Underreporting in HIV-Related High-Risk Behaviors: Comparing the Results of Multiple Data Collection Methods in a Behavioral Survey of Prisoners in Iran

Abstract

We explored the potential of using three indirect methods including crosswise, proxy respondent method, and network scale-up (NSU) in comparison with direct questioning in collecting sensitive and socially stigmatized HIV-related risk behavior information from Iranian prisoners. Participants reported more sexual contact in prison for their friends than they did for themselves. In men, NSU provided lower estimates than direct questioning, whereas in women NSU estimates were higher. Different data collection methods provide different estimates and collectively offer a more comprehensive picture of HIV-related risk behaviors in prisons.

Keywords

HIV stigma behavioral surveys data collection methods prisoners population size estimation Iran

Introduction

Behavioral data, if collected and analyzed accurately, can aid in HIV prevention efforts and prevent further transmission of HIV. Such data are regularly collected from certain key populations who are at high risk for HIV infection (e.g., injection drug users, female sex workers, and men who have sex with men; Zablotska, Kippax, Grulich, Holt, & Prestage, 2011).

In some countries, there are other subpopulations who are disproportionally affected by the HIV epidemic. In Iran, prisoners are one of those subpopulations that are considered a high-risk group for HIV infection (Haghdoost, Mirzazadeh, Shokoohi, Sedaghat, & Gouya, 2013; Navadeh et al., 2013). HIV prevalence is about 2% and only one in eight has ever been tested for HIV and only 20% have sufficient knowledge about HIV (Navadeh et al., 2013). Globally, there are certain prevention challenges specific to inmate populations (Spaulding et al., 2002): lack of awareness about HIV, lack of resources for HIV testing and treatment given limited funding resources, rapid turnover among jail communities and inmate concerns about privacy, and fear of stigma that prevent them from disclosing their high-risk behaviors (Dolan et al., 2015; Hammett, 2006; Shahbazi, Farnia, Rahmani, & Moradi, 2014). Yet, jails and prisons continue to be potent targets for public health interventions (Spaulding et al., 2009) and there is a need to address these challenges.

Self-reported data collection is the most common data collection method in behavioral studies and is subject to underreporting when the data being collected involve socially stigmatized behaviors (Mirzazadeh et al., 2013). Although, underreporting of stigmatized behaviors has been explored in previous studies, this study is the first to our knowledge to examine this paradigm among prison inmates. We explored the potential of three indirect methods of data collection to provide the best estimates of socially stigmatized HIV-related risk behaviors among inmates in three prisons in Iran. The design of the study was to conduct questionnaire-based interviews, using crosswise, proxy respondent method (PRM), and network scale-up (NSU) methods. The collected information was then compared with the findings from direct questioning.

Method

For the period, August through October of 2013, 265 prisoners from three prisons (Shiraz = 70 males, Hamedan = 70 males, and Qazvin = 70 males and 55 females) were recruited into the study. After obtaining informed consent, we applied a systematic random sampling method, using the prisoner’s personal identification codes as the sampling frame. This gave us a sample proportional to the overall number of prisoners in every ward in each prison.

The trained interviewers approached the selected study participants. Data on risk behaviors were collected using direct and indirect methods in the order listed below:

Method 1: Close and Randomly Selected Friends (PRM)

This method involves asking the respondents about the high-risk behaviors of their closest friend or randomly selected friends. Therefore, this method did not reveal personal behaviors of the respondent (Gilpin et al., 1994; Rwanda Biomedical Center/Institute of HIV/AIDS, 2012). We defined “close friend” as a person with whom the respondent had a close friendship, discussed personal information, shared meals, as well as received social support. Every respondent was asked about the HIV risk behavior of their closest friend and the degree of certainty they have in their responses and recall (as a proxy measure of information transparency). To randomly select a friend of the respondent, two separate lists of 45 common male and female names were provided to the respondent—one for male prisoners and one for female prisoners. Each one of the two lists was randomly grouped and listed on nine different cards. The respondents were asked to randomly select a card and check whether he or she knows anyone listed on the card. If they did not know anyone on the card they were handed, another card was selected by the respondent. This process was repeated until the respondent selected a name. If a respondent provided two or more names, the one that was considered the “closest friend” to the respondent was chosen. The respondents were instructed not to disclose the name of the friend. They were then only questioned about the HIV risk behavior of the “closest friend” selected, and the degree of certainty they had about their responses and recall.

