1 Introduction

Social capital – defined as “the ability of people to work together for common purposes in groups and organizations” (Fukuyama 1995, p. 10) and as “those persistent and shared values and beliefs that help a group overcome the free rider problem in the pursuit of socially valuable activities” (Guiso et al. 2011)—is an intangible but important driver of human behaviour. An extensive literature has found it to be positively associated with economic and financial development (Tabellini 2010; Guiso et al. 2004), firms’ financial performance (Lins et al. 2017), borrowers’ loan conditions and repayment (Hasan et al. 2020), economic mobility (Chetty et al. 2022; Güell et al. 2018), political accountability (Nannicini et al. 2013), and health outcomes (Folland 2007; Kim et al. 2011; Xue et al. 2020), and negatively associated with crime rates (Lederman et al. 2002; Buonanno et al. 2009; Akçomak and Ter Weel 2012) and road accidents (Nagler 2013).

In a recent work, Castriota et al. (2023; CRT hereafter) show the importance of social capital in a new context, hit-and-run road accidents. They analyse the decision to flee after a fatal accident involving pedestrians in the United States. They find that social capital at the county level is negatively associated with the decision to flee, with an increase in social capital by one standard deviation associated with a decrease in the probability of hit-and-run by around 10.5%. They exploit the availability of the ZIP code of the driver when knownFootnote 1 to implement a falsification test supporting a causal interpretation of the result. In particular, they show that the negative and statistically significant relationship between the probability of fleeing and local social capital is present only for local drivers and not for those who are out-of-county.

In this paper, we study the relationship between social capital and hit-and-run accidents in the Italian context and find that an increase by one standard deviation in social capital at the provincial level is associated with a lower probability of hit-and-run by around 12.5%, a figure comparable to the one found for the US. One limitation of CRT is that US data include only fatal accidents. Whether an accident results in a fatality may be partly endogenous to the decision to flee, given that in several circumstances a fleeing driver is likely to delay the emergency response, thus reducing survival rates (Tay et al. 2009),Footnote 2 and also endogenous to social capital, if social capital has an impact on the quality of local health services. One important advantage of using Italian data is that records also include accidents with injuries. There are, however, also disadvantages in that, for instance, Italian records do not specify the ZIP code of drivers or information on whether fleeing drivers have been subsequently identified, making it unfeasible to repeat some of the analysis performed in CRT. From this point of view, the analysis conducted here and the analysis in CRT can thus be seen as complementing each other.

Another contribution of this paper is that we show the robustness of the negative relationship between fleeing an accident and social capital in a different institutional context. CRT have been the first to look at this issue and have documented it for the US. There is no guarantee, however, that this holds beyond the—important but clearly not universal—American case, and it is thus important to extend the analysis to different countries. The Italian case is particularly relevant, considering the large and long-lasting differences in social capital existing within the country—in particular between the Centre-North and the South—(Bigoni et al. 2016; Bracco et al. 2015), differences that have greatly inspired the literature, starting at least with Banfield (1958). Indeed, several of the contributions listed in the opening paragraph concern the Italian case and many more exist in the literature on social capital.Footnote 3 Thus, another contribution of this paper is to assess the replicability of a recent empirical finding in an institutional context, Italy, that is different from the widely studied US case and very relevant for the literature on social capital.

Finally, the paper contributes to our understanding of hit-and-run in Italy, an issue that is of high policy relevance given the dire consequences of such acts, often triggering indignation in the public opinion and calls for strong policy interventions.Footnote 4 Castriota and Tonin (2023) find that the likelihood of Italian drivers fleeing after an accident increases by around 20% with darkness and the consequent decrease in the probability of detection. They suggest that policies increasing the likelihood of apprehension could be effective in reducing hit-and-run despite it being an unplanned decision taken under strict time constraints and dramatic psychological conditions. This paper goes beyond formal sanctions and explores instead the role of the social environment, as summarized by the concept of social capital, as a potential deterrent for hit-and-run.

