Trip Generation Models for Transportation Impact Analyses of Shopping Centers in Croatia

Abstract

For effective transportation planning, land use, travel behavior, and infrastructure capacity should be optimized to support sustainable urban growth and reduce congestion. Every new site development generates traffic volume, which can affect the quality of traffic flow in the surrounding road network. Therefore, trip generation, which predicts future travel demand, is a crucial step in the traditional four-step transportation model. In this context, the main objective of this study is to develop a model for estimating vehicle trip generation due to the construction of a shopping center, which is a significant traffic generator. The survey was conducted in Split (Croatia) at five existing locations, and linear regression analysis was used to develop models for different time periods. The results indicated that vehicle trips are strongly correlated with the gross floor area of shopping centers, with a high coefficient of determination. Additionally, this study presents a comparison of measured traffic volumes with estimates using ITE Trip Generation Manual equations. The findings suggest that these vehicle trip estimates should be reduced by approximately 40%. Since no previous studies have been conducted on the impact of land use on trip generation in the Republic of Croatia, the developed models represent a first step in creating a database that should be expanded with new data. Estimating the traffic generated by a new site development is a crucial component of traffic management, as it helps planners and engineers assess its impact on the surrounding road network and implement necessary measures to ensure efficient and safe traffic flow.

1. Introduction

Transportation planning is the process of designing and managing transportation systems to ensure efficient, safe, and sustainable movement of people and goods. It involves analyzing the existing infrastructure, forecasting future needs, and developing strategies to improve mobility while considering environmental, economic, and social factors [1]. By accurately predicting future traffic volume, planners and engineers can design infrastructure and regulations that ensure long-term functioning and efficiency.

Transportation models are an integral part of this process and serve as an analysis tool for transportation planners, helping decision makers in evaluating alternative proposals [2]. The traditional model of transportation planning is a four-step travel demand model, which consists of trip generation, trip distribution, mode choice (modal split), and traffic assignment [3]. This model, in addition to being used in urban and regional traffic studies, is also used in conducting transportation impact studies for new site developments [4].

Trip generation is the first step of the four-step prediction model, and the quality of trip generation estimates directly affects the quality of the estimates of the other model phases (trip distribution, mode choice, and trip assignment). According to [4], the need for a transportation impact study is suggested when the new facility generates 100 added vehicle trips during the peak hours of the adjacent street network or during the peak hours of the facility. In this regard, estimating the number of trips generated by a new development is crucial for making appropriate decisions that balance development needs with transportation capacity, safety, and mobility.

In the United States, the Institute of Transportation Engineers (ITE) Trip Generation Handbook [5] and the corresponding Trip Generation Manual [6] are the primary sources of data and methods for estimating vehicle trips in conducting transportation impact studies. The ITE Trip Generation Manual was first published in 1976 and is now in its eleventh edition, published in 2021, with trip generation data for more than 170 land uses [7]. Research on the number of trips generated for different land uses is continuously conducted in the USA, and the existing database is updated with new findings [8,9].

For each land use, the ITE Trip Generation Manual presents the independent variable or variables that appear to explain the variation in the number of generated trips. An independent variable is defined as a measurable and predictable unit describing the studied trip generator (e.g., gross floor area, Gross Leasable Area, number of employees, dwelling units, number of seats) [5]. Many factors, such as socioeconomic development, motorization rate, different transportation mode availability, and population habits, influence trip generation. Therefore, it is desirable for each country or region to conduct its own research to develop trip generation models that reflect its own local conditions. Accordingly, although this manual is globally recognized as the primary reference for estimating future trip demand, many countries have developed their own trip generation models [10,11,12,13,14].

In the Republic of Croatia, as well as in neighboring countries, there is no legal obligation to conduct traffic impact analyses before constructing a new facility. This could explain why there has not been any research on trip generation models that consider local characteristics and resident habits. Despite the lack of legal regulation, planners, investors, and local authorities often request traffic impact studies from transportation specialists to analyze the accessibility of new facilities, the quality of traffic flow (level of service), and the sustainability of the transportation system. In Croatia, especially in growing urban areas, there is increasing pressure on transport infrastructure due to the expansion of commercial zones, including large shopping centers. Shopping centers are among the largest trip generators, attracting a high number of visitors. These developments often lead to increased traffic volume, parking demand, and challenges in managing vehicle access, particularly during peak periods. However, the tools used in traffic impact analysis are often based on foreign data sources, such as the ITE Manual, which offers useful benchmarks, but it is based on data from North America, which do not reflect the travel behavior, motorization rate, and modal split typical of Croatian cities. This presents a research gap in the availability of context-specific data for this region of Europe; this was a motivation for this study. Furthermore, there is no standardized national trip generation database or guideline currently in use, which creates inconsistencies in planning and makes it difficult for local authorities to evaluate the real impact of new developments. This study aims to address this problem by providing empirical, context-specific trip generation data for shopping centers in Croatia, contributing to more reliable and locally based urban mobility planning.

The construction of a new shopping center will significantly increase traffic volume on the nearby existing transportation network, requiring an impact assessment to ensure adequate infrastructure. Therefore, the main objective of this study is to define reliable models for estimating the number of vehicle trips generated by local shopping centers during peak periods. As a part of this study, traffic counts on the access roads to five shopping centers in Split (Croatia) was performed and peak time periods, and corresponding traffic volumes were identified. Linear regression analysis was used to develop models for predicting the number of vehicle trips for different time periods. The formulated models have a high coefficient of determination, making them suitable for future estimates of vehicle trip generation for shopping centers. Additionally, a comparison was made between the measured values and the values obtained by the models from the American ITE Manual [6], which highlights how local conditions and population habits can significantly influence trip generation estimates. Given the continued reliance on international models that may not accurately represent regional travel behavior, the findings presented here offer valuable contributions for traffic management and urban planning. By understanding how shopping center characteristics influence trip generation, transportation planners can better predict traffic volumes, optimize road design and signal timing, and ensure sufficient parking capacity in the area of a shopping center. Additionally, these insights support sustainable transport strategies by identifying opportunities to integrate public transportation, improve pedestrian and cycling access, and reduce car dependency.

