using both ML and QML,4 are reported in Table 2. The presence of autocorrelation in the time series suggested the estimation of the dynamic models, including different lags of the dependent variable (Table 3).

Based on the data collected and described above, the equations for each island and for the entire archipelago were estimated. The variables included in the final estimations are: dummies to control for different levels of exposure observed over the week and for specific public/school holidays; hours of sunshine; precipitation in liters per square meter; mean daily temperature in Celsius; a dummy to control for the implementation of a safety law through which drivers are entitled to a given amount of "points" that are subtracted when committing traffic violations (CPP)5; a linear trend, usually included in similar exercises in order to control for improvements in the infrastructure; and the variables that account for the stock of the resident population (DIHP residents) and of tourists (DIHP tourists). Interaction and quadratic forms of the DIHP (both residents and tourists) were also considered but turned out to be non-significant and therefore were not included in the final estimations presented here.

Bearing in mind that the reference observation is the number of crashes on Sunday, in all cases except for Menorca, the signs of the estimated parameters for the day-of-the-week dummies follow the expected relationship, which indicates that there is a lower level of exposure on Sundays. Regarding the dummy variables for holidays, up to nineteen were initially considered, but only five turned out to be significant, indicating a lower level of accidents on important religious, national and local holidays. The other dummy considered (CPP) was also significant in the cases of Mallorca and the entire archipelago, with a negative sign that shows the effectiveness of the law approved by the Spanish Government. The linear trend, as in similar applications, was significant and negative for all cases under study.

In reference to the weather, both sunshine and precipitation were significant in most equations and followed the expected relationships, based on similar exercises (Eisenberg, 2004; Andreescu and Frost, 1998; Brijs et al., 2008; Fridstrom et al., 1995; Hermans et al., 2006). Thus, more hours of sun and more precipitation could be associated with higher numbers of accidents. Meanwhile, the average temperature was significant only in one equation, showing a positive relation with accident counts.

The population stock was highly significant in all cases, indicating a positive relationship between the population and crashes. The distinction between the tourist and resident populations shows, in a direct way, the impact of tourists in determining the level of accidents.

When comparing the ML and QML models in terms of the log-likelihood and the Akaike Information Criteria (AIC), the statistics do not differ much from one model to another. Thus, each island should find better statistics and the most suitable estimation. Meanwhile, the comparison between static and dynamic alternatives shows a slight improvement for the latter; hence the dynamic equations are preferred in terms of the AIC.

⁴ The variance is computed using the estimate obtained fromWooldridge's (1990) over-dispersion test.

⁵ This variable takes the value 1 for all days since February 1, 2006 and 0 otherwise.

