The Digital Sin City: An Empirical Study of Craigslist s Impact on Prostitution Trends

Transcription

1 The Digital Sin City: An Empirical Study of Craigslist s Impact on Prostitution Trends Probal Mojumder Carlson School of Management University of Minnesota mojum003@umn.edu Completed Research Paper Anindya Ghose Stern School of Business New York University aghose@stern.nyu.edu Abstract Jason Chan Carlson School of Management University of Minnesota jchancf@umn.edu Online prostitution has flourished due to presence of online intermediaries. Despite the illegality of selling sexual services online, the Section 230 of Communications Decency Act shields websites from liability for unlawful postings by third parties. Consequently, the websites like Craigslist have become a haven for prostitution-related ads. In this paper, we quantify the economic impact of Craigslist s entry on prostitution incidence, and identify potential pathways in which the website affects the sex industry. Using a national panel data for 1,796 U.S. counties from 1999 to 2008, our results suggest that entry of Craigslist to a county is related to percent increase in prostitution cases. In addition, our analyses investigates four key market mechanisms, namely the relationship to commercial vice, geographical trends, spillover effects, and enhanced policing efficacy. Our results contributes broadly to the emerging literature that researches the societal challenges associated with online intermediaries and Internet penetration. Keywords: Craigslist, Classified ad websites, Online Intermediaries, Prostitution, Vice Crimes, Online Platforms, Commercialized Vice Thirty Seventh International Conference on Information Systems, Dublin

2 Introduction In recent years, the Internet is transforming the sex trade. Perer (2012) portrays the Internet as a virtual red-light district which plays the intermediary role of facilitating information flow between sex workers and buyers, making it easier for these individuals to strike mutually satisfactory deals, and engage in paid sexual transactions. Even in the United States, where prostitution and its facilitation are illegal everywhere except Nevada, the marketing and arrangement of commercial sex is moving online. In particular, online classified advertising sites and underground entertainment sites have become the go-to avenues to advertise and set up transactions for paid sexual services. Despite the illegality of selling sexual services online, the Section 230 of Communications Decency Act (CDA) shields websites from liability for unlawful postings made by third parties. Consequently, the websites like Craigslist, often known as the Wal-Mart of online sexual services, have become a haven for prostitution-related ads. 1 Craigslist is one of the largest online classified service website that offers a dedicated section for prostitution ads in the early 2000s. Advertisements for sexual services were initially posted under the section Erotic Services, which was subsequently renamed as Adults Section in 2009 after numerous complaints made against Craigslist. The website faces intense public scrutiny over the years, as many believed that the website s operators have knowingly allowed ads that offer paid sexual services to be posted on the site. The continuous pressure from a group of state attorneys general finally led to the closure of the Adults Section in September Despite the anecdotal accounts that link Craigslist to prostitution acts, no studies known to date has formally examined and quantified the relationship between the site entry and prostitution trends. Moreover, the relationship between the site entry and prostitution trends is theoretically ambiguous. On the one hand, literature from platform economics posit that the presence of online platforms is likely to increase prostitution incidence through intermediation effects which streamline the market clearing process (Bakos 1997, Il-Hann and Terwiesch 2003). On the other hand, the trust literature argues that online platforms are prone to issues of market failure due to the prevalent trust issues of transacting in online environments (Granovetter 1985). In particular, trust issues inherent in online transactions are exacerbated by the risks of legal persecution of participating in sexual solicitation, which in turn can deter users from participating in the platform-enabled prostitution market. On top of the theoretical ambiguous relationship, the mechanisms underlying the link between site entry and prostitution trends are not well-understood. To fill these gaps, the present research attempts to quantify the economic impact of Craigslist s entry on prostitution incidence, and shed light on the potential pathways in which site entry could affect the market for sex workers. Using a national panel data set, we examine the longitudinal relationship between Craigslist s entry and prostitution trends in 1,796 U.S. counties from 1999 to To empirically identify the entry effects on prostitution acts, we rely on a natural experiment setup inherent in Craigslist s expansion in the United States. During its expansion, Craigslist was launched in different locations in a staggered fashion across the years. Thus, in a given year, there is a subset of counties with Craigslist and another set of counties that do not. Given that each Craigslist site is location-specific, the prostitution ads posted on the site cater to the population living within the location. This unique entry setting allows for the econometric comparison of treated counties with control counties. Exploiting the natural experiment framework, we run differencein-difference panel regressions of prostitution incidence on Craigslist s entry with county and year fixed effects, and include controls to account for demographic, socioeconomic, and crime-related factors that may affect prostitution trends. Further, we perform a series of robustness checks which include analyzing a subsample of counties with active prostitution cases, considering alternative model specifications, using propensity score matching to identify comparable treatment and control counties, and performing falsification checks to check for spurious results. In addition to the empirical checks, we investigate various mechanisms, namely the site entry relationship to commercial vice, geographical trends, spillover effects, and enhanced policing efficacy. Our empirical analysis reveals that the entry of Craigslist holds a positive relationship with prostitution trends, which equates to a percentage increase in prostitution cases. We find that sign and significance of our main results is robust under various model specifications. Falsification tests suggest that the relationship between site entry and prostitution cases did not arise spuriously, and that a pre-entry trend 1 A Craigslist blog post revealed that there are at least 700,000 prostitution ads posted on the site within a year. See manual-screening-matters/, accessed on June 17, Thirty Seventh International Conference on Information Systems, Dublin

