Prison Rape: A Critical Review of the Literature
Gerald G. Gaes and Andrew L. Goldberg
National Institute of Justice
March 10, 2004
EXECUTIVE SUMMARY
The opinions expressed in this paper are those of the authors’ and should not be
construed as the opinions or the policy of the National Institute of Justice, the Office of
Justice Programs, or the U. S. Department of Justice.
Prison Rape: A Critical Review of the Literature – Executive Summary
This executive summary covers the highlights of the report Prison Rape: A Critical
Review of the Literature, which analyzes obstacles and problems that must be overcome
to effectively measure sexual assault at the facility level. Each bold heading in this
summary refers to the same bold heading contained in the larger report.
Federal Legislation. The Prison Rape Elimination Act of 2003 calls for research and
policy changes to minimize sexual victimization of incarcerated juveniles and adults. The
Act also calls for a zero tolerance policy; national standards for the detection, prevention,
reduction, and punishment of prison rape; collection of data on incidence; and
development of a system to hold prison officials accountable. Also, the Bureau of Justice
Statistics is to design a methodology to assess the prevalence of prison sexual assault and
monitor adult prisons, jails, and juvenile facilities. In the findings section of the public
law, there is a claim from unnamed experts that a conservative estimate of victimization
suggests that 13 percent of inmates in the United States have been sexually assaulted.
Defining Sexual Victimization – Prevalence and Incidence. Research should
distinguish various levels of sexual victimization from completed rapes to other forms of
sexual coercion. Any measurement process will have to distinguish between the
prevalence and incidence of the events. Prevalence refers to the number of people in a
given population who have ever had a sexual assault experience. Incidence refers to the
number of new cases. This distinction is important, because prevalence can be high, but
the number of new cases is low due to some kind of intervention or enforcement of
policy.
Prison Rape Literature. Aside from one study conducted by the Bureau of Justice
Statistics (BJS) in 1997, all other studies conducted in the United States included fewer
than 50 prisons in total. In 2000, BJS reported there were 1,668 federal and state prisons.
There has also been one study of sexual victimization in a jail system. In 1999, the
Bureau of Justice Statistics reported there were 3,365 jails in the United States.
Studies Involving Primarily Men, or Men and Women. Studies by StruckmanJohnson, Struckman-Johnson, Rucker, Bumby, and Donaldson (1996), StruckmanJohnson and Struckman-Johnson (2000), Davis (1968), Nacci and Kane (1982, 1983,
1984), Saum, Surratt, Inciardi, and Bennet (1995), Tewksbury (1989), Maitland and
Sluder (1998), Wooden and Parker (1982), Lockwood (1980), Toch (1977), Hensley,
Tewksbury, and Castle (2003), Carroll (1977), Chonco (1989), Moss, Hosford, and
Anderson (1979), Butler, Donovan, Levy, and Kaldor (2002), Fuller and Orsagh (1977),
Butler and Milner (2003), Forst, Fagan, and Vivona (1989), and the Bureau of Justice
Statistics (1997) reported on primarily male samples, or a combination of female and
male samples. The Butler and Milner and Butler et al., studies were conducted as part of
a larger health assessment in the prison system in New South Wales, Australia. Details of
each of these studies are covered in the full report.
March 10, 2004
1
Studies Involving Exclusively Women – Coerced Sex among Women. StruckmanJohnson and Struckman-Johnson (2002), and Alarid (2000) reported on exclusively
female samples. These studies are reviewed in detail in the full report. There is also a
great deal of research on consensual sex among women that is mentioned, but not
reviewed in the report.
U.S. National Probability Sample of Rape during Incarceration. The only attempt at a
U.S. national probability sample of adults in state and federal prisons was conducted by
BJS in 1997. In that study, 0.45 percent of men and 0.35 percent of women prisoners
reported they had experienced an attempted or completed rape during a previous
incarceration.
U.S. National Probability Sample of Forced Sexual Activity among Youth in
Juvenile Facilities. There has also been a national probability sample of youth living in
juvenile facilities because they are accused or convicted of a crime. The Survey of Youth
in Residential Placement (SYRP) was sponsored by the Office of Juvenile Justice and
Delinquency Prevention. Over 7,000 juveniles participated (75 percent response rate) and
detailed questions about forced sex were asked. The results will be released soon.
Summary of Prison Rape Estimation Studies. Aside from the New South Wales and
BJS studies, most other research papers report survey return rates of 50 percent or less.
Many response rates are 25 percent or lower. The prevalence estimates in this research
range from 0 to 40 percent. When the data are limited to definitions that involve primarily
assault or a completed sexual victimization, most of the prevalences were 2 percent or
less typically referring to the entire period of incarceration. When forms of sexual
pressure are included, these estimates increase to an upper limit of about 21 percent or
less except for a couple of prisons. National and system probability samples which are
designed to give an estimate of victimization for the entire jurisdiction, reported sexual
victimization rates of 2 percent or less. There are few incident studies, and these have
little, or no, information on how to construct an appropriate denominator to get a
percentage or rate. A “back of the envelope” estimate places this at no more than 2
percent in a given year, based primarily on the one jail system study conducted in the
1960’s and as low as 0.69 percent based on one prison study. Women’s victimization
percentages appear to be lower than men’s.
These studies use different methods to establish the level of victimization (questionnaires,
interviews, informants, administrative records); they use different questions, and they use
different time frames. Definitions vary widely from rape to sexual pressure. Some of
these estimates rely on self-reported victimizations, while others are based on the
perceptions of inmates and staff on the overall level of victimization in the prison. These
latter estimates always appear higher than self reports, and it is unclear what these latter
estimates mean since there is no presumption that inmates or staff actually witness all of
the sexual assaults they claim are occurring. Most studies fail to report how long the
sexual assault victim has been in prison making it difficult to compare prisons across
jurisdictions, due to the likelihood of different exposure periods.
March 10, 2004
2
A Meta-analysis of Prison Sexual Assault Studies
In an effort to get a summary estimate of the level of sexual victimization, a metaanalysis was conducted to provide a calculation of an average estimate over all of the
studies, even though any single study may not meet conventional levels of statistical
significance. Results of the meta-analysis indicate an average prison lifetime sexual
assault prevalence of 1.91 percent. This means that 1.91 percent of inmates have
experienced a sexual victimization over a lifetime of incarceration. This estimate is
based primarily on studies which report completed victimizations, although it
incorporates some studies which also include serious attempts of sexual assault and one
study that includes sexual pressure.
Social Desirability of Responses and the Nature of Sensitive Questions. Prison sexual
assault surveys are similar to surveys conducted in the community eliciting information
on other sensitive behaviors. Survey participants tend to underreport behaviors that are
perceived to be against society’s norms (socially undesirable), that invade privacy, and
that may be disclosed to third parties despite precautions by researchers to protect
confidentiality.
Study Procedures and the Problem of Selection Bias. There are often very low
response rates in these studies and researchers usually do not report differences between
those that choose to be surveyed and those refusing. Nor do any of the research reports
make adjustments to the victimization estimates based on differences in characteristics of
the respondents. Such adjustments could easily change the level of sexual victimization,
either to a higher or lower percentage.
Recall and Telescoping. Most of the surveys conducted among inmates ask respondents
to recall events since their initial incarceration. With such long periods of recall, it is
likely respondents forget details or telescope events by placing them in a more recent
time frame than they actually occurred. Most studies do not use techniques to help
inmates place the time of an event in context of other life events. This is particularly
important if a researcher wants to establish a prevalence rate that may refer to a given
incarceration or specific time frame.
Interview Modes. To date, most of the studies have used either interviews or paper and
pencil self-administered questionnaires to record events. New methods are now available
that allow a respondent to answer questions through a computer assisted survey format
(CASI), or one that also includes an audio version where instructions and questions are
asked by the computer, and the respondent answers these questions directly into the
computer using touch screens (audio-CASI). This latter technique has been used in other
surveys of sensitive information (drug use, sexual behavior, legal abortions), and has
been shown to elicit more reliable and higher incidence response rates. The computer
intervention removes the shame and embarrassment of the interview setting, and helps to
insure the confidentiality of the response. There is another methodology called
randomized response that has also been used to insure confidentiality of response.
March 10, 2004
3
Neither method has been tried in the prison sexual victimization domain. The full report
covers the use of these methods to measure other sensitive and stigmatized behavior, and
evidence on the reliability and validity of these methods.
The Problem of Validity. Unlike some other assessments of sensitive and stigmatized
behaviors such as sexual practices and legal abortions, there is no way to directly
measure the veracity of the self-reported prison sexual victimization. We propose two
models that use other information about drug use, the level of blood borne infectious
disease, and the level of sexual victimization to try to establish the validity of the data at
the individual or institution level, after a large scale survey has been conducted. This
method will not provide an independent validity check on the actual proportion of sexual
victimization. It will, however, provide some assurance that the relative ranking of
prisons, from best to worst, has some validity.
Sample Size and Question Wording. If researchers are interested in completed or
serious victimizations, and these are relatively rare, the sample sizes needed to establish
the level of sexual assault in a particular prison will have to be fairly large and more
costly than if the study were designed to measure the jurisdictional level of victimization.
Adjustments to the Prison Rape Estimates and the Ranking of Problematic Prisons.
The legislation recognizes that to report the best and worst prisons, jails, and juvenile
facilities with respect to their ranking on sexual victimization, there will have to be some
adjustment in the rankings to “level the playing field.” For example, it is unfair to
compare prisons that contain different inmate security compositions. Adjusting the
victimization rates to make prisons appear equivalent is a technically difficult problem.
Since there are consequences to low rankings, the adjustments and resulting rankings will
also be controversial.