Me tho d	Ba lea rics Co eff . (p lue ) -va		Ma llor Coe ff. ( alu e) ca p-v		Me Coe ff. ( alu e) nor ca p-v		Pit ius Coe ff. ( alu e) es p-v
	ML	QM L	ML	QM L	ML	QM L	ML	QM L
Int ept erc	( 0) 0.9 084 0.0	( 0) 0.9 084 0.0	( 4) 0.1 606 0.5	( 4) 0.1 648 0.5	( 0) −1 .42 86 0.0	( 1) −1 .38 85 0.0	( 0) −2 .12 55 0.0	( 0) −1 .62 99 0.0
Mo nda y	0.0 626 ( 0.0 3)	0.0 626 ( 0.0 3)	0.1 086 ( 0.0 0)	0.1 029 ( 0.0 0)	−0 .32 14 ( 0.0 1)	−0 .28 43 ( 0.0 5)	0.0 659 ( 0.4 8)	0.1 172 ( 0.1 7)
Tu esd ay	0.0 576 ( 0.0 5)	0.0 576 ( 0.0 5)	0.0 993 ( 0.0 0)	0.0 992 ( 0.0 0)	−0 .56 38 ( 0.0 0)	−0 .55 57 ( 0.0 0)	0.0 952 ( 0.3 1)	0.0 858 ( 0.3 1)
dn esd We ay	( 2) 0.0 363 0.2	( 2) 0.0 363 0.2	( 3) 0.0 682 0.0	1 ( 3) 0.0 68 0.0	( 0) −0 .37 88 0.0	( 2) −0 .35 71 0.0	( 1) 0.0 356 0.7	( 8) 0.0 475 0.5
Th day urs	0.1 029 ( 0.0 0)	0.1 029 ( 0.0 0)	0.1 643 ( 0.0 0)	0.1 643 ( 0.0 0)	−0 .15 84 ( 0.2 1)	−0 .13 33 ( 0.3 4)	−0 .06 65 ( 0.4 9)	−0 .00 03 ( 0.9 9)
Fri day	0.1 398 ( 0.0 0)	0.1 398 ( 0.0 0)	0.1 791 ( 0.0 0)	0.1 791 ( 0.0 0)	−0 .09 69 ( 0.4 3)	−0 .05 21 ( 0.7 0)	0.1 073 ( 0.2 5)	0.1 372 ( 0.1 0)
urd Sat ay	( 0) 0.0 954 0.0	( 0) 0.0 954 0.0	( 0) 0.1 119 0.0	( 0) 0.1 119 0.0	( 9) −0 .21 29 0.0	( 2) −0 .21 95 0.1	( 1) 0.1 150 0.2	( 4) 0.1 685 0.0
of shi Ho urs sun ne	0.0 059 ( 0.0 2)	0.0 059 ( 0.0 2)	–	–	–	–	0.0 119 ( 0.1 4)	–
Pre cip itat ion	0.0 070 ( 0.0 0)	0.0 070 ( 0.0 0)	0.0 049 ( 0.0 0)	0.0 046 ( 0.0 1)	0.0 212 ( 0.0 0)	0.0 209 ( 0.0 0)	0.0 048 ( 0.1 3)	–
Av tem atu era ge per re	( 8) 0.0 050 0.0	( 8) 0.0 050 0.0	–	–	–	–	–	–
CPP	( 0) −0 .16 21 0.0	( 0) −0 .16 21 0.0	( 0) −0 .18 86 0.0	( 0) −0 .18 91 0.0	–	–	( 7) −0 .03 79 0.6	–
