Environmental context and socio-economic status drive plastic pollution in Australian cities

Marine plastic pollution has rapidly become one of the pressing environmental concerns of the 21st century. Because of its adverse impacts on wildlife, human health, and the economy, it has been identified as one of the main targets within the UN's sustainable development goals (SDGs) (UN General Assembly 2015). One target of SDG 14 is to 'reduce marine pollution of all kinds, particularly from land-based activities.' Approximately 80% of marine plastic pollution is thought to originate from mismanaged waste on land (Derraik 2002). Global modelling suggests that between 19 and 23 million tonnes of debris enter aquatic ecosystems annually (Borrelle et al 2020). In Australia, it has been estimated that there are more than 120 million pieces of debris around the country's coastline (Hardesty et al 2017).

To determine where to best implement interventions to reduce waste leakage and evaluate which of the myriad of available approaches will work most effectively, it is useful to understand the underlying drivers of mismanaged waste. Typically, estimations of mismanaged plastic waste rely heavily on local population density (e.g. Jambeck et al 2015, Lebreton and Andrady 2019). However, there are many other factors which could potentially influence the distribution of litter, including socio-economic status (SES) (Matsunaga and Themelis 2002), environmental variables such land cover (Sakti et al 2021), and waste and transport infrastructure (Willis et al 2017).

While there is a growing body of literature that has quantified the economic impacts of litter (Jang et al 2014, Leggett et al 2018, Newman et al 2015, Watkins and ten Brink 2017), our knowledge about the role of SES as a driver of this litter is relatively sparse. Hardesty et al (2021) analysed debris across 118 different countries and reported that debris was negatively correlated with national wealth, and positively with the value of infrastructure in the local area (Hardesty et al 2021). A few publications discuss the relationship between self-reported attitudes about littering and education or economic status (Santos et al 2005, Widmer and Reis 2010, Slavin et al 2012, Eastman et al 2013). Most conclude that higher levels of education or income correlate with a decreased propensity for littering (but see Arafat et al 2007). Similarly, while there have been attempts to correlate physical forcing factors with debris distribution on beaches and on the sea floor (Thornton and Jackson 1998, Galgani et al 2000, Ribic et al 2010, 2011, 2012), few have investigated the effects of environmental factors on the distribution of litter in the inland terrestrial environment.

To address this knowledge gap and help identify where to best target interventions to reduce waste leakage from land, we analysed data from urban areas around Australia. We asked whether environmental and social factors can explain the distribution of litter in the terrestrial environment at a national scale. We considered four indices of SES (economic advantage, economic disadvantage, economic resource, and education and occupation), as well as measures of infrastructure (roads, rail networks, and landfills), population density, land use, site type, and time (years since 2007).

2.1. Debris data

We compiled data gathered by Keep South Australia (SA) Beautiful and Keep Australia Beautiful (hereafter referred to as KAB) from 2007 to 2017 as part of the National Litter Index (NLI). The NLI is conducted bi-annually at a set of fixed locations in cities around Australia (figure 1). Survey locations are distributed across a variety of site types defined by KAB (beaches, car parking areas, recreational parks, residential areas, roadside retail strips, shopping centres, highways, and industrial zones). The site locations remain constant from year to year, except in rare instances where the site usage changes (e.g. from residential to industrial) or where landform changes make it impossible to re-survey the same area. Locations are kept anonymous to ensure an unbiased sample from year to year.

Zoom In Zoom Out Reset image size

At each location, trained observers count and classify all of the litter that they see within the specified survey area. Surveyed areas range in size between 71 to 9600 m², with an average of 1526 m². In the annual NLI reports (Keep Australia Beautiful 2019), results are reported as the number of items/1000 m², so we have used this convention for comparability.

2.2. Driving variables (covariates)

2.3. Statistical analyses

A total of 983 sites were sampled over the 11 year period, with the vast majority of the sites (946) being sampled either 18 (n = 946) or 17 (n = 20) times.

