Main findings and interpretation
In this study, GSV was found to be generally less sensitive than on-street observation in identifying various aspects of alcohol-related content in the urban streetscape environment. The sensitivity difference between GSV and on-street surveying was largest when looking for visible drinkers, followed by alcohol-related advertising, and then venues with evidence of alcohol sales. The lower sensitivity of GSV for visible drinkers may be due to the small size of alcoholic beverages in the imagery. Similarly, some advertising materials are difficult to identify in GSV images with limited resolution and ability to zoom. This picture was supported by the subjective experience of researchers: small item detection was felt to be difficult using GSV, whereas large static objects were generally easy to identify and categorize (although GSV observation still found only 86 % of the venues that on-street observation found). Compared to visible drinkers and alcohol-related advertising, there was greater concordance between the total number of venues identified between both methods, suggesting that GSV may provide useful data on large, static components of the built environment (e.g., alcohol retail venues and large signs).
While surveying with GSV showed a decreased survey time compared to the on-street method, it was only statistically significant when travel time was also included in the analysis. However, it is plausible that further familiarization with the technique could enable greater reduction in time taken to survey compared to surveying on foot. It should also be noted that in this study the researchers’ workplace was based within six km of the street segments, so travel times were relatively small.
Another reason for the difference in results is that GSV’s in-built blurring algorithm, although designed to maintain anonymity of people photographed, often blurs items close to the face which then limits the detection of people consuming alcoholic beverages. GSV also performs automatic blurring of a large proportion of signs, which is a feature intended to blur vehicle license plates (at least in the New Zealand setting). This seems to have rendered detection and categorization of some signage difficult.
The range in GSV image capture dates and times may account for some of the observed variation when compared with on-street observation. Though the majority of imaging was taken within three months of the observations, the oldest images were from 2008. It is also possible that some venues, signage or seating areas for drinkers may have changed even in the few months between the dates of image capture used in our GSV surveys and on-street surveys. In terms of the wider use of GSV, if it was to be used for monitoring of policy, e.g., advertising regulations, then frequent new imaging would be required to ensure up to date information is provided. Within the region studied, images had generally been updated on two occasions since the initial establishment of GSV coverage in 2008.
The time of day of GSV image capture was also variable. Even within daylight hours the presence of portable signage and seating can vary, and without knowing exact times of image capture it is impossible to truly match on-street observation times to image capture times when comparing survey methods. This may particularly affect the number of visible drinkers seen by GSV, compared to that seen by the 1430 h to 1730 h on-street observation, as there is likely to be a higher prevalence of drinking in this mid-late afternoon period, compared to across the whole day.
Images captured on GSV may be further limited by the means by which GSV images are acquired. Vehicles used to capture GSV images may be unable to access pedestrian-only areas of streets or areas with difficult road access, resulting in suboptimal image angle or quality.
The fact that GSV images were also taken during daylight hours makes it broadly comparable to on-street observations, but exclusive use of daylight data may limit our ability to draw universally applicable conclusions. Since many bars and restaurants neither open nor display portable signage until late afternoon, and not all our data collection overlapped with this period, we suspect that our findings may not reflect the maximal amount of exposure, and hence the maximum potential to influence behaviors. However, exposure to large, static alcohol and tobacco-related imagery in the streetscape is unlikely to be significantly affected by this.
Study strengths and limitations
A notable strength of this study was it being the first (to our knowledge) to rigorously compare GSV with on-street observation for alcohol-related imagery in the urban streetscape. It was able to do this for a wide range of imagery and in a range of different types of streets and suburbs. The observers were also all very familiar with alcohol-related imagery in the urban environment studied. The study results should have reasonable generalizability to other developed country urban environments – but some cities might have different alcohol-related imagery (e.g., if they have legal constraints on alcohol advertising and sponsorship – which hardly exist in the New Zealand setting).
Nevertheless, this study could have been more rigorous if the time of day, day of week and season were exactly matched between the GSV image date and the on-site observations (this was not feasible in this study for logistic reasons). Improvements could also have been made in the standardization of street segment lengths for surveying (for example, Google Maps universally rounds distances to the nearest 50 m, meaning we were only able to standardize street length to 400 ± 50 m). Variations in the lengths of street segments selected for surveys may have contributed to some of the variation in counts of observed environmental factors across street segments.
Potential future research
Future study in this area could also take into account both the need for prior knowledge of signage and brands in order to identify alcohol-related advertising, and the time of data collection. The level of observer knowledge could be a consideration for future research attempting to use similar methods. Prior knowledge was especially helpful when using GSV, as signs were often blurry, unclear or of poor resolution. The requirement of background knowledge may suggest that for surveys using GSV in international or cross-cultural contexts, additional observer training might be important. Furthermore, our study was limited to data collection during daylight hours only, and further studies could include evening periods.
It is worth noting that “footpath views” in GSV were occasionally available in our study, were of a much higher image quality and yielded a more accurate representation of the pedestrian experience. If this footpath view were to become more universal in urban settings, and image quality and frequency were increased, GSV could become be a more valid tool for surveying urban streetscapes for health determinants such as alcohol-related imagery.
Further studies could also investigate to see if the exclusion of older GSV images was a way to improve the reliability of evidence, compared to on-street observation.