Summary. The task framed by the Prison Elimination Act of 2003 presents problems of
estimation, validity, and bias. The correctional setting amplifies the problems
encountered when researchers measure sensitive and stigmatized behaviors in the
community. Most of the literature has been concerned with adult prisons. While there are
difficulties encountered in prisons, there will be additional problems in jails and juvenile
facilities. Jails have high turnover rates. To get compliance from adolescents, in most
jurisdictions you need the consent of their parents. While the task is a formidable one, it
is worth the effort, even if prison rape is a relatively rare event. The data can be used to
raise or allay concerns depending on the results of the jurisdiction. The survey results can
be used to train staff and inmates. The data may lead to better classification of victims
and assailants which will help to reduce the level of sexual assault. The American
Correctional Association has already promulgated new standards that address prevention,
detection, and records collection associated with sexual assault. Because there is no
validity check on the outcomes, there will probably always be some controversy
associated with the results of a facility-based estimate. The adjustments to the estimates
required by the public law will probably amplify that controversy. Furthermore, there are
critics of correctional administration and some researchers who argue that prison sex is
part of a subculture of sexuality that is not commonly understood by most analysts doing
March 10, 2004
4
work in this domain. They argue that to fully understand the level of sexual victimization,
one must first understand the language and subcultural definitions used by the confined.
The data may also lead to a more objective understanding of the actual level of prison
sexual victimization that will either support or invalidate the assumptions inherent in the
Rape Elimination Act that make it appear prison rape is endemic in American
correctional institutions. However, since there is no independent assessment of the
validity of the self-reported incidents, there may well be dissatisfaction with the results of
a national probability assessment regardless of the outcome.
March 10, 2004
5
Prison Rape: A Critical Review of the Literature
Gerald G. Gaes and Andrew L. Goldberg
National Institute of Justice
March 10, 2004
The opinions expressed in this paper are those of the authors’ and should not be construed as the
opinions or the policy of the National Institute of Justice, the Office of Justice Programs, or the
U. S. Department of Justice. We want to express our appreciation to Tim Hughes for critiquing
and commenting on this manuscript.
Please cite this paper as Gerald G. Gaes and Andrew L. Goldberg. (2004) Prison Rape: A
Critical Review of the Literature, Working Paper, National Institute of Justice, Washington, D.C.
Prison Rape: A Critical Review of the Literature
Recent federal legislation has called for research and interventions to address the problem
of prison rape. This paper critically reviews the published research on prison sexual victimization
and places this research in the broader context of measuring sensitive and stigmatized behaviors.
The paper is intended to offer substantive suggestions on the best ways to measure the
prevalence and incidence of sexual victimization in prison, to explore problems that will be
encountered in assessing and interpreting results of a national survey of prisons and jails, and to
summarize the prior and current literature. While there are some similarities in measuring prison
rape and sensitive behaviors in the community such as abortion, drug use, and homosexual
behavior, the prison context also changes the nature of the measurement problem.
First, we review the findings and goals of The Prison Rape Elimination Act of 2003
("The Prison Rape Elimination Act of 2003," 2003(c)(3) Data Adjustments). The second section
of this paper discusses definitions of incidence and prevalence. Part of the confusion that arises
in representing the quantity and rate of prison and jail rape is that different authors have used
different definitions. The next section reviews the current literature on estimating the amount of
prison and jail victimization. There have been very few attempts to measure sexual victimization
in prison. Only the Bureau of Justice Statistics (BJS) has attempted to study this topic using a
national probability sample in the United States, although the Bureau of Prisons conducted a
national probability sample of prisons under its jurisdiction (Nacci & Kane, 1982, 1983, 1984).
There has also been a health survey conducted among inmates confined in New South Wales,
Australia that included sexual victimization, and was designed as a probability sample for that
jurisdiction (Butler & Milner, 2003). It is only with such a sample that we can ever attempt to
understand the scope of the problem. We summarize the level of sexual victimization by
March 10, 2004
1
reporting the results of a meta-analysis. The measurement problems for these surveys are
formidable given the stigmatization associated with sexual victimization, as well as the fact that
many of the previous attempts to measure victimization have resulted in large unit non-response.
This is the term of art used by survey statisticians when individuals refuse to participate in the
survey. This should be distinguished from item non-response which refers to questions that
respondents either intentionally or inadvertently fail to answer. We elaborate on these problems
in the paper. In the subsequent section, we review some of the literature on different modes that
have been used to elicit reporting of stigmatized behavior in different national probability sample
surveys that have been used to study sensitive behavior such as illicit drug use, abortions, and
homosexuality. In the following section on validity, there is a discussion of possible ways to
assess whether the self report data gathered from the surveys can be compared to some objective
measures to give us greater confidence in the veracity of the survey data. We then briefly cover
the problem of question wording, and relate it to whether the sample sizes will be large enough
to detect sexual victimization at the facility level. In the section on institution adjustments, we
review recent research that directly addresses the problem outlined in the legislation. The Act
requires the prisons to be rank ordered so that the institutions with the best and worst levels of
sexual victimization will be highlighted, but recognizes this requires a “level playing field.”
Certain types of prisons will have higher victimization rates because of their security status,
other dimensions of institutional operations, and the type and composition of the inmates. In the
last section of this paper, we summarize the problems and issues.
March 10, 2004
2
Federal Legislation
In the findings section of The Prison Rape Elimination Act of 2003 ("The Prison Rape
Elimination Act of 2003," 2003), the public law states that “Insufficient research has been
conducted and insufficient data reported on the extent of prison rape.” (p. 2). The findings
section of this bill also asserts that according to conservative estimates of experts, nearly 13
percent of the inmates in the United States have been sexually assaulted in prison. Under current
levels of imprisonment, this would imply that about 200,000 inmates now in prison have been
sexually assaulted. The findings also assert that prison staff are unprepared by their training to
“…prevent, report, or treat sexual assaults.”; that prison rape goes unreported; that prison rape
contributes to the transmission of infectious diseases such as HIV, tuberculosis, hepatitis B and
C; that rape victims pose a public safety problem because they are more likely to commit crime;
that the interracial nature of sexual assault causes racial tensions both in prison and in the
community; that rape exacerbates violence within prison; that members of the public and
government are largely unaware of the epidemic proportions of prison rape and the daily horror
of rape victims; that victims of prison rape are less likely to successfully reintegrate into their
communities upon release from prison; that the high levels of prison rape violates prisoner’s
rights under the Cruel and Unusual Punishment clause of the Eighth Amendment of the U. S.
Constitution; and that prison rape undermines other government efforts to promote public health,
public safety, salutary race relations, and economic sufficiency.
To address these issues the Act calls for a zero-tolerance standard; an effort to make this
a top priority in every prison system, national standards for the detection, prevention, reduction,
March 10, 2004
3
and punishment of prison rape; increased data on the incidence; standardization of definitions for
collecting data; a system that holds prison officials accountable to detect, prevent, and punish
prison rape; and a reduction in the costs of prison rape on interstate commerce.
The Act acknowledges that when the results of the survey of sexual victimization are
reported, the data must be adjusted so that prisons are judged on a level playing field. The
legislation also allows the Bureau of Justice Statistics, the agency responsible for obtaining
prison rape data, leeway in determining how rape should be defined, although clarifying
language under the Definitions Section of the bill are quite explicit. Rape means “…the carnal
knowledge, oral sodomy, sexual assault with an object, or sexual fondling of a person forcibly,
or against that person’s will (Sec.10.Definitions.(9)(A) ).” Fondling is defined as “…the touching
of private body parts for the purpose of sexual gratification.” (Sec.10.Definitions.(11) )
Defining Sexual Victimization – Prevalence and Incidence
There are two distinct ways in which researchers characterize the extent to which some
attribute is present in a population. These terms and concepts have been borrowed from the study
of disease. Epidemiologists distinguish between prevalence and incidence. Prevalence is the
total number within a population infected with a particular disease at a given point in time. As an
example, a researcher may be interested in the number of inmates currently in our nation’s prison
infected with HIV. Incidence refers to new cases of a disease or other phenomenon in a specified
period of time. Using the HIV example, researchers may try to estimate how many new cases of
HIV infection have occurred within a calendar year. Prevalence rates or percentages are
expressed as the number of cases of a disease present in a population at a particular time divided
March 10, 2004
4
by the total number of individuals in the population at that time. Incidence proportions are
expressed as the number of new cases, divided by the number of individuals at risk during that
time period.
Because most diseases have a limited duration, or they kill their host, incidence and
prevalence rates can indicate very different pictures of the disease cycle. For example, If a
disease has a long duration (such as HIV), and it was spread widely in 2002, even if the number
of new cases has declined dramatically, there will still be a high prevalence in 2003. Conversely,
a disease that is easily transmissible, but has a short duration may have a high incidence and low
prevalence. Sometimes epidemiologists distinguish between first and total incidence. First
incidence is the initial occurrence of an event such as the first occurrence of a cold, an accident,
or a rape. There are of course some phenomena which can only have a first occurrence. There is
no second incident of an HIV infection. Total incidence allows the individual to be counted more
than once during some time frame. In a given year, if we count the number of rapes, and if
someone is raped more than once that person is counted each time the assault occurs. Total
incidence is also sometimes called the attack rate. There are actually comparable prevalence
concepts. Lifetime prevalence is the number of people who have ever experienced an event.
Point prevalence is the number of people who have experienced an event in a given time frame.
Point and lifetime prevalence have both been used to characterize prison rape without explicitly
recognizing the difference in meaning of the two statistics.
When these concepts are translated into areas of interest outside of disease, we must
recognize that the phenomenon we are dealing with may not have a cycle of infection. This is the
case with sexual victimization. Someone who has been raped will always have that attribute.
Nonetheless it is still important to distinguish between incidence and prevalence for the same
March 10, 2004
5
reason it is important in the epidemiology of disease. A prevalence rate may be high at the same
time the number of new cases (incidence) is declining, or equally compelling, prevalence may be
low while incidence is increasing.