Method 2: NSU

This method involves measuring the prevalence of HIV risk behaviors in a population using the respondents’ social network. The general concept behind this method is the equivalence of the proportion of individuals with high risky behaviors within one’s social network to the number of people with those high-risk behaviors within a defined population. By asking questions about an acquaintance—a person other than the respondent—the interview takes on some anonymity, allowing the responses to be honest without fear of stigma or other negative consequences for the respondent or his or her friends (Bernard et al., 2010; Johnsen, Bernard, Killworth, Shelley, & McCarty, 1995; Killworth, Johnsen, McCarty, Shelley, & Bernard, 1998). Therefore, the respondents were asked about the number of people/inmates they had known face-to-face (acquaintance) within their ward during the past 6 months. Then, the respondents were questioned about the number of people they knew who were engaged in risky behaviors, such as “ever used drugs,” “ever injected drugs,” and “ever had extramarital sexual contact” in and out of prison. They were also asked about the number of people/inmates with whom they had meals or from whom they had mutual financial support over the course of the past 6 months inside the prison and, again, the number of those who had risky behaviors inside and outside prisons.

Method 3: Crosswise

The concept of crosswise method is to pair ones answers on the risky behaviors with a randomly selected known proportion (here picking an envelope out of 10). The process makes it impossible to discover the individual’s response to the sensitive attributes, for example, drug injection in prison. In this method, all respondents were asked to pick one envelop from the 10 provided envelopes. Only one of the envelopes had a card labeled as “no,” all others as “yes.” Asked not to disclose the label of the selected card, they were then questioned about their risky behaviors. The respondents should have responded with an answer equal to the label on the card or not. At the end of the interview, they were asked to replace the card in the envelope and slide it into the roster in a way that it could not be distinguished (Jann, Jerke, & Krumpal, 2012; Yu, Tian, & Tang, 2008).

Method 4: Direct Questioning

Finally, the respondents were questioned directly about their own risky behaviors. This is a common method that is used in collecting information on risky behaviors in behavioral surveys (Amon et al., 2009; Navadeh et al., 2013; Phillips, Gomez, Boily, & Garnett, 2010).

The study protocol and procedures were reviewed and approved by the research review board of the Kerman University of Medical Sciences (K/93/162). The study was piloted for refining the method and standardizing the questionnaire.

Three groups of one field supervisor and one interviewer were trained in a one-day workshop to implement the survey. During the workshop, the direct and indirect methods of questioning were explained, and then practiced by role-playing. During the implementation process and before commencing the study, every interviewer was asked to complete two to five questionnaires under the supervision of his or her supervisor.

Data Analysis

Data analysis was done using STATA (StataCorp., 2011). We defined the information transparency bias as the proportion of respondents who reported that they were not aware of their friends’ or acquaintances’ risky behaviors. Furthermore, we reported the information transparency bias for every sensitive risky behavior by sex and type of acquaintances, either the close friend or the randomly selected friend. We examined the differences between men and women in regard to the information transparency bias by chi-square test (or Fisher exact test, if required).

In the NSU method, we asked the participant about the number of inmates they knew over the past 6 months prior to the interview (large network). To deal with the outliers, the large network size of 10 is given to those who reported the size “below 10” and 100 to those who reported the size “above 100.” We also asked for the number of inmates the study participants had meals with over the past 6 months (meal network). Any report on the meal network size as “zero” was replaced by “one.” We also replaced all reported meal networks above 30 with 30. Using the average of reported large and meal network sizes ( $\bar{C}$ ), and the number of acquaintance they knew who were engaged in any of the risky behaviors (m), we calculated the proportion of prisoners having such risky behaviors as (Equation 1; Killworth et al., 1998)

P_{N S U} = \sum_{1}^{i} \frac{m_{i}}{\bar{C}} .

The calculation was done for the large network size and for the meal network size separately. The 95% confidence intervals (CIs) were calculated using a bootstrap technique with 100 iterations (Shokoohi, Baneshi, & Haghdoost, 2012).

In the crosswise method, we applied the following formula to estimate the proportion π of risk behaviors among the study participants (Equation 2):

π = \frac{q + p - 1}{2 p - 1} .

Here, q is the proportion of persons who gave identical answers (either yes or no) to the pairs of sensitive and nonsensitive questions. As explained earlier, the nonsensitive question in each paired questions was the randomly selected card labeled as “yes or no” (from a roster of 10), which we already know the amount which is 10% (p). Given the binomial distribution, we calculated the 95% CI for π. Based on the following equations, the 95% CIs were estimated (Equations 3 and 4; Jann et al., 2012; Yu et al., 2008):

S E_{π} = \sqrt{\frac{q (1 - q)}{(N - 1) {(2 p - 1)}^{2}}} .