The best-fit GAM model (lowest AIC value) included a number of socio-economic and infrastructure factors (table S1, figure 2). The coefficients for all terms were significantly different from zero, except for road distance, population residuals, and several of the state terms. Leaving any of these terms out increased the AIC, indicating that these terms, although some were not significantly different from zero at a p < 0.05 level, still made an important contribution to the fit of the model. The deviance explained by the model was 32.5%, meaning that the model explained over 32% of the variability in the data.

Zoom In Zoom Out Reset image size

For the continuous variables, relative area was the strongest predictor of debris load. Aside from this artefact of the sample design, the next strongest predictors of debris load were the log of the population density within 5 km, the index of economic disadvantage, and the survey year. The index of economic disadvantage measures not only the economic status of individuals, but more broadly measures their 'access to material and social resources, and their ability to participate in society' (Australian Bureau of Statistics 2011). A lower score on this index indicates greater disadvantage, and a higher score indicates a lack of disadvantage (though not necessarily advantage). The higher the index of economic disadvantage, the less litter reported.

We also measured three different socio-economic residuals, to understand the relative influence of education, resources (more narrowly defined as access to economic resources), and advantage (incorporates not only measures of disadvantage, but also economic and social indicators of advantage). Only one, the residual value between education and disadvantage, was significant, and improved the model. This residual was positively correlated with debris levels, meaning that higher-than-expected education indicators at a given level of social disadvantage was correlated with higher levels of debris. Similarly, the residual for advantage/disadvantage was positive (though not significant at the p < 0.5 level), while the residual for resource/disadvantage yielded a negative correlation, though was also not significant at the p < 0.5 level.

For the categorical variables, the strongest predictors of litter were site type and land use. Highways and retail strips had elevated levels of litter relative to other site types (figure 3). In contrast, recreational parks, residential areas, and beaches were relatively low. Land use was also important in determining loads. Areas classified as 'water' (including creeks and wetlands), had the highest debris loads, followed closely by 'production from the natural environment', which is land used for primary production where there is little change to the native vegetation (Australian Bureau of Agricultural and Resource Economics and Sciences (ABARES) 2016).

Zoom In Zoom Out Reset image size

A number of measures of urbanization are relevant for litter loads. Population density within 5 km was positively correlated with litter (on a log scale), while population density within 50 km was negatively correlated (also log scale). Litter levels are higher near rail stations and closer to roads (though this measure was not significant). Sites with a lower road density surrounded by regions with higher density of roads have higher levels of debris. Additionally, litter levels were higher closer to waste management facilities. However, these factors (other than population density within 5 km) have relatively low effect sizes, suggesting they are less important than socio-economic variables, land use, and site type.

Even taking into account the variability captured by the other terms in the model (site type, population density, economic index, etc), the political unit (state/territory) was an important predictor of the density of litter at a site. Queensland (QLD), the Northern Territory (NT), Western Australia (WA), and Tasmania (TAS) had relatively higher litter counts while the Australian Capital Territory, New South Wales (NSW), SA, and Victoria (VIC) had lower litter counts based on KAB surveys across all years (figure 4).

Zoom In Zoom Out Reset image size

Finally, the total count of debris found on surveys declined by about 0.67% per year over the time period between 2007 and 2017, though not uniformly in each state.

The National Litter Index is conducted in states and territories around Australia to understand litter loads, particularly in urban areas (most surveys take place near large cities) (figure 1). The results from NLI surveys are encouraging, with a significant decline in litter levels across the country recorded over the past decade. This decline, however, appears to be driven by two states in particular. VIC and WA show a significant negative relationship between year and amount of litter, while litter levels in QLD and TAS have increased over time. NSW, NT, and SA show no significant trend over time.

