Drawing a Probability Sample of Female Street Prostitutes in Los Angeles County.

Female prostitutes are known to play an important role in the epidemiology of certain sexually transmitted diseases (STDs), both because of the sexual activity involved in their work and because some female prostitutes inject drugs or have sex with men who inject drugs (Darrow, 1990; Miller, Turner, & Moses, 1990; Plummer & Ngugi, 1990). Nevertheless, little of what is known about the size of this population or their risk behaviors has been derived from careful scientific study. Most studies of prostitutes rely on samples of convenience, typically recruiting in jails, STD clinics, and methadone maintenance programs (Astemborski, Vlahov, Warren, Solomon, & Nelson, 1994; Cohen, Alexander, & Wofsy, 1988; Gellert, Maxwell, Higgins, Pendergast, & Wilker, 1993; Khabbaz et al., 1990). A few studies also include outreach recruitment of respondents in areas known for street prostitution (Cohen et al., 1988; Khabbaz et al., 1990; Tabet, Palmer, Wiese, Voorhees, & Pathak, 1992). The biases in these convenience samples are unknown.

The usual way to minimize sampling bias is through the use of probability sampling techniques. However, the nature of commercial sex work makes that approach especially difficult. Because prostitution is an illicit activity, registries or rosters of prostitutes are not available, and screening of persons in the general population who are willing to admit to such activity is inefficient and unlikely to yield satisfactory coverage of the target population. Recruitment of subpopulations found in convenient locations, such as jails and methadone maintenance clinics, may not be representative of the larger population.

Prostitutes can make themselves available to their clients in various ways, including being present in locations where clients are likely to seek them out or notice them. When these locations are public places, prostitutes are also potentially available for sampling by researchers. One way to conduct probability sampling of prostitutes is to take a two-stage sampling approach, in which the first stage consists of sampling locations and time slots from a spatial-temporal sampling frame, and the second stage involves screening and sampling eligible persons at locations and time slots sampled in the first stage. This general approach has been used with success to draw probability samples of other difficult-to-reach, stigmatized populations, such as homeless persons (Koegel, Burnam, & Morton, 1996; Rossi, Fisher, & Willis, 1986). Applying the approach to sample prostitutes extends the method to an illicit population. One obvious challenge is the possible reluctance of prostitutes to acknowledge engaging in illicit activity, a reluctance that could limit the effectiveness of screening to identify cases. Another is the pronounced spatial mobility of the street population in response to law enforcement activities and fluctuations in client availability.

The Los Angeles Women's Health Risk Study (LAWHRS) is the first study of its kind to attempt to draw a probability sample of the prostitute population in a single major metropolitan area. Our goal was to sample prostitutes in a way that permits empirically based estimation of important population characteristics. Separate sampling frames were constructed for street and offstreet segments of the population; however, probability sampling proved infeasible for the offstreet prostitutes because of the difficulty of defining sampling frames for women who solicited clients via advertisement or referrals and our restricted access to those who solicit clients in private places associated with prostitution.

In contrast, we were successful in defining the sampling frame and in contacting and interviewing street prostitutes because they were accessible and could be recruited while they were soliciting clients in public places. This paper describes the methods that were used to construct the sampling frame and to sample and conduct field interviews with street prostitutes. It also describes the field disposition of the street sample. Preliminary data on the relationship between HIV antibody status, prostitute characteristics, and risk and preventive behaviors can be found in Berry, Kanouse, Duan, and Lillard (1992) and in Kanouse, Berry, Duan, Richwald, and Yano (1992). We have also compared the characteristics of the probability sample of street prostitutes with those in convenience samples recruited in jails, drug treatment programs, and STD clinics (Berry, Duan, & Kanouse, 1996).

METHOD

We constructed a sampling frame by systematically interviewing informants knowledgeable about street prostitution in various parts of Los Angeles County. Our primary sampling unit was the "area-day-shift," where area refers to the place of solicitation, day to the day of the week, and shift to a six-hour time period within the day. We randomly sampled area-day-shifts every week for 36 weeks and sent field teams of interviewers, drivers, and phlebotomists to these locations at the selected times. In each area, the field staff randomly selected women on the street, screened them for study eligibility, conducted interviews, and took blood samples. Field staff also conducted an enumeration of all women who could have been approached for screening in that area-day-shift.

