There are few amenities as integral to British culture as pubs. Once described as ‘the heart of England’ (Pepys, 1661 in Fletcher, 2013), these pubs serve alcohol and food. Beyond feeding British alcoholism, pubs are important places for sociality: having a pint constitutes a ‘social act’, giving a ‘sense of inclusion…independent of our status’ (King, 2008: 3). Thus, pubs act as a ‘social leveller’ by attracting customers from various socioeconomic classes (ibid). However, pubs are associated with criminality: alcohol consumption increases proclivity to commit crime (Wheeler, 2018), violent tendencies (Exum, 2006), and the victim’s vulnerability to assault (Testa & Parks, 1996). Therefore, excessive alcohol consumption within and surrounding pubs may explain spatial correlations between pubs and crime (Gruenewald, 2007).
I aim to investigate the correlations between crime rates and pub locations. To confirm the spatial perspective of pubs as ‘social levellers’, I will investigate correlations with demographics of areas surrounding pub locations.
Kernel Density Estimation (KDE) is used as a spatial smoothing technique to visualise the spatial pattern of pubs. KDE is a non-parametric technique for density estimation: a known density function is ‘averaged across observed data points to create a smooth approximation’ (Cerrito, 2010: 149). Fig 1 shows that the 10% highest density of pubs occurs in Zone 1, where south Camden, south Islington and west Westminster meet, possibly due to the high population of childless residents, and high tourist activity. The City of London has been omitted from analysis because of its anomalous nature, with an extremely high pub rate at 114 per ten thousand residents.
Fig 2 visualises each Borough’s pub numbers normalised to its legal population. Aforementioned Zone 1 boroughs of Camden, Islington and Westminster have the highest pub rates at 5–6 per ten thousand residents. An outlier is Bromley, with a rather high pub rate (3–4), despite only ¼ of its land area being within the 75% catchment boundary in Fig 1. This is due to the large land area of the borough, so pubs look sparse despite the high number.
Westminster experienced 79,213 crimes in 2019, highest out of all boroughs, due to its large night-time economy with vices like alcohol and drugs, and high number of visitors at 26 million annually (City of Westminster, 2018). To investigate the spatial association of pubs and demographic factors, I will look at the Lower Layer Super Output Area (LSOA) level. Pertinent to this investigation is Westminster, where most crimes and top 10% pubs density are, at 5.07 pubs per 10,000 residents (Fig 1; Table 1), reflecting the size of its night-time economy. Geographically weighted regression (GWR) will be applied with an adaptive bandwidth to account for spatial non-stationarity across wards (Fotheringham et al, 2002).
Most crimes occur in southeast Westminster. Nightlife is rampant here: most of Westminster’s nightlife outlets, including pubs, are located here. This high-crime area coincides with the top 25% pub density. Centre-west of Westminster is Hyde Park, a large green space, accounting for the few crimes. However, crime data accuracy is questionable; crimes are often underreported since it ‘does not always serve the victim’s best interest to report’ (Myers, 1980: 23).
Most LSOAs have a positive correlation between pubs and crime, except for northwestern Westminster with a coefficient of -0.2–0.0. Fig 4 shows that globally, each pub located in Westminster increases crime numbers by 1090; this is statistically significant at alpha-level=0. GWR coefficients vary geographically: for LSOAs within 50% density of pubs, every pub located here increases crime number by at least 200. Here, the GWR model explains about 72.0–87.1% of variance in crime numbers.
Westminster is mostly a childless borough. About three-quarters of Westminster’s area have childless households at proportions of above 70%, which is much higher than London’s median proportion of 54%. The highest percentage of childless households occur in southeastern Westminster, at 70%-90% childless. The lowest proportions accrue near the northwest at around 40%-60% childless.