This study is organized in six sections, as follows. Following the introduction, Section 2 presents a literature review on trip generation databases and studies in different countries. Section 3 details the research methodology, including data collection and the applied statistical method, while Section 4 presents the results and discussion. In Section 5, the main conclusions, limitations, and future study recommendations are presented.

2. Literature Review

This review of the available literature includes references that are relevant to trip generation databases and procedures used in different countries as well as providing details on studies that are related to the estimation of the trip generation characteristics of shopping centers.

2.1. Trip Generation Databases

As mentioned earlier, the ITE Trip Generation Manual is a widely recognized reference in transportation engineering that is used to estimate the number of trips generated by various land uses and developments. It provides a summary of trip generation data collected from studies throughout the United States that are continuously conducted. In addition to this manual, several states in the USA have conducted trip generation studies and defined trip generation rates that reflect local characteristic of their area [15,16]. The results of the studies indicate that trip generation rates were generally lower than comparable rates in the ITE Manual.

While the United States relies heavily on the ITE Trip Generation Manual, some other countries have developed their own models and guidelines tailored to local travel behavior, land use patterns, and multimodal planning principles.

In the United Kingdom and Ireland, the TRICS (Trip Rate Information Computer System) is a national system for trip generation analysis, which is used for estimating how different types of developments impact road networks [10]. This database contains traffic survey data from thousands of sites across the UK and parts of Ireland. The New Zealand Transport Agency provides a comprehensive research report on trips and parking related to land use in New Zealand [17]. It updates and expands on a previous 2001 report [18], providing new data on travel modes, trip purposes, and comparisons with international databases.

The TRIPS (Trip Rate Information Processing System) [11] is Australia’s first national database for land use-based trip and parking generation surveys. Perera and Bell [19] provide a summary of insights gained from the background research underpinning the TRIPS project. This study explores the challenges in existing methodologies and the impact of creating a centralized data resource for land use-based trip generation surveys. The authors point out a strong need for the development of national procedural guidance to enable a harmonized method of data collection and reporting by practitioners.

The Abu Dhabi Trip Generation and Parking Rates Manual (ADTGM) [12] was developed by the Department of Transport to support sustainable urban growth and effective transport planning. Based on extensive data collection across nearly 400 land use types, the manual provides reliable estimates of traffic and parking demand for new developments. It follows international best practices that have been adapted to local conditions.

De Gruyter states in his study [20] that a software program for estimating trip generation known as Ver Bau has been developed in Germany (also used in Austria and Switzerland), but only limited information about this program is available in English. The Ver_Bau software developed by Bosserhoff [14] enables traffic volume forecasts for urban development planning projects. Traffic volume is estimated using an integrated method for all modes of transport for residential, commercial, retail, and leisure uses, as well as other traffic-intensive facilities such as educational institutions.

In the Netherlands, the CROW institute provides guidance for estimating trip generation with a strong focus on modal split and sustainable mobility. The CROW is a knowledge institute for transport, infrastructure, and mobility, and their publication contains trip rates and parking rates for numerous activities [21]. In the new publication, CROW publication 317, all key figures on parking and traffic generation are clearly presented for seven main groups: living, working, shopping and groceries, sports and relaxation, catering and recreation, healthcare and (social) facilities, and education. In total, approximately 100 destinations are described [13].

To create a database on the impact of land use on surrounding traffic, a trip generation study needs to be conducted for each specific land use type. Since shopping centers are locations that attract a significant number of visitors, their impact has been analyzed in various studies.

2.2. Trip Generation Studies on Shopping Centers

Andrade et al. [22], in their study, evaluated the validity of ITE trip generation models for shopping centers in Brazil. Based on data collected from 16 locations in Rio de Janeiro, they concluded that ITE estimates were well above the results for local shopping centers. Goldner and Portugal [23] developed regression models based on studies of shopping centers in Brazil and Spain, including trip generation and modal choices. The model obtained relates the Gross Leasable Area (GLA) at the shopping centers to the number of trips attracted.

In their study, Kikuchi et al. [24] present the trip attraction rates of the shopping centers in Northern New Castle County in Delaware based on eighteen analyzed locations. They developed two approaches to compute the trip attraction rate: microscopic and macroscopic. Their conclusion is that the microscopic approach, which deals with the relationship between the trip attraction rates of individual stores and the shopping center as a whole, gives better results compared to the macroscopic approach.

Mamun et al. [25] developed two macroscopic trip generation models with different independent variables, based on data for six medium- and small-sized shopping malls in Dhaka. The variables included in the models are the following: gross floor area (GFA), the number of parking spaces, the number of restaurants, and the total number of stores. The authors concluded that the total number of stores is better predictor than GFA. Al-Masaeid et al. [26] developed a trip- and parking-generation model for shopping centers in Jordan and found that GFA and the number of employees are the most significant variables in trip generations. A study was conducted on selected eighteen shopping centers in Mumbai Metropolitan Region in India [27]. The authors developed trip generation models based on regression analysis with five different variables. They also stated that the Gross Leasable Area is highly correlated with the built-up area, the number of stores, the number of stores plus kiosks, and the number of parking spaces.