Fieldwork on the street segment began in May 1990 and was completed in February 1991. A probability sample of 998 street prostitutes was recruited and interviewed. Because of legal uncertainties, blood sampling did not begin until September 1990. Blood samples were obtained from 638 women interviewed, representing 85% of those from whom a sample was sought. (There was a 4% refusal rate, and 11% of the attempts were unsuccessful because of failure to locate a vein.)

Sampling Frame

Our target population consisted of women soliciting customers on the streets of Los Angeles County, a 4,000 square mile area. We used a spatial-temporal sampling frame (such as that used to sample museum visitors or shoppers in a mall) to define places and times when solicitation was likely to take place during the period of study. We narrowed the frame to areas (street segments and adjacent side streets, parking lots, etc.) where street solicitation was known to occur.

We began by compiling an extensive list of possible areas of street prostitution activity, identified in two ways. First, we interviewed a broad range of expert informants, including officers, sergeants, and lieutenants in divisional vice in all 18 Los Angeles City Police Department precincts, persons at all ranks from officers to captains at each of the county's 37 other municipal police departments and 17 county sheriff stations. We also interviewed sexually transmitted disease (STD) field investigators in each of the 23 districts served by the Los Angeles County Department of Health Services and two independent health departments serving the cities of Long Beach, Pasadena, and Vernon. Ethnographic informants, including outreach workers, current and former prostitutes, and persons familiar with minorities or other subcultures and institutions, were asked to identify areas (street segments) where prostitution activity occurs and to estimate the extent of this activity (number of women) at various times of day. Altogether, we conducted over 200 interviews with expert informants and ethnographic consultants for this purpose. Second, using licensing records, we mapped the locations of marker establishments for adult entertainment or gambling that might attract prostitution activity in the surrounding area. Marker establishments included gambling clubs, hostess dance halls, adult bookstores, adult theaters and arcades, and strip clubs and topless bars licensed as "adult entertainment" under city or county ordinances.

We compiled information from these sources into a data file of possible street segments. For the most part, informants drawn from different backgrounds (law enforcement, health departments, and ethnographic sources) tended to agree on the general locations of prostitution activity. We defined street segments as the street and sidewalks along the street between two cross streets. A segment was usually several blocks long and included side alleys. Because law enforcement sources were the most comprehensive in identifying street segments, we used their reports as our basic source, supplemented with information from STD field investigators and other sources where available. We made two or more independent on-site visits to observe the level of prostitution activity at each segment. Quantitative data from these various sources were combined using a weighted average that gave lesser weight to higher estimates, to reflect our experience in studies of other populations that expert informants tend to overestimate population size.

The timing of prostitution activity varied across areas. Some areas were most active in the evenings, others during the lunch hour, and so forth. Activity also varied by day of the week. We divided each day of the week into four six-hour shifts (5 a.m. to 11 a.m., 11 a.m. to 5 p.m., 5 p.m. to 11 p.m., and the late-night shift, 11 p.m. to 5 a.m.). We estimated probable levels of prostitution activity for each area-day-shift based on informants' reports and the results of our own inspection. Low activity was defined as fewer than two sex workers per area-day-shift, medium activity as two to six workers, and high activity as more than six workers. We then constructed a sampling frame composed of area-day-shifts where and when prostitution activity occurs. We sampled area-day-shifts from this frame with a probability proportional to size, where size was the estimated level of prostitution activity in each area-day-shift "cell." Area-day-shifts with low levels of estimated activity ([is less than] 2 women by weighted average) were not sampled.

Our estimates of the level of prostitution activity were updated every two weeks during the 36-week field period to reflect actual field experience and new information provided by informants, which was combined with previous information but weighted more heavily because it was more recent. We also periodically canvassed informants and continually canvassed respondents to identify new sample areas. The dynamic nature of the frame was represented in the sampling frame itself by dividing the study period into 38 assignment periods. We maintained transaction journals to keep track of all modifications and updates to the sampling frame by assignment period. Additions during the course of the study mainly extended or modified the boundaries of known locations, but did not add new locations.

Selection and Screening of Women

The geographic layout of sample areas allowed field staff to cruise each area by proceeding in a loop--for example, by starting on one side of the street, proceeding to the end of the street segment defining that sample area, then turning around and returning on the other side of the street. To sample women randomly, we first chose a random starting point within the segment (either the middle or one of the ends of a street) on a randomly chosen side of the street and randomly selected a direction in which to cruise (clockwise or counterclockwise). We then approached the first woman seen. If this first encounter did not lead to an interview, the field team would proceed in the same direction until they encountered another potential subject. The procedure was repeated until the field team had obtained an interview, or until they reached the starting point without obtaining an interview. After obtaining an interview, the field team continued to cruise the rest of the area to complete enumerating the women present, without attempting to approach or interview them.