The global coefficient is 0.07, statistically significant at alpha=0. With every 1% increase in proportion of childless households in Westminster, pub numbers increase by 0.07. East Westminster experiences a 0.3–0.5 increase in pub numbers with each 1% increase in childless household proportion. For most of this area, the model explains 50.3%-70.3% of variance in pub numbers. This model has a moderate to high explanatory power for areas with 50% pub density. Anomalous LSOAs, Westminster 004B and 004E at R-squared -9.38 and -57.5 respectively, were removed to prevent skewing the range.
The proportion of white residents experiences a large geographical variation, with the proportion generally increasing from northwest to southeast. Northeast Westminster mostly has 23.6%-49.3% of white residents, while majority of southeast Westminster has a much higher proportion at 61.6%-85.8%. Jenks optimisation method was used to provide a more meaningful visualisation, since the values have a high variance of 414 (Jones, 2010).
The global coefficient is 0.02, which is not statistically significant at alpha=0.05. Westminster experiences geographical variation in the correlation, with the centre and northeast regions having a positive correlation while the rest experience a negative correlation. The explanatory power of this model is low, ranging from 10%-40% for most regions. An anomaly would be the small, eastern LSOA: each 1% increase in white residents’ proportion decreases pub incidences by 0.06 to 0.08; the model has a high explanatory power for this LSOA at 60–70%. Altogether, this model has high explanatory impotence; white residents’ proportion is unlikely to significantly affect the number of pubs in the area without accounting for confounders.
Westminster is a very wealthy borough. Most LSOAs have a mean annual household income of at least £47,800, which is 63% higher than UK median of £29,400 and 35% higher than UK mean of £35,300 (ONS, 2019). In fact, most southern LSOAs earn about £64,900-£140,700 a year. In the eastern region where most pubs are located, income ranges from £47,800-£107,700. However, mean incomes tend to favour high-earners since income distribution has a positive skew (Chiripanhura, 2011), under-representing lower-earning households in this statistic.
The global coefficient of 0.06 is not statistically significant at alpha=0.5. Northeastern, southern, and some of eastern Westminster experience negative coefficients of income on pub incidences. In the eastern region of its perimeter, every £10,000 increase in mean annual household income tends to result in about 0.2–0.6 fewer pubs in the area. A small area southeast has a positive coefficient of 0.0 to 0.4. This model explains some variance in pub numbers, with local R-squared of mostly 0.3 to 0.5.
Conclusion
This analysis reveals that in Westminster, crime does indeed amass near pubs, and that more pubs are likely to be sited in areas with more childless households, possibly due to lack of time and freedom constraints imposed by childcare. However, there may be confounders not investigated in this study that contribute to the prior correlations. In contrast, no statistically significant correlation has been found between pubs and white residents’ proportion (Fig 8), and between pubs and income (Fig 10). Hence, pubs are unlikely to be a ‘social leveller’ in Westminster. This analysis can help business owners select suitable pub locations to maximise appeal and access to customers.
However, non-resident office workers congregate around the clock in Westminster, so residency may not accurately reflect pub customer demographics. Modifiable areal unit problem may cause statistical bias in GWR, since maps were limited by LSOAs’ varying sizes and populations. Choropleth maps may misrepresent visual information should they suffer from area-size bias, wherein LSOAs with large administrative units are overemphasised (Dykes et al, 2002; Skowronnek, 2015), as exemplified by Hyde Park’s LSOA having bigger visual weight than surrounding LSOAs despite its low population. Removing LSOAs with R-squared < 0 risks cherrypicking data; instead, reasons for negative values should be analysed. R-squared cannot be used to determine coefficient estimates and bias in predictions, nor to indicate the goodness-of-fit of data to model (Faraway, 2016). Assessing residual plots and using multivariate GWR analysis can increase models’ explanatory power.
List of figures
Fig 1. KDE of pubs. Catchment boundaries indicate 10%, 25%, 50% and 75% KDEs.
Fig 2. Pub rates per borough.
Table 1. Boroughs with highest pub rates.
Fig 3. Crimes in thousands and KDE of pubs.