3 leading to an increase in prostitution trends was not present. In our examination of the underlying mechanisms, our analysis reveals that the increase in prostitution from Craigslist s entry is likely a result of new market generated by independent prostitutes and traditional vice organizations. In addition, we find that the impact of Craigslist s entry tends to manifest as new prostitution activity in both counties with no prior cases of prostitution and those with past history of prostitution. Further, we find that the entry of Craigslist generate spillover effects to neighboring counties without Craigslist. Finally, we observed that entry of Craigslist not only increased prostitution trends, but also helped increase policing efficacy towards making prostitution arrests. Our paper aims to make a few key contributions. First, our study contributes broadly to the emerging literature on the societal challenges associated with online intermediaries and Internet penetration (Chan and Ghose 2014, Chan et al. 2016, Cooper 2013). While the use of the Internet holds both positive and negative impacts towards the society, extant work has largely focused on examining the former while overlooking the downsides of the Internet. Responding to the call for research on the societal challenges and drawbacks of digital technologies (Fichman et al. 2015, Majchrzak et al. 2012), our investigation of Craigslist s impact on prostitution trends seeks to bring about greater awareness on the vulnerabilities introduced by online technologies, so that proper technological design and policy changes can be introduced. Second, by assessing the longitudinal impact of a major classified ad site on prostitution trends, our study not only empirically validates the link between Craigslist and prostitution, but also quantifies the entry effect at the societal level. In quantifying the site impact on prostitution trends, our findings also add to the literature of platform economics (Rochet and Tirole 2003) by illustrating that platform dynamics can induce an increase in market participation even for illicit markets, wherein participants are faced with substantial costs of arrests and criminal charges. Third, by shedding light on the underlying mechanisms that govern the growth in prostitution trends facilitated by Craigslist s entry, our study provides actionable, policy-relevant insights needed for the effective implementation of strategies and interventions for curbing the proliferation of the underground sex trade. In particular, the awareness of the mechanisms of the platform-enabled prostitution serves to expose how and where the sex trade would flourish in the Internet era, allowing for a more efficient deployment of enforcement resources. Study Context and Related Literature Craigslist Craigslist started in early 1995, and is the leading online classified service provider in the United States in the 2000s. With the launch of the Erotic Services section in 2001, Craigslist website served as a portal for prostitution-related advertisements (Farley et al. 2013). By 2005, Craigslist had become a virtual prostitution venue, hosting 25,000 new prostitution-related ads every ten days in the United States alone (Farley 2006). A typical prostitution ad can either explicitly indicate the provision of sexual services or implicitly suggest similar services via sensual massage services. These solicitation ads tend to include provocative photographs and provide information such as hourly rates, phone numbers, descriptions of worker s body measurements, and keywords like busty, or fantasy girl (Sandoval 2009). In protest of prostitution related ads, the community and government took critical actions against Craigslist. Specifically, two victims of online sex trafficking sent an open letter to Craig Newmark, the founder of Craigslist, as an outcry against the prostitution activities enabled by the site. In that letter, the two victims describe the horrors they suffered as they were being sold for sex through the adult services section of Craigslist. In March 2009, Cooke county Sheriff Dart sued Craigslist and provided evidence that the site was being used to facilitate prostitution of child and adult victims and that it was a public nuisance (Dart vs. Craigslist, 2009). In response, Craigslist made token changes to the site by announcing that minors should not use its erotic services section, providing links to anti-trafficking websites, and charging ten dollars to users who post in the adult service section (Sarno 2009). Following various prostitution-related complaints on Craigslist, seventeen state attorneys general sent a letter to the owners of Craigslist in August 2010, which requested the immediate removal of the adult services section of Craigslist. A month after the request, Craigslist voluntarily replaced its adult services section with a black text box that read censored. Several days later, the section was removed entirely (Doft and Robertson 2010, Lindenberger 2010). Despite actions taken to suppress the posting of prostitution ads on Craigslist, prostitution ads started to appear in other sections of the site (e.g., personals section), and in Craigslist sites outside of United States (e.g., Shenzhen, China; Lucknow, India; and Newcastle, Australia) Thirty Seventh International Conference on Information Systems, Dublin