The interviewer introduced herself and said she was from RAND. She then started the interview:

   We're talking to working girls about their health and possible risks to
   their health like AIDS. We're not from the police and we can pay you for an
   interview. Could you answer some questions for me now? Have you traded any
   kind of sex, including sex talk or B and D (bondage and discipline) for
   money or drugs or anything else of value in the past 12 months? (If yes):
   Have you been interviewed already by the Los Angeles Women's Health Risk
   Study?

Respondents who acknowledged eligibility and said that they had not previously been interviewed were given an informed consent form explaining that participation in the study involved (a) taking part in a 45-minute interview about the way they work, their life experiences, and their health, and (b) providing a blood sample to be tested for exposure to the AIDS virus, syphilis, and hepatitis B. They were told that they would not be asked to disclose their name, address, or other identifying information. To preserve the respondent's anonymity, the informed consent form was signed by the interviewer, who certified that the respondent had read the form or had it read to her or that the interviewer had discussed all points on the consent form with the respondent.

The blood test was voluntary; women who declined the blood test but completed the interview were paid the same $25 for participation. The opportunity to receive a high-quality, free, anonymous blood test was used as one of the incentives to participate, although as part of the informed consent procedure women were also provided with specific information on alternative sites where they could obtain free anonymous testing.

Women were counseled about the personal health benefits of knowing their serostatus and encouraged to get their blood tested, whether the testing was done through the study or elsewhere.

Enumeration of Sample Area

To provide a basis for estimating sampling probabilities, we enumerated the women available for sampling in each area-day-shift where we sought interviews. We enumerated most women on the street, except for those that were obviously not prostitutes or not working at the time (e.g., women carrying groceries or pushing strollers). We preferred to rely on self-screening rather than interviewer judgment in determining potential eligibility for the study; however some women could not, in the interviewer's judgment, be safely or discreetly approached, such as women in the midst of a large group of men. Some women refused screening or denied eligibility for the study. Interviewers were asked, however, to record their judgments on a scale estimating the likelihood that a sampled woman who refused screening or denied being eligible was actually a prostitute. Demographic characteristics were also recorded for all women who were screened.

A completed interview cycle consisted of an interview and an enumeration of the women in the sample area. If no interview was completed during a tour of the sample area, the enumeration simply represented a count of the number of women seen in the sample area during that interview cycle. Field staff were usually able to complete three or four interview cycles during a six-hour shift.

Interview and Blood Sample

Interviews were conducted by trained interviewers working in teams. About two thirds were experienced STD outreach workers, and the remaining third were from various backgrounds. Although we used both male and female interviewers, most interviewer teams consisted of two women. Interviews were conducted in various locations on or accessible from the street, including bus stop or park benches, fast food restaurants, laundromats, parking lots, or interviewers' cars. Unlike some studies of street prostitutes in other cities, we did not routinely use a study van for conducting interviews and collecting blood samples. Working out of backpacks made interviewers less identifiable and therefore less vulnerable as carriers of cash and needles.

We obtained blood samples from most women interviewed during a portion of the study. Blood samples were drawn by trained phlebotomists using procedures especially adapted to difficult field situations. Study participants received pretest counseling regarding the blood test after they had completed the interview. Those who provided a blood sample could obtain their test results by calling RAND to arrange for an appointment. Test results and post-test counseling were given in person and not made available by telephone. Results were stored and retrieved using a personalized identification code constructed at the time of the interview, which was reproducible if lost.

RESULTS

Sample Coverage

Of 164 potential street areas named by informants or associated with marker establishments, 111 were judged on the basis of informant estimates of activity and pre-fieldwork enumeration visits to have sufficient prostitution activity for inclusion in the study. Of the remaining 53 areas, 51 were judged to have little or no activity (based on informant estimates and enumeration visits) making it cost-ineffective to study them, and two were low-activity areas in remote parts of the county that were not included for reasons of access and cost.