Tre nd	−0 .00 02 ( 0.0 0)	−0 .00 02 ( 0.0 0)	−0 .00 02 ( 0.0 0)	−0 .00 02 ( 0.0 0)	−0 .00 03 ( 0.0 0)	−0 .00 03 ( 0.0 0)	−0 .00 01 ( 0.1 4)	−0 .00 02 ( 0.0 0)
DIH P to uri sts	7 ( 5 × 10− 0.0 0)	7 ( 5 × 10− 0.0 0)	7 ( 9 × 10− 0.0 0)	7 ( 9 × 10− 0.0 0)	6 ( 9 × 10− 0.0 0)	6 ( 9 × 10− 0.0 0)	6 ( 8 × 10− 0.0 0)	6 ( 8 × 10− 0.0 0)
esi den DIH P r ts	6 ( 0) 1 × 10− 0.0	6 ( 0) 1 × 10− 0.0	6 ( 0) 3 × 10− 0.0	6 ( 0) 3 × 10− 0.0	0−5 ( 9) 1.1 × 1 0.0	0−5 ( 4) 1.1 × 1 0.1	0−5 ( 0) 1.3 × 1 0.0	0−5 ( 0) 1.1 × 1 0.0
Ho lida
ys Au t 1 5 gus	−0 .39 06 ( 0.0 2)	−0 .39 06 ( 0.0 2)	−0 .44 07 ( 0.0 5)	–	–	–	−1 .21 03 ( 0.0 4)	–
obe Oct r 1 2	( 2) −0 .43 33 0.0	( 2) −0 .43 33 0.0	( 8) −0 .32 91 0.0	( 9) −0 .32 74 0.0	–	–	–	–
ber 8 De cem	−0 06 ( 0.0 1) .57	−0 06 ( 0.0 1) .57	–	−0 .44 00 ( 0.0 5)	–	–	–	–
De ber 26 cem	−0 .61 14 ( 0.0 1)	−0 .61 14 ( 0.0 1)	−0 .99 19 ( 0.0 0)	−0 .99 11 ( 0.0 0)	–	–	–	–
ber De 31 cem	( 1) −0 .81 07 0.0	( 1) −0 .81 07 0.0	( 2) −0 .69 64 0.0	( 2) −0 .69 53 0.0	–	–	–	–
atio isti Equ tat n s cs
Adj . R- d squ are	0.2 592	0.2 595	0.1 943	0.1 94 1	0.0 42 7	0.0 444	0.1 846	0.1 862
LR tist ic sta	11 02. 289	58 777 3	73 8.0 664	85 737 6.8	43 3.0 24 1	36 360 6.3	46 4.3 44	55 777 .25
bab ilit ( t) Pro LR sta y	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000
Me de an p. v ar.	7.9 339	7.9 339	6.6 326	6.6 326	0.4 216	0.4 216	0.8 647	0.8 91 1
S.D . de p. v ar.	3.5 579	3.5 579	3.0 337	3.0 338	0.9 968	0.9 968	1.0 719	1.1 029
2) ind (ps eud LR o-R ex	0.0 794	0.9 787	0.0 569	0.9 859	0.0 924	0.9 88	0.0 89	0.9 013
AIC	5.0 156	5.0 148	4.8 036	4.8 033	1.6 732	1.7 339	2.3 626	2.4 019