Rakić et al. [28] proposed a model to estimate the traffic volume generated by shopping centers in medium-sized towns. A trip generation study was conducted at eight different locations in Bosnia and Herzegovina and the Republic of Serbia. By applying regression analysis to the established database, the authors compared models by combining multiple independent variables. Based on the obtained results, they concluded that the Gross Leasable Area (GLA) of the shopping center is the most significant predictor and that the model with that independent variable provides the best estimate of trip generation.

Although previous studies have explored trip generation rates for shopping centers, they are based on local datasets for related countries and may not reflect travel behavior in Croatia. Our study addresses a specific gap in the literature related to trip generation analysis of shopping centers in the Croatian context, using empirical data collected onsite. It also provides an insight into the comparison of measured traffic volumes and ITE vehicle trip estimates for shopping centers.

3. Materials and Methods

The applied research methodology is based on a structured approach consisting of several fundamental steps, and it is commonly used for conducting trip generation studies. The sequence of the main activities carried out in this research is presented in the workflow diagram shown in Figure 1.

The first three steps are described in more detail later in this chapter. Step 4 involves a statistical analysis of the collected data to assess the feasibility of developing predictive models for trip generation, as well as the presentation of the resulting models. The field-collected and analyzed data provide more accurate and locally relevant insights compared to relying solely on external sources, such as the ITE Trip Generation Manual.

In addition to developing models tailored to local conditions, this study also includes a comparison between the measured vehicle trip data and the corresponding estimates derived from the ITE models presented in [6], as described in Step 5. Step 6 highlights the practical traffic planning benefits of the developed models. Based on the results obtained, several conclusions were drawn.

Steps 4, 5, and 6 are further explained in the Results and Discussion Section (Section 4.1, Section 4.2, and Section 4.3, respectively), while step 7 is addressed in the final Section 5 of this study, Conclusions.

3.1. Selection of Study Locations

A shopping center is an integrated group of commercial establishments that is planned, developed, owned, and managed as a unit [6]. It is primarily intended for shopping and contains many stores offering a variety of products. However, beyond shopping, modern shopping centers often feature restaurants, movie theaters, gyms, banks, children’s amusement parks, and spaces for hosting various events. Today, shopping centers have evolved beyond just being places for shopping; they have become sites for social interaction and meetings.

With their wide range of offerings, shopping centers attract a significant number of visitors with various travel purposes, making them one of the largest trip generators. Although there are multiple transportation options to get to shopping centers, private vehicles are the most commonly used mode of transport. This highlights the need for well-planned access roads with sufficient capacity to ensure smooth and efficient traffic flow. Additionally, it is crucial to provide adequate parking areas adjacent to the shopping center with a sufficient number of parking spaces to meet visitor demand. Good accessibility and adequate parking facilities are key factors that further increase the attractiveness and trip generation potential of shopping centers.

Therefore, when planning and selecting a location for a new shopping center, it is essential to analyze the transportation network in the surrounding area to assess accessibility to the location and the capacity of the existing road infrastructure to accommodate the additional trips generated by the shopping center.

Since a significant number of visitors travel using private vehicles, it is advisable to conduct traffic impact studies before the construction of a facility. These studies help in determining the impact of newly generated trips on the traffic flow of the adjacent local road network. High trip generation can lead to increased vehicular traffic, especially during peak hours, which may result in longer travel times, greater delays, and traffic congestion. This requires careful traffic management to reduce congestion, prevent accidents, and optimize traffic flow. Road and traffic management includes various types of measures to improve the quality of traffic flow such as construction of new roads or intersections, improvement of the existing street network, new signal regulation or optimization, public transport integration, and safety measures. The results of a traffic impact study should be used to determine the strategies and measures required to satisfy traffic demand as well as preserve an acceptable level of service at a new location. The fundamental data for such traffic analyses are the number of trips generated by the planned shopping center, specifically the expected number of vehicles on the existing network. In the Republic of Croatia, the construction of shopping centers is a relatively recent occurrence. The first shopping center, the Importanne Center, was opened in 1994 in downtown Zagreb, near the railway station. In other major cities across Croatia, the first shopping centers were built approximately twenty years ago, around the early 2000s. The transition and democratic changes in Croatia had a significant impact on the retail sector and the purchasing power of the population. As a result, shopping centers have become key destinations for daily trips related to shopping, entertainment, recreation, and social interaction. Since there has been no systematic research in Croatia on the number of trips attracted by different land uses, the results presented in this study represent the first step in this direction. This study outlines the development of a predictive model for estimating the number of vehicle trips generated as a consequence of shopping center construction.

A trip generation study should include the land use types to be surveyed, the number of survey sites, the survey period, the independent variable data, and the traffic counting methodology. To establish a local trip generation rate or equation, it is recommended to survey at least three sites (preferably five) [5]. For the purposes of this study, five shopping centers were selected in the city of Split and its surrounding area. The city of Split has a population of approximately 161,000, with an additional 90,000 residents living in nearby smaller cities and municipalities (Solin, Kaštela, Trogir, Omiš, and Podstrana), resulting in a total population of over 260,000 that gravitates towards these shopping centers [29]. Among the five selected shopping centers, identified as (1) Mall of Split, (2) City Center One, (3) Joker, (4) Salona Mall, and (5) Emmezeta, only one center (3) is located in the city center of Split, whereas the remaining four (1, 2, 4, and 5) are situated on the outskirts. The locations of the shopping centers analyzed and included in the study are shown in Figure 2.

3.2. Data Collection

The construction and opening years of the selected shopping centers vary between 2002 and 2016, and, at the time of the present study, all centers were fully operational. According to the selection criteria, each development had to be at least two years old [5]. All shopping centers have the same operating hours, from Monday to Saturday, opening at 9:00 AM and closing at 9:00 PM. Due to legal regulations, Sunday operations are limited to 16 working Sundays per year, with the specific Sundays being decided by the management of each center.