4 (Louie 2010). Furthermore, following Craigslist s closure of the adult services section, other online platforms (e.g., Backpage.com, MyRedBook, and Men4RentNow) assumed the role of facilitating online solicitation (Chansanchai 2011, Chen 2010). In particular, after the closure of Craigslist s adult section, the number of prostitution-related ads on Backpage increased tremendously in 2010, suggesting the possibility of sex workers migrating to Backpage for solicitation (Feyerick & Steffen, 2012). Based on these observations, the selling of sexual services on classified ad sites is likely to persist beyond the closure of Craigslist s adult section. Thus, it is imperative to investigate the link between Craigslist and prostitution despite the closure of the prostitution section, as study findings would still serve as relevant policy inputs for curbing prostitution trends facilitated by other similar online intermediaries. 2 We next discuss the theoretical foundations underlying the phenomenon under study. Rational Choice Theory and Prostitution Based on Becker s (1974) framework on the economics of crime, the decision to engage in a criminal act is a rational process in which individuals make choices that maximizes their utility. Accordingly, under the rational choice theory, an individual practices conscious decision making and would choose to commit a crime if the total payoffs of the illicit act are higher than its associated costs of legal punishment. Similarly, in making a choice to participate in the prostitution market, the providers and clients alike would weigh the relative benefits and costs when deciding whether to engage in paid sexual transactions. From the supplyside perspective, the main benefits of participating in this market lie largely in the financial gains. 3 From the demand-side perspective, the incentive to participate in the market is mainly for the satiation of sexual desires. Common to both providers and clients, the costs of market participation involve legal punishment that comes in the form of fines and imprisonment. In the United States, the severity of punishment for engaging in prostitution varies across states and the type of prostitution offense committed. With the introduction of an online intermediary that facilitates the matching of the supply and demand of sex services, the payoffs and costs faced by participants in the prostitution market are subjected to several important changes that may influence the existing prevalence of prostitution. For instance, the literature in platform economics argues in favor of an increase in prostitution incidence, as intermediation effects and online modes of transactions are likely to enhance matching efficiency and reduce the risk of detection, which in turn induces greater market participation. Yet, it may be counter-reasoned that the online intermediation of sexual services is prone to market failures as trust issues of transacting in the online environment impede the successful matching of market participants. Further, there might not be available mechanisms to prevent such market failure, as participants cannot rely on authorities or governing bodies to resolve transactional disputes given the illegal nature of engaging in the prostitution market. These competing perspectives produce a theoretically ambiguous prediction of the relationship between Craigslist entry and prostitution trends. We expound each of these views in more detail next. Extant literature on platform economics posits that online intermediaries bring about reduced search costs (Bakos 1997, Il-Hann and Terwiesch 2003), facilitate online communication across spatial and temporal boundaries (Bailey and Bakos 1997, Jin and Robey 1999), and provide a broader base of product offerings (Brynjolfsson et al. 2003, Jin and Robey 1999). Consequently, the introduction of online intermediaries can affect the outcomes of related industries in significant ways. In particular, past studies have documented the impacts of online intermediaries on the newspaper industry (Seamans and Zhu 2013, Gurun and Butler 2012), the real estate industry (Kroft and Pope 2014), and also the sexual services industry (Logan and Shah 2013). Similarly, the entry of the Craigslist platform into specific locations is likely to increase local incidence of prostitution by facilitating the market matching process between the sex workers and clients in the area. First, by consolidating the demand and supply of sexual services within a location, participants of the Craigslist platform experience a reduction in search costs. In particular, sex workers need not to walk the streets to solicit customers, and clients need not travel to red-light districts to search for sexual services. In addition to eliminating search costs, a reduction in cost also comes in the form of reduced probability of 2 We chose Craigslist because its entry patterns are likely to be exogenous which helps to facilitate a cleaner empirical identification as compared to other sites which charges posting fees for prostitution ads in the study period. More details are provided in our empirical strategy section. 3 We understand that this can be an over-simplification of the situation faced by service providers, as sex workers may not always have aligned interests with the agency or the pimps, especially in the case of involuntary prostitution. However, this distinction is not crucial in this study as we are examining the aggregate prostitution trend induced by the entry of Craigslist, which is ultimately driven by financial motives. Thirty Seventh International Conference on Information Systems, Dublin

5 arrests and detection as the sex workers and clients soliciting on Craigslist can work out transactional details using online messages and s (Cunningham and Kendall 2011), thereby allowing them to avoid physical exposure to enforcement agents during the solicitation process. Further, given that the posting of solicitation ads on Craigslist is free, sex workers and escorts could avoid advertising cost, 4 which is otherwise present when posting ads in traditional medium (e.g., classified ads in newspapers). By lowering the overall cost of solicitation, the entry of the Craigslist platform is likely to increase market participation by spurring the demand and supply of sex workers who might not have participated in the past due to risk and cost considerations (Holt and Blevins, 2014). Second, sex workers and clients communicating on the online platform can plan sexual transactions with the affordance of Internet technologies. Past research in ecommerce has found that effective site features are able to facilitate greater purchase intentions (Hausman and Siekpe 2009). In the sales of sexual services, online technologies allow clients to assess key information of sex workers before committing into a purchase. For example, through the photos sent online, clients are able to perceive not only the facial attractiveness of workers (e.g., a profile headshot) but also any preferred attributes via their nude shots. At the same time, clients are able to compare service attributes across workers concurrently in online settings where transaction information (e.g., price, service duration, types of sexual