We successfully completed interviews in 79 of the 111 areas included in the study. In the remaining 32 areas, once in the field we found virtually no prostitution activity. In 120 visits to these 32 areas, we identified a total of 65 potentially eligible women (mean = 0.5 per visit) and approached 26 for screening (0.2 per visit). Although we did not sample day shifts in which we expected (by weighted average) to find fewer than two eligible women per visit, many areas that turned out to have low activity were sampled early in the study based on higher weighted estimates. In 419 interview visits to 57 areas with fewer than 2.0 potentially eligible women per interview visit, we identified 272 potentially eligible women, approached 186 women for screening (68 percent), and completed interviews with 90 women (48 percent of those approached). This completion percentage is lower than in the sample as a whole (61 percent of those approached), and the number of potentially eligible women per visit is considerably lower (0.6 vs. 3.7). Altogether, our interview teams made 1,033 visits to the sample areas, for a yield of about 1.0 interview per visit. Approximately two to six teams operated each day. While there was continual law enforcement activity, there was no extraordinary activity affecting the operation of street prostitution during the period of the study.

Use of informants proved to be by far the superior strategy for identifying areas of prostitution activity. Of the 79 sample areas where we were able to complete interviews, only three were identifiable solely by the presence of marker establishments; collectively, these three areas accounted for only 1 percent of all interviews completed. In contrast, 54 areas were identified solely by informants; collectively, these areas accounted for 67 percent of all interviews completed.

Although we made an effort to sample active areas at various shifts, not all shifts were sampled in each area. Many areas of street prostitution in Los Angeles County also tend to be associated with drug dealing and gang activity. Consequently, certain sample areas and the late-night shift were initially assigned a selection probability of zero because of safety concerns. However, we did make a total of 141 enumeration-only visits in these areas during the late-night shift and carried out limited interviewing: 90 interviews were completed in high-risk areas and 11 were completed during the late-night shift using double teams for increased safety.

Interviews With Street Sample

As Figure 1 shows, results of the street field operation were quite successful. Although for safety reasons we were not able to approach about a third of the women who were selected, the vast majority of the women we approached answered the screening question. More than three quarters acknowledged being eligible for the study, and we were able to complete an interview with 89 percent of those eligible. A major question of interest is the completion rate--that is, of all women in the sample area who met our eligibility criteria and were approached, what percentage actually completed an interview? Our data do not permit a precise answer but do permit us to set bounds. A lower bound is provided by assuming that all women who refused screening or said that they were ineligible were simply refusing to participate. In that case, the completion rate is 998 interviews out of 1,629 eligible women approached for the first time (see Figure 1), or 61 percent. An upper bound may be set by assuming that all women who refused screening or denied eligibility were in fact ineligible, in which case the completion rate is 89 percent (998 women interviewed/women of the 1,119 screened as eligible). The completion rate probably lies closer to the lower than to the upper bound. Interviewers judged that about 3% of the women who denied being eligible were almost certainly working, 56% were probably working, and only 8% were probably not working. The remaining 34% fell in an uncertain middle category, could be working--can't rule it out. Women who claimed they were ineligible or who refused to be interviewed were more likely to be older, Black as opposed to White or Hispanic, and more often appeared to be homeless. We encountered very few pimps during sampling, so they were not a factor in refusals.

[Figure 1 ILLUSTRATION OMITTED]

Characteristics of Respondents

As Table 1 shows, about half of the women we interviewed were over age 30, and very few admitted being under age 18. This was not because of any age restrictions on sampling. Because of the special nature of this survey and its target population, we sought and obtained approval from both RAND's Institutional Review Board and that of the Department of Health and Human Services to waive requirements for parental consent of potential respondents under age 18, who can be considered emancipated minors. Thus, we were prepared to interview minors but found few of them on the streets to interview, based on interviewer observation of the apparent age of women screened and respondent self-reports. Los Angeles has several effective outreach programs whose goal is to keep runaway teenagers from turning to prostitution for survival.

Table 1. Characteristics of Respondents
                                              Percentage of
Characteristics                                Respondents

Age
  18 or younger                                      2
  19-25                                             22
  26-30                                             28
  31-40                                             39
  41 or over                                         8
Race/Ethnicity
  White                                             17
  Black                                             69
  Hispanic                                           9
  Asian                                             <1
  Mixed/other                                        4
Other Characteristics
  Ever arrested for prostitution offenses           68
  Ever in drug detoxification                       29
  Ever visited STD clinic                           46

Note. Based on respondents' self-reports. Interviewers' judgments regarding approximate age and race/ethnicity of the respondent, made at the time of screening, agreed closely with respondent self-reports.