More generally, this paper contributes to a strand of literature investigating the role of social capital in situations characterized by extreme and stressful conditions. A number of studies show the importance of social capital in coping with natural disasters (Aldrich 2012): Liu et al. (2022) investigate the role of social capital in earthquake recovery in Nepal; Munasinghe (2007) discusses the role of social capital by comparing the behaviour of citizens after the 2004 tsunami in Asia and the 2005 hurricane Katrina in New Orleans; Zahnow et al. (2019) study a flood in Brisbane, Australia. Frey et al. (2010) analyse the sinking of the Titanic – a quasi-natural experiment where passengers were suddenly put in a life-and-death situation – and find that “social norms of helping” such as “women and children first” persevered also during such a tragic event and influenced the probability to survive. On the other hand, Elinder and Erixson (2012) study 18 maritime disasters, finding instead evidence of “every man for himself” behaviour. All these examples concern disasters involving whole communities or large groups. The case of hit-and-run is clearly different, as it is an extreme and stressful condition directly involving in most of the cases just the driver and the victim,Footnote 5 thus social capital may have a different—possibly more subdued—role.

The rest of the paper is organized as follows. Section 2 describes the database on road accidents and social capital, Sect. 3 provides descriptive statistics, Sect. 4 discusses the results of the econometric analysis, while the last section concludes.

2 Data

Accident data come from the Italian Institute of Statistics (ISTAT). We have access to comprehensive records of all road accidents involving injured or deceased individuals over the period 2000–2016. The dataset contains detailed information regarding each collision, encompassing aspects such as location (Province), time of occurrence, number of participants and their respective attributes, types of vehicles involved, weather conditions, road types, as well as the total count of injured and deceased individuals. Notably, the data also includes information on whether an individual has fled after an accident. Following CRT (2023), we focus on accidents with pedestrians because in these cases serious damages to the vehicle and serious injuries to the driver are rare events, therefore they seldom prevent the driver from fleeing. Hence, the variation in the decision to run away is independent of such confounders.

When it comes to measuring social capital, the literature offers various approaches due to the multifaceted nature of the concept (Dasgupta and Serageldin 2000; Christoforou 2013). One approach to operationalize social capital uses trust towards individuals and institutions, usually measured through survey questions (Glaeser et al. 2000; Guiso et al. 2008; Akçomak and Ter Weel 2009; Lins et al. 2017). Another perspective involves considering social capital in terms of membership in groups, networks, and voluntary associations (Putnam et al. 1994; Alesina and La Ferrara 2002; Sobel 2002) or higher voter turnout (e.g., Putnam 2000; Cartocci 2007; Buonanno et al. 2009). Furthermore, the strength of social capital can be captured through voluntary blood donations, as proposed by Guiso et al. (2004).

In this paper, to facilitate a comparison of our findings with other research on this subject and in particular the work by CRT (2023), we developed social capital indexes that resemble as much as possible those used in "The Geography of Social Capital in America" project (2018) at the county level. These indexes have also been employed in studies conducted by Ding et al. (2020), Borgonovi et al. (2021), and Milosh et al. (2021). Specifically, we constructed a provincial-level social capital index called Social Capital, which comprises the following sub-indexes: Community Health based on data about members of sport associations, blood donors and newspaper-readers per thousand inhabitants; Institutional Health captured by the voter turnout in referenda; and Collective Efficacy, that is the number of voluntary homicides per thousand inhabitants. Additionally, we formulated an alternative measure, Social Capital (2), by excluding the collective efficacy variable from its computation. The reason behind this exclusion is that previous works concerning Italy have not commonly considered this aspect (Cartocci 2007; Righi 2013; Nifo and Vecchione 2014; Calcagnini and Perugini 2019). Note that a higher score on these social capital indexes implies a higher level of social capital in the respective provinces. It is important to point out that, although we tried to build for Italy social capital indicators similar to the ones used for the US by CRT (2023), perfect adherence was not possible.