All shopping centers are located in the vicinity of major roads and four of them are on the outskirts. This strategic placement ensures that visitors from surrounding areas traveling to the Split region do not significantly burden the city center’s traffic network. The entrances to garages and parking lots designated for these centers have separate access routes, making it easy to accurately determine the number of vehicles whose drivers’ primary destination is the shopping center.

Table 1. Some characteristics of analyzed shopping centers.

Location	Shopping Center	Opening Year	GLA * (m2)	Parking Spaces
1	Mall of Split	2016	61,000	2320
2	City Center One	2010	57,000	2700
3	Joker	2007	35,000	780
4	Salona Mall	2002	22,000	720
5	Emmezeta	2003	42,000	905

* Gross Leasable Area.

presents the key characteristics of the shopping centers included in the study, such as their names, opening years, total Gross Leasable Area (GLA), and the number of parking spaces. GLA refers to the total floor area of a building for which the tenant pays rent and that is designed for the tenant’s occupancy and exclusive use. The size of the shopping center, measured in square feet or square meters of leasable space, is a primary factor in determining trip generation rates [30].

In the American ITE Trip Generation Manual, as well as in numerous previous studies conducted in various countries that have examined trip generation rates of shopping centers [26,27,28], it has been established that Gross Leasable Area (GLA) has the greatest influence on trip attraction. Although other potential influencing variables have been analyzed (number of employees, number of parking spaces, number of households, and number of stores), GLA remains the dominant factor. This finding is not surprising, as it is evident that a larger leasable area generally implies a greater number of stores or other attractive amenities, which in turn leads to more employees and a higher number of parking spaces, meaning that these independent variables correlate with each other. Therefore, in this study, GLA was considered the primary independent predictor variable for forecasting the number of vehicle trips generated by shopping centers. Additionally, GLA was used as the predictor variable for comparison with trip generation data from the ITE Manual [6].

When conducting transportation impact studies, the analysis should focus on the time period when the combined traffic volumes from the research site and adjacent street network are at their maximum [5]. The peak hour of the generator and the peak hour of the surrounding road network often do not occur at the same time, which is also the case for shopping centers. For this reason, separate trip generation models were defined for both the peak hour of the generator and the peak hour of the adjacent network to provide accurate estimates of generated trips. Since travel habits and trip purposes differ between weekdays and weekends, the number of trips on Saturdays was also analyzed.

To determine the correlation between the number of trips made by private vehicles (dependent variable) and the Gross Leasable Area (GLA) of the shopping center (independent variable), it was necessary to count vehicle trips at the selected shopping centers. Each of the observed shopping centers has two access roads to the entrances/exits to the garage or parking lot. For the purposes of this study, vehicle counts were conducted manually on all access roads to the parking area to ensure full coverage of all vehicle’s entrances/exits. Figure 3 shows the locations where traffic counts were carried out to record the number of vehicles entering and exiting each analyzed shopping center. These points are marked with red arrows.

Traffic counting was conducted in May and June 2023, from 3:00 PM to 9:00 PM, on a Thursday as a typical weekday and on a Saturday as a typical weekend [31]. For one of the locations (City Center One), data were used from the counting that was carried out during two working days and two Saturdays in May 2018 (before the COVID-19 pandemic) [32]. Data on vehicle arrivals and departures were recorded in 15 min intervals to determine both the peak 15 min period and the peak hour of the generator (shopping center). The morning peak hour of the adjacent traffic network is from 7:15 AM to 8:15 AM, when most people arrive at their workplace, and the afternoon peak hour is from 3:30 PM to 4:30 PM, when people depart from their workplace. Since all shopping centers open at 9:00 AM, they are closed during the morning peak hour of the adjacent network. Therefore, the traffic counting period was selected from 3:00 PM to 9:00 PM, covering both the afternoon peak hour of the surrounding street network and the working time of the shopping centers. By conducting traffic counting between 3:00 PM and 9:00 PM, it was possible to determine the number of generated vehicle trips during the peak hour of the generator (shopping center), which, under local conditions, occurs in the afternoon hours and the number of attracted vehicle trips during the afternoon peak hour of the adjacent street network. Based on the collected data, graphs were created showing the number of vehicles entering and exiting in 15 min intervals. This allowed for the identification of the following data:

• The peak hour for each shopping center;
• The corresponding number of vehicle trips during this peak period;
• The number of vehicles generated during the peak hour of the adjacent street network.

Table 2. Data on vehicles volume during the peak hour of adjacent street network on a weekday.

Location	Shopping Center	Peak Hour Adjacent Street Network	Vehicles Total	Vehicles Entering	Vehicles Exiting
1	Mall of Split	3:30–4:30 PM	1203	583 (48%)	620 (52%)
2	City Center One	3:30–4:30 PM	1392	752 (53%)	660 (47%)
3	Joker	3:30–4:30 PM	590	304 (52%)	286 (48%)
4	Salona Mall	3:30–4:30 PM	401	227 (56%)	174 (44%)
5	Emmezeta	3:30–4:30 PM	564	267 (47%)	297 (53%)

presents the vehicle entry and exit data during the peak hour of the adjacent street network on a weekday. Additionally, for all locations, the percentage of vehicles entering and exiting the shopping center is provided, which is essential for the directional distribution of vehicle trips.

It is evident that larger shopping centers, such as Mall of Split and City Center One, typically attract more vehicle trips. Furthermore, based on the collected traffic volume data, the peak hour of the generator is not exactly the same for all analyzed shopping centers. However, for all locations, it occurs within the time period from 5:30 PM to 8:00 PM.

