Online Sex Seeking has been repeatedly associated with risky sexual behaviour, but also a variety of risk reduction practices, including increased HIV testing, and so called seroadaptive strategies which require gbMSM to be aware of their HIV status.
So, given the increasing proportion of gbMSM who report meeting partners online, there is a need to understand the spatial density of app users so that regional planning of sexual health resources can be optimized and tailored interventions for these men can be appropriately targeted.
As such, previous methodology pioneered in Atlanta has used Geosocial Networking Apps themselves, to estimate the spatial distribution of users by sampling a number of locations via a sex-seeking app popularly used by gay and bisexual men. At each sampling location they then counted the number of participant, recorded their distance, and identified their race or ethnicity. As you can see in the third panel here, they found that this information could be used to better understand the population of app-users and hoped that these methods could be used to better tailor services to these men.
Leveraging these methods, we sought to use a popular geosocial networking app to describe the spatial density of app using gbMSM in Vancouver and to model app-user density using administrative data from the 2016 Canadian Census.
To do so, a grid of sampling coordinates was generated across Metro Vancouver starting with a random location in Downtown Vancouver. The order at which these locations were sampled was then randomly assigned to control for spatial bias introduced by this method.
We then counted the number of individuals within 1-mile of our sampled points – which were programmed in using a geolocation emulator that allowed us to specify our exact coordinates.
As we were cautious about user privacy, no user data or characteristics were recorded, only the number users within 1-mile of each location was sampled.
This data collection took place on a Monday, Tuesday and Wednesday night in November 2016, between 5 PM and 11 pm. Therefore, in addition to randomizing the order at which sampling occurred, we also included a control term in our multivariate analyses that accounted for an interaction between sampling time and day.
Other variables from the Canadian Census were also modeled, including population size, relationship status, average age, median income, government transfers, employment levels, immigration and visible minority data, and educational attainment.
To match these dissemination area-level variables to our sampling unites, we calculated a population-area weighted value for each census variable by taking the sum of the products for the unweighted census variable for each dissemination area, multiplied by the proportion of the sampling radius accounted for by the DA and again by population-weight of the DA . This resulted in an imputed population-area weighted value for each sampling radii.
As you can see, most sampled units had relatively low user density. With a couple of large counts in the downtown area. Therefore, a multivariable poisson regression was used to identify the independent and adjusted factors associated with user density. Backwards AIC selection was used to identify the best-fitting model and compared against a model that only included population density and our time/date control variable.
As you can see on this map, coloured by natural breaks, population density was highlight skewed.
So, the distribution, as one might expect, appears to be largely related to the general population density. Indeed, we found that for each 1000 person increase in male population density, app user density increased by 1.8 persons per square kilometer.
Improving on this model, however, we found that for each 1000 person increase in population density the number of app users increased by only 1.49; for each 1-percent increase in the proportion of DA residents receiving government transfers app-user density increased by .79 users/sq-km; for each 1-percent increase in the proportion of males who were single, app user density increased by 1.14 users/sq-km. for each 1-percent increase in the proportion of males who were immgrants, app user density increased by 1.07 users/sq-km. for each 1-percent increase in the proportion of males who were visible minorities, app user density increased by .98 users/sq-km.
This model performed significantly better than the model which only included population density. The pseudo-R-squared increased by 15%, the number of model outliers decreased four-fold, and the AIC was minimized.
By showing that administrative data and geosocial networking applications and be leveraged to improve care delivery, these findings support the need to tailor geographically imbedded services to meet the needs of gbMSM and those who seek sex online, as well as the equitable distribution of health resources. While these data alone are not the only tool available to public health planners and community advocates, they may provide an additional source of data which can be consulted as health messaging is deployed and geosocial networking aps are leveraged for the betterment of gbMSM health.
While this study has several obvious limitations, some of these can be addressed by research that examines the acceptability of collecting publicly available data from user profiles, compares app-user densities across several platforms, examines how user density varies over the course of the day and week, assess the generalizability of our findings to other metropolitan and rural areas, and examine the effect that various levels of geography and measurement have on model results.
In conclusion, our findings have highlighted the potential utility of matching geosocial networking data to administrative data, which may help public health and community-based organizations better understand the spatial density of app-users and thus better tailor and target their services to this population. Indeed, we observed that several variables from administrative datasets might be leveraged to better understand the distribution of app-users in Metro Vancouver.