Fig 4. GWR of pubs on crimes, in thousands, in Westminster.
Fig 5. Proportion of childless households and KDE of pubs.
Fig 6. GWR of proportion of childless households on pub incidences.
Fig 7. White residents’ proportion and KDE of pubs.
Fig 8. GWR of white residents’ proportion on pub incidences.
Fig 9. Mean household annual income estimates and pub locations.
Fig 10. GWR of income on pub incidences.
References
Brantingham, P., & Brantingham, P. (1995). ‘Criminality of place’ in European journal on criminal policy and research, 3, 3, 5–26.
CEHI (n/d) ‘Introduction to Hotspot Analysis’ (WWW) Houston: WHO, Children’s Health Environmental Initiative. (Retrieved 18 April 2020 https://www.cdc.gov/dhdsp/maps/GISX/training/module3/files/3_hotspot_analysis_module.PDF)
Cerrito, P. (2010) Cases on Health Outcomes and Clinical Data Mining: Studies and Frameworks: Studies and Frameworks. Pennsylvania: IGI Global.
Chiripanhura, B. (2011) ‘Median and mean income analyses: Their implications for material living standards and national well-being’ in Economic & Labour Market Review, 5, 45–63.
City of Westminster (2018) ‘The evening and night time in Westminster (6pm — 6am): Night Time Commission Consultation Response’ (WWW). London: City of Westminster. (Retrieved 21 April 2020 https://committees.westminster.gov.uk/documents/s26213/Item%204-%20Night%20Time%20Commission%20Consultation%20Response.pdf)
Dykes, J. A., (2002) Creating information-rich virtual environments with georeferenced digital panoramic imagery. In Fisher, P., and Unwin, D., (eds.) Visual exploration of virtual environments. New York: Taylor & Francis.
Exum, M. L. (2006) ‘Alcohol and aggression: An integration of findings from experimental studies’ in Journal of Criminal Justice, 34, 2, 131–145.
Faraway, J.J. (2016) Linear Models with R. Florida: CRC Press.
Fletcher, R. (2013) ‘Pubs are ‘the heart of England’ (WWW) UK: The Journal. (Retrieved 10 April 2020 http://www.thejournal.co.uk/culture/restaurants-bars/pubs-the-heart-england-4414305)
Fotheringham, A. S., Brunsdon, C., and Charlton, M. E. (2002) Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. Wiley, Chichester.
Gruenewald, P.J. (2007) ‘The spatial ecology of alcohol problems: niche theory and assortative drinking’ in Addiction, 102, 6, 870–9.
Jones, R. (2010) ‘GIS Data Classifications in Cartographica’ (WWW) Northern Virginia: ClueTrust. (Retrieved 20 April 2020; https://blog.cartographica.com/gis-data-classifications-in-cartographica.html)
King, G. (2008) The enduring appeal of the local. Oxford: Social Issues Research Centre.
Myers, S.L. (1980) ‘Why are crimes underreported? What is the crime rate? Does it “really” matter?’ in Social Science Quarterly, 61, 1, 23–43.
Office for National Statistics (ONS) (2019) ‘Average household income, UK: Financial year ending 2019 (provisional)’ (WWW) UK: Office for National Statistics. (Retrieved 22 April 2020 https://www.ons.gov.uk/peoplepopulationandcommunity/personalandhouseholdfinances/incomeandwealth/bulletins/householddisposableincomeandinequality/financialyearending2019provisional)
Skowronnek, A. (2015) Beyond choropleth maps: A review of techniques to visualize quantitative areal geodata. Vancouver: UBC InfoVis Group.
Testa, M. and Parks, J.M. (1996) ‘The role of women’s alcohol consumption in sexual victimization’ in Aggression and Violent Behavior, 1, 217–34.
Wheeler, A.P. (2018) ‘Quantifying the Local and Spatial Effects of Alcohol Outlets on Crime’ in Crime & Delinquency, 65, 6, 845–871.