Incidence rates are bounded by a time frame. How many new cases have there been in
one week, one month, one year? Prevalence rates of rape are not typically time bounded, but as
indicated above they can be by collecting data to provide point prevalence. One could ask a
sample from the current population of all prisoners whether they have ever been sexually
assaulted in prison – the lifetime prevalence of sexual assault in prison. Or one could ask
whether they have been sexually assaulted within there current period of incarceration – a point
prevalence estimate, or as the legislation requires, whether they have been assaulted in the last
year.
To compare incidence or prevalence over time or across populations, the statistics must
be expressed as a proportion or a rate. Thus, one must have an appropriate denominator. In the
case of prevalence, the denominator is the number occurring within the population during the
reference time period. For example, one might interview every inmate in a county jail on a given
day and the denominator is the number of prisoners in that jail. Incidence proportions or rates
require that the analyst be able to count the potential number of people exposed to the risk of
sexual assault. For a specific prison, if we were to count all instance of rapes occurring in
calendar year 2002, we would have to start with everyone in that prison on January 1, 2002, and
also count every new prisoner admitted to prison during that one year period. Every one of those
inmates would be potential targets of a sexual assault during that time period.
One other distinction that should be clarified is the rate of victimization versus the rate of
incidents. The former refers to the number of persons who have been victimized. The latter refers
March 10, 2004
6
to the number of times a sexual assault has occurred. If one person is raped 10 different times,
then there will be a single victim and 10 sexual assaults.
In addition to the technicalities of measuring prevalence and incidence, researchers must
establish a coherent, consistent definition of the phenomenon. In Table 1, we have recorded the
different definitions of sexual victimization researchers have used along with other information
on each study. There is language in the Prison Rape Elimination Act of 2003 that can also be
used to guide definitions. The operational definitions should distinguish completed sexual
assaults, from attempted victimizations, from pressure to engage in sex. We cover this issue in
detail after the review of relevant research.
Prison Rape Literature
In this section, we review studies that have been conducted to ascertain the level of prison
sexual victimization. The Bureau of Justice Statistics reported that there were 1,668 state and
federal prisons in the United States in the year 2000 (Stephan & Karberg, 2003), and with the
exception of one attempt at a national probability sample by BJS, the sexual victimization studies
have been conducted in fewer than 40 – about 2.4 percent of all prisons in the U.S. While the
Prison Rape Elimination Act calls for an assessment in jails as well, only Davis’ (1968; 2000)
study was conducted in a jail. In 1999, there were 3,365 jails operating in the United States
(Stephan, 2001). The studies listed and described in this section are also summarized in Table 1.
Studies Involving Primarily Men, or Men and Women
March 10, 2004
7
Struckman-Johnson and Struckman-Johnson and Colleagues
Two of the studies cited most often by advocates of legislation and/or policy to reduce or
eliminate prison rape were those conducted by Cindy and David Struckman-Johnson and their
colleagues (Struckman-Johnson & Struckman-Johnson, 2000; Struckman-Johnson, StruckmanJohnson, Rucker, Bumby, & Donaldson, 1996).
The procedures used in both studies were quite similar. The first study was conducted in
1994 using inmates under the custody of the Nebraska Department of Corrections housed in four
institutions. The second study was conducted in 7 prisons in the Midwest. The Midwestern states
were not identified. In both studies, paper and pencil surveys were distributed (mailed or hand
delivered) to the inmates who then completed them at their convenience. In the Nebraska study,
the institutions were two maximum and one minimum security men’s facility as well as one
women’s prison. In the “Midwest” study, all of the prisons housed male inmates. Many of these
facilities had mixed custody responsibilities housing maximum, medium, and minimum security
inmates in different housing units. There was also a long-term maximum security segregation
facility and a minimum security facility included in the study. The total number of available
participants in the first study was 1,801 inmates (1,708 men and 93 women) and 714 staff. In the
second study, the total available sample was 7,032 male inmates and 1,936 security staff. The
distribution of surveys was accompanied by informed consent forms. In the Nebraska study,
28.7 percent of the inmate surveys were returned and usable (n=516) and 37 percent of the staff
surveys were returned and usable (n=264). In the Midwest study, 25 percent of the inmate
(n=1,788) and 25 percent of the staff (n=475) surveys were returned and usable.
March 10, 2004
8
To obtain an estimate of the prevalence of sexual victimization, inmates were asked “In
the prison you are in now, about what percentage of inmates do you think have been pressured or
forced to have sexual contact against their will?” The choices available to the respondents were
0%, 1%, 5%, 10%, and then increments of 10% up to 100%. Both staff and inmates were asked
to estimate this prevalence and inmate estimates averaged 19 percent. Staff estimates were, on
average, 15 percent. This varied by facility. In men’s maximum security facility A, the respective
estimates were staff 19 percent, inmates 19 percent. In men’s maximum facility B, the estimates
were staff 16 percent, inmates 26 percent. In a men’s minimum security facility, the estimates
were staff 11 percent, inmates 16 percent. In the women’s facility, the estimates were staff 8
percent, inmates 3 percent.
The key question in the 1996 study was “Since the time you have been in a Nebraska
prison, has anyone ever pressured or forced you to have sexual contact (touching of genitals,
oral, anal, or vaginal sex) against your will? A person could respond “yes,” “no,” or “not sure.”
The estimate of victimization based on this question was 20 percent (104 out of 516
respondents). There were 3 women out of 42, or 7.1 percent and 101 men out of 474, or 21.3
percent who responded “yes.” For inmates who answered “yes” to this question, a skip pattern
probed about details of the victimization. These victim targets indicated they had experienced, on
average, 9 episodes of pressured or forcible sex. Of the 101 men targeted, 51 percent were
victims of anal sex; 8 percent were victims of oral sex. The perpetrator was described as a staff
member 18 percent of the time; however, most of the time the perpetrator was an inmate. Of the
three women, two had been fondled and one groped. There were 75 inmate victims who provided
descriptions of their sexual coercion. Fifty five percent of these were completed anal, oral, or
vaginal sex that was forced by one or more perpetrators. Five percent were pressured into
March 10, 2004
9
completed sexual victimization. The remaining were either incomplete sexual acts or involved
only touching. Only 29 percent of the male victims said they disclosed the worst incident to staff
at the institution. Based on data reported in Table 3 of their study, there were 76 men and one
women who reported attempted or completed oral, vaginal, or anal sex. These data allow the
construction of an estimate with sexual pressure excluded. Based on data in this table, 16 percent
of the men and 0.23 percent of the women were sexually victimized. This is the number we use
in our meta-analysis. The other cases were instances of sexual pressure or “unknown.”
In the 2000 study, the same key questions as indicated above were asked to establish
estimates of sexual coercion. The overall victimization percentage was 21 and the facility by
facility prevalence estimates appear in Table 1. They vary between 4 and 41 percent depending
on facility, and whether the estimate was made by staff or inmates. In Table 1 of that study, the
researchers list different categories of assaults based on rape, force, or pressure. Row 7 of the
table calculates the percentage of inmates who reported a worst-case incident of rape for each of
the 7 prisons. The weighted average prevalence for the 7 prisons is 7.6 percent. For this study,
this is the result we use in our meta-analysis.
Other than Wooden and Parker (1982), the Struckman-Johnsons and their colleagues
have recorded and reported the highest levels of sexual assault. This may be because the
particular institutions they evaluated had high rates of assault, or it could be due to their
particular methodology. In both of these studies, questionnaires were sent to the prisons and
distributed by staff. Each inmate could mail the completed questionnaire back in a pre-addressed,
postage paid envelope. This is probably not the most reliable way to collect these type of data.
Inmates who participated in this study could have discussed the survey with each other before
returning the questionnaire. Unfortunately, allowing inmates to fill out these self administered
March 10, 2004
10
questionnaires back in their cells, or at their convenience, also raises suspicions about collusion
and lack of independence in filling out the answers, especially when one considers how low the
response rates were in these studies. In a later section of this paper, we discuss in more detail the
low levels of response for many of the studies reviewed in this paper.
The Davis Report on Philadelphia Jails in 1968
Alan J. Davis, a Chief Assistant District Attorney conducted an investigation of the
Philadelphia Jail system as a result of allegations of widespread sexual victimization (Davis,
1968). Davis, along with the cooperation of the Philadelphia police interviewed 3,304 inmates,
and 562 employees. These investigators took 130 written statements and 45 polygraph tests. The
investigation lasted from June 1, 1966 though July 31, 1968. A sample of about 5 percent of the
approximately 60,000 inmates passing through the system during this time period was identified
for participation. Of the 3,304 interviewed inmates, 97 victims were identified, or 2.9 percent.
Since some of these inmates were repeatedly victimized. There were actually 156 separate
assaults involving 176 aggressors. Interviews were conducted during a two week period, July 15,
to July 31, 1968 in three separate Philadelphia facilities. There were also 561 staff interviews in
these three institutions.
Davis expresses his findings sometimes as if they were incidence rates and sometimes as
if they were prevalence rates. The 97 documented victims represent a prevalence rate of 2.9
percent of the total number of inmates interviewed. Thus, we can see, of the inmates in the
Philadelphia jail system in this time period, 2.9 percent had a documented sexual victimization at
some point in their jail experience. Later in the report, Davis discusses the reluctance of inmates
March 10, 2004
11
to talk about a sexual victimization. He makes a back-of-the-envelope estimate, and conjectures
that there were 2,000 sexual assaults in a 26 month time frame. This would yield an incident rate
of about 3.3 percent (2,000/60,000) although this estimate includes repeated victimizations of the
some of the same inmates. Therefore, this is an incidence estimate of sexual assaults, not sexual
assault victims.
Lie detector tests were given to some victims and some assailants and certain staff
informants. All but one of the 26 staff who was asked to take a polygraph test refused. Of the 48
inmates who were asked, 7 refused. Ten of the 41 inmates who submitted to a polygraph test
showed indications of deception in their narration of the facts. This was the only study to attempt
a validity check of self-reported sexual victimization.