More than 80 percent of the women we interviewed were minority--mainly Black or Hispanic. This reflects the fact that most areas of street prostitution activity in Los Angeles County are in minority neighborhoods. Sixty-eight percent of the women had been arrested for prostitution-related offenses: The subset of these who were arrested in Los Angeles County during the study period would have been potentially available for study through a jail-based sample, although our interview did not inquire about recent arrest history. As Table 1 shows, fewer women would have been available through recruitment in STD clinics or drug detoxification centers. Differences in demographic characteristics and risk behavior among those potentially recruitable in these disparate settings are analyzed elsewhere (Berry et al., 1996).

Reinterviews

We estimate that approximately 84 women were interviewed more than once and that reinterviews account for some 10 percent of the 998 interviews: thus, interviews were actually completed with about 900 different women. To arrive at this estimate, we used an identified set of replicate interviews as a training set to develop a metric for measuring the "distance" between two interviews based on answers to key questions and similarity in identification codes. We applied this metric to the entire set of interviews to identify likely matches. We assumed that the distribution of similarity scores for the closest matching interview was a mixture of scores from true matches and false matches, then we applied a method developed by Belin (see Belin & Rubin, 1995) to estimate the proportion of false matches. The difference between this proportion and unity is an estimate of the proportion of women who were interviewed more than once.

DISCUSSION

Although female street prostitutes are such a small proportion of the total female population that they certainly qualify as a "rare population" (Kalton & Anderson, 1986), it is feasible to use screening methods to identify them because they work in geographically concentrated areas, where they often make up a large proportion of women on the streets. Their need to bring attention to their availability to potential clients makes it possible to ascertain by observation whether prostitution activity is occurring in a given area, although appearance and behavior do not always provide sufficient information to determine whether a particular woman is a prostitute, especially when she is not actively working.

Data from a probability sample will, when suitably weighted, support inferences about the work force of prostitute women from which the sample was drawn (Duan, Kanouse, Berry, & Carson, 1995). However, it is important to be clear about the population represented. First, women who solicit clients only in places other than the street (e.g., bars or cantinas) are not represented. Similarly, women who work through escort services or in private establishments, such as massage parlors, are excluded. Although it is possible to develop sampling frames for women working in some of these other venues, our field experience led us to conclude that it is far more difficult to contact and interview women working off the street, and obtaining a probability sample of such women proved infeasible in our study (Lever & Kanouse, 1998).

The spatial-temporal sampling frame is most suitable for studying the female prostitute work force, measured in person-time units. Because the probability of a woman being sampled for study is directly related to the time she spends soliciting clients on the street, women who work longer hours are more likely to be sampled. Such women also contribute more to the work force than do women who work fewer hours. In contrast, the headcount population of female prostitutes consists of all women who work on the street in a defined period of time, regardless of how active they are. The unit of observation for the headcount population is the person, whereas the unit of observation for the work force population is the person hour, or other unit of person-time. Because the sampling frame used in this study captures women only at times and places when they are likely to be available for work or working versus engaged in private activities, it is more easily weighted to represent the work force than to represent the headcount population. However, it can be used to make inferences about the latter when allowances are made for censoring as a result of turnover in the work force (new entrants, retirements, etc.) during a period of field observation (Duan et al., 1995).

The two principal challenges we anticipated in achieving a representative sample were difficulty in getting women to acknowledge eligibility and difficulty constructing an accurate sampling frame and maintaining it over the course of the study. Both of these challenges proved to be surmountable. Although we were not able to validate respondent self-reports of eligibility or ineligibility, 61 percent or more of eligible women agreed to participate. It is possible that some of the women we interviewed were not prostitutes and therefore not eligible for the study. We believe, however, that the number of such interviews is probably quite small, given the stigma of prostitution, the length of the interview, the collection of a blood sample, and the detailed questioning about practices with clients.

Constructing a spatial-temporal sampling frame also proved to be quite feasible. Because we sought to cover a 4,000 square mile area encompassing dozens of civil governments, use of expert informants was a highly labor intensive enterprise. Nevertheless, we found that experts with overlapping areas of jurisdiction were in substantial agreement regarding street segments and adjacent areas in which street prostitution activity occurs. Although it was necessary to update the sampling frame multiple times over the course of the study, these updates were generally based on field reports of shifts in activity from one known area to another known area rather than on new intelligence about activity in previously unidentified sample areas.