95 % CI : π \pm 1.96 \times S E_{π} .

Results

Demographic of Study Participant and Their Friends

A sample of 265 participants was included in the study (210 males, and 55 females). The average age of male participants was 35.3 years (95% CI = [34.1, 36.6]) and 33.1 years (95% CI = [30.4, 35.6]) for the females. Male participants were serving longer prison sentences than female prisoners (2.5 vs. 1.5 years, p < .01). Moreover, the male prisoners reported “past history of incarceration” more than the females (59.2% vs. 34.5% p < .01). The female prisoners were twice more likely to be illiterate or be able to read or write than male prisoners (9.4% vs. 18.1%). Instead, one third (32.7%) of women received a high school diploma, much higher than men (17.1%). Male and female prisoners differed significantly according to marital status. 33.2% of males were single (never married), whereas only 7.3% of females had never been married. Female prisoners were more likely to be married and not living with their spouse (14.5% vs. 0.9%) or identified as a widow (18.2% vs. 1.4%). 16.2% of the males and 9.1% of the female prisoners participated in the recent biobehavioral survey in 2013 (Table 1).

Table 1.

Demographic Characteristics of the Study Participants by Sex (N = 265).

Variables	Men (n = 210)	Women (n = 55)
M age (year)	35.3 [34.1, 36.6]	33.1 [30.6, 35.6]
Duration of current incarceration (year)	2.5 [2.1, 2.9]	1.5 [0.95, 2.1]
Level of education (%)
Illiterate or only able to read/write	9.4 [5.4, 13.4]	18.1 [7.8, 28.5]
Primary school	32.2 [25.8, 38.5]	0.2 [9.2, 30.7]
Guidance school	35.5 [29.0, 42.0]	23.6 [12.2, 35.0]
High school and diploma	17.1 [11.9, 22.1]	32.7 [20.1, 45.2]
University degree	5.6 [2.5, 8.8]	5.4 [0.0, 11.5]
Current marital status (%)
Single (never married)	33.2 [26.7, 39.5]	7.3 [0.3, 14.2]
Married, live with spouse	47.9 [41.1, 54.6]	40.0 [26.8, 53.1]
Married, live without spouse	4.3 [1.5, 7.0]	1.8 [0.0, 5.3]
Sigheh (temporary marriage)	0.9 [0.0, 2.2]	14.5 [5.0, 23.9]
Divorced	12.3 [7.8, 16.7]	18.2 [7.8, 28.5]
Widower/widow	1.4 [0.0, 3.0]	18.2 [7.8, 28.5]
History of past incarceration (%)
Never	40.7 [34.1, 47.4]	65.4 [52.7, 78.1]
At least once	59.2 [52.5, 65.9]	34.5 [21.8, 47.2]
Participated in the recent (2013) biobehavioral survey	32 (16.2%)	5 (9.1%)

Note. Numbers in square brackets are 95% confidence intervals.

Information Transparency Bias

Table 2 presents the information transparency bias of every HIV risk behavior collected, separated by sex and type of selected friend (closest friend or randomly selected friend). Information transparency was consistently lower in male prisoners, when they were questioned about their closest friend instead of the randomly selected friend. The difference was as low as 3.3% for “injecting drug during last incarceration” and as high as 13.3% for “extramarital sex during past 12 months.” This pattern was not observed in female prisoners.

Table 2.

The Information Transparency Bias in Different HIV-Related Risk Behaviors by Sex (N = 265).