service provided, travel requirements such as in-calls or out-calls) are readily available within the ads or communications (Häubl and Trifts 2000). Furthermore, through specialized online review sites (e.g., TheEroticRview.com, NaughtyReviews.com), clients are also able to seek information on the service quality of the workers and also check if the workers have a history of scamming customers based on experiences of other clients. Similarly, sex workers can screen out risky clients before physically meeting them (Brooks 2009), through an online search of the client s name or telephone number in online databases of individuals with records of violence or criminal involvement. Thus, given the availability of online affordances, sex workers and clients transacting via Craigslist platform are better informed than before, which will likely to increase the matching efficiency and heighten the prostitution prevalence in locations served by Craigslist. Third, as an online aggregator, online platforms tend to offer a broader variety of products and services compared to the physical stores. Compared to the traditional prostitution market in which the ethnicity of the sex worker is highly dependent on the location where the service is offered (Dank et al. 2014), the workers operating on the Craigslist platform are not limited to a specific ethnicity. Moreover, as reported in Milrod and Monto (2012), online platforms are likely to host a variety of sex workers by which a broad assortment of paid sexual services may be solicited, including unique and niche sexual services that may be difficult to locate in the traditional prostitution market. These services can span the sexually disenfranchised communities like LGBT, or S/M, or fetish, who actively use internet as a safe means to look for their unique sexual requirements (Cooper et al. 1999). Arguably, the broader diversity of workers along with the wider selection of service types would facilitate more matches in a market marked with highly heterogeneous preferences. Overall, under the various arguments listed above, the use of Craigslist in solicitation is likely to lead to an increase in prostitution incidence. Specifically, the Craigslist-induced prevalence of prostitution can either come from 1) individuals who have not participated in this market previously to enter the market, and also from 2) existing market participants who increase their frequency of solicitation as a result of reduced market frictions. However, it may be counter-argued that the intermediation effects from Craigslist may not result in an increase in prostitution incidence. While the intermediation effect of Craigslist may reduce the friction inherent in seeking sex workers and willing clients, it does not eliminate the legal risk involved in solicitation. In particular, it is possible for policing agencies to rely on the solicitation information posted on Craigslist to uncover prostitution activity and plan sting operations to arrest the parties involved in solicitation. Using Craigslist, sting operations may be conducted in two main ways. In the first approach, police agents act as clients responding to the Craigslist ads posted by the sex worker. The arrest process involves setting up a date to meet the sex worker, physically meeting the worker at a motel, and establishing a price for her services. Once a price is set, the undercover police agents can make a legal arrest. In the second approach, decoy ads are posted on Craigslist to solicit clients who are willing to pay for sexual services. Officers pose as sex workers and would meet clients at a hotel. Similar to the first approach, an arrest is made for the client when he offers money in exchange for sex (Howard 2008, Shively et al. 2013). 4 Craigslist only started charging a posting fee for escort ads in The site has allowed the free posting of such ads since the launch of this section in Thirty Seventh International Conference on Information Systems, Dublin

6 These web-based sting operations have been in operation since the early 2000s when Craigslist was launched (Table 1, Shively et al. 2013), and has the effect of heightening the risk of transacting on the platform. Thus, the intermediating effect of the Craigslist platform may not be achieved as market participants on the site do not enjoy a greater ability to evade legal detection and prosecution. Despite the online affordances that participants enjoy when transacting on Craigslist, the online environment is marked with high level of uncertainty that can impede successful transactions from taking place (Dimoka et al. 2012). While participants on Craigslist platform may have easy access to information on the characteristics of sex workers and their services, the authenticity of the information communicated is questionable. Further, excessive information and equivocality in online markets can lead to heightened consumer uncertainty, which subsequently reduces consumer trust in electronic markets (Grover et al. 2006). Given the ease of posting fee-less ads on Craigslist, the escort section is typically filled with large volumes of solicitation ads that can hinder decision making. Furthermore, Craigslist does not have institution-based mechanisms like online ratings and reviews for assessing the past behavior of the parties, making it difficult for sex workers and clients to make proper judgments on whether the transacting partners would honor their role in the transaction (Dellarocas 2003). Given that the online environment allows users to assume false identities easily (Cao et al. 2012), law enforcement entities may pose as transacting parties to launch sting operations on market participants, which further worsens information asymmetry issues in this situation (Dennis et al. 2008). While it may be possible to check the credibility of sex workers from third-party review sites, such review sites may also be used by law enforcement agencies to set up elaborate sting operations, which removes the advantages that online affordances provide to the transacting parties. Given these counter reasons, it is theoretically unclear if the introduction of Craigslist would indeed reduce the transaction costs and risk of participating in the prostitution market. Data To investigate the impact of Craigslist s entry on prostitution trends, we constructed a nationwide panel data consisting of 1,796 U.S. counties across 1999 to 2008 using data from various sources including The Erotic Review (TER) website, FBI s Uniform Crime Report (UCR) database, Craigslist, and U.S. Census Bureau. Our main dependent variable is Prostitution iy, which is the number of prostitutes having reviews in TER website in county i for year y. 5 TER is a review site that allows customers to rate their experiences with female sex workers. We construct two separate variable using TER dataset. First, the All Years variable considers the sex worker active for all subsequent years since her first year of review. Second, the active years variable keep the sex worker active within the first and last year of review. We further considered a sub set of sex workers with variable using Craigslist, when their review text mentions the words Craigslist, CL, or Craigs. These are those special case of prostitutes who surely used Craigslist ads to promote their business. We also use female prostitution arrests and female indoor prostitution arrests as other measure of prostitution. 