The data comes from a variety of sources: information on members and teams affiliated with sports associations (CONI and EDP),Footnote 6 blood donors and donations, and no-sportive newspaper circulation has been retrieved from Cartocci (2007), while we have consulted the Italian Ministry of the Interior website to obtain data on the referendum participation. The number of voluntary homicides is collected by ISTAT. The social capital indicators are time invariant and the data refer to the period 1990–2016. See Table 5 in the Appendix for details on the variables and sources employed to define the index and the sub-indexes at the provincial level.Footnote 7

Information pertaining to additional control variables at aggregate level such as unemployment rate and vehicle density is sourced from ISTAT (Italian National Institute of Statistics) and the Ministry of Transportation and Infrastructures; while information on spending on road infrastructure from Openpolis Foundation.Footnote 8

As mentioned above, social capital is a multifaceted concept that relates to dimensions like personal relationships, cooperative norms and trust, social network support and civic engagement. While our measures capture some of these dimensions, it is important to assess the robustness of the relationship we study to alternative measures of social capital. To do this, we use three different social capital indexes employed and developed in the investigation of the Italian context by Calcagnini and Perugini (2019)—referred to, in this work, as Social Capital CP -, Calcagnini et al. (2019)—referred to as Social Capital CGP-, and Fini et al. (2011)—referred to as Social Capital FGSS. The first indicator is constructed by Principal Component Analysis (PCA) on four variables, at the provincial level, assessing these three facets of social capital: trust, social norms, and networks. Specifically, Calcagnini and Perugini (2019) choose two dimensions of generalized trust in institutions and society, namely, the duration of first-instance ordinary court proceedings and the incidence of car thefts (Micucci and Nuzzo 2003). They use blood donations, measured as the number of blood bags per million inhabitants, to proxy civic engagement and people's inclination to cooperate, as done in Guiso et al. (2004). For social inclusion and networks, following the approach of Mattisson et al. (2015), they utilize the length of bike lanes.

Regarding the Social Capital CGP measure, it was developed through PCA on 26 variables that capture networks, trust, and social norms at the provincial level. Specifically, with respect to networks, akin to Putnam et al. (1994), Calcagnini et al. (2019) gauge social capital characteristics related to social participation and networks using metrics such as the number of tickets sold for public events or museums, and the number of libraries. Additionally, they include variables associated with firm innovation, such as the number of patents or individuals with a doctorate, recognizing that innovative firms are more likely to engage in information exchange and mutual interaction (Hauser et al. 2007). Furthermore, variables related to economic development, including the employment rate, the number of registered firms, and the length of cycling lanes, are selected. According to Jordan et al. (2010), economic development can enhance individuals' ability to participate in community and associational activities, thereby fostering higher levels of social capital. Turning to trust, they consider variables related to education, which is seen as a direct promoter of social capital accumulation by aiding individuals in developing civic skills and cognitive capacities that facilitate participation in groups and associations (Milligan et al. 2004). Additionally, given that interpersonal altruism supports trust (Glaeser et al. 2000), they include variables such as the number of cooperatives, voluntary members, or not-for-profit organizations. Moreover, recognizing the link between health status and social capital in various studies (Fiorillo 2008), variables related to health status and health care institutions are also included. With respect to social norms, they use variables reflecting civic and political participation (Guiso et al. 2004).Footnote 9

Lastly, Social Capital FGSS is assessed at the regional level by combining variables associated with the concept of free-riding behaviours in market transactions, as outlined by Micucci and Nuzzo (2005). The indicator is specifically formulated by incorporating data on crime rates, bankruptcy rates, forgery rates, and instances of commercial fraud.

In Sect. 4, we will also explore the relationship between hit-and-run and these alternative measures of social capital.

3 Descriptive Statistics

Table 1 provides a comprehensive overview of the variables used in the econometric analysis, along with the main summary statistics. The analysis covers the period from 2000 to 2016, during which there were in Italy a total of 320,835 accidents involving pedestrians. Among these accidents, 2% (6766 accidents) were classified as hit-and-run incidents, where the driver fled the crash scene.