Table 3. Data on vehicles volume during the peak hour of shopping center on a weekday.

Location	Name	Peak Hour Weekday	Vehicles Total	Vehicles Entering	Vehicles Exiting
1	Mall of Split	7:00–8:00 PM	1925	927 (48%)	998 (52%)
2	City Center One	6:30–7:30 PM	2073	1041 (50%)	1032 (50%)
3	Joker	6:00–7:00 PM	788	366 (46%)	422 (54%)
4	Salona Mall	6:00–7:00 PM	538	305 (57%)	233 (43%)
5	Emmezeta	6:00–7:00 PM	1007	476 (47%)	531 (53%)

presents the number of vehicle trips during the peak hour of the shopping center on a weekday, while

Table 4. Data on vehicles volume during the peak hour of shopping center on a Saturday.

Location	Name	Peak Hour Saturday	Vehicles	Vehicles Entering	Vehicles Exiting
1	Mall of Split	6:45–7:45 PM	1831	910 (49%)	921 (51%)
2	City Center One	6:15–7:15 PM	1810	937 (52%)	873 (48%)
3	Joker	5:30–6:30 PM	594	353 (59%)	241 (41%)
4	Salona Mall	6:00–7:00 PM	585	328 (55%)	257 (45%)
5	Emmezeta	5:30–6:30 PM	815	414 (51%)	401 (49%)

provides the corresponding data for a Saturday.

According to graphical data on popular times from the websites of each of the assessed shopping centers, the highest number of visitors (persons) occurs on Saturday afternoons. However, when comparing these visitors’ popular times data and data on field-counted vehicles from Table 3 and Table 4, an opposite trend is observed: a higher number of vehicles was recorded on weekdays than on Saturdays, suggesting that peak visitor times do not necessarily correspond to peak hours of vehicle volumes attracted by a shopping center. This can be explained by the following factors: (1) on Saturdays (a weekend day), the vehicle occupancy rate is higher due to family trips to shopping centers; (2) on Saturdays (a weekend day), a higher number of visitors use public transport and non-motorized transport (walking, cycling), especially among the younger population who do not own cars. These factors contribute to a larger number of visitors on Saturdays overall, despite the vehicle count being lower compared to weekdays.

Since the primary objective of this study is to develop a model for estimating the number of generated vehicle trips that could potentially impact the surrounding road network, the number of visitors (person trips) was not specifically recorded. Additionally, the measured vehicle trip data were compared with the estimated number of vehicle trips obtained by applying the ITE Trip Generation Manual equations.

3.3. Applied Statistical Analysis

In the context of trip generation, regression analysis is the most commonly used approach for developing predictive models based on land use characteristics such as floor area, number of dwelling units, and number of employees [6]. It is a mathematical–statistical procedure used to determine an appropriate functional relationship between one dependent variable and one or more independent variables [33]. Regression techniques enable a quantitative expression of dependence (correlation), and the resulting model can be used to predict data for which no direct measurements exist.

In this study, simple linear regression (with a single independent variable) was used to estimate the number of vehicle trips generated by shopping centers, using Gross Leasable Area (GLA) as the predictor. The regression model enables the prediction of traffic volumes during peak hours by analyzing how variations in GLA influence trip generation. The general form of the simple linear regression model is given by the following equation [33]:

Y = αX + β + e

(1)

where

Y = dependent variable (number of vehicle trips);

X = independent variable (GLA);

α and β = regression coefficients that need to be determined using the least squares method, which minimizes the sum of the squared differences between the observed values and the predicted values to ensure the smallest possible error in estimating the relationship between Y and X;

e = the residual errors, representing the differences between the actual observed values (the counted number of vehicle trips) and the values estimated by the regression model (the model predicted number of vehicle trips).

To evaluate the quality of the regression model, the coefficient of determination (R²) was used. This coefficient defines the proportion of the variation in the dependent variable (Y) that can be explained by the independent variable (X). The coefficient of determination (R²) is calculated according to the following formula [33]:

$${\mathrm{R}}^2=1-{\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{y}}_{\mathrm{i}}-{\widehat{\mathrm{y}}}_{\mathrm{i}})}^2}{\sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{y}}_{\mathrm{i}}-\overline{\mathrm{y}})}^2}}$$

(2)

where

${\mathrm{y}}_{\mathrm{i}}$ = the observed value of dependent variable for observation i;

${\widehat{\mathrm{y}}}_{\mathrm{i}}$ = the predicted value of the dependent variable for observation i, as estimated by the regression model;

$\overline{\mathrm{y}}$ = the average of all observed values $y_i$;

n = the total number of observations.

In the context of the regression analysis, R² is a statistical measure of how well the regression line approximates the actual data. It ranges from 0 to 1, with values closer to 1 indicating that the chosen independent variable effectively explains the variation in the dependent variable. In this study, R² reflects the percentage of variance in the number of vehicle trips that can be explained by the variance in the independent variable (GLA) [6].

To verify whether the predictor variable (X) is statistically significant in the model, a t-test was conducted, by comparing the empirical and critical (tabulated) values at a chosen level of significance. The following test statistic was used:

$$\mathrm{t}={\displaystyle \frac{\hat{\mathrm{\alpha }}}{\mathrm{S}\mathrm{E}\left(\hat{\mathrm{\alpha }}\right)}}$$

(3)

where

$\hat{\mathrm{\alpha }}$ = estimated regression coefficient (slope of the independent variable);

SE($\widehat{\alpha }$) = standard error of the regression coefficient.

A p-value below the chosen significance level (typically 0.05, meaning a 95% confidence interval is accepted) indicates that the independent variable has a statistically significant influence on the dependent variable. The results of the regression analysis performed on the created database are presented in the following chapter.