Risk behaviors	Men (n = 210)		Women (n = 55)		p value men vs. women
Risk behaviors	Close friend	A random selected friend	Close friend	A random selected friend	Close friend	A random selected friend
Drug use ever	2.3 [0.3, 4.4]	5.7 [2.5, 8.9]	5.4 [0.0, 11.5]	1.8 [0.0, 5.3]	.36	.31
Injection drug use ever	2.8 [0.5, 5.1]	11.9 [7.5, 16.3]	5.4 [0.0, 11.5]	3.6 [0.0, 8.6]	.40	.081
Drug use during last incarceration	2.8 [0.5, 5.1]	6.2 [2.9, 9.5]	1.8 [0.0, 5.3]	0	.99	.077
Injection drug use during last incarceration	1.4 [0.0, 3.0]	4.7 [1.8, 7.6]	0	1.8 [0.0, 5.3]	.99	.46
Extramarital sex during past 12 months	14.4 [9.6, 19.2]	27.7 [21.6, 33.8]	7.2 [0.3, 14.2]	7.2 [0.3, 14.2]	.18	.001
Same-sex sexual contact ever	13.3 [8.7, 17.9]	24.4 [18.5, 30.2]	1.8 [0.0, 5.3]	7.2 [0.3, 14.2]	.013	.005
Sexual contact during last incarceration	6.6 [3.2, 10.0]	14.3 [9.5, 19.1]	1.8 [0.0, 5.3]	7.2 [0.3, 14.2]	.32	.25
HIV test ever	36.1 [29.6, 42.7]	45.9 [39.1, 52.7]	41.8 [28.6, 55.0]	36.3 [23.4, 49.2]	.43	.22

Note. Transparency bias = % “do not know” response to the questions asking about the risk behaviors of the study participants’ (close or random selected) friend. Numbers in square brackets are 95% confidence intervals. Bold value are significant p-value under cut-off 0.05.

Risk Behaviors Estimates by Different Methods

Prevalence of drug use, injection drug use, sexual risk, and history of HIV testing are given in Table 3. According to the findings, past history of drug use was the most frequent of risky behaviors reported by both men and women. The behavior was consistently higher in men using both direct and indirect methods. The crosswise method produced the lowest estimate for ever having used drugs, and higher estimates were found for “close” or randomly selected friends.

Table 3.

The Estimates of HIV-Related Risk Behaviors Reported by Prisoners About Themselves or Their Friends by Different Data Collection Methods (N = 265).

Risk behaviors	The participant by direct questioning	Close friend	A random selected friend	Crosswise method	Network scale-up
Risk behaviors	The participant by direct questioning	Close friend	A random selected friend	Crosswise method	Large network^a	Meal network^b
Male prisoners
Ever used drug	74.2 [68.3, 80.2]	73.1 [67.1, 79.2]	77.6 [71.8, 83.5]	65 [55.1, 74.9]	77.2 [69.7, 84.7]	63.7 [53.2, 74.3]
Ever injected drug	15.3 [10.3, 20.2]	10.8 [6.5, 15.1]	9.7 [5.4, 14.1]	21.3 [7.5, 35.0]	20.1 [15.1, 25]	6.9 [4.6, 9.1]
Using drug during last incarceration	38.1 [31.4, 33.7]	42.1 [35.3, 48.9]	44.8 [37.8, 51.9]	37.5 [25.1, 49.9]	47.2 [38.9, 55.6]	38.8 [29.5, 48]
Injecting drug during last incarceration	3.8 [1.2, 6.4]	0.9 [0.0, 2.3]	1.0 [0, 2.4]	15.0 [0.8, 29.2]	0.5 [0, 1.1]	1.0 [0.4, 1.7]
Extramarital sex during past 12 months	12.9 [8.3, 17.4]	13.4 [8.4, 18.5]	19.2 [12.8, 25.5]	17.5 [3.5, 31.5]	12.1 [8.4, 15.7]	11.6 [7.7, 15.4]
Ever same-sex sexual contact	12.8 [8.2, 17.4]	12.6 [7.7, 17.5]	16.4 [10.6, 22.2]	20.0 [6.2, 33.8]	8.0 [5.4, 10.5]	7.0 [4.3, 9.7]
Sexual contact during the last incarceration	3.8 [1.2, 6.4]	4.1 [1.2, 6.8]	10.6 [6, 15.1]	11.3 [0.0, 25.7]	4.3 [2.9, 5.7]	3.3 [1.8, 4.8]
Ever had an HIV test	46.1 [39.4, 52.9]	58.2 [49.7, 66.6]	58.4 [49.2, 67.6]	50.0 [38.7, 61.3]	—	—
Female prisoners
Ever used drug	47.2 [33.8, 60.6]	59.6 [46.1, 73.1]	57.4 [44, 70.7]	33.8 [8.6, 58.9]	58.4 [47.4, 69.3]	50.0 [34.6, 65.4]
Ever injected drug	3.6 [0.0, 8.6]	1.9 [0.0, 5.7]	15.0 [5.3, 24.8]	41.3 [17.4, 65.1]	10.2 [5.4, 15]	3.6 [0.4, 6.9]
Using drug during last incarceration	12.7 [3.7, 21.6]	16.6 [6.5, 26.7]	30.9 [18.5, 43.2]	35.0 [10.1, 59.9]	22.2 [10.4, 34]	18.2 [6.4, 30]
Injecting drug during last incarceration	1.8 [0.0, 5.3]	0.0	1.8 [0.0, 5.4]	37.5 [13.0, 62.0]	1.1 [0.1, 2.1]	0.0
Extramarital sex during past 12 months	10.9 [2.5, 19.2]	37.2 [23.7, 50.7]	35.2 [21.9, 48.6]	22.5 [0.0, 49.3]	50.7 [37.3, 64]	52.8 [9.1, 96.5]
Ever same-sex sexual contact	0.0	7.4 [0.3, 14.4]	17.6 [7, 28.2]	22.5 [0.0, 49.3]	0 [0, 0]	4.7 [0.0, 9.7]
Sexual contact during last incarceration	0.0	5.5 [0.0, 11.7]	13.7 [4.1, 23.3]	23.8 [0.0, 50.4]	5.6 [3.2, 8.1]	3.8 [0.0, 8.3]
Ever had an HIV test	38.1 [25.1, 51.1]	50.0 [32.2, 67.7]	54.2 [37.4, 71.1]	28.8 [2.9, 54.6]	—	—