Table 1 Description and summary statistics of the variables used in the estimations
Full size table

Looking at Fig. 1, the distribution of hit-and-run pedestrian accidents over the country is heterogeneous and clear patterns cannot be identified. Provinces with high and low hit-and-run rates are present in the North, in the Centre and in the South of the country. According to Table 2, the rate of hit-and-run for pedestrian crashes is at the highest mainly during the night (specifically from midnight to 4 am), during the weekend and in clear weather conditions. Also, the rate is higher in urban areas and on non-local streets. Given that there are many factors, beside social capital, that may be related to the propensity to flee after an accident, in what follows we also describe the rich set of control variables we use in the econometric analysis.

Fig. 1
figure 1

Percentage of pedestrian hit-and-run accident in Italy, by province

Full size image
Table 2 Descriptive statistics
Full size table

Concerning social capital indexes, the main right-hand-side variables of interest, these are standardized to make results easier to interpret, so they all have mean zero and standard deviation one. Figure 2 shows the Social Capital Index by province. As it has been widely documented by the literature on social capital, there is a geographical divide, with the Centre-North part of the country displaying higher levels of social capital. Studies such as Banfield (1958) and Putnam et al. (1994) indeed observed huge differences in social capital between the North and the South and used them to explain the economic divide.

Fig. 2
figure 2

Social Capital Index Scores in Italy, by province

Full size image

As control variables, beside those related to the time when the accident happened, we use the type of street where the accident occurred (Urban), and the weather conditions (Clear Weather). We also account for road characteristics (Two way Road) and whether the accident took place during a holiday period (Holiday). These variables play a role in influencing the driver's decision to flee or stay at the accident scene, as indicated by previous studies (Solnick and Hemenway 1994, 1995; Johnson 1997; Tay et al. 2008, 2009; MacLeod et al. 2012). For instance, a crash in a rural area, under poor lighting, rain, and wind, could lead drivers to flee as there is a lower likelihood of being identified by witnesses. The variable Holiday partly captures alcohol and drugs consumption, which can affect the propensity to flee given the higher penalties under these conditions. We also control for local economic and infrastructural conditions by using the provincial unemployment rate (Unemployment Rate); the local per capita spending on road infrastructures (Spending on Safety); and the vehicle density (Vehicle Density). Moreover, we also use as controls, variables capturing the seriousness of the accident (the number of dead or injured pedestrians) and, for the large subset of accidents involving only one pedestrian, the characteristics of the victim, namely gender (Female Pedestrian) and age (Age Pedestrian), as they may also influence the driver's behaviour.

4 Econometric Model

To test whether and to what extent the social capital endowment, measured at the province level, is associated with the likelihood of fleeing after a fatal accident involving a pedestrian, we estimate the following logit model, closely resembling the one estimated in CRT:

$$Prob\left({HitandRun}_{i}=1|X\right)= F\Bigg({\beta }_{0}+{\beta }_{1}{Social Capital}_{p}+\phi {X}_{i}+ \zeta {Z}_{p}+\sum_{h=1}^{23}{\gamma }_{h}{Hour}_{h}+\sum_{d=1}^{6}{\gamma }_{d}{Day}_{d}+\sum_{m=1}^{11}{\omega }_{m}{Month}_{m}+\sum_{t=2000}^{2015}\sum_{s=1}^{3}{\alpha }_{ts}{{Year}_{t}*Macro Area}_{s} \Bigg)$$
(1)

The dependent variable is a dummy equal to 1 if the driver flew after the accident. The key regressor of the analysis is the social capital index (Social Capital), or alternatively the Social Capital (2), and its sub-indexes, in the province p where the accident occurred. As a robustness check, we use the alternative indexes mentioned above. Furthermore, we include a vector of control variables, \({X}_{i}\), that contains characteristics of the context in which the crash occurs and its characteristics such as Urban, Clear Weather, Holiday, and Twoway Road, Total Deaths and Total Injured. The vector Z include variables at the provincial level as Unemployment Rate, Spending on Safety, and Vehicle Density. Lastly, we add dummy variables to control for time and space fixed effects (hour of the day, day of the week, month of the year, and the interaction between year and macro-areas – that is North, Centre and South of Italy). As a robustness check, we have a specification controlling for characteristics of the accident victim (sex, age and age square) when there is a single one.