Based on our regression model (table S1), the data also paint a picture of high debris loads in relatively socio-economically disadvantaged areas, similarly to what was reported by Hardesty et al (2021) in a multinational study of 188 countries. At first glance, this might indicate a difference in littering behaviour at different income levels. However, the index of socio-economic disadvantage measures not just income level, but includes a broader measure of the ability to access social resources. The index includes indicators such as the number of houses without internet, the percent of people with poor English skills, and a number of educational parameters. When we investigated the relationships between economic resource and disadvantage, and education and disadvantage, we saw that where income levels were higher than expected given a certain level of disadvantage, litter counts were lower (though not statistically significantly so). In contrast, when education levels were higher than expected given a certain level of disadvantage, litter counts were higher. This suggests that wealth or income are likely the primary drivers behind lower litter levels in areas of high income, as opposed to education. This observation could also result from municipal services (such as street cleaning) being more widely available in areas of higher SES.

Higher litter counts were also found in transitory human use sites (shopping areas, retail strips, highways, car parks). This was especially true when those site types occurred within areas designated as water. Together, these results describe two general types of litter sources; littering from near transitory locations and potentially illegal dumping around watercourses. Both of these landuses/site types attract more litter in socio-economically disadvantaged areas (Hardesty et al 2016). Thus, one would expect to find particularly high litter loads near retail strips, shopping centres, and industrial areas with a creek or wetland nearby in an area of high socio-economic disadvantage.

There are several different components that may influence peoples' tendency to litter. Proximity to sources of potential litter, such as shops selling single use drink containers, may increase the likelihood of littering (Willis et al 2019). In this study we found that sites that are sources for litter (retail strips, shopping centres) have significantly higher litter. By placing additional receptacles close to these sources, mismanaged litter may be reduced. Research has shown that the absolute number of litter receptacles does not alone decrease the propensity for litter, but that the placement of these receptacles matters a great deal. If receptacles are available close to where a person needs to dispose of a single use item, this can greatly reduce the likelihood of litter (Schultz et al 2013).

Second, the attachment that an individual has to the surrounding environment can heavily influence littering behaviour. 'High quality' neighbourhoods, e.g. those characterized by green spaces, well-kept private residences, and attractive vistas, have lower littering rates than 'low quality' neighbourhoods, even after individual demographic attributes are controlled for (Weaver 2015). In this study, residential areas, beaches, and parks had significantly lower levels of litter than other site types. This could be related to the relatively higher cultural or aesthetic value of these sites over those that do not have such value (e.g. industrial areas, parking lots).

The cultural or aesthetic value is also affected by the normative context of the site, or the influence of local norms and standards, which can play a large role in decisions regarding littering (Ajzen 1991, Willis 2021). Where litter already exists, individuals are more likely to litter (Cialdini et al 1990). In transitory areas such as carparks and highways there may be less emphasis on clean-up campaigns or regular cleaning, which could provide a perceived social influence resulting in increased littering.

These social norms may differ in areas of low and high socio-economic disadvantage, leading to differences in the amount of environmental litter. Perhaps there is better waste infrastructure or more money spent on cleaning public spaces in wealthier areas, or perhaps resident rate payers in these areas expect more cleaning. High litter counts in areas of low SES may also be attributable to what Pepper and Nettle call the 'behavioural complex of deprivation' (2017). They hypothesize that lower levels of individual control over one's life in areas of low SES (due to lack of resources, education, etc) leads to a shorter decision-making horizon, and therefore to a constellation of resulting behaviours that are not necessarily advantageous in the long term, but that are an appropriate and relevant response to the prevailing conditions. We suggest that litter and illegal dumping could be seen as outcomes of this type of short-term decision making, which would explain the strong correlation with low SES levels.

Establishing effective interventions in this context is complex, though might include financial incentives, additional cleaning aimed at reducing social cues to litter, or interventions that affect the balance of short and long-term opportunities. For instance, access to improved health or dental care could shift littering patterns, along with many other issues arising from anti-social behaviour. Previous research found that economic incentives such as container deposit legislation was most effective at reducing beverage container litter in areas of lower SES (Schuyler et al 2018).