4. Results and Discussion

4.1. Models for Predicting Vehicle Trips Generated by Shopping Center During Peak Periods

The primary objective of this study was to investigate the number of vehicles attracted by existing shopping centers and to analyze the feasibility of formulating a mathematical model that could be used to predict the number of vehicles during the peak hour for newly planned shopping centers in the future. A trip generated by shopping centers’ attractiveness can be realized using various modes of transportation (private vehicles, bicycles, walking, or public transport). However, this study focused only on vehicle trips, as an increase in vehicle volume on the existing road network can affect traffic flow quality, congestion, and overall safety for all road users. For this purpose, traffic counts were conducted, and a database was created for the application of statistical analysis methods and the development of a predictive model.

Trip generation data can be determined in two ways:

• By applying the average trip generation rate of the relationship between the number of trips and a specific independent variable, or
• By applying the developed mathematical (regression) equation.

The average trip generation rate is defined as the average number of trips generated by a particular land use type over a specified time period (e.g., per day or during peak hours) in relation to a relevant independent variable. A regression equation with an R² of at least 0.75 is preferred because it indicates the desired level of correlation between the trips generated by a site and the value of independent variable [5].

For the five analyzed locations, a trip generation estimation model was formulated based on linear regression analysis. Linear regression analysis and appropriate statistical tests were performed in JASP 0.19.3 software [34]. As previously mentioned, some earlier studies have used multiple regression analysis with several independent variables (number of employees, number of stores, number of parking spaces, etc.). However, it was found that Gross Leasable Area (GLA) has the greatest influence on the number of vehicle trips generated by shopping centers [26,27,28]. Due to its strong correlation with trip generation, GLA was selected as the primary independent variable in this study. This approach aligns with the methodology used in the ITE Trip Generation Manual [6], where GLA is also considered the key predictor variable for estimating vehicle trips associated with shopping centers.

Linear regression analysis was performed for the peak hour of the adjacent street network on weekday, followed by the peak hour of the generator on weekday and on Saturday. The results of the performed analysis are given in

Table 5. Linear regression models summary.

Model	Time Period	Independent Variable	R	R2	Standard Error
1	Peak hour—adjacent street	GLA	0.919	0.845	284.571
2	Peak hour—weekday	GLA	0.961	0.923	315.618
3	Peak hour—Saturday	GLA	0.932	0.869	381.684

. For each model, the correlation value R, the coefficient of determination R², and standard error of the estimate are shown.

The obtained coefficients of determination (R²) indicate that all models have good agreement with the measured data, meaning that a substantial proportion of the variability in the dependent variable (vehicle trips) can be explained by the independent variable (GLA). According to the Trip Generation Handbook [5], a regression equation is recommended when R² is at least 0.75. Since all models meet this criterion, they are considered reliable for predicting vehicle trip generation. Additionally, the results of the Analysis of Variance (ANOVA), presented in

Table 6. ANOVA results.

Model		Sum of Squares	df	Mean Square	F	p
1	Regression	648,121.078	1	648,121.078	16.305	0.027
	Residual	119,248.922	3	39,749.641
	Total	767,370.000	4
2	Regression	1.764 × 106	1	1.764 × 106	36.084	0.009
	Residual	146,689.475	3	48,896.492
	Total	1.911 × 106	4
3	Regression	1.423 × 106	1	1.423 × 106	19.896	0.021
	Residual	214,527.064	3	71,509.021
	Total	1.637 × 106	4

, confirm that there is a statistically significant relationship between Gross Leasable Area (GLA), as the independent variable, and the number of vehicle trips, as the dependent variable.

The significance tests, conducted with a significance level of α = 0.05, indicate that changes in the independent variable significantly influence changes in the dependent variable and the null hypothesis of no relationship between the dependent and independent variables can be rejected. For all analyzed models, the p-value is less than 0.05, confirming that there is a statistically significant relationship between the observed variables.

The values of the regression coefficients (Unstandardized Coefficient) and their corresponding statistical significance (p-value) are shown in

Table 7. Model coefficients.

Model		Unstandardized	Standard Error	Standardized	t	p
1	(Intercept)	−261.223	284.571		−0.918	0.426
	GLA	0.025	0.006	0.919	4.038	0.027
2	(Intercept)	−534.257	315.618		−1.693	0.189
	GLA	0.041	0.007	0.961	6.007	0.009
3	(Intercept)	−489.790	381.684		−1.283	0.290
	GLA	0.037	0.008	0.932	4.461	0.021

for each of the models. These values confirm the hypothesis that the independent variable significantly contributes to the explanation of the variability of the dependent variable. The p-value for the intercept (constant) is greater than 0.05, but this does not affect the validity of the model as a whole.

The mathematical equations of the created models for estimating vehicle trips for shopping centers, based on the results of the regression analysis, are presented in

Table 8. Mathematical equations of models.

Model	Time Period	Model Equation
1	Peak hour of adj. street traffic—weekday	VT 1 = 0.025 GLA—261.223
2	Peak hour of SC 2—weekday	VT = 0.041 GLA—534.257
3	Peak hour of SC—Saturday	VT = 0.037 GLA—489.790

¹ Vehicle trips; ² shopping center.

.

Figure 4, Figure 5 and Figure 6 show the measured data along with the corresponding regression equations for each analyzed peak period. The scatter plot with a regression line in Figure 4 illustrates the relationship between the Gross Leasable Area (GLA) of shopping centers and the number of vehicle trips during the peak hour of adjacent street traffic on a typical weekday. The positive regression coefficient (0.025), which represents the slope of the regression line, indicates that, for every 1000 m² increase in GLA, the number of vehicle trips is expected to increase by approximately 25 during the peak hour of adjacent street traffic.