6 We construct this variable by summing up the number of prostitution arrests across all the reported arrest-level incidents in a county for a given year. We consolidated the annual number of female prostitution arrests cases from the FBI s National Incident-Based Reporting System (NIBRS) dataset under Uniform Crime Reports (UCR), which is a nationwide, cooperative statistical effort of nearly 18,000 federal law enforcement agencies from cities, universities and colleges, counties, and states reporting on crimes brought to their attention. Given the consistency in crime reporting standards established over the years across various enforcement agencies, the crimes reported in the UCR achieves high levels of accuracy and reliability. Using pre-entry prostitution figures as a counterfactual against post-entry prostitution trends, across treated and control counties, we are able to infer whether site entry has an impact on prostitution trends. Based on the TER data, the average count of prostitutes with reviews is 0.99 cases per county per year, while the FBI data suggests 0.21 cases per county per year for prostitutes arrested indoor. However, we also note that the prostitution incidence is highly skewed across locations (i.e., standard deviation of 14.1 for TER data and 25.0 for FBI data). The largest count of reviewed sex workers was 601 from Miami, Broward 5 We used sex worker s phone numbers to identify county information by mapping phone number s NPA-NXX area codes with county FIPS. 6 We measured indoor prostitution when location of arrests was mentioned as Hotel/Motel or Residence/Home in FBI s NIBRS. Thirty Seventh International Conference on Information Systems, Dublin

7 County, Florida in 2006, yet the largest count of female prostitutes arrested indoors was 178 as reported in Memphis, Shelby County, Tennessee in Our main independent variable is Craigslist Entry iy, which is a dummy variable indicating the entry of Craigslist in county i in year y. To examine the entry timing of Craigslist into a location, we collected data on the years in which new sites are launched through Craigslist s website. 7 In Figure 1, we can witness a staggered entry pattern of Craigslist site into different counties, and find a large degree of variation in entry years, with 2006 having maximum cases of site entry. Finally, in Figure 2, we report average prostitution incidence for TER active years dataset. We find that counties with Craigslist showcase an increasing trend in prostitution, and is significantly greater than counties lacking Craigslist. Figure 1. Distribution of Craigslist s Year of Entry Figure 2. Yearly Trend of Average Prostitution Incidence (For Counties with & without Craigslist) To account for potential confounding factors that influence prostitution and Craigslist entry in each county, we include demographic characteristics, socioeconomic factors, and crime-related trends from various sources as control variables. The U.S. Census Bureau provides annual county-level information on agegroup proportions, ethnicity proportions, population size, poverty, employment proportions, and annual 7 Collected from Accessed on 18 June Thirty Seventh International Conference on Information Systems, Dublin

8 income. FBI s databases provide county-level information on various crime trends for each year, and annual number of police employees. For analysis purposes, we created the population proportion of the age group 15-19, 20-39, and 40-59, which captures the proportion of sexually active individuals in a location. Arguably, the proportion of sexually active individuals can influence the demand and supply of sexual services. We also control for the ethnicity proportion of White, African American and Asian. We also include population size, as a covariate to account for potential increase in demand and supply of sexual services due to availability of more people at a county. Further, to control for the general economic growth in a county, we rely on proportion of employment and annual income. The economic growth of a county brings more disposable income per capita, which leads to greater consumption of goods and services, and hence a subsequent demand for entertainment with vice industry and prostitution. In addition, the lack of employment can steer individuals towards prostitution for financial gains, especially for youth (Eisenberg and Lazarsfeld 1938). Data on individuals below poverty level is obtained from the U.S. Census Bureau. The Census Bureau constructs poverty data via yearly surveys on the level of poverty at each county. The Annual Social and Economic Supplement (ASEC) to the Current Population Survey (CPS) is the official source of national poverty estimates. The yearly poverty data is used to control for the fact that individuals may enter into prostitution for financial reasons (Monroe 2005). In addition, the FBI database Law Enforcement Officers Killed and Assaulted (LEOKA) provides information on the annual number of police employees at each county. We control for number of police employees for each location as police availability serves as a deterrent against prostitution activities (Weitzer 1999). Further, we tabulated the total number of all crimes in each county using data from the FBI s UCR, to control for the intensity of crime outbreak for each location. Extant work has found that non-financial incentives of joining the online sex industry include the need to feel different, to be validated, to be in control, and as an alternative to self-harming (Jonsson et al. 2015), which are inextricably related to adverse life experiences and domestic violence. We account for these factors by controlling