Table 3Dynamic estimations for crash counts.

tho d Me	lea eff . (p lue ) Ba rics Co -va		llor ff. ( alu e) Ma Coe ca p-v		ff. ( alu e) Me Coe nor ca p-v		ff. ( alu e) Pit ius Coe es p-v
	ML	QM L	ML	QM L	ML	QM L	ML	QM L
Int ept erc	0.7 645 ( 0.0 0)	0.7 645 ( 0.0 0)	0.3 465 ( 0.1 8)	0.3 465 ( 0.1 9)	−1 .27 51 ( 0.0 1)	−1 .26 40 ( 0.0 2)	−1 96 ( 0.0 0) .57	−1 98 ( 0.0 0) .57
Da 1 y	0.0 693 ( 0.0 2)	0.0 693 ( 0.0 2)	0.1 147 ( 0.0 0)	0.1 147 ( 0.0 0)	−0 .33 36 ( 0.0 1)	−0 .35 95 ( 0.0 1)	0.1 267 ( 0.1 3)	0.1 264 19 ( 0.1 3)
Da 2 y	( 4) 0.0 603 0.0	( 4) 0.0 603 0.0	( 0) 0.1 054 0.0	( 0) 0.1 054 0.0	( 0) −0 .57 71 0.0	( 0) −0 .58 14 0.0	( 0) 0.1 059 0.2	( 1) 0.1 055 0.2
Da 3 y	( 8) 0.0 389 0.1	( 8) 0.0 389 0.1	( 2) 0.0 726 0.0	( 2) 0.0 726 0.0	( 0) −0 .38 17 0.0	( 1) −0 .39 62 0.0	( 5) 0.0 637 0.4	( 6) 0.0 636 0.4
Da 4 y	0.1 141 ( 0.0 0)	0.1 141 ( 0.0 0)	0.1 715 ( 0.0 0)	0.1 715 ( 0.0 0)	−0 .17 50 ( 0.1 6)	−0 .20 26 ( 0.1 5)	0.0 142 ( 0.8 7)	0.0 137 ( 0.8 7)
Da 5 y	1 ( 0) 0.1 46 0.0	1 ( 0) 0.1 46 0.0	( 0) 0.1 893 0.0	( 0) 0.1 893 0.0	( 1) −0 .10 28 0.4	( 4) −0 .13 63 0.3	( 7) 0.1 509 0.0	( 7) 0.1 513 0.0
Da 6 y	( 0) 0.0 944 0.0	( 0) 0.0 944 0.0	( 0) 0.1 078 0.0	( 0) 0.1 078 0.0	( 1) −0 .20 45 0.1	( 8) −0 .19 46 0.1	( 2) 0.1 923 0.0	( 2) 0.1 925 0.0
Ho of shi urs sun ne	0.0 053 ( 0.0 4)	0.0 053 ( 0.0 4)	–	–	–	–	–	–
Pre cip itat ion	0.0 070 ( 0.0 0)	0.0 070 ( 0.0 0)	0.0 052 ( 0.0 0)	0.0 052 ( 0.0 0)	0.0 210 ( 0.0 0)	0.0 210 5 ( 0.0 0)	–	–
CPP	( 0) −0 .10 05 0.0	( 0) −0 .10 05 0.0	( 0) −0 .10 23 0.0	( 0) −0 .10 23 0.0	–	–	–	–
nd Tre	−0 .00 01 ( 0.0 0)	−0 .00 01 ( 0.0 0)	−0 .00 01 ( 0.0 0)	−0 .00 01 ( 0.0 0)	−0 .00 03 ( 0.0 0)	−0 .00 03 ( 0.0 0)	−0 .00 02 ( 0.0 0)	−0 .00 02 ( 0.0 0)
DIH P to uri sts	0−7 4 1 ( 0.0 0) ×	7 ( 4 × 10− 0.0 0)	0−7 5.1 × 1 ( 0.0 0)	0−7 5.1 × 1 ( 0.0 0)	5 ( 1 × 10− 0.0 0)	5 ( 1 × 10− 0.0 0)	0−6 6.8 × 1 ( 0.0 0)	0−6 6.8 × 1 ( 0.0 0)
den DIH P r esi ts	0−6 ( 0) 1.1 1 0.0 ×	0−6 ( 0) 1.1 × 1 0.0	0−6 ( 0) 1.6 × 1 0.0	6 ( 0) 2 × 10− 0.0	5 ( 0) 1 × 10− 0.1	5 ( 9) 1 × 10− 0.1	5 ( 0) 1 × 10− 0.0	5 ( 0) 1 × 10− 0.0
Acc ide (− 1) nts	( 0) 0.0 121 0.0	( 0) 0.0 121 0.0	( 0) 0.0 191 0.0	( 0) 0.0 191 0.0	( 6) −0 .05 19 0.1	( 7) −0 .04 50 0.2	1 ( 6) 0.0 36 0.0	1 ( 7) 0.0 36 0.0