According to Davis, the likelihood of being approached for sex or sexually assaulted in
the Philadelphia Correctional System if you were small in stature was almost certain.
Sexual assaults are epidemic in the Philadelphia Prison System. Virtually every slightly built young man
committed by the courts is sexually approached within a day or two after his admission to prison. Many of
these young men are overwhelmed and repeatedly “raped” by gangs of inmate aggressors. Others are
compelled by the terrible threat of gang rape to seek protection by entering a “housekeeping” relationship
with an individual tormentor. Only the toughest and more hardened young men – and those few so
obviously frail that they are immediately locked up for their own protection – escape penetration of their
bodies. (Davis: 17)
Davis goes on to recount narratives of sexual victimizations to sensitize the reader to the
“raw, ugly, and chilling” (Davis: 18) nature of these assaults.
The facts that were gathered about sexual assault were documented by institutional
records, polygraph examination, or other methods. Davis argues that the true victimization rate
March 10, 2004
12
was much higher than reported because inmates refused to cooperate in the investigation. Davis
attributes this lack of cooperation to fear of retaliation, the shame associated with disclosure, and
the mistrust of prison and criminal justice officials. Davis also argues that much of the putative
consensual homosexual sex that occurred was actually a continuation of the victimization of
intimidated inmates.
Nacci and Kane , 1982 – a Study of the Federal Bureau of Prisons
Nacci and Kane have several reports on a study of sexual victimization conducted within
the Federal Bureau of Prisons (Nacci & Kane, 1982, 1983, 1984). This study was prompted as a
result of an unusual number of homicides at one the Bureau’s federal penitentiaries, and there
was some evidence to indicate that many homicides were related to consensual and
nonconsensual sexual activity.
This is one of the few studies on the topic with a sound approach to the sampling
methodology in its attempt to draw a sample representative of all inmates under the custody of a
jurisdiction. A two stage probability sample was used to first randomly select 17 prisons from the
federal system, and then randomly select 330 inmates from those prisons proportional to the
institution population. In addition, every correctional officer in each of the 17 prisons was
eligible to complete a survey as well. The inmate survey, however, was an interview conducted
by “ an articulate, black ex-offender (Nacci & Kane, 1982: 2).” The inmate surveys contained
over 300 items including questions of sexual victimization. Of the inmates contacted, 64 percent
decided to volunteer for the interview. Given the sensitivity of the questions, this is a respectable
response rate. In their 1982 paper, Nacci and Kane (1982), reported that the respondents were
March 10, 2004
13
similar to non respondents with the exception that there were more African Americans and fewer
whites among the participants. The authors also reported high reliabilities for the survey items
and a coherent structure to the items. Although the 1982 paper did not include any statistics to
support those contentions, the authors mentioned they conducted factor, cluster, and reliability
analyses of the data.
In their 1982 report, Nacci and Kane report a number of conclusions, but do not provide
data or statistical evidence to support these assertions. Using the Kinsey study (Kinsey, Pomeroy,
& Martin, 1948) as a baseline, they compare prisoner interview response to the sexual practices
of men documented in the Kinsey study. Prisoners were less accepting about certain sexual
practices than the Kinsey sample regarding mate swapping and homosexuality; however, the
prisoner respondents were more accepting of sex before marriage, and “exotic” sexual practices
than the Kinsey sample. A number of questions were asked about sexual behavior while in
prison. Among the interviewees, 28 percent stated they had a homosexual experience some time
in their lives; 25 percent had homosexual experiences as an adult; 12 percent had a homosexual
experience in their current institution; 20 percent of inmates housed in the highest security levels
claimed they had a homosexual experience in their current institution; 29 percent had been
propositioned for sex in their current institution; 7 percent were “seduced” by inmates bearing
gifts; 2 percent had taken money for performing sex; and 1.8 percent were in a long standing
love relationship.
Victimization was evaluated by a response to a question about whether anyone had
forced or attempted to use force to get the inmate to perform sex against his will. In response to
this question, 9 percent had been targeted sometime in their prison career (state or federal
March 10, 2004
14
prisons), 2.0 percent were targeted in a federal prison, 0.6 percent had to perform an undesired
sexual act in a federal prison, and 0.3 percent had been raped.
Saum, Surratt, Inciardi, and Bennet, 1995
Saum, Surratt, Inciardi, and Bennet (1995) conducted a study of sexual victimization in
the Delaware prison system among inmates in a therapeutic community (TC) located in one of
the state prisons. The authors claim that since the treatment staff had an excellent rapport with
the inmates, this would promote more honest responding to sensitive questions. Of the 106 TC
inmates who had been in the program 30 days or more, 101 volunteered to participate in an
interview. The inmates were asked to respond to questions about sexual activity in prison that
had occurred prior to their assignment to the TC. They were asked to report on sexual activities
that they had witnessed, heard about, or participated in during the previous year of their current
incarceration. There were about 1,350 inmates housed in this institution at the time, and the
authors of this study made no attempt to show the extent to which TC inmates were
representative of the entire population. While there was a high degree of cooperation among the
TC inmates, it is not known what degree of cooperation the researchers would have gotten had
they tried to interview a sample from the entire population.
About 51 percent of the inmates reported having heard about consensual sex in the prior
year and 25 percent claimed to have witnessed consensual sex. Almost 60 percent of inmates had
not heard about a rape occurring in the prior year, 3 percent said they had seen one rape and 1
percent had seen two rapes. When asked to estimate how often rape occurred, almost 30 percent
March 10, 2004
15
said once a month. When asked to report their own victimization, only one inmate reported
being raped, while 5 inmates indicated there had been a rape attempt.
Tewksbury, 1989
Tewksbury (1989b) conducted a study in the Lebanon State Correctional Institution in
Ohio. He was primarily interested in more generic sexual activities of inmates, but his self
administered questionnaire (SAQ) also included items about coerced sexual behavior. He
collected 150 surveys but does not report how many inmates were housed at this particular
institution. He recruited inmates in many different ways and does not discuss any kind of
sampling strategy. Because one of his recruiting strategies was to enlist inmates after a college
program class, it is not surprising that 84 percent of the survey respondents were enrolled in a
college program while only 19.5 percent of the inmate population was enrolled in a college
program. The research sample also over-represented inmates who were white and those who
were never married.
The SAQ included the item, “How many times have you been raped in this prison?’ No
one reported a rape. Another question asked about coerced sex, “While in this prison, how many
times has another male tried to have sex with you using threats or force?” Among the
respondents, 4.5 percent answered affirmatively and 7.4 percent of those inmates indicated this
had happened one or more times. Inmates were also asked to make estimates of the percentage
of inmates in this prison who had been sexually assaulted. Inmates estimated this number to be
14 percent.
March 10, 2004
16
Tewksbury (1989a) also wrote a second paper on this same sample in which his primary
concern was an analysis of the fear of sexual assault. Taller inmates were less likely to perceive a
threat; however, heavier inmates were more likely to fear a sexual assault. Whether or not the
inmate’s incarceration crime was violenct had no impact on perceived fear. Neither did the
inmate’s race, number of friends in the institution, or religion.
Maitland and Sluder 1996, 1998
Maitland and Sluder (1998) assessed all forms of victimization (assault, sexual assault,
threats) in a Midwestern, medium security prison that had an average daily population of 1,100,
and held primarily youthful inmates. They used a self administered questionnaire. Unfortunately
they did not use probability sampling techniques in soliciting volunteers for their study. A total
of 111 inmates completed a survey out of 150 who attended classes on the day of the study.
Although they had a response rate of about 74 percent, there was little analysis to compare the
student inmates to the remaining population. The two variables they report indicated their sample
had the same racial composition as the overall population, but was somewhat younger. The SAQ
included a victimization question, “During this sentence has anyone forced sexual activity on
you? “ Only 0.9 percent of the respondents said “yes.” A second question asked whether
“…anyone made sexual comments to you that made you feel uncomfortable (Maitland & Sluder,
1998: P. 63, Table 3)?” Of the 111 respondents, 16.2 percent answered yes to this question.
Maitland and Sluder also used the survey to measure the General Well-Being scale, an
index to measure fear of victimization, a prison stress scale, indicators of social support, a scale
assessing anomie, and gang affiliation. Administrative data were also collected including age,
March 10, 2004
17
race, height, weight, IQ, education level, marital status, number of incarcerations, number of
times on probation, length of current sentence, and current offense. Maitland and Sluder report
that they used these measures in bivariate analyses to distinguish victims from nonvictims. Since
there were so few victims of forced sex, no statistical distinction could be made. Maitland and
Sluder also reported that these variables did not distinguish between victims and nonvictims with
respect to sexual comments either.
This study was also reported in a Federal Probation article (Maitland & Sluder, 1996). In
that article, the General Well Being scale was considered the dependent variable and
victimization and other covariates were the independent variables. In a multivariate analysis of
the data, Maitland and Sluder found that the more an inmate was victimized, the lower his sense
of general well being. The variable was a composite measure of all types of victimizations, and
therefore, there was no independent test of sexual victimization, although victimization of any
type would probably lower general well being.
Wooden and Parker, 1982
Wooden and Parker (1982) investigated sexual behavior and victimization in a California
medium security prison during 1979 and 1980. The prison held 2,500 inmates. This prison was
used to house self-avowed homosexual and vulnerable inmates in single cells. The researchers
distributed over 600 questionnaires to a random sample of inmates and received back 200. Of
those, 14 percent claimed to have been the victim of sexual assault, although the definition also
included pressure to engage in sexual activity. This is an unusually high level of sexual
victimization in comparison to most studies, but may be explained by the question wording and
March 10, 2004
18
the unusual composition of the inmate population. The researchers also distributed a
questionnaire and interviewed a targeted subpopulation of self-defined homosexuals. Employees
were also interviewed. One of the researchers was serving a four-year sentence at the facility.