for the prevalence of runaways and offenses made against family recorded under the FBI s UCR. In addition, we extracted the individual count of commercialized vice, drinking under influence, and vandalism to be used in various tests. We constructed all log transformed variables using log(x+1) to include zero values. Table 1 provides the descriptive statistics of our data. Table 1. Summary Statistics Variables Obs. Mean Std. Dev. Min Max Craigslist Entry Log(TER Prostitution (All Years)) Log(TER Prostitution (Active Years)) Log(TER Prostitution (All Years; using Craigslist)) Log(TER Prostitution (Active Years; using Craigslist)) Log (FBI Female Prostitution Arrests) Log (Commercial Vice) Log (Drinking Under Influence) Log (Vandalism) Age Proportion Age Proportion Age Proportion White Proportion Black Proportion Asian Proportion Log (Population Size) Log (Poverty) Employed Proportion Log (Income) Log (Police Officers) Log (Offense Against Family) Log (Runaway) Thirty Seventh International Conference on Information Systems, Dublin

9 Empirical Methodology Main Analysis The expansion of Craigslist into different counties over time represents a natural experiment setup that allows the comparison of the difference in prostitution incidence before and after site entry for counties with Craigslist to the same difference in prostitution for counties that has yet to have Craigslist sites. Craigslist s entry is likely to be exogenous for a few reasons. The site did not charge users to post or read prostitution related ads within the study period, which meant that the business of Craigslist is not driven by profits from ads in the erotic/adults service section (Stannard 2008). Thus, the entry decision of Craigslist is unlikely correlated with the prevailing demand trends for paid sexual services in these locations. Furthermore, Craigslist as a classified advertisement portal had multiple ad sections including jobs services, housing services and for sale services. In fact, Craigslist draws its revenue only from the jobs section in six major US cities and the brokered apartment section in New York City. We exploit the variation in Craigslist s entry across counties and years in the natural experiment as the basis for identifying entry effects on prostitution trends. To test for site entry effects, we estimate a difference-in-difference regression of the form: ln(prostitution iy ) = α i + β y + g γ iy + p Craigslist iy + e iy, (1) where i indicates counties and y refers to year, y = 1999,, 2008; Prostitution iy is the number of prostitutes with reviews or arrests for county i in year y; α i is a vector of 1,796 county fixed effects; β y is a vector of year fixed effects; γ iy is a vector of county-year demographics features, socioeconomic indicators, and crime-related factors, including age proportion, racial proportion, population size, employment proportion, income level, poverty level, number of police employees, and the general crime trends; Craigslist iy is the binary indicator for Craigslist entry, i.e., Craigslist iy = 1 if the county has Craigslist in a particular year, and equals zero otherwise; and e iy is the error term. In the above specification, the coefficient p is the difference-in-difference estimate of the effect of Craigslist s entry on the incidence of prostitution. If p > 0, then site entry is linked to an increase in prostitution trends. The county level fixed effects control for time-invariant differences, while year fixed effects control for macroeconomic shocks. These fixed effects help to compare prostitution incidence in counties across time periods. Since Craigslist enters cities, the use of counties as the unit of analyses is appropriate, as it avoids capturing effect of Craigslist across multiple locations. Further, to account for serial correlation in the data, which can exists due to mismatches between unit of observation and level of variation, we clustered the error terms at the county level (Bertrand et al. 2004). Finally, we tackle heteroscedasticity through population by weighting our regressions by population size. On top of our main analysis, we conducted several checks to assess the robustness of the results to different model assumptions and specifications. First, we use unweighted county and time fixed effect regressions to examine whether the main effect might be affected without population weights. Second, to allow for a more comparable set of counties to be used in our difference-in-difference setup, we restrict our analysis to a subsample of counties which have at least one count of prostitution taking place during the study period. Third, to further derive a more comparable set of counties for analysis, we rely on propensity score matching to arrive at a set of counties that are statistically similar in terms of the demographic, socioeconomic and crime indicators. Falsification Checks To assess the possibility of the entry variable picking up significant effects spuriously, we run two falsification tests. First, we repeat our analysis using alternative crimes as our dependent variables. For this purpose, we picked crimes that are unlikely to bear any relationship with the entry of Craigslist and prostitution trends. Under this test, we expect the entry coefficients not to pick up any effect in the falsification regressions involving such crimes. In executing this set of falsification checks, we choose driving under influence and vandalism as alternative dependent variables because Craigslist is unlikely to be involved facilitating these crimes. Second, the relationship between Craigslist s entry and prostitution may be driven by unobserved confounding factors that are already present prior to the actual entry of Craigslist. We assess the possibility Thirty Seventh International Conference on Information Systems, Dublin