Acc ide (− 2) nts	0.0 095 ( 0.0 0)	0.0 095 ( 0.0 0)	0.0 102 ( 0.0 0)	0.0 102 ( 0.0 0)	−0 .02 61 ( 0.4 5)	−0 .02 89 ( 0.4 8)	0.0 476 ( 0.0 1)	0.0 476 ( 0.0 1)
ide (− 3) Acc nts	( 9) 0.0 032 0.1	( 9) 0.0 032 0.1	1 ( 1) 0.0 07 0.0	1 ( 1) 0.0 07 0.0	( 8) −0 .04 86 0.1	( ) −0 .04 02 0.3	( 1) 0.0 304 0.1	1 ( 2) 0.0 30 0.1
ide (− 4) Acc nts	( 0) 0.0 103 0.0	( 0) 0.0 103 0.0	( 1) 0.0 079 0.0	( 1) 0.0 079 0.0	( 4) 0.0 266 0.3	( 7) 0.0 320 0.3	( 1) 0.0 195 0.3	( 2) 0.0 195 0.3
Acc ide (− 5) nts	0.0 054 ( 0.0 3)	0.0 054 ( 0.0 3)	0.0 094 ( 0.0 0)	0.0 094 ( 0.0 0)	−0 .03 90 ( 0.2 7)	−0 .03 50 ( 0.3 5)	−0 .00 54 ( 0.7 8)	−0 .00 52 ( 0.7 9)
Acc ide (− 6) nts	0.0 044 ( 0.0 7)	0.0 044 ( 0.0 7)	0.0 078 ( 0.0 1)	0.0 078 ( 0.0 1)	−0 .05 30 ( 0.1 6)	−0 .04 57 ( 0.2 7)	0.0 346 ( 0.0 7)	0.0 346 ( 0.0 8)
ide (− 7) Acc nts	( 7) 0.0 033 0.1	( 7) 0.0 033 0.1	( 3) 0.0 063 0.0	( 3) 0.0 064 0.0	( 6) −0 .05 32 0.1	( 6) −0 .04 75 0.2	( 4) −0 .02 31 0.2	( 5) −0 .02 32 0.2
Ho lida ys
Au t 1 5 gus	−0 .48 11 ( 0.0 0)	−0 .48 11 ( 0.0 0)	−0 .30 43 ( 0.0 8)	−0 .30 43 ( 0.0 9)	–	–	−1 .16 83 ( 0.0 2)	−1 .16 91 ( 0.0 3)
obe Oct r 1 2	( 2) −0 .41 36 0.0	( 2) −0 .41 36 0.0	–	–	–	–	–	–
ber 8 De cem	−0 .61 36 ( 0.0 1)	−0 .61 36 ( 0.0 1)	−0 .52 26 ( 0.0 2)	−0 .52 26 ( 0.0 2)	–	–	–	–
De ber 26 cem	−0 .59 36 ( 0.0 1)	−0 .59 36 ( 0.0 1)	−0 .96 01 ( 0.0 0)	−0 .96 01 ( 0.0 0)	–	–	–	–
ber De 31 cem	( 1) −0 .73 51 0.0	( 1) −0 .73 51 0.0	( 4) −0 .60 20 0.0	( 4) −0 .60 20 0.0	–	–	–	–
atio isti Equ tat n s cs
Adj . R- d squ are	0.2 737	0.2 739	0.2 042	0.2 045	0.0 485	0.0 466	0.2 039	0.2 042
LR tist ic sta	11 73. 666	83 180 2	78 9.5 542	11 814 38	44 2.7 436	32 22. 762	62 4.6 829	90 613 .91
bab ilit ( t) Pro LR sta y	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000	0.0 000
Me de an p. v ar.	7.9 369	7.9 369	6.6 66	6.6 66	0.4 227	0.4 227	0.8 898	0.8 898
S.D . de p. v ar.	3.5 617	3.5 617	3.0 352	3.0 352	0.9 979	0.9 979	1.1 028	1.1 028
2) ind (ps eud LR o-R ex	0.0 847	0.9 849	0.0 617	0.9 899	0.0 946	0.4 256	0.0 934	0.9 373
AIC	4.9 939	4.9 929	4.7 863	4.7 854	1.6 772	1.7 206	2.3 933	2.3 926