The co-researcher participated in the study by distributing questionnaires and by participant
observation.
The authors note in the introductory chapter to their book that this prison was atypical
and may not represent a true picture of inmate behavior that would generalize to the California
Department of Corrections population at that time. Although they also suggest that victimization
may have been higher in maximum security prisons than the medium security prison in their
study. The major problem with this study is that it was conducted in a prison in which a large
number of homosexuals were placed and this could bias the results in favor of finding a
victimization effect.
Lockwood, 1980
Lockwood (1980) studied sexual victimization in the New York State prison system in
1974 and 1975. The primary intent of his book, Prison Sexual Violence, was to describe the
characteristics of sexual assault and sexual pressure by interviewing inmates and analyzing
background data of 107 targets of sexual aggression and 45 sexual aggressors. Lockwood also
conducted a small study of 89 randomly sampled offenders in two New York State prisons of
which 15 percent refused to participate. Only 1 of the 76 respondents had been sexually assaulted
while 28 percent had been the target of sexual pressure. While the rate of sexual victimization in
the New York State Department of Corrections may be rare, when Lockwood focuses on the
March 10, 2004
19
actual incidents, over 1/3 of the 148 he analyzed involved violence. Most of the aggressors were
black and the targets were white. There was no analysis of the representativeness of the sample.
Toch, 1977
Toch (1977) discusses victimization in the New York State prison system in his book
Living in Prison: The Ecology of Survival. In the chapter on inmate victimization, he notes that
the extreme form of victimization is rape, but that it is very rare in prisons. He reports that 28
percent of the inmates he interviewed were victimized. This is the same proportion as
Lockwood’s study, and it is not clear whether the two books cover the same sample. The
characteristics about the incidents reported by Toch indicated that most of the victims were white
and most of the aggressors were black. Aggressors had assault histories. There is no report of a
rape and this is different from Lockwood’s sample where he reported one sexual assault. Toch
uses a lot of inmate narrative to characterize the social and cultural context of sexual aggression
and the fear these incidents evoke in many of the threatened sexual targets. Because it is unclear
whether the Toch data represent a new sample, his study was left out of summary Table 1.
Hensley, Tewksbury, and Castle, 2003
Hensley, Tewksbury, and Castle (2003) interviewed 174 Oklahoma inmates who agreed
to participate in their study from a random sample of 300 (100 inmates each were randomly
selected from a minimum, medium, and maximum security institution). Thus, the response rate
was 58 percent. The study was conducted from August 1998 to May 1999. Whites were
March 10, 2004
20
underrepresented at each security level and Native Americans were overrepresented in the
minimum security sample. There is no other assessment of sample representativeness. A 44-item
interview instrument was used; however, the exact wording of the interview items does not
appear in the journal article. The inmate was asked if he had been sexually threatened or sexually
assaulted. Two of the inmates who reported they had been sexually threatened (13.8 percent of
the sample) said that they had been raped (1.2 percent). The question appears to have been
unbounded. Inmates were to indicate if they had ever experienced sexual victimization at any
point in their current or any previous incarceration. The targets were more likely to be
homosexual or bisexual than the remainder of the sample who were not sexually coerced. The
targets were young (median age 18.5 years) and were approached, on average, 143 days after
their incarceration.
Carroll, 1977
Carroll’s (1977) study of Eastern Correctional Institution, a maximum security state
prison of only 200 prisoners was more an ethnography than a systematic assessment of the level
of sexual victimization. Carroll spent a 15-month period in 1970 and 1971 doing participant
observation of the facility. He used unstructured interviews and many conversations to get some
idea of the level of victimization and the antecedents of these actions. Carroll estimated there
were 40 or more sexual assaults per year, but he does not say whether these were repeated or
new victimizations. This translates into a one-year prevalence of 20 percent if these assaults
involved different victims. Carroll states that he never observed an act of sexual victimization,
and that these were informant reports that were not verified. If true, this would be the highest
March 10, 2004
21
estimate of sexual assault prevalence found in any study. Carroll’s paper focuses on the
interracial nature of the assaults (black on white) and the historical, social, and cultural
antecedents of the prison setting that led to the asymmetrical racial nature of sexual predation in
this prison. There is no way to know whether or not these social facts were unique, or whether
his finding is spurious because it is based on exaggerations of informants.
Chonco, 1989
Chonco’s (1989) study is a descriptive analysis of sexual assault in what was
characterized as a pre-release center in a large Midwestern state. He interviewed 20 white, 19
black, and one Mexican-American inmate, each interview lasting one to one and a half hours.
Chonco’s main interest was to explain why African American inmates chose whites as their
victims in a sexual assault. Various explanations have been offered based on cumulative
discrimination, the rage resulting from psychological emasculation, the perception that whites
are weak and sexually attractive, and the result of having been arrested, tried, convicted and
confined in a white dominated criminal justice system. Since there are no estimates of
victimization contained in this paper, it does not appear in Table 1. Chonco argues that perceived
weakness and naivety, rather than race was a more salient factor leading toward victimization.
He describes how potential targets are set up, probed, and tested prior to an assault. His
informants claim that inmates were afraid to report assaults once they had occurred for fear of
retaliation, and that correctional officers would only intervene if they had witnessed an attack.
Moss, Hosford, and Anderson, 1979
March 10, 2004
22
Moss, Hosford, and Anderson (1979) studied the characteristics of 12 individuals out of
1,100 at a federal correctional institution who were segregated for having raped other inmates.
This data, had it included new admissions during the year would produce a one-year incidence
rate. The rate is no higher than 1.1 percent (12/1,100). However, there would have been a sizable
number of admissions during the year and the denominator should be larger than the number of
inmates on a given day. While these authors used a discriminant analysis to compare rapists and
a randomly selected group of non-rapists, the samples were so small that the results are not
reliable. This is one of the few studies that used administrative records to isolate known
aggressors. However, it is not clear when these inmates had committed their assault. The 1.1
percent is an upper bound of the one-year incidence estimate and the actual estimate is probably
much lower since the denominator should be higher and these assaults may have occurred in
previous years and in other institutions.
Fuller and Orsagh, 1977
In 1977, Fuller and Orsagh reported on a study of general victimization within ten
institutions of the North Carolina prison system (Fuller & Orsagh, 1977). They used three
sources of data. The first was based on administrative records collected from the 10 prisons in
the last quarter of 1975. The second source was personal interviews with the superintendents of
these institutions. The last source was a stratified sample of 400 inmates drawn from six of the
prisons. The main purpose of the study was to assess overall levels of victimization, and to this
end, the authors compared the three sources of information. For all types of victimization, the
March 10, 2004
23
administrative records indicated a 1.7 percent incidence, the superintendent interviews indicated
a 3.4 percent incidence, and the inmate self-reports revealed a 5.8 percent incidence. There was
one instance of rape according to the administrative records; however, the superintendents
indicated there were 31 incidents of sexual assault in the last year. Fuller and Orsagh incorrectly
computed the incident rate for the year as 31 incidents divided by the standing population of
4,495 inmates, or .69 percent. As we explained at the beginning of this paper, an incident rate
must be based on the population at risk, and because prisons have many transfers the at-risk
population consists of the standing population at the beginning of the year and all admissions
during that year. There was no indication they had assessed sexual victimization with their
inmate self-report method. It appears the definitions used in assessing the administrative records
may have been a completed rape, while the administrators were estimating assaults. Furthermore,
it is not clear what the superintendent’s subjective estimate of inmate sexual victimization
means. Was this a guess, speculation, or was it based on some alternative data source? While we
have included this study in Table 1 as a study of incidence, we did not include it in the metaanalysis of prevalence we report on later in this paper.
Butler, Donovan, Levy, and Kaldor 2002; Butler and Milner, 2003
As part of a broader health survey, inmates in New South Wales, Australia were
interviewed about sexual practices in 1996 and 2001. Data from 789 inmates collected in 1996
indicated that 2 percent of women and 2 percent of men had engaged in nonconsensual sex
(Butler, Donovan, Levy, & Kaldor, 2002). The results reported from the 2001 survey were even
lower (Butler et al., 2002; Butler & Milner, 2003). A cross-sectional random sample of inmates
March 10, 2004
24
stratified by age, sex, and Aboriginality was conducted. The sample represented 10 percent of
the men and 34 percent of the women held in full-time custody. The survey was conducted by
using nurses who were not assigned to a given facility to interview inmates. During this
assessment, blood and urine samples were drawn, and in addition to health measures, a mental
health assessment was also taken. This latter testing was done by psychology masters degree
students. The response rate for the 2001 survey was 85 percent. The sample of men was 745 and
for women it was 163. Within these groups, 0.4 percent of males and 1 percent of females
reported nonconsensual sex within the last year. Inmates were also asked if they had ever been
sexually harassed or threatened with sex by another inmate. The percentages responding
affirmatively were almost equivalent for men, 4.6 percent, and women, 4.7 percent. The majority
of these cases involved verbal harassment only. These same inmates also reported on sexual
assaults of other inmates. Among women, 23 percent indicated this happened, while 15 percent
of the men said that they were aware of sexual assaults of other men in the previous 12 months.
This seems typical of studies which ask about personal victimization, as well as the perception of
the extent to which sexual victimization occurs. The latter estimate is typically much larger than
the disclosure of a personal victimization. These latter estimates are always higher than the self
reports, and it is unclear what they mean since there is no presumption that inmates or employees
actually witness all of the sexual assaults they claim are occurring. One might argue that victims
are ashamed or fearful of reprisals. Or, one might surmise that estimates of someone else’s
victimization exaggerate the actual occurrence.