Similar patterns are observed in the other two models. Figure 5 and Figure 6 show the relationships between the shopping center GLA and the number of vehicle trips during the peak hour of the shopping center for a typical weekday and for Saturday, respectively. According to the model for the shopping center peak hour on a weekday (Figure 5), each additional 1000 m² of GLA results in an expected increase of approximately 42 vehicle trips. For Saturday’s peak hour (Figure 6), the model predicts an increase of approximately 37 vehicle trips for every additional 1000 m² of GLA.

The presented results are consistent with the traffic counts on which the developed models are based (Table 2, Table 3 and Table 4). The number of vehicle trips is higher during the shopping center’s peak hour than during the peak hour of the adjacent street network. Additionally, the number of trips is greater on weekdays than on weekends, which can be attributed to higher vehicle occupancy rates (e.g., family trips to shopping centers), more frequent use of public transport, and non-motorized modes of travel during the weekend.

All coefficients of determination (R²) for the proposed models have high values, indicating a strong dependence between the dependent and independent variables. Given the characteristics of the analyzed locations, the proposed models can be applied to shopping centers with a Gross Leasable Area (GLA) ranging between 20,000 and 70,000 square meters.

To define the directional distribution of the estimated generated trips, the average distribution of the measured vehicle entry and exit data (as shown in Table 2, Table 3 and Table 4) was used. The average directional distribution for the peak hour of adjacent street traffic is 51% vehicles entering and 49% exiting. For the peak hour of the generator, this distribution comprises 50% entering and 50% exiting vehicles for weekdays, while for Saturday, at the peak hour, this distribution comprises 53% entering and 47% exiting vehicles.

Although the number of locations included in this study meets the criteria outlined in the ITE Handbook [5], the authors are aware that a larger dataset would provide an even more accurate representation of the impact of GLA on the number of vehicles attracted to a shopping center zone. Due to the limited availability of consistent data, the regression analysis in this study was conducted using five data points, each representing a shopping center. While this approach provides a preliminary insight into potential relationships, statistical measures such as R-squared values and p-values may be affected by the small sample size. As a result, the outcomes should be interpreted as exploratory rather than conclusive. This limitation is noted, and future studies with larger datasets are recommended to validate and strengthen the findings. The presented results are an initial step in traffic demand modeling and database creation in the Republic of Croatia, as there have been no previous studies on the number of trips generated by different land uses in this area. Furthermore, these findings serve as an incentive for further research to expand and update the database, not only for shopping centers but also for other trip generators (such as residential complexes, hospitals, and large supermarkets) whose construction could significantly impact traffic flow on the existing road network. Based on the estimated future traffic volume caused by new site development, infrastructure measures can be implemented to maintain an adequate level of service and traffic flow quality.

4.2. Comparison of Measured Trip Generation Data with Data According to the ITE Generation Manual

The majority of trip generation studies for different land uses are conducted in the USA, with the results published in the ITE Trip Generation Manual. Due to the lack of locally specific data for Croatia, traffic impact analyses have often relied on data from this publication, adjusted by reducing the values by a certain percentage to better reflect local conditions. The differences arise due to different motorization rate between Croatia and the USA, as well as socio-demographic characteristics and different travel habits of the population. The motorization rate in the USA for 2022 is 775 vehicles per 1000 inhabitants. This value refers to the number of registered light duty vehicles [35] per 1000 inhabitants [36]. For the same year and for the same type of vehicle, the motorization rate for Croatia is 475 vehicles per 1000 inhabitants [29,37], which is 39% lower than in the USA. There are also differences in the modal distribution of trips, as some European and other countries have a higher usage percentage of public transportation and non-motorized modes than the USA [38].

Table 9. ITE equations for estimating vehicle trips [6].

Time Period	Independent Variable X	ITE Equation	R2
Peak hour of adj. street traffic—weekday	1000 Sq. Feet GLA	Ln(T 1) = 0.74 Ln (X) + 2.89	0.82
Peak hour of SC 2—weekday	1000 Sq. Feet GLA	Ln (T) = 0.72 Ln (X) + 3.02	0.76
Peak hour of SC—Saturday	1000 Sq. Feet GLA	Ln (T) = 0.79 Ln (X) + 2.79	0.87

¹ Average vehicle trips endings; ² shopping center.

shows the regression analysis equations used to estimate vehicle trips for different peak periods according to the ITE Trip Generation Manual, 10th edition [6]. The independent variable in these equations is GLA expressed in 1000 square feet.

A comparison of the number of vehicle trips measured at the observed shopping center locations and the number of vehicle trips estimated by ITE regression equations (Table 9) is shown in

Table 10. Comparison of measured vehicle trips and estimates according to the ITE equation for the peak hour of the adjacent street traffic.

Location	GLA	Measured Vehicle Trips	ITE Estimate	Difference	Deviation
1	61,000	1203	2187	984	45.00%
2	57,000	401	2060	688	33.09%
3	35,000	590	1450	860	59.31%
4	22,000	1392	1028	627	61.01%
5	42,000	564	1659	1095	66.01%
				Mean	52.88%

,

Table 11. Comparison of measured vehicle trips and estimates according to the ITE equation for the peak hour of the generator—weekday.

Location	GLA	Measured Vehicle Trips	ITE Estimate	Difference	Deviation
1	61,000	1925	2188	263	12.02%
2	57,000	2073	2084	11	0.51%
3	35,000	788	1467	679	46.27%
4	22,000	538	1050	512	48.76%
5	42,000	1007	1672	665	39.79%
				Mean	29.47%

and

Table 12. Comparison of measured vehicle trips and estimates according to the ITE equation for the peak hour of the generator—Saturday.