Table 3, columns 1–5, shows the results of the baseline regression (marginal effects and standard errors corrected for heteroskedasticity). Column 1 includes the overall social capital index (Social Capital) while, in columns 2–5, we replace the main indicator with the alternative social capital measure (Social Capital (2)) and the three sub-indexes (Collective Efficacy, Community Health, Institutional Health). Except for Collective Efficacy, which is not significant, the level of social capital has a negative and robust association with the probability of fleeing. Looking at column 1, one standard deviation increase in Social Capital is associated with a reduction in the probability of hit-and-run of around 12.5% (that is, a decrease of 0.0025 percentage points over a baseline of 0.02). The same results are obtained when focusing on accidents with only one pedestrian and adding controls related to the victims’ characteristics (columns 6–10). With regard to the control variables, hit-and-run appears more likely to occur in urban areas, during holidays, with good weather conditions and with higher vehicle density. The positive coefficient for urban areas when looking at accidents with only one pedestrian involved is against the hypothesis we formulated, but in line with what found in other papers discussed in the previous sections (Solnick and Hemenway 1995; MacLeod et al. 2012; Castriota et al. 2023). The positive coefficient for clear weather is also unexpected. Here, the literature is rather mixed: Castriota et al. (2023) find a negative coefficient; Tay et al. (2009) report an insignificant coefficient, while Castriota and Tonin (2023) a positive one. A possible explanation for a positive coefficient could be, as noticed in Tay et al. (2008), that “factors that reduce the likelihood of blame or fault will also increase the likelihood of reporting crashes. Such mitigating factors include wet weather […]”. Indeed, in our dataset rain is the dominant cause for lack of clear weather.Footnote 10 Concerning other control variables, there is a lower likelihood to flee after the crash on two-way roads and in provinces with higher unemployment rates and per-capita spending on road infrastructures. Moreover, the number of dead and injured pedestrians in the accident show a negative coefficient, thus the probability of flying away decreases with the seriousness of the accident. Concerning the victim characteristics, female sex negatively influences the choice to leave the scene of the crash, while age has a non-linear effect in that the likelihood of fleeing is lower for young and old victims (the inflection point is at approximately at the age of 38).

Table 3 Estimation results
Full size table

In Table 4 we report the results obtained by using the three alternative measures of social capital discussed in Sect. 2. In columns 1–3 we consider the whole sample, while in columns 4–6 we focus on accidents with only one pedestrian. The findings related to the Social Capital CGP are in line with the previous results. Indeed, focusing on column 1, one standard deviation increase in Social Capital CGP is associated with a reduction in the probability of hit-and-run of around 11% (that is, a decrease of 0.0022 percentage points over a baseline of 0.02). The second alternative measure of social capital reported in column 2, that is, Social Capital CP, has a stronger impact, in that a one standard deviation increase is associated with a reduction in the probability of hit-and-run of around 15%. Lastly, results are also confirmed by accounting for Social Capital FGSS, defined in this case only at the regional level.

Table 4 Estimation results using alternative measures of Social Capital
Full size table

5 Conclusions

Using microdata on road accidents with injuries or fatalities involving pedestrians, we have shown a negative relationship between the likelihood of a driver fleeing and measures of social capital at the provincial level in Italy. The magnitude of this relationship is very similar to recent evidence from the US.

These results point to a role of social capital in guiding the behaviour of drivers in the dramatic circumstances immediately following an accident in which a pedestrian has been killed or injured. This suggests that social capital exerts its influence even in a fast, instinctive, and emotional setting, or "System 1" in the language of Kahneman (2011).

In terms of policy implications, these findings complement the existing literature and suggest that interventions targeted to promote civic norms and pro-social behaviour could thus have a beneficial impact across a wide range of domains, including road accidents. There may also be a role in campaigns aimed at reducing hit-and-run behaviour for appeals leveraging on social capital, for instance by underlining the impact of hit-and-runs on local communities or underlining the moral duty that people have towards each other to provide assistance in case of an accident. This, of course, would be a complement and not a substitute to more “traditional” measures like penalties and increased detection probabilities.