Forst, Fagan, and Vivona, 1989
March 10, 2004
25
The only published study on juvenile sexual victimization that we found was primarily
designed to contrast the experience of youth in training schools as opposed to those who were
sent to prison as a result of transferring the case to criminal court (Forst, Fagan, & Vivona,
1989). The sample was drawn from youth adjudicated for violent offenses in four urban juvenile
courts. There was no description or statistical analysis on the extent to which this sample was
representative of the violent juvenile population in these jurisdictions. There were 59 youth
adjudicated in juvenile court who were subsequently sent to the traditional training school. These
juveniles were, on average, 15.7 years old at the time of their offense. Training school
participants in the study were interviewed when they were released. They had spent, on average,
2.9 years at the training school. There were 81 youth from these same jurisdictions who were
transferred to criminal court, and who were sent to a state prison as a result of a conviction. This
group averaged 16.1 years at the time of their offense. They were interviewed in prisons where
they had served, on average, 1.8 years of a 29 year average sentence. All of the participants in
the study were asked questions about staff assistance, case management services, the social
climate of the facilities, and victimization experiences. For the purpose of this review, the
important question was “Has anyone attempted to sexually attack you or rape you?” Among the
training school sample, the prevalence was 1.7 percent, while the prevalence for the prison
sample was 8.6 percent. Because this is a unique study of youth victimization, it is not included
either in Table 1 or the meta-analysis we report later. It also suggests that youth may be
particularly vulnerable in adult institutions.
Studies Involving Exclusively Women -- Coerced Sex among Women
March 10, 2004
26
Although there have been only a handful of studies on sexual victimization of men, there
are even fewer studies regarding women. Most of the research on women’s sexuality in prison
has been on consensual behavior (Ford, 1929; Giallombardo, 1966; Greer, 2000; Halleck &
Hersko, 1962; Heffernan, 1972; Hensley, Tewksbury, & Wright, ; Mitchell, 1969; Nelson, 1974;
Otis, 1913; Owen, 1998; Propper, 1978; Propper, 1981; Propper, 1982; Selling, 1931; Ward &
Kassenbaum, 1964, 1965). In some of this research, there is an indication of subtle coercion and
cooptation, the fuzzy gray area between consensual and coerced sex inside of prison. In addition
to the study of one women’s prison cited in Struckman-Johnson et al., (Struckman-Johnson et al.,
1996), in which 3 of 42 women (7 percent) were either groped or fondled, Struckman-Johnson
and Struckman-Johnson (2002) report on 3 additional women’s prisons. This was actually data
collected at the same time these researchers were collecting information in 7 male prisons. The
same procedures were employed and the same questions elicited the victimization results. As
depicted in Table 1, women inmates responded to the question “Since the time you have been in
a Nebraska prison, has anyone ever pressured or forced you to have sexual contact (touching of
genitals, oral, anal, or vaginal sex) against your will?” (Responses were “yes,” “no,” or “not
sure”). In institution 1, 27 percent responded yes; in institution 2, 9 percent reported
affirmatively; and in institution 3, the “yes” response was 8 percent. The worst incident resulting
in a rape occurred among 5 percent of the women in the first institution; however, there were no
reported rapes in the second or third prisons. When asked to estimate how many women were
forced or pressured into sex in their current facility, the institution percentages were 21, 11, and
13. There was no information on the representativeness of these three samples.
As reported above in the Butler and Milner study in New South Wales (Butler & Milner,
2003), 1 percent of the women reported nonconsensual sex within the last year. In another study,
March 10, 2004
27
Alarid (2000) used letters from an inmate in a southern jail/prison system who kept in weekly
contact with her. Excerpts from these letters indicated that sexual assaults that did occur were
rarely reported. Secondly, the inmate informer asserts that most women capitulate to sex through
pressure, and it appeared that rape was not very common.
U. S. National Probability Sample of Rape during Adult Incarceration
In the 1997 Survey of Inmates in State Correctional Facilities (Bureau of Justice
Statistics, 1997), there were several questions about sexual victimization prior to the current
incarceration. If inmates indicated there had been sexual contact against their will, they were
asked a series of questions including whether or not they were raped and whether the incident
took place while incarcerated. To establish this prevalence, one has to be able to compute, the
proportion of the sample with a prior incarceration to get an appropriate denominator. In fact,
55.1 percent of the sample had a prior incarceration. With the appropriate denominator, 0.45
percent of males and 0.35 percent of females reported a completed rape in a prior incarceration.
This information was elicited in the BJS survey by an interviewer. To minimize the potential
sensitivity of the question in the context of the current prison term, inmates were asked about
incidents prior to their current incarceration. These data represent population estimates of the
entire prison system at the time of this survey. There were 1,409 State prisons included in the
sample design. The sample was conducted in two stages. In the first stage, 275 prisons were
sampled. In the second stage, inmates were sampled. The 13 largest male and 17 largest female
prisons were sampled with certainty. The remaining prisons were stratified by region, facility
March 10, 2004
28
type (confinement and community based), security level, and size of the population. A total of
14,285 prisoners were interviewed in the state survey with a 92.5 percent response rate.
U.S. National Probability Sample of Forced Sexual Activity among Youth in Juvenile
Facilities
The Prison Rape Elimination Act also calls for investigation of sexual victimization of
youth. There has already been a great deal of work sponsored by the Office of Juvenile Justice
and Delinquency Prevention (OJJDP). The Survey of Youth in Residential Placement (SYRP) is
a 10 percent national probability sample of youth, ages 10 to 20, living in juvenile facilities
because they are accused or convicted of a crime (Office of Juvenile Justice and Delinquency
Prevention, 2004). OJJDP insured that SYRP was designed to address a variety of issues
involving youth in these facilities including their needs, services, safety, health, security,
accountability, and expectations for the future. SYRP was developed as an audio-computer
assisted survey instrument (audio-CASI). We discuss the merits of this survey methodology later
in this paper. More than 7,000 juveniles were interviewed, and the response rate was 75 percent.
The administrators of the survey took precautions to insure confidentiality.
The SYRP approach to eliciting information on sexual victimization was to start with a
global question, “Since you have been in this facility, has anyone forced you to engage in sexual
activity.” If the answer was “Yes,” then the victim was asked “How many times has this
happened? Please enter a number” and “Who did this to you?” with possible responses “Another
March 10, 2004
29
resident that I know,” “Another resident that I don’t know,” “A staff member of the facility,”
“Someone else.” Respondents could choose more than one assailant type. Youth were also asked
whether a weapon was used during the threat, and if so, what type of weapon. The victim was
also asked, “Did this person put any part of their body inside you?” For each incident, victims
were also asked about injuries, medical care, whether the incident was reported to a staff
member, counselor, teacher, or someone else who could help them. Finally, they were asked if
anything had been done to stop this from happening again. When the data are made available,
the SYRP results will give the first insight into the prevalence of sexual victimization among
youth in juvenile placement.
Summary of the Prison Rape Estimation Studies
Putting aside the one national probability sample conducted by BJS and the one
conducted by the Corrections Health Service in New South Wales, the other studies represent
only a fraction of the incarcerated adult male and female population. Most of the studies have
extremely poor (less than 50 percent) response rates. When non-response is high, there is
typically little or no effort to compare the respondents to the non-respondents, and no effort to do
post survey adjustments to the responses using stratification procedures. There is no effort to
gain an understanding of why inmates are not responding. Are they embarrassed? Are they afraid
of retaliation? Are respondents trying to embarrass the prison administrators? There is no study
which compares survey modes in an effort to understand the best method to collect these
sensitive and stigmatized data.
March 10, 2004
30
For every researcher, there is a difference in the definition of victimization. What does
sexual pressure mean? To make sense of the problem researchers will have to provide more
definitive operational definitions of rape. Recalling each incident may be difficult for someone
who has been serially raped; however, for each occurrence a prisoner should be asked to
characterize precisely the victimization. The survey should include: the nature of a completed
assault, the extent of injuries, the nature of an attempted, but uncompleted assault and associated
injuries, the nature of pressure (threat, intimidation, and constant invitation), the number of
assailants, and characteristics of the event (place, time, setting).
When we limit the studies to those that focus on assault or completed assault, the range is
from 0 to 16 percent, although most of the prevalence estimates (typically lifetime prevalence)
are 2 percent or less. When forms of pressure are included, the lifetime prevalence is 21 percent
or less, although in at least one institution the result was 40 percent. The few studies that include
incident information do not report enough information to construct a good denominator, or do
not indicate the time frame of the estimate. “Back of the envelope” incident estimates are
typically less than 2 percent, but this number is even less certain than the prevalence estimates.
Carroll’s study indicated a high level of incidence, but was based on only a few informants.
Women’s prevalence of sexual victimization appears to be lower than men’s, but there have only
been a few studies. To meet the legislative requirements, the Bureau of Justice Statistics must
produce one-year prevalence rates, and these are rarely, if ever, reported. Few studies indicate
the amount of time the prevalence covers. A prison lifetime exposure may mean 5 years in some
jurisdictions and 10 years in others. Exposure is crucial to our understanding of the nature of the
problem if we are to make reasonable comparisons among prisons and jurisdictions.
March 10, 2004
31
Meta-analysis of Prison Sexual Assault Studies
To provide a summary estimate of sexual assault, the prison lifetime prevalence estimates
were analyzed with meta-analysis methods (Lipsey & Wilson, 2000; Rosenthal, 1994; Shadish &
Haddock, 1994). The Wooden and Parker estimate that was included in this analysis was based
on sexual pressure, the only result reported in their study. The other estimates are based on either
a completed rape, or serious sexual assault, although some of the question wordings are vague.