Note. Numbers in square brackets are 95% confidence intervals.

Large network size ( $\bar{C}$ ) for male prisoners was 56 and for female prisoners was 37.

Meal network size ( $\bar{C}$ ) for male prisoners was ≈5 and for female prisoners was ≈ 2.

In response to direct questioning, past history of injection drug use was reported by 15.3% of men and 3.6% of women. The estimates were surprisingly high, using the crosswise method (21.3% in men and 41.3% in women). Drug use during last incarceration ranged between 37.5% (crosswise method) and 47.2% (NSU-large network). This was quite lower in females, between 12.7% (direct questioning) and 35.0% (crosswise method).

History of injection drug use during last incarceration varied between 0.5% (NSU-large network) to 15.0% (crosswise method). This was also quite different in female prisoners, between 0% (close friend and NSU-meal network) and 37.5% (crosswise method).

Extramarital sex in during the last year was acknowledged by 12.9% of men and 10.9% of women. Respectively, these figures increased to 19.2% using random selected friend method and to 52.8% using the NSU-meal network method. Same-sex sexual contact ever, ranged from 7.0% (NSU-meal network) to 20.0% (crosswise) in men. Sexual contact within prison was reported by 3.8% men and 0% of women. The crosswise method provided the highest estimates for both men (11.3%) and women (23.8%).

Regarding the HIV testing history, although 46.1% of men reported that they have been tested at some time for HIV, they mentioned that the prevalence of testing is higher among their friends (58.2% for close and 58.4% for a random selected friend). The same pattern was seen among women while the ever HIV testing reported as 38.1% and for their friends as 50.0% and randomly selected friend as 54.2%.

Discussion

The results indicate that there is a high level of HIV-related risk behavior estimates, depending on the methods used for data collection. In settings like prisons, risk behaviors are vastly underreported by prisoners even when the study is completely anonymous, informed consent is verbal, and no HIV tests were administrated.

Overall, prisoners reported a higher level of risk behavior for their closest friends and even more for a friend selected at random rather than for themselves. This effect is called “prestige bias” or “social desirability bias” in the literature (Gregson, Zhuwau, Ndlovu, & Nyamukapa, 2002). Social desirability bias states that people are more open in discussing the stigmatized risk behaviors of their friends rather than themselves. We also observed this phenomenon among prisoners and their acquaintances in prison. This is the basic rationale for applying a proxy respondent or NSU methods to measure risk behaviors or estimate the number of those who are engaging in such behaviors, rather than direct questioning methods.

The challenge in such indirect data collection methods is the information transparency bias. As we saw, inmates are not openly talking to their friends about their risk behaviors, especially sexual risk behaviors. Drug-related behaviors are less stigmatized (Haji-Maghsoudi, Haghdoost, & Baneshi, 2014) and so the transparency biases are lower. As expected, close friends are more transparent in disclosing their HIV risk behaviors to each other. Men were more conservative than women in disclosing extramarital and same-sex sexual contact than women. This might be so as such heavily stigmatized behaviors are more common among men. These behaviors had a higher level of information transparency bias and were more underreported than women in our study (Mirzazadeh et al., 2013; Phillips et al., 2010). In addition, we observed in our study population that HIV status is considered to be a very personal and stigmatized issue. Only one out of three study participants knew about their “close friends” HIV status. The disclosure of HIV status is also reported as very low in other settings and especially among sexual partners (Shelley et al., 2006).