Our approach to random selection--selecting the first woman encountered while proceeding along a randomly chosen path within the sample area--was arrived at only after we had unsuccessfully tried an alternative approach of first enumerating women in the sample area and then randomly selecting one to interview. In this population, women who are not approached shortly after they are first seen are often difficult to find, because of their mobility within the sample area and their propensity to leave the scene with clients.

We found that a substantial proportion of women approached for screening are willing to acknowledge eligibility and to be interviewed, even though our field staff had limited means for establishing credibility and gaining trust. This contrasts sharply with the substantial time investment typically made in developing trust and rapport in ethnographic studies, and demonstrates that it is possible to collect detailed and sensitive behavioral data from this illicit population through survey methods.

One advantage of carrying out a study such as this one is that the systematic data collection required to identify a sampling frame can itself yield interesting insights on the overall scale and patterns of prostitution activity, the policies of regulatory and law enforcement agencies, and the interplay between the two. A major disadvantage is the substantial effort required to mount a probability sample compared with using a convenience sample. Comparison of results obtained from our overall sample with those that might have been obtained from subgroups of women who were recruitable in other venues--for example, women who had been arrested, gone to STD clinics, or been in methadone maintenance programs--suggests that some convenience samples are apt to be more biased than others. Most street prostitutes in Los Angeles County, and presumably elsewhere, have arrest histories and are potentially recruitable while in jail. A convenience sample drawn from jails would be more similar to the overall population than, say, convenience samples drawn from methadone maintenance programs or STD clinics (Berry et al., 1996). Moreover, including questions on recent work history and arrest history when collecting data on a sample of jailed women could provide a basis for weighting sample data to be more representative of all women who are ever arrested for prostitution, thereby reducing some of the bias in studies based on convenience samples.

To assure the safety of our field staff, we adopted several policies in the field that affected the field results. The most obvious example is that we did not send interview teams to area-day-shifts that we considered too dangerous, which means that women working in those areas at certain times were not represented in our study. A second example is that we did not attempt to record information on the demographic characteristics of women not approached for screening. Although we would like to know more about their characteristics, it would have been difficult to accomplish this in the field without drawing attention to the interviewers and raising concern about what was being recorded.

Several lessons emerged from this study that may help other researchers design similar studies at lower cost. First, mapping the locations of marker establishments added very little to our list of areas of street prostitution activity. Second, although we gained much information from public health investigators and ethnographic informants about prostitution in off-street locations, these sources added little to what we learned from law enforcement sources regarding street prostitution. This might not be the case in cities where anti-prostitution laws are not vigorously enforced. Third, although prostitution activity is widely dispersed throughout Los Angeles County, most of the activity is in a few highly active areas. It should therefore be possible to recruit a large fraction of a probability sample from a smaller number of areas and reduce the cost of covering numerous areas of low activity, although they should be included if the sample is intended to be representative of all areas.

In summary, field results for the street segment of the Los ,Angeles Women's Health Risk Study demonstrate that a spatial-temporal sampling frame can be constructed and field sampling techniques used to draw a probability sample of female street prostitutes. Doing so requires first constructing a comprehensive sampling frame and, second, developing workable methods for drawing a probability sample of women and recruiting them into the study. We developed a comprehensive sampling frame by interviewing representatives from all law enforcement and public health agencies in the county. Inasmuch as every area of the county lies within the jurisdiction of a law enforcement agency, it is highly unlikely that significant levels of street-based prostitution activity could be sustained in any area for long without coming to the attention of law enforcement personnel. Our informants' knowledge of areas of prostitution activity proved to be quite accurate, in that field observations generally corroborated informants' reports. To check on the completeness of our sampling frame and to update it based on field results, we asked women we interviewed to describe other areas where they solicited clients. Women named other areas in our sampling frame but did not identify any areas that we did not know about. We conclude that the sample areas identified for the study are a reasonably complete list and cover most of the street-based prostitution activity in Los Angeles County.

REFERENCES

Astemborski, J., Vlahov, D., Warren, D., Solomon, L., & Nelson, K. E. (1994). The trading of sex for drugs or money and HIV seropositivity among female intravenous drug users. American Journal of Public Health, 84, 382-387.

Belin, T. R., & Rubin, D. B. (1995). A method for calibrating false-match rates in record linkage. Journal of the American Statistical Association, 90, 694-707.