The average effect size was first computed assuming a fixed effects model. In Figure 1, the
individual study proportions and their 95 percent confidence intervals are represented. The
prevalence for women is shown in red. The men’s prevalence is depicted in black. Some of the
confidence intervals span 0 and by themselves those studies should not be considered statistically
significant. These statistical tests rarely appear in the original studies. Under the assumption of a
fixed effects model, the average effect size is weighted by the inverse variance estimates. Using
proportions, this yields an average weighted estimate of .00448 (0.5 percent), with a standard
error of .000427, and 95 percent confidence intervals of .00364 to .00532. To ascertain whether
the assumption of a fixed effects model was appropriate, a homogeneity test statistic, Q, was
computed. The test statistic was Q=372.62, df =18. Since Q is distributed approximately as χ2,
this high value indicated that these studies failed the homogeneity test, and a fixed effects
assumption was inappropriate. The average effect size was re-computed using an estimate of the
between studies variance in addition to the within studies variance. The method of moments
estimate of the between studies variance was used (Lipsey & Wilson, 2000: 119). The
recalculation of the weighted average of prevalence yielded an estimate of .0191(1.91 percent), a
standard error of .00277, with 95 percent confidence intervals of .0137 to .0246. This is a
March 10, 2004
32
statistically significant effect. By definition, this is the most conservative estimate that could be
calculated from the available studies. If we included sexual assault estimates from studies that
used definitions of sexual pressure, this result would have been higher. The heterogeneity in
effect sizes indicates that there are factors that might explain the variability between studies –
possibly composition of the population, definitions of sexual assault, survey methods, and rates
of unit nonresponse. However, because there are so many problems with the individual studies,
we focus instead on reviewing these problems, and suggest ways to improve future estimates,
rather than explore the different factors that account for the heterogeneity. In the next several
sections of this report, some of the major obstacles inherent in this research are reviewed and
discussed.
Social Desirability Responses and the Nature of Sensitive Questions
Tourangeau, Rips, and Rasinski (2000) discuss the self report of sensitive behaviors in
their text entitled The Psychology of Survey Response. They identify three important
psychological dimensions to question sensitivity, social (un)desirability of the response, the
intrusiveness of the inquiry, and the perception of disclosure to third parties. As these authors
note, “Sensitive questions ask, in effect, whether we have violated [certain] norms.(Tourangeau
et al., 2000: 257)” If respondents are concerned enough about representing themselves in a
positive manner, they may distort their responses. Intrusive questions invade privacy.
Respondents may be unwilling to disclose simply because they perceive their privacy has been
violated. Third party disclosure is an additional dimension to the sensitivity of a question. This is
especially true in a prison or other criminal justice setting where admission to any kind of sexual
March 10, 2004
33
behavior is a rule violation, if not a crime. As discussed later in this paper, the influence of social
(un)desirability may be reduced by different modes of administering surveys.
Study Procedures and the Problem of Sample Selection Bias
Because there were such low response rates by both staff and inmates in many of these
studies, it is incumbent upon the researchers to demonstrate that their returned samples were
representative of the inmate population. There must be some accounting of the possibility of
selection bias, namely that inmates and staff who returned the surveys may have over- or underrepresented the incidence and prevalence of sexual victimization.
This is especially important when the topic is such a sensitive one. Returning to the
Struckman-Johnson studies, in the Nebraska study, Struckman-Johnson and Struckman-Johnson
compared the characteristics of the inmates who returned the surveys to the total inmate
population using age, race, most severe crime type, most severe crime, and minimum sentence in
years. The researchers assert that the returned survey sample was similar to the inmate
population at these facilities on age, most severe crime (murder, sex offense, aggravated assault,
robbery, drug related), minimum sentences, and average time in prison. The authors noted
differences in most severe crime type (against persons, drug-related, against property, public
order) and race. The returned survey sample had a higher proportion of whites than the inmates
in the Nebraska facilities, and the returned sample also had a higher proportion of offenders who
committed crimes against persons than other inmates in those facilities. The authors reported
these differences and similarities without conducting statistical tests. The data also indicate that
the return sample had an older age composition than the inmate population in those facilities and
March 10, 2004
34
possibly longer sentences. Longer sentences would increase the risk period of sexual assault.
There was no attempt in the second study to assess the degree to which the returned survey
sample was representative of the extant inmate population in the 7 facilities at the time of the
study.
The fact that the characteristics of the sample of inmates returning the surveys were
different than the characteristics of the inmate population at that time is a major flaw in the first
study. The second study also raises questions of validity since no attempt was made to assess the
representativeness of the sample. Because of such a low unit response rate, and a lack of
information to compare the survey and population characteristics, it is impossible to know how
to adjust the victimization estimates to make them valid estimates of the actual level of
victimization. Are the survey respondents more likely or less likely to report sexual victimization
than the prisoners who chose not to return the surveys? On the one hand, it may be that inmates
who chose not to return the surveys were more likely to be sexual victims and were fearful of
reprisals, or were ashamed to admit to the assault. On the other hand, inmates who did return the
surveys may have been more inclined to embarrass the administration by claiming or
exaggerating unwanted sexual approaches. This is also a problem with many of the other studies
reported in this literature.
Ever since the original Kinsey study (Kinsey et al., 1948), it has been widely reported
that about 10 percent of the U. S. male population has homosexual preferences. This was the
percentage among the men participating in the Kinsey interviews of sexual practices. Only one
scientific national survey has ever been conducted of sexual practices among 18 to 59 year olds
living in the U. S., and that study found that only 2 percent of the male respondents reported
homosexual preferences (Michael, Gagnon, Laumann, & Kolata, 1994). In this case, the Kinsey
March 10, 2004
35
method of soliciting interviewees increased the likelihood that they would get homosexual
respondents. Other well publicized but non-scientific sexual practice surveys have also led to
confusion about the rate of homosexuality in the United States. The Hite Report, the Redbook
survey, and the Masters and Johnson study were all well publicized, but highly unscientific
approaches to soliciting information on the estimates of sexual practices (Hite, 1976; Masters &
Johnson, 1966; Sadd, 1975). It is unknown how the methods of eliciting a sample affected the
estimate of prison sexual victimization in the studies reviewed in this paper.
Catania et al., (Catania, Gibson, Chitwood, & Coates, 1990) reviewed some of the
methodological problems in AIDS research that resulted from selection bias in the kinds of
people volunteering to participate in sex research. The authors referred to this selection artifact
as participation bias. Catania et al., hypothesize that if people refuse to participate because of
stigmatization, then the participants represent a less risky population and the level of HIV
transmission will be overestimated. If nonparticipation is the result of someone being less
committed to participate because he or she does not practice risky behavior, then the biased
estimate will over-represent the possible risk of HIV transmission. One solution to this problem
is to study the reasons people will not respond. Clearly, the challenge for researchers in prison
sexual victimization is to preclude, if possible, low response rates. If there is high unit nonresponse, then researchers must assess the extent to which their samples represent the population
being studied.
Recall and Telescoping
March 10, 2004
36
One of the problems in asking respondents to recall events in the past, even abhorrent
events, is that they tend to forget or to telescope those events. Forgetting would lead to the
underreporting of past sexual assaults. Telescoping is the phenomena that occurs when a person
recalls an event but misplaces it in a different time period than it actually happened. In studies of
household expenditures, lengthening the recall period from one month to even three months can
cause telescoping (Neter & Waksberg, 1964) where respondents are more likely to recollect
current expenditures that actually occurred at an earlier time frame. This is an example of
forward telescoping. While a vicious, brutal sexual attack may be a vivid memory, questions
about pressured or co-opted sex may be more problematic. Sudman and Bradburn (1973) have
found that some procedures that reduce telescoping increase forgetting, while other procedures
do the opposite. While calendar methods have been shown to aid the dating of past events,
survey research methodologists argue that questions should be bounded within recent time
frames (Converse & Presser, 1986).
Interview Modes
Recent work on modes of interviewing indicate that self administered questionnaires are
better at eliciting sensitive behavior than interviews, and that new computer assisted techniques
may further increase the reporting of these behaviors.
Interviewer versus Self-Administered Questionnaires
March 10, 2004
37
Many of the researchers conducting sexual assault studies argue that a major strength of
their anonymous self-administered survey methodology is that it eliminates face-to-face
interviews which suppress the self reporting of sexual victimization because this information is
sensitive (stigmatized). A National Academy of Sciences study asked survey respondents their
preferences for types of surveys (National Academy of Sciences, 1979). Respondents to this
survey said they preferred the interview to self administered mail-in surveys, or telephone
surveys. Interviews were preferred because they are more personal; they allow the interviewee to
ask clarifying questions; and people feel it is a more trusting situation. In many situations,
researchers have combined methods to elicit sensitive information. In the only study of sexual
practices that has involved a national probability sample of the United States (Michael et al.,
1994), respondents were interviewed for about an hour and a half. The most sensitive questions
about sexual practices were elicited by asking respondents at that point in the interview to
answer the questions by filling out a self administered questionnaire, and sealing the answers in
an envelope.
There has also been published work on designing the best interviewing techniques to
elicit information on human sexuality. Catania (1999) has reviewed and summarized this work.
In his synthesis, he finds that there are two key variables that pose a threat to distortions of
presentation of self in an interview, threat to self esteem and emotional distress. In his article he
reviews factors associated with the respondent, the interviewer, the task, and the context that
may minimize threats to self esteem and distress.
There have been a number of recent studies comparing different modes of interview
techniques comparing self-reported responses concerning sensitive information. Aquilino (1994)
is typical of a number of investigations comparing interviewer versus self administered
March 10, 2004
38
questionnaires (SAQs) for sensitive data. In this study, Aquilino also investigated the response
patterns for a telephone interview. This mode was the least sensitive. Respondents were asked to
provide information on alcohol and illegal drug use. The more sensitive the type of question, the
more likely the SAQ mode elicited higher use rates. The mode effects were larger for AfricanAmericans than they were for white respondents. Jones and Forrest (1992) investigated the self
reporting of abortions. In this study, independent estimates of abortions reported by clinics to the
Alan Guttmacher Institute showed that self reporting of abortion was underreported on the
National Survey of Family Growth, the National Surveys of Young Women, and the National
Longitudinal Surveys of Work Experience of Youth. Jones and Forrest (1992) found that SAQs
could be used to increase the reporting accuracy of abortion. Turner et al., (Turner, Lessler, &
Devore, 1992) also found that self-administered questionnaire items elicited higher self-reported
drug use than an interviewer mode.