And with that, I want to thank our institutional partners, and those who provided data and or funding for this project as well as all of you for your time today. Thank you.
Spatial Density of Gay Men In Metro Vancouver
CAN GEOSOCIAL NETWORKING APPLICATIONS
AID IN HEALTH SERVICE DELIVERY EFFORTS?
KG. Card, J. Gibbs, NJ. Lachowksy, BW. Hawkins, M. Compton, J. Edward,
D. Ho., T. Salway, M. Gislason, R. Hogg
0.50 1.00 2.00
14 Studies | Yang et al., 20141.35 (1.13-1.62)
Sought Sex Online
15 Studies | Liau et al., 20061.68 (1.18-2.40)
11 Studies | Lewnard et al., 20141.24 (1.01-1.52)
Meta-Analytic Odds for CAS, by Online Sex Seeking
Fig 4. Increasing Prevalence of Online Sex Seeking
2.1% per year
R² = 0.339
2000 2002 2004 2006 2008 2010 2012 2014
Delaney et al. (2014) “Using a Geosocial Networking Application
to Calculate the Population Density of Sex-Seeking Gay Men for
Research and Prevention Services.” - Atlanta
Sampled Locations Sampling Areas Relative User Density, By Race
• Use a popular geosocial networking application to
describe the spatial density of app-using gbMSM in
• Use 2016 Canadian Census data to identify factors
associated with increasing app-user density.
User 1 User 2 User 3
User 4 User 5 User 6
User 7 User 8 User 9
User 10 User 11 User 12
• Distance set to feet
to maximize sampling
• Counted Number of
users within <1-mile
of sampling location.
• No data collected
from profiles and
only record of counts
• Dissemination Area Characteristics
• Male Population Density
• Percent of population that is male.
• Percent of male population that is not married.
• Average age of males, in years from Census.
• Median income of males, per $1000 CAD.
• Percent of male income provided by government transfers.
• Percent of males that are unemployed.
• Percent of males that are immigrants to Canada.
• Percent of males who identify as a visible minority.
• Percent of males with a postsecondary education.
𝑾 = 𝒙 ∗
𝒙 = Unweighted Census Value for a Given Dissemination Area.
𝑷𝒊 = The Population of the Given Census Dissemination Area
𝑷 𝒕 = The Total Population for all Census Dissemination Areas
captured by the sampling radius.
𝑨𝒊 = The area of the Given Census Dissemination area captured
by the sampling radius.
𝑨𝒕 = The total area of the sampling radius.
Population-Area Weighted Value for a Given Sampling Radius (W)
Prop. of Sampling
Area in DA
User Density (Users/km2)
p-value = 0.2804
App User Density = Population Density(1000*x) + Time*Day
Variable β SE Z-score P-value
Male Population Density (per 1000) 1.494 0.025 16.351 <0.0001
% Receiving Government Transfers 0.799 0.024 -9.523 <0.0001
% of Males Not Married 1.139 0.010 12.457 <0.0001
% of Males who are Immigrants 1.068 0.012 5.290 <0.0001
% of Males who are Visible Minorities 0.977 0.006 -3.643 <0.0001
*Controlling for Interaction Between Time of Day and Day of Week
McFadden’s Pseudo-r2: 0.75
Census-Level Factors Associated with App-User Density*
• Understanding the spatial density of app users can potentially…
• Ensure that existing services are appropriately tailored to
gbMSM and those who seek sex online,
• Advocate for the equitable distribution of health care
• Empower community-based organizations to better deploy
public service announcement and health service
advertising within communities and municipalities.
• Empower community- and municipal- health organizations
to better leverage geosocial networking apps for the
improvement of gbMSM health.
Limitations and Future Research
Future research should:
• Study the acceptability of data collection techniques that
examine user characteristics and not just user counts.
• Compare densities of using a variety of apps.
• Explore how user densities vary over the course of the day
• Examine app-user densities in more rural regions and other
• Asses whether modifiable areal unites impact study
estimates by comparing results across multiple geographic
levels (e.g., census dissemination areas and other relevant
administrative units, such as health service delivery areas).