In Figure 1, using a schematic diagram, we present the interaction of the two biases, social desirability and transparency, in direct and indirect questioning techniques we used to collect the HIV risk behaviors. As it is obvious, when we directly ask about the participants’ own behaviors, they may not disclose their truth behaviors, as it is not socially acceptable. This bias may be less when asking about their friends, close friends, random anonymous friends, or their overall network. On the contrary, the transparency bias is increasing, as they might have less information about their accountancies’ (i.e., their network) risk behaviors, random friend, and close friend. It should be noted that the effect of the two biases for different data collection methods may not be equal and they may not cancel out each other completely. Their magnitudes of effects need to be measured and considered carefully when interpreting the results.

Figure 1.

An illustrative graph on the trend of social desirability and transparency biases from direct questioning methods (asking people about their own behavior) to indirect techniques (asking about the behavior of their acquaintances in their network—for example, network scale-up).

Overall, the NSU method produced the lowest estimates of high-risk behaviors in our study population. In fact, this could be partly explained by the information transparency bias discussed (Salganik et al., 2011; Zheng, Salganik, & Gelman, 2006). The amount of such biases reported as 40% (20%-50%) in FSW (Mirzazadeh et al., 2013). Another bias that might affect the NSU estimates is that people with high-risk behaviors might have fewer connections with other people in the community (degree ratio; Salganik et al., 2011). This can lead to a significant underestimation of risk behaviors. Crude NSU estimates need to be adjusted for these two biases to provide more accurate estimates. In addition, the NSU is a complex method and difficult to clarify to participants regarding the social network and social network size. We used two definitions of the network, the overall social network and the meal network. We did not find any meaningful discrepancies between the two NSU estimates. A Rwanda NSU study concluded that using meal network size provided more accurate estimates (Rwanda Biomedical Center/Institute of HIV/AIDS, 2012). In our study, we asked about the networks within prison, and discovered that the meal and overall social networks were almost comparable, no meaningful difference was found between the two.

Among the methods used to estimate the prevalence of risk behaviors, the crosswise method provided the highest estimates that are hard to believe. Moreover, we observed a differential bias correction with crosswise among women and men. For example, ever injection was reported by 15.3% of men and using crosswise, it is estimated to be 21.3%. In females, 3.6% reported ever injection while crosswise estimate was unbelievably high as 41.3%. We observed the same pattern for injection during last incarceration. For other variables, such as ever drug use and HIV test, where the reported prevalence was considerably high, the crosswise estimates decreased. This can happen if participants randomly answered to the paired sensitive/insensitive questions. This could partly explain the high crosswise estimates of ever injection among women and the small difference with men’s estimates. Such differential correction of crosswise method between men and women, or better to say, rare and common variables, was also observed in a study of illicit drug use among students (Shamsipour et al., 2014). Given the complexity of this method, we had a difficult time demonstrating to study participants how this method works. Unfortunately, the card game and the smart card answer technique were not effectively understood by participants, potentially leaving some room for error. This limitation has been reported partially in other studies who have applied the crosswise method (Jing, Qu, Yu, Wang, & Cui, 2014). To our knowledge, our study is the first to implement this method among prisoners.

Conclusion

HIV-related risk behaviors are significantly underreported by inmate populations, especially in settings where they are still heavily stigmatized and considered illegal. Different data collection methods produce different estimates, and it is crucial to carefully understand the data collection process in terms of potential biases and also the level of complexity of the methods (Gregson et al., 2002). Findings need to be triangulated and corrected for such biases before being used for decision making.

Footnotes

Acknowledgements

The authors would like to thank all the field supervisors for their contribution in designing, implementing, and monitoring the data collection process. They express their gratitude to Azam Valipour, the wonderful program assistant who contributed to staff training, supervision, and study implementation. This collaborative work was implemented by the supports from Iranian Prison Organization, and the authors thank particularly Dr. Shahbazi and Dr. Vaezi, for their valuable continuous support.

Authors’ Contribution

A.M., A.D., A.S., and A.H. were responsible for study concept and design. M.S. and A.M. performed data analysis. S.N. and J.J. produced interpretations and findings. A.M., S.N., and M.S. drafted the article and all authors provided critical feedback and approved the final version.