Computer Assisted Interviews and Self-Administered Surveys
Recent work has been conducted on video- and audio-computer assisted self-interviewing
(video-CASI, audio-CASI) as a mode of surveying to tap into sensitive information. Audio-CASI
includes both video and audio components. Respondents sit in front of a computer after receiving
training, and respond to questions in privacy.
Williams et al., (2000) compared audio-CASI to face-to-face interviewing in eliciting
drug and sexual behavior at 10 sites recruiting participants from an HIV risk reduction study that
targeted drug offenders. These were self-reported heroin or cocaine users (later confirmed by
urinalysis). Every respondent was interviewed a second time within 48-72 hours of the first
March 10, 2004
39
interview. This allowed the researchers to use a crossover design in which ¼ of the respondents
were interviewed with the audio-CASI mode twice, ¼ were interviewed face-to-face twice, ¼
crossed over from audio-CASI to face-to-face, and ¼ crossed over from face-to-face to audioCASI. This is a very powerful and sensitive experimental design. Urinalysis tests confirmed the
validity of the self-reported drug use. Eighty percent of the self-report and lab test results agreed.
When complex questions about number of times of drug use were compared, the face-to-face
interview produced higher reported risky behaviors. When the questions were more
straightforward, the two response modes produced similar estimates. The authors argue that
highly trained interviewers may be necessary to guide a respondent through difficult types of
responses. In this study, however, they were able to use a validity check to confirm the accuracy
of self-reported drug use.
Newman, DesJarlais, Turner, Gribble, Cooley, and Paone (2002) compared audiocomputer assisted self-interviewing (audio-CASI) with face to face interviews probing risky
behaviors among participants in a needle exchange program. They found that using the audioCASI system increased the reporting of risky behaviors such as whether they were HIV positive,
or had rented or sold their works (needles and injection paraphernalia) in the last 30 days.
However, questions that had “emotional stress” content had higher self-report outcomes for the
face-to-face method. Newman et al., reasoned that such questions require an empathetic listener
to elicit such responses.
DesJarlais et al., (1999) also found that audio-CASI elicited higher response percentages
of sensitive questions. The sample was also composed of needle injecting users who were
exchanging their works and were approached about participating in an interview. There were 724
audio-CASI and 757 face-to-face interviews. As an example of the differences in response rates,
March 10, 2004
40
audio-CASI interviewees reported having sex with a same sex partner in the 30 days prior to the
interview 10 percent of the time. Face-to-face respondents reported such behavior 5 percent of
the time.
Similar underreporting in face-to-face interviews as compared to audio-CASI has been
demonstrated by other authors in a population of people who have a low risk for AIDS (Miller,
Gribble, Mazade, & Turner, 1998; Turner, Forsyth, & O'Reilly, 1998; C. F. Turner, L. Ku et al.,
1998). For example, Turner et al., (C. F. Turner, L. Ku et al., 1998) demonstrated that audioCASI increases reporting above self-administered questionnaires (SAQ’s) in an administration of
the 1995 National Survey of Adolescent Males (NSAM). The stigmatized behaviors involved
HIV-risk behaviors, drug use, and interpersonal violence among 15 to 19 year old men.
Previously, sensitive questions were asked by having respondents fill out a paper questionnaire,
and then sealing the responses in an envelope so that the interviewer could not see the answers.
When the two modes were compared, SAQ versus audio-CASI, there were much higher response
percentages to stigmatized questions about male-male sexual encounters among the audio-CASI
interviewed youth. For example, youth were more likely to admit to anal and oral intercourse
with another male when interviewed with audio-CASI. The overall male-male sex percentages
were 1.5 percent with the SAQ modality and 5.5 percent with the audio-CASI method. Even
telephone administered audio-CASI seems to elicit higher reporting rates than a normal
telephone interviews.
Tourangeau and Smith (1998) summarized research comparing six modes of data
collection methods for sensitive behaviors: paper and pencil personal interviews (PAPI), paper
and pencil self administered questionnaires (SAQs), Walkman-administered questionnaires
(ASAQ), computer assisted personal interviews (CAPI), computer assisted self-administered
March 10, 2004
41
interviews (CASI), and audio computer assisted self-administered interviews (audio-CASI or
abbreviated as ACASI). As these authors point out, the major hypothesis guiding most of the
literature on sensitive topics such as self reported drug use and sexual behavior is that there is
deliberate misreporting. They review a great deal of research to show that SAQs have higher
reporting rates of sensitive behaviors than personal interviews in studies of illicit drug use,
sexual behavior, alcohol consumption, and abortion reporting. Tourangeau and Smith also
reviewed mode research they conducted on the National Survey of Family Growth. This survey
addresses issues involving pregnancy, contraception, fetal and infant deaths including abortions,
sexually transmitted diseases, and infertility. The study was intended to examine different modes
to improve the reliability and accuracy of the survey responses. The modality experiment had
five variables: whether the questionnaire began with items about medical conditions or
pregnancies; whether the interviewer was a NORC employee or nurse; whether the interview
was done inside or outside the respondent’s home; whether the interviewer asked questions, or
the questions were self-administered; and whether the data were collected by paper or computer.
Thus, the data were measured with CAPI, PAPI, CASI, and SAQ techniques.
When asked to report about their sex partners, women who completed self administered
questionnaires reported more sex partners than women who responded to an interviewer. This
was true whether the period of time was one year or lifetime. The computer assisted self
administration method (CASI) elicited the highest number of reported sex partners in the past
year and past five years. There were no differences in the reporting of lifetime drug use, although
the researchers suggest that mode effects are more likely with shorter time horizons. Item nonresponse was higher for self-administration, and lower for computer administration, although the
March 10, 2004
42
lowest proportion of questions answered was 94.6 percent for the SAQ. The CAPI proportion
was 99.2 percent.
In a second experiment, these authors reported the modes effects of CAPI, CASI, and
ACASI on sensitive questions about drug use and sexual practices. ACASI produced the highest
reported one-year, five-year, and lifetime number of sex partners for men and women. As is
typically the case, there was a sex difference in the number of reported sex partners -- men
reporting more than women. ACASI also elicited the highest level of reported drug use. This
second study is reported in detail in Tourangeau and Smith (1996). Table 5 in the study shows
ratios of reporting of sensitive behaviors comparing ACASI to CAPI and CASI to CAPI. The
ratio of ACASI to CAPI is always above 1.00, with two exceptions, over 17 sensitive items. For
example, the percentage reporting anal sex was 4.21 times higher for ACASI than for CAPI.
ACASI proportions were often higher then CASI proportions. For example, among respondents
reporting use of cocaine in the past year, the percentage for ACASI respondents was 5.4, while
for CAPI it was 1.9, and for CASI it was 2.6. Item context also affected how women and men
responded to the questions regardless of whether the interview mode was ACASI, CASI, or
CAPI. The authors conclude that ACASI methods increase the level of sensitive item reporting
by increasing the legitimacy of the context, and by increasing privacy, thus lowering
embarrassment and the normative features of an interview setting.
Turner and his colleagues (Charles F. Turner et al., 1998) have also reviewed the theory
and evidence regarding the use of audio-CASI as a mode of response. Some of their work has
already been covered in this section. Turner et al., argue that an interview creates a social
interaction that introduces a social context to the collection of the data. Computer assisted self
March 10, 2004
43
administration has the advantage of removing this social context which may especially have
implications for sensitive questions. They state that:
By removing the requirement that respondents divulge sensitive, stigmatized, or counter normative
behaviors to another human, CASI procedures may substantially reduce the extent to which response
accuracy for such measurements is compromised by the social presence of the human interviewer. (Charles
F. Turner et al., 1998: 457)
Audio features of CASI can reduce literacy problems, increase standardization of the
presentation of questions, and provide for multilingual administration. Literacy is an important
issue with an inmate population and may have been a problem with the self-administered
questionnaires used by previous researchers conducting inmate sexual victimization studies. In
this paper, Turner et al., report that audio-CASI increased the reporting of abortions in the NSFG
above an interviewer administered questionnaire. In this study, a subset of the women who were
interviewed was also asked to respond to the questionnaire by ACASI. Among sexually active
females, age 15 to 44, an additional 4.5 percent reported having had one or more abortions in the
ACASI mode. In this book chapter, Turner et al., also report on the results of ACASI in an
administration of the NSAM. We have already reviewed that data showing men are much more
likely to admit to homosexual sex using ACASI than a SAQ.
This literature supports the use of audio-CASI as a mode that overcomes literacy
problems, and stigmatization, while still allowing the researcher control of the interview
environment.
Randomized Response
March 10, 2004
44
Another technique for eliciting sensitive information from respondents is called
randomized response (Fox & Tracy, 1986). To insure respondent anonymity, an interviewee is
given two statements. The two statements can both be sensitive such as “I have had an abortion,”
or “I have never had an abortion.” Or, one of the statements is innocuous such as “My birthday
is in November.” The “trick” behind this approach is that the respondent is asked to
affirm/disconfirm either statement one or statement two based on a random draw from a
procedure that determines which statement to answer. For example, the respondent may be given
a box containing red and green marbles. If a red marble is chosen, he or she responds to the first
statement. If a green marble is chosen, he or she responds to the second statement. Only the
respondent knows the color marble that has been chosen, and, therefore only the respondent
knows whether he or she answered the first or second statement. The underlying proportion can
be imputed because the researcher knows the probability of choosing a red versus a green
marble, and the researcher also knows that the probability that the individual’s birthday occurs in
November is p=1/12.
The success of the method relies on whether the respondent actually believes the choice
of answering the statement is actually reliant on a random process. In Fox and Tracy’s review of
the research comparing randomized response and traditional methods, the two methods did not
always yield different results. When the randomizing process was credible, the randomized
response method...
Purchase answer to see full
attachment