Location	GLA	Measured Vehicle Trips	ITE Estimate	Difference	Deviation
1	61,000	1831	2737	906	33.11%
2	57,000	1810	2595	785	30.24%
3	35,000	594	1765	1171	66.35%
4	22,000	585	1223	638	52.17%
5	42,000	815	2038	1223	60.02%
				Mean	48.38%

for the peak hour of adjacent street and the peak hour of the generator on a weekday and on a Saturday, respectively. These disparities highlight the need for locally developed trip generation models that account for Croatia’s specific transportation and behavioral patterns instead of relying solely on adjusted values from the ITE Manual.

This study, like some previous research [12,13,17], has confirmed that the measured vehicle trip values are significantly lower than those estimated using equations from the ITE Trip Generation Manual. For the observed shopping centers in Croatia, the average deviation ranges between 30 and 50%, meaning that values estimated using ITE equations should be reduced by this percentage to better reflect local conditions. It can be observed that the smallest deviations are for shopping centers with larger leasable areas.

Trip generation estimates are generally lower than in the USA due to differences in urban planning, motorization rates, transportation infrastructure, and travel behavior. European cities are more compact with mixed-use developments, where residential, shopping, and recreational areas are closer together, reducing the need for driving. Additionally, the high use of public transport and non-motorized modes of travel results in fewer vehicle trips compared to more car-dependent countries like the USA. The USA has historically invested heavily in car-friendly infrastructure, including extensive highway systems and large parking facilities, which supports higher vehicle usage and trip generation.

In this study, data and estimates for shopping centers were analyzed and compared; however, the general conclusion can also be applied to other land uses. In the context of a traffic impact study, if it is necessary to estimate vehicle trips generated by the new site development, estimates derived from the ITE Generation Manual may be utilized because there are no national databases for trip generation estimates for different land uses in Croatia. However, the resulting values should be lowered by 30–50%, depending on the observed peak period, the type of the facility, and the judgment of the traffic planner. The mean difference of 40% also corresponds to a difference in the rate of motorization, which is 39% lower than that in the USA.

4.3. Traffic Management Implications

The findings of this trip generation study offer several practical implications for urban planners and transportation authorities, especially in conducting transportation impact analysis. Understanding the relationship between shopping center characteristics (e.g., Gross Leasable Area) and trip generation rates can aid in predicting traffic volumes, which is essential for planning access roads, designing nearby intersections, and ensuring optimal signal timing. Secondly, these data can be used to generate parking demand estimates, allowing planners to better allocate parking space based on predicted vehicle trips, particularly during peak shopping periods. From a policy perspective, this data-driven approach supports more sustainable mobility planning, including the integration of public transport, pedestrian access, and cycling infrastructure into shopping center design. By incorporating trip generation data into strategic planning, transportation planners can improve traffic flow quality, reduce congestion, and enhance the overall efficiency of site developments.

5. Conclusions

In many countries, studies are being conducted on trip generation rates and equations for different land uses, and the obtained data are used in traffic analyses and impact studies to assess how new developments affect traffic conditions and flow quality in their surroundings. In the Republic of Croatia, there has been no systematic research on vehicle trip generation related to specific land uses. Therefore, this study focused on shopping centers, as they are among the largest trip generators. Five locations with different Gross Leasable Areas (GLAs) were selected, where data were collected on the number of vehicles entering and exiting the shopping center zone during a typical weekday and on a Saturday. Peak periods and corresponding traffic volumes were identified. Based on the formed database of traffic volume and shopping center characteristics, a linear regression analysis was conducted, and models were developed to estimate vehicle trips for the peak hour of adjacent street traffic and for the peak hour of the generator (shopping center) on a weekday and on a Saturday. All developed models have a high coefficient of determination (R²), indicating a strong relationship between the dependent and independent variables. These models can be applied to estimate future vehicle trips for the construction of new shopping centers with a Gross Leasable Area (GLA) ranging from 20,000 to 70,000 square meters. Vehicle trips estimation is crucial for analyzing the impact on traffic flow and is used for planning purposes and traffic management in the context of the overall design of shopping centers and their surrounding infrastructure.

The study also includes a comparison of measured values with the estimated vehicle trips based on the American ITE Trip Generation Manual [6], which provides equations for predicting the number of trips generated by various land uses. The comparison results indicate that trip estimates from [6] should be reduced by 30 to 50%, depending on the observed peak period. The average difference of 40% corresponds to a difference in the rate of motorization, which is 39% lower than in the USA, where private vehicle use is more dominant as a mode of transportation. In contrast, European cities rely more on public transportation and non-motorized travel modes compared to the USA, leading to fewer vehicle trips in similar land use developments.

The collected data and formulated models represent an initial step in defining the estimate of vehicle trip generation due to the construction of a new shopping center in Croatia. However, additional data from other locations with different population sizes should be collected to expand the database, improving the accuracy and broader applicability of the presented models.

Furthermore, it is advisable to conduct trip generation studies for other land uses, especially major trip generators such as new residential zones, hospitals, student campuses, and other facilities whose construction generate vehicle trips that can significantly impact the adjacent road network and traffic flow quality. The developed models would reflect local socio-demographic characteristics and travel behavior, providing a more accurate estimation of trip generation in the transportation planning process.

Additional research can focus on the choice of transportation mode to a specific facility, conducting a survey among visitors to determine how improved public transportation connectivity and better cycling infrastructure would influence their choice of transport. This would further contribute to reducing traffic congestion and environmental pollution, promoting more sustainable urban mobility solutions.

Good planning and spatial management can significantly influence the necessary infrastructure to meet traffic demand. Additionally, effective traffic management, including the development and promotion of micromobility and public transportation, can contribute to the goals of sustainable transport and environmental protection.