Table 4 Simulation of the impact of tourism on accidents.

| Method | ML | | QML | | | |------------------|---------|---------|---------|---------|--| | | Static | Dynamic | Static | Dynamic | | | Balearic Islands | −15.80% | −14.1% | −15.80% | −14.1% | | | Mallorca | −14.00% | −13.4% | −14.00% | −13.5% | | | Menorca | −27.80% | −28.5% | −28.10% | −28.9% | | | Pitiuses | −40.40% | −41.8% | −40.40% | −40.2% | |

Using the estimation results, since the population was split into residents and non-residents, simulations were carried out to evaluate the impact of tourism on road traffic accidents. From the data, we can see that the average number of daily traffic accidents within the observed period is 7.9 for the Balearics, 6.6 for Mallorca, 0.4 for Menorca and 0.9 for the Pitiuses. We then reestimated the equation with the substituted coefficients obtained from the estimation results under a hypothetical scenario in which the non-resident population was zero (Table 4).

It is worth noting that the percent decrease is higher in the cases of Menorca and the Pitiuses, which can be explained by the fact that during the high season, on both islands, tourists outnumber the resident population. Furthermore, differences between Menorca and the Pitiuses could be explained by the different typology of tourists that often visit the islands, with more family-oriented tourism in the case of Menorca. In any case, we note that all the estimations carried out yielded similar results in the estimation of the impact of tourism on crash counts.

6. Discussion and conclusions

The increase in tourist mobility in many destinations implies a rise in tourism-associated externalities, among which road accidents are one of the most significant. Within the transport literature, the analysis of road traffic accidents has resulted in many efforts aimed at understanding their causes. Meteorological conditions and the level of exposure have been pointed to as the most important factors, while the role of the tourism has been neglected.

Using a case study of the Balearic Islands, we relate the number and variation in the amount of accidents to the most common set of determining variables used in the reference literature and also include a new explanatory variable that incorporates tourism activity. Our results show how weather-related variables play a key role in the determination of road traffic accidents and also how tourism can be associated with a significant amount of the accidents that take place in the Balearics.

From the estimated results, different simulations were formulated to show the impact of tourism on road accidents. The simulation considered showed how, ceteris paribus, the level of accidents would decrease in 15.8% for the Balearics. These results should be connected to the fact that population in the Balearics grew from 875,000 in 2000 to 1 million in 2006, while the number of tourist arrivals grew from 11 to 12.5million over the same period. This gives a constant rate of about 12.5 tourists per inhabitant in the sample.

Tourism is often regarded as an important source of externalities, and this study provided empirical evidence to prove that, indeed, the presence of tourists leads to an increase in the number of road traffic accidents. The present study provides a quantification of this relationship for the case of the Balearic Islands. We obtain the expected positive relationship according to which the increase in population due to the presence of tourists leads to an increase in road accidents. The application of the model should be extended to other tourist areas to assign a tourism responsibility to crash counts. However, bearing in mind that the data used in this study, which benefits from the isolated characteristics of the islands, are not ideal, the application of this model in other destinations should be exercised cautiously in the sense of working with an improved dataset. This improvement should focus directly on making a distinction of participants (tourists and residents) in accidents and estimating the real use of rental cars.

Role of funding source

We gratefully acknowledge financial support from the Spanish (SEJ2007-67910) and local governments though the project entitled "Externalidades del Transporte Rodado en Economías Turísticas" and Catedra Sol Meliá de Estudios Turísticos.

Acknowledgements

The authors acknowledge help provided by David Martínez and the Centre de Recerca Econòmica in the estimation and/or provision of the databases. The authors would like to thank the valuable and enriching comments from Prof. Dr. Tom Brijs (Transportation Research Institute, Belgium), Daniel Eisenberg (Health Management and Policy, School of Public Health, University of Michigan, USA) and Lasse Fridström (Institute of Transport Economics, Norway), none of whom share any responsibility for any flaws in the article.

References

Andreescu, M.P., Frost, D., 1998. Weather and traffic accidents in Montreal, Canada. Climate Res. 9, 225–230.

Bentley, T., Meyer, D., Page, S., Chalmers, D., 2001. Recreational tourism injuries among visitors to New Zealand: an exploratory analysis using hospital discharge data. Tourism Manage. 22, 373–381.

Brijs, T., Karlis, D., Wets, G., 2008. Studying the effect of weather conditions on daily crash counts using a discrete time-series model. Accid. Anal. Prev. 40, 1180–1190.

Cameron, A.C., Trivedi, P.K., 1990. Regression-based tests for over dispersion in the Poisson model. J. Econometrics 46 (3), 347–364.

Cameron, C., Trivedi, P.K., 2005. Microeconometrics: Methods and Applications. Cambridge University Press, New York.

Cameron, C., Trivedi, P.K., 2001. Essentials of count data regression. In: Baltagi, B.H. (Ed.), A Companion to Theoretical Econometrics. Blackwell, Oxford (U.K.), pp. 331–348.