Berry, S. H., Duan, N., & Kanouse, D. E. (1996). Use of probability versus convenience samples of street prostitutes for research on sexually transmitted diseases and HIV risk behavior: How much does it matter? In R. Warnecke (Ed.), Health survey research methods (DHHS Publication No. [PHS] 96-1013, pp. 93-97). Hyattsville, MD: Centers for Disease Control and Prevention and National Center tot Health Statistics.

Berry, S. H., Kanouse, D. E., Duan, N., & Lillard, L. A. (1992). Risky and non-risky sexual transactions with clients in a Los Angeles probability sample of female street prostitutes. Paper presented at the Eighth International Conference on AIDS, Amsterdam, The Netherlands.

Cohen, J., Alexander, P., & Wofsy, C. (1988). Prostitutes and AIDS: Public policy issues. AIDS and Public Policy Journal, 3. 17-22.

Darrow, W. W., & the Centers for Disease Control Collaborative Group for the Study of HIV-1 in Selected Women. (1990). Prostitution, intravenous drug use, and HIV-1 infection in the United States. In M. A. Plant (Ed.), AIDS, drugs, and prostitution (pp. 18-41). New York: Routledge.

Duan, N., Kanouse, D. E., Berry, S. H., & Carson, S. (1995). A probability sample of street prostitutes in Los Angeles Courty. Unpublished manuscript.

Gellert, G. A., Maxwell, R. M., Higgins, K. V., Pendergast, T., & Wilker, N. (1993). HIV infection in the women's jail, Orange County, California, 1985 through 1991. American Journal of Public Health, 83, 1454-1456.

Kalton, G., & Anderson, D. W. (1986). Sampling rare populations. Journal of the Royal Statistical Society, A, 149, Part 1, 65-82.

Kanouse, D. E., Berry, S. H., Duan, N., Richwald, G., & Yano, E. M. (1992). Markers for HIV-1, hepatitis B. and syphilis in a probability, sample of street prostitutes in Los Angeles County, California. Paper presented at the Eighth International Conference on AIDS. Amsterdam, The Netherlands.

Khabbaz, R. E, Darrow, W. W., Hartley, T. M., Witte, J., Cohen, J. B., French, J., Gill, P. S., Potterat, J., Sikes, R. K., Reich, R., Kaplan, J. E., & Lairmore, M. D. (1990). Seroprevalence and risk factors for HTLV-1/II infection among female prostitutes in the United States. Journal of the American Medical Association, 263, 60-64.

Koegel, P., Burnam, A., & Morton, S. C. (1996). Enumerating homeless people: Alternative strategies and their consequences. Evaluation Reviews; 20, 378-404.

Lever, J., & Kanouse, D. E. (1998). Using qualitative methods to study the hidden world of offstreet prostitution in Los Angeles County. In J. E. Elias, V. L. Bullough, V. Elias, & G. Brewer (Eds.), Prostitution: On whores, hustlers, and johns. Amherst, NY: Prometheus Press.

Miller, H. G., Turner, C. F., & Moses, L. E. (1990). AIDS: The second decade. Washington, D.C.: National Academy Press.

Plummer, F. A., & Ngugi, E. N. (1990). Prostitutes and their clients in the epidemiology and control of sexually transmitted diseases. In K. K. Holmes, P.-A. Marsh, P. F. Sparling, P. J. Wiesner, W. Cares, Jr., S. M. Lemon, & W. E. Stamm (Eds.), Sexually transmitted diseases (2nd ed., pp. 71-76). New York: McGraw Hill.

Rossi, P. H., Fisher, G. A., & Willis, G. (1986). The condition of the homeless in Chicago. Amherst, MA: Social and Demographic Research Institute, University of Massachusetts.

Tabet, S. R., Palmer, D. L., Wiese, W. H., Voorhees, R. E., & Pathak, D. R. (1992). Seroprevalence of HIV-1 and hepatitis B and C in prostitutes in Albuquerque, New Mexico. American Journal of Public Health, 82, 1151-1154.

Manuscript accepted June 29, 1998

Supported by Grant R01 HD24897 from the National Institute of Child Health and Human Development.

Address correspondence to David E. Kanouse, Ph.D., RAND, 1700 Main Street, Santa Monica, CA 90407; e-mail: [e-mail]

David E. Kanouse, Sandra H. Berry, and Naihua Duan RAND, Santa Monica, CA

Janet Lever RAND, Santa Monica, CA and California State University, Los Angeles

Sally Carson, Judith F. Perlman, and Barbara Levitan RAND, Santa Monica, CA