Finally, the relative ease (or lack thereof) of littering can also influence individual behaviour. (Ajzen 1991). In other words, people may litter more in areas where there is less fear of reprisal. Previous research has shown that individuals tend to litter more in areas with fewer people, perhaps because of the perceived risk of being observed (Bator et al 2011). In the current study, sites with less infrastructure (fewer roads) surrounded by more densely populated areas tended to have higher litter loads, as did areas near water. Similar results were reported in a national survey of coastal sites in Australia (Hardesty et al 2017). This may indicate the availability of a suitable dumping area (e.g. wetland, river), coupled with a lack of guardianship, which facilitates littering (Cohen and Felson 1979). Interventions targeting these behaviours might include increased signage indicating that the areas are monitored, or campaigns emphasizing the ubiquity of enforcement of litter laws.

On a state level, the states with the lowest levels of litter (on a per capita basis) were VIC and NSW. This was true even when we accounted for population density, site-level socio-economic factors, transportation networks, and land use. Interestingly, VIC and NSW are the two wealthiest states in Australia (Pariona 2019). Thus, this state effect could be due to a combination of policy differences and long-term effects of wealth or infrastructure investment.

The relationship between population density and mismanaged waste reported from these data is complex. At a local level, increasing population (on a log scale) is positively correlated with litter, but at a regional level, increasing population density (log scale) is correlated with decreasing levels of litter. This may be a result of economies of scale, where enhanced waste management solutions are more efficient in larger communities, and also to the lower availability of land in densely populated regions (Matsunaga and Themelis 2002). In these areas, diverting waste from landfill becomes a more critical concern, and other solutions such as recycling become more economically viable, leading to decreased mismanagement of waste.

It is worth noting that that population density is not the strongest predictor of mismanaged litter in Australian cities, despite many global models relying heavily on population density as a predictor variable (Jambeck et al 2015, Lebreton et al 2017). This pattern is true not only in Australia, but in a study of seven global countries (Schuyler et al 2021). Incorporating a broader suite of predictor variables will certainly improve global models of mismanaged waste.

We found a significant negative correlation between the amounts of litter found (per unit area), and the relative area searched at each site. Because site sizes ranged significantly in area, from 71 to 9600 m², we suggest this could be a result of changes in search efficiency during surveys as areas increase. Given the range in site areas, search effort (number of person hours) would have to increase by over 100-fold to maintain constant effort per unit area. Additionally, NLI protocols do not specify a search pattern, so it would be much easier to miss items of debris at larger sites. These findings emphasize the importance of designing and conducting surveys in such a manner that they do not create noise in the data, but also point to the potential consequences of not accounting for sampling effort when modelling data.

One limitation of the National Litter Index is that surveys are conducted primarily in urban areas and do not extend to rural communities. Ideally, surveys would also be conducted in regional areas to understand whether the trends in less populated towns are similar to those in urban centres. In any case, with a better understanding of the drivers of litter distribution, decision makers are poised to more effectively target interventions aimed at decreasing environmental litter.

The results of this research enhance our understanding of the factors that drive litter distribution within the terrestrial environment. Knowing how socio-economics, population, land use, site type, and other variables affect litter levels provides a useful road map of where and how to best target interventions to decrease the flow of mismanaged waste into the environment. This research also has implications for modelling litter loads on land, and subsequent input to the ocean. We demonstrate that accurately estimating total debris loads on land requires a more complex source function than population density alone.

We are extremely grateful to Keep South Australia Beautiful and Keep Australia Beautiful for generously sharing this data set. We would like to thank McGregor Tann and the numerous people that were involved in collecting the data. Kimberley Opie was instrumental in compiling the GIS data layers that we used as covariates. Thanks also to the three reviewers whose insightful comments strengthened the manuscript

The data generated and/or analysed during the current study are not publicly available for legal/ethical reasons but are available from the corresponding author on reasonable request.

This research was supported by the Australian Packaging Covenant Organisation and the Commonwealth Scientific Research Organisation.