Carey, M., Aitken, M., 1996. Motorbike injuries in Bermuda: a risk for tourists. Ann. Emerg. Med. 28 (4), 424–429.

Clift, S., Page, S.J., 1996. Health and the International Tourist. Routledge, London. Dionne, G., Desjardins, D., Laberge-Nadeau, C., Maag, U., 1995. Medical conditions, risk exposure, and truck drivers' accidents: an analysis with count data regression models. Accid. Anal. Prev. 27 (3), 295–305.

Dickinson, J.E., Robbins, D., 2008. Representations of tourism transport problems in a rural destination. Tourism Manage. 29, 1110–1121.

Dickinson, J.E., Robbins, D., Fletcher, J., 2009. Representation of transport a rural destination analysis. Ann. Tourism Res. 30 (1), 103–123.

Eisenberg, D., 2004. The mixed effects of precipitation on traffic crashes. Accid. Anal. Prev. 36, 637–647.

Fridstrom, L., Ifver, J., Ingebrigsten, S., Kulmala, R., Krogsgard-Thomsen, L., 1995. Measuring the contribution of randomness, exposure, weather, and daylight to the variation in road accident counts. Accid. Anal. Prev. 27, 1–20.

Fridstrom, L., 1999. Econometric Models of Road Use, Accidents, and Road Investment Decisions, Volume II: An Econometric Model of Car Ownership, Road Use, Accidents, and Their Severity (Essay 3). Institute of Transport Economics, Oslo, Norway.

García-Ferrer, A., de Juan, A., Poncela, P., 2006. Forecasting traffic accidents using disaggregated data. Int. J. Forecasting 22, 203–222.

Golob, T.F., Recker, W.W., Levine, D.W., 1990. Safety of freeway median high occupancy vehicle lanes: a comparison of aggregate and disaggregate analyses. Accid. Anal. Prev. 22 (1), 19–34.

Gourieroux, C., Monfort, A., Trognon, C., 1984a. Pseudo-maximum likelihood methods: theory. Econometrica 52, 681–700.

Gourieroux, C., Monfort, A., Trognon, C., 1984b. Pseudo-maximum likelihood methods: applications to Poisson models. Econometrica 52, 701–720.

Gujarati, D.N., 2004. Gujarati: Basic Econometrics, Fourth edition. The McGraw–Hill Companies, p. 2004.

Gursoy, D., Jurowski, C., Uysal, M., 2002. Resident attitudes a structural modeling approach. Ann. Tourism Res. 29 (1), 79–105.

Haldrup, N., Hylleberg, S., Pons, G., Sansó, A., 2007. Common periodic correlation features and the interaction of stocks and flows in daily airport data. J. Bus. Econ. Stat. 25, 21–32.