Author’s Note

Jennifer P. Jain is now affiliated to University of California, San Diego, School of Medicine, San Diego, USA

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The Global Fund, Iran, has funded the study. The University of California, San Francisco’s International Traineeships in AIDS Prevention Studies (ITAPS), U.S. NIMH, R25MH064712.

Author Biographies

Ali Mirzazadeh, MD, PhD, is an assistant professor, Department of Epidemiology and Biostatistics, University of California, San Francisco. He is an infectious disease epidemiologist.

Mostafa Shokoohi, MS, is a doctoral student at the Western University of Ontario, London, Canada. An MSc in epidemiology, he was a research assistant in the Regional Knowledge Hub for HIV Surveillance, a World Health Organization collaboration center in Iran.

Soodabeh Navadeh, MD, PhD, Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences. Her research focus is HIV surveillance and methods to deal with unmeasured factors in epidemiological studies.

Ahmad Danesh, MD, PhD, is an assistant professor of epidemiology, Golestan University of Medical Sciences. His main interest is controlling HIV infection and drug addiction by the opioid maintenance treatment in prison settings.

Jennifer Jain, MPH, is a coordinator with the University of California, San Francisco, Prevention and Public Health Group and the Department of Epidemiology and Biostatistics. She is the site coordinator for the Vaccination if Prevention (VIP) Vaccine Trial and the research assistant for the U-Find-Out study.

Abbas Sedaghat, MD, MPH, is the national AIDS program manager, Ministry of Health and Medical Education, Tehran, Iran.

Marziyeh Farnia, MD, is a physician and researcher affiliated with Iran Prisons Organization, Health and Treatment Department, Tehran, Iran.

AliAkbar Haghdoost, MD, PhD, is a physician and professor of epidemiology based at the Kerman University of Medical Sciences (KMU). He has several years of research experience in the field of infectious disease epidemiology, medical education, public health, and modeling analysis.

References

Amon

Brown

Hogle

MacNeil

Magnani

Mills

. . . Sow

C. K.

(2009). Behavioral surveillance surveys, BSS. Guidelines for repeated behavioral surveys in populations at risk of HIV. Washington, DC: Family Health International.

Bernard

Hallett

Iovita

Johnsen

Lyerla

McCarty

. . . Stroup

D. F.

(2010). Counting hard-to-count populations: The network scale-up method for public health. Sexually Transmitted Infections, 86(Suppl. 2), ii11-ii15.

Dolan

Moazen

Noori

Rahimzadeh

Farzadfar

Hariga

(2015). People who inject drugs in prison: HIV prevalence, transmission and prevention. International Journal of Drug Policy, 26(Suppl. 1), S12-S15. doi:10.1016/j.drugpo.2014.10.012

Gilpin

E. A.

Pierce

J. P.

Cavin

S. W.

Berry

C. C.

Evans

N. J.

Johnson

Bal

D. G.

(1994). Estimates of population smoking prevalence: Self- vs. proxy reports of smoking status. American Journal of Public Health, 84, 1576-1579.

Gregson

Zhuwau

Ndlovu

Nyamukapa

C. A.

(2002). Methods to reduce social desirability bias in sex surveys in low-development settings: Experience in Zimbabwe. Sexually Transmitted Diseases, 29, 568-575.

Haghdoost

A. A.

Mirzazadeh

Shokoohi

Sedaghat

Gouya

M. M.

(2013). HIV trend among Iranian prisoners in 1990s and 2000s: Analysis of aggregated data from HIV sentinel sero-surveys. Harm Reduction Journal, 10, 32. doi:10.1186/1477-7517-10-32

Haji-Maghsoudi

Haghdoost

A. A.

Baneshi

M. R.

(2014). Selection of variables that influence drug injection in prison: Comparison of methods with multiple imputed data sets. Addict Health, 6, 36-44.

Hammett

T. M.

(2006). HIV/AIDS and other infectious diseases among correctional inmates: Transmission, burden, and an appropriate response. American Journal of Public Health, 96, 974-978. doi:10.2105/ajph.2005.066993

Jann

Jerke

Krumpal

(2012). Asking sensitive questions using the crosswise model: An experimental survey measuring plagiarism. Public Opinion Quarterly, 76, 32-49.

10.

Jing

Wang

Cui

(2014). Estimating the sizes of populations at high risk for HIV: A comparison study. PLoS ONE, 9(4), e95601.