10 of pre-entry effects that may induce an increase in prostitution trends regardless of Craigslist s presence. To perform this check, we include three years of pre-entry indicators as placebos along with four years of post-entry indicators with the year of entry as reference point to capture potential inter-temporal entry effects as follows: j ln(prostitution iy ) = α i + β y + g γ iy + j p j Craigslist iy + e iy, (2) where j { 3, 2, 1, 1, 2, 3, 4}, indicating whether year y is the j th year since Craigslist s entry in county i. In the presence of a pre-entry effect, the placebo indicators would produce positive and significant coefficients. In addition, the coefficients of the post-entry indicators would reflect the change in effect sizes over time. Results Main Analysis Table 2 presents the main results for our empirical analysis. In the basic specification where only demographic, socio-economic and crime related controls are captured (Model 1), we see that the binary entry variable yields a positive and significant coefficient. This estimate on average represents a percentage increase in prostitution trends attributed to Craigslist s entry. 8 Our county and time fixed effect regression results are shown in Models 4 to 8. The county fixed effect regression in Model 2 continues to show that the binary entry variable yields a positive and significant coefficient with estimate representing a percentage increase in prostitution trends. With a year fixed effects in Model 3, the coefficient for site entry is also positive and significant. Model 4 reports the result with both county and time fixed effects, and it is observed that the site entry coefficient is positive and significant. Here, we get an effect size of percentage increase in prostitution trends with TER all years data. For TER active years data, we find percentage increase due to Craigslist entry, and percentage increase for the using Craigslist subgroup. Finally, in Model 8 we get 7.68 percentage increase in female prostitution arrests due to Craigslist entry. Table 2. OLS and Fixed Effects Regressions Showing Impact of Craigslist Entry on Prostitution TER TER TER Prostitution Prostitution Prostitution TER Prostitution (All Years) (All Years; (Active (Active using Years; using Years) Craigslist) Craigslist) FBI Female Prostitution Arrests OLS FE FE FE FE FE Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Craigslist Entry 0.518*** 0.406*** 0.515*** 0.392*** 0.237*** 0.199*** 0.162*** 0.074*** (0.07) (0.05) (0.07) (0.05) (0.03) (0.03) (0.03) (0.02) Age Proportion *** 7.787*** *** *** 7.103*** 6.828*** 5.879*** (0.90) (1.58) (0.91) (1.56) (1.18) (1.17) (1.03) (0.46) Age Proportion *** *** ** 0.843** (0.41) (0.81) (0.42) (0.80) (0.61) (0.46) (0.40) (0.24) Age Proportion * 2.444*** ** * 1.457*** 1.268*** (0.56) (0.83) (0.57) (0.85) (0.63) (0.50) (0.43) (0.20) White Proportion *** *** *** *** *** ** (0.13) (2.35) (0.14) (2.33) (1.73) (1.40) (1.23) (0.58) Black Proportion 0.317** ** 0.268* * (0.15) (2.20) (0.15) (2.21) (1.63) (1.29) (1.13) (0.51) Asian Proportion 6.480*** *** (2.44) (5.16) (2.40) (5.29) (3.99) (3.52) (3.11) (1.01) Log (Population Size) *** *** 0.886*** 0.433*** 0.378*** 0.109* (0.06) (0.26) (0.06) (0.25) (0.19) (0.14) (0.12) (0.07) 8 In interpreting the effect sizes, we used the technique mentioned in Lin et al. (2013). Accordingly, 1 unit increase in Craigslist entry, i.e. from 0 to 1, prostitution cases increase by the percentage 100(e^β-1), where β is the coefficient of Craigslist entry and e equals Thirty Seventh International Conference on Information Systems, Dublin

11 Log (Poverty) 0.205*** 0.269*** 0.145** 0.256*** 0.165*** ** (0.05) (0.06) (0.06) (0.06) (0.04) (0.03) (0.03) (0.03) Employed Proportion 1.326*** *** 1.835*** *** *** *** *** * (0.47) (0.46) (0.52) (0.49) (0.39) (0.28) (0.25) (0.21) Log(Annual Income) 0.338*** * * (0.11) (0.14) (0.12) (0.06) (0.04) (0.04) (0.03) (0.03) Log (Police Officers) 0.079*** *** ** (0.02) (0.03) (0.02) (0.03) (0.02) (0.02) (0.02) (0.01) Log (Offense Against Family) * (0.01) (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.01) Log (Runaway) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.00) (0.01) County Fixed Effect Time Fixed Effect R-Squared F-Stats Observations The dependent variables for Model 1 to 8 are the log number of prostitution cases. Robust standard errors clustered by counties are reported in parentheses below coefficient values. All models have binary entry regressor named Craigslist Entry. * p < 0.10, ** p < 0.05, *** p < Results from the robustness models are reported in Table 3. 9 In Model 1, under the assumption of homoscedasticity, the un-weighted regression model produces positive and significant coefficients. Further, to get comparable treatment and control observations, we consider counties that show a non-zero count of prostitution in Model 2. Results from the use of the sub-sample produce positive and significant effect of site entry for both TER and FBI data. Finally, in Model 3 after using a propensity score matching approach, we find that the positive relationship between Craigslist s entry and prostitution persists across all data sets, with an effect size of percent for TER all years data. Table 3. Robustness Checks Un-Weighted Regression Prostitution Only Counties Propensity Score Matching Variables Model 1 Model 2 Model 3 Craigslist Entry: TER Prostitution (All Years) 0.341*** 0.589*** 0.146*** (0.04) (0.11) (0.04) TER Prostitution (Active Years) 0.204*** 0.377*** 0.066*** (0.03) (0.09) (0.02) TER Prostitution (All Years; using Craigslist) 0.170*** 0.220*** 0.055*** (0.02) (0.07) (0.02) TER Prostitution (Active Years; using Craigslist) 0.137*** 0.177*** 0.030* (0.02) (0.06) (0.02) FBI Female Prostitution Arrests 0.063*** 0.142* 0.030* (0.02) (0.09) (0.02) Observations The dependent variables for Model 1 to 3 are the log number of prostitution cases. Robust standard errors clustered by counties are reported in parentheses below coefficient values. An alternative sub-sample based on positive prostitution count in a county is used in Model 2. Model 3 uses propensity score matching with caliper size of All models have binary entry regressor named Craigslist Entry. All control variables reported in Table 2 are included. * p < 0.10, ** p < 0.05, *** p < For the tables, hereafter we only report the coefficients for Craigslist Entry variable to keep tables concise. All tables include full set of control variables in the actual regressions. Thirty Seventh International Conference on Information Systems, Dublin