Hall, D.R., 1999. Conceptualizing tourism transport: inequality and externality issues. J. Transp. Geogr. 7, 181–188.
Hermans, E., Wets, G., Van den Bossche, F., 2006. Frequency and severity of Belgian
road traffic accidents studied by state space methods. J. Transp. Stat. 9 (1), 63–76. Howard, R.W., 2009. Risky business? Asking tourists what hazards they actually encountered in Thailand. Tourism Manage. 30, 359–365.
Hutchings, C.B., Knight, S., Reading, J.C., 2003. The use of generalized estimating equations in the analysis of motor vehicle crash data. Accid. Anal. Prev. 35, 3–8. Jovanis, P.P., Chang, H.-L., 1989. Disaggregate model of highway accident occurrence
using survival theory. Accid. Anal. Prev. 21 (5), 445–458. Keay, K., Simmonds, I., 2006. Road accidents and rainfall in a large Australian city.
Accid. Anal. Prev. 38, 445–454. Leggat, P.A., Leggat, F.W., 2003. Reported fatal and non-fatal incidents involving
tourists in Thailand, July 1997–June 1999. Travel Med. Infect. Dis. 1, 107–113. Levine, N., Kim, K.E., Nitz, L.H., 1995. Daily fluctuations in Honolulu motor vehicle accidents. Accid. Anal. Prev. 27, 785–796.
Lindberg, G., Ajo, P., Brito da Silva, A., Krupp, R., Osório Nunes, M.R., Peirson, J., Schneglberger, G., Smolders,W., 1999. Calculating transport accident costs. Final report of the expert advisors to the High Level Group on Infrastructure Charging (Working Group 3). Sweden.
Lindberg, G., 2001. Traffic insurance and accident externality charges. J. Transp. Econ. Policy 35 (3), 399–416.
Lord, D., Persaud, B.N., 2000. Accident prediction models with and without trend: application of the generalized estimating equations procedure. Trans. Res. Rec. 1717, 102–108.
Lord, D., 2006.Modelingmotor vehicle crashes using Poisson-gammamodels: examining the effects of low sample mean values and small sample size on the estimation of the fixed dispersion parameter. Accid. Anal. Prev. 38, 751–766.
Martin, J.L., 2002. Relationship between crash rate and hourly traffic flow on interurban motorways. Accid. Anal. Prev. 34, 619–629.
Noland, R.B., Oh, L., 2004. The effect of infrastructure and demographic change on traffic-related fatalities and crashes: a case study of Illinois county-level data. Accid. Anal. Prev. 36, 525–532.
Page, S.J., Meyer, D., 1996. Tourist accidents an exploratory analysis. Ann. Tourism Res. 23 (3), 666–690.
Page, S.J., 2009. Current issue in tourism: the evolution of travel medicine research: a new research agenda for tourism? Tourism Manage. 30 (2), 149–157.
Palmer-Tous, T., Riera-Font, A., Rosello-Nadal, J., 2007. Taxing tourism: the case of rental cars in Mallorca. Tourism Manage. 28, 271–279.
Quddus, M.A., 2008. Time series count data models: an empirical application to traffic accidents. Accid. Anal. Prev. 40, 1732–1741.
Riera, A., Mateu, J., 2007. Aproximación al volumen de turismo residencial en la Comunidad Autónoma de las Illes Balears a partir del cómputo de la carga demográfica real. Estudios Turísticos 174, 59–71.
Ryan, C., 1997. The Tourist Experience, 1st ed. Cassell, London.
UNWTO, 2008. Tourism Highlights 2008 Edition.
UNWTO, 2009. Facts and Figures Information, Analysis and Know-How. Historical Perspective of World Tourism, Retrieved April 02, 2009 from http://unwto.org/facts/eng/historical.htm.
Van den Bossche, F.,Wets, G., Brijs, T., 2005a. Predicting Road Crashes Using Calendar Data. Transportation Research Institute, Belgium.
Van den Bossche, F., Wets, G., Brijs, T., 2005b. Role of exposure in analysis of road accidents: a Belgian case study. Trans. Res. Rec. 1908, 96–103.
Wang, X., Abdel-Aty, M., 2006. Temporal and spatial analyses of rear-end crashes at signalized intersections. Accid. Anal. Prev. 38, 1137–1150.
Wilks, J., 1999. International tourists, motor vehicles and road safety: a review of the literature leading up to the Sydney 2000 Olympics. J. Travel Med. 6, 115– 121.
Wilks, J., Watson, B., Faulks, I.J., 1999. International tourists and road safety in Australia: developing a national research and management programme. Tourism Manage. 20, 645–654.
Wooldridge, J.M., 1990. Quasi-likelihood methods for count data. In: Pesaran, M.H., Schmidt, P. (Eds.), Handbook of Applied Econometrics, vol. 2. Blackwell, Malden, MA, pp. 352–406.
Wooldridge, J.M., 2000. Introductory Econometrics: A Modern Approach, 2nd ed. South Western College Publishing, Thomson Learning.

rosselloroadaccidentstourismcase2011.md 70 KB History Raw