11.

Johnsen

E. C.

Bernard

H. R.

Killworth

P. D.

Shelley

G. A.

McCarty

(1995). A social network approach to corroborating the number of AIDS/HIV + victims in the US. Social Networks, 17, 167-187.

12.

Killworth

P. D.

Johnsen

E. C.

McCarty

Shelley

G. A.

Bernard

H. R.

(1998). A social network approach to estimating seroprevalence in the United States. Social Networks, 20, 23-50.

13.

Mirzazadeh

Mansournia

M. A.

Nedjat

Navadeh

McFarland

Haghdoost

A. A.

Mohammad

(2013). Bias analysis to improve monitoring an HIV epidemic and its response: Approach and application to a survey of female sex workers in Iran. Journal of Epidemiology & Community Health, 67, 882-887. doi:10.1136/jech-2013-202521

14.

Navadeh

Mirzazadeh

Gouya

M. M.

Farnia

Alasvand

Haghdoost

A. A.

(2013). HIV prevalence and related risk behaviours among prisoners in Iran: Results of the national biobehavioural survey, 2009. Sexually Transmitted Infections, 89(Suppl. 3), iii33-iii36.

15.

Phillips

A. E.

Gomez

G. B.

Boily

M. C.

Garnett

G. P.

(2010). A systematic review and meta-analysis of quantitative interviewing tools to investigate self-reported HIV and STI associated behaviours in low- and middle-income countries. International Journal of Epidemiology, 39, 1541-1555.

16.

Rwanda Biomedical Center/Institute of HIV/AIDS. (2012). Estimating the size of populations through a household survey. Available at: https://dhsprogram.com/pubs/pdf/FR261/FR261.pdf

17.

Salganik

M. J.

Mello

M. B.

Abdo

A. H.

Bertoni

Fazito

Bastos

F. I.

(2011). The game of contacts: Estimating the social visibility of groups. Social Networks, 33, 70-78.

18.

Shahbazi

Farnia

Rahmani

Moradi

(2014). Trend of HIV/AIDS prevalence and related interventions administered in prisons of Iran—13 years’ experience. Iranian Journal of Public Health, 43, 471-479. Retrieved from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4433728/pdf/IJPH-43-471.pdf

19.

Shamsipour

Yunesian

Fotouhi

Jann

Rahimi-Movaghar

Asghari

Akhlaghi

A. A.

(2014). Estimating the prevalence of illicit drug use among students using the crosswise model. Substance Use & Misuse, 49, 1303-1310. doi:10.3109/10826084.2014.897730

20.

Shelley

G. A.

Killworth

P. D.

Bernard

H. R.

McCarty

Johnsen

E. C.

Rice

R. E.

(2006). Who knows your HIV status II? Information propagation within social networks of seropositive people. Human Organization, 65, 430-444.

21.

Shokoohi

Baneshi

M. R.

Haghdoost

A. A.

(2012). Size estimation of groups at high risk of HIV/AIDS using network scale up in Kerman, Iran. International Journal of Preventive Medicine, 3, 471-476.

22.

Spaulding

A. C.

Seals

R. M.

Page

M. J.

Brzozowski

A. K.

Rhodes

Hammett

T. M.

(2009). HIV/AIDS among inmates of and releasees from US correctional facilities, 2006: Declining share of epidemic but persistent public health opportunity. PLoS ONE, 4(11), e7558. doi:10.1371/journal.pone.0007558

23.

Spaulding

A. C.

Stephenson

Macalino

Ruby

Clarke

J. G.

Flanigan

T. P.

(2002). Human immunodeficiency virus in correctional facilities: A review. Clinical Infectious Diseases, 35, 305-312. doi:10.1086/341418

24.

StataCorp. (2011). Stata Statistical Software: Release 12. College Station, TX: StataCorp LP.

25.

Tian

Tang

(2008). Two new models for survey sampling with sensitive characteristic: Design and analysis. Metrika, 67, 251-263.

26.

Zablotska

I. B.

Kippax

Grulich

Holt

Prestage

(2011). Behavioural surveillance among gay men in Australia: Methods, findings and policy implications for the prevention of HIV and other sexually transmissible infections. Sexual Health, 8, 272-279. doi:10.1071/sh10125

27.

Zheng

Salganik

M. J.

Gelman

(2006). How many people do you know in prison? Using overdispersion in count data to estimate social structure in networks. Journal of the American Statistical Association, 101, 409-423.