12 Falsification Results In our first falsification result where we examine the impact of Craigslist on non-related crimes, 10 we observe that binary entry variables do not hold statistically significant relationships with the incidence of drinking under influence and vandalism. This test provides supporting evidence in favor of a noncoincidental effect captured by the Craigslist entry variable in the main regressions in Table 4. Table 4. Falsification Test Using Alternative Crimes Driving Under the Influence Vandalism Variables Model 1 Model 2 Craigslist Entry (0.03) (0.03) R-Squared F-Stats Observations The dependent variables for Model 1 and 2 are the log number of DUI and Vandalism, respectively. Robust standard errors clustered by counties are reported in parentheses below coefficient values. All models have binary entry regressor named Craigslist Entry. All control variables reported in Table 2 are included. * p < 0.10, ** p < 0.05, *** p < In our second falsification test, we examine whether prostitution incidence are already increasing in the time periods prior to Craigslist entry. Results of this falsification test are reported in Table 5. For this result, we report TER prostitution dataset for using Craigslist sub set. Across all models, we observe that the three pre-entry placebo variables did not pick up any pre-entry effect, indicating that the positive relationship between prostitution trends and Craigslist entry observed in the main analyses is unlikely to be an artifact from events that occur in periods prior to Craigslist s entry. In addition, across all models, the entry indicators for post entry periods are all positive and significant. Specifically for Model 2, we find that the post-entry coefficients hold effect sizes ranging from to percentage increase in prostitution trends. Consistent across all models, the Craigslist estimates are generally increasing over time. Such a result is expected as the intermediating impact of the Craigslist platform is likely to grow as more users became aware of the site which would in turn attract further users to participate in the Craigslist-enabled prostitution market. Table 5. Falsification Test Using Pre and Post Entry Craigslist Indicators TER Prostitution (All Years; using Craigslist) TER Prostitution (Active Years; using Craigslist) Variables Model 1 Model 2 Craigslist Entry (0.02) (0.01) Craigslist Entry (0.02) (0.02) Craigslist Entry (0.03) (0.02) Craigslist Entry *** 0.118*** (0.03) (0.03) Craigslist Entry *** 0.194*** (0.04) (0.04) Craigslist Entry *** 0.527*** (0.11) (0.10) Craigslist Entry *** 1.130*** (0.30) (0.27) R-Squared F-Stats Observations Result table for the first falsification is available with authors on request. Thirty Seventh International Conference on Information Systems, Dublin

13 The dependent variables for Model 1 and 2 are the log number of prostitution cases. Robust standard errors clustered by counties are reported in parentheses below coefficient values. All models have binary entry regressors named Craigslist Entry±i, where i (-i) is i th year lead (lag) from year of entry. All control variables reported in Table 2 are included. * p < 0.10, ** p < 0.05, *** p < Potential Mechanisms and Related Trends From the analyses thus far, results suggest that the entry of Craigslist has an effect of increasing the incidence of prostitution. However, it is less clear what underlying mechanisms are driving this positive relationship. To understand the mechanisms behind the rise in prostitution levels due to Craigslist s entry, we next examine four specific questions related to the market for sex workers induced by site entry. Relationship with Commercial Vice It is important for policy makers and enforcement agencies to understand whether organized vice groups are involved in the prostitution activities on Craigslist. The legal considerations and guidelines for curbing prostitution operated by organized rings can differ qualitatively from those for curbing Craigslist-related prostitution that are a result of independent sex workers who enter the sex trade voluntarily. In the former scenario, commercialized vice groups are likely to capitalize the online channel as a new avenue to peddle prostitutes to clients, while the latter involve new market entrants who begin providing paid sexual services after the launch of Craigslist. While extant work reports that a majority of offline sex workers are associated with organized prostitution rings (Farley et al. 2013, Helfgott 2008, Faugier and Sargeant 1997), the current literature does not provide a clear view of the market structure of online prostitution enabled by platforms. On the one hand, it may be argued that a majority of the solicitation ads are posted by willing adult participants who are not part of organized vice groups, in the attempt to capitalize on the financial gains of providing sexual services in this market. On the other hand, McAllister (2011) indicates that organized prostitution rings are making use of the Internet as a portal to solicit customers. To address this dichotomy in possibilities, we assess and compare the impact of commercial vice activity on prostitution with Craigslist s impact on prostitution. To do so, we regress prostitution on both commercial vice activity and Craigslist s entry with the same set of fixed effects in Model 1. And, in Model 2, we add the interaction term of commercial vice and Craigslist entry, along with controls and fixed effects. In the full model that includes the interaction term, the coefficient of Craigslist entry indicates that the impact of the site in increasing prostitution trends in locations where organized vice activities are low or absent. This estimate is likely to capture the size of the market operated by independent sex workers. A significant and positive interaction term indicates that the increase in prostitution trends is observed in areas with Craigslist presence and high commercial vice activity, which is suggestive of a likely involvement of organized vice groups in the Craigslist-enabled prostitution market. In conducting this analysis, we rely on the data from FBI s UCR database which tracks the incidence of organized prostitution activities arrests. Table 6 reports the results of this analysis. In Models 1a and 2a, we find that the Craigslist entry is positive and significant. More importantly, we note that the interaction term and for the Craigslist variable are both positively significant in Models 1b and 2b. Taken jointly, these estimates suggest that the prostitution market enabled by Craigslist is made up of both independent sex workers and commercial vice organizations. A comparison of the coefficient sizes suggest that the prostitution by commercial vice groups are at least two times larger than that made up by independent providers. Table 6. Craigslist and Commercial Vice on Prostitution TER Prostitution (All Years; using Craigslist) TER Prostitution (Active Years; using Craigslist) Variables Model 1a Model 1b Model 2a Model 2b Log(Commercial Vice) * * (0.05) (0.04) (0.05) (0.04) Craigslist Entry 0.199*** 0.079*** 0.162*** 0.059** (0.03) (0.03) (0.03) (0.02) Log(Commercial Vice) * Craigslist Entry 0.201*** 0.172*** (0.02) (0.02) Thirty Seventh International Conference on Information Systems, Dublin