U.S. Department of Justice Office of Justice Programs Bureau of Justice Statistics A BJS Internet Document Summary of Human Subjects Protection Issues Related to Large Sample Surveys June 2001, NCJ 187692 ---------------------------------------------- The full formatted report is available in .pdf format at- ---------------------------------------------- By Joan E. Sieber U.S. Department of Justice Office of Justice Programs John Ashcroft Attorney General Bureau of Justice Statistics Lawrence A. Greenfeld Acting Director Report of work performed under a BJS purchase order to Joan E. Sieber, Department of Psychology, California State University at Hayward, Hayward, California 94542, (510) 538- 5424, e-mail jsieber@csuhayward.edu. The author acknowledges the assistance of Caroline Wolf Harlow, BJS Statistician and project monitor. Ellen Goldberg edited the document. Contents of this report do not necessarily reflect the views or policies of the Bureau of Justice Statistics or the Department of Justice. This report and others from the Bureau of Justice Statistics are available through the Internet Table of Contents 1. Introduction Limitations of the Common Rule with respect to survey research 2. Risks and benefits of participation in sample surveys Standard risk issues, researcher responses, and IRB requirements Long-term consequences Background issues 3. Procedures to protect privacy and maintain confidentiality Standard issues and problems Confidentiality assurances and their consequences Emerging issues of privacy and confidentiality 4. Other procedures for minimizing risks and promoting benefits Identifying and minimizing risks Identifying and maximizing possible benefits 5. Procedures for responding to requests for help or assistance Standard procedures Background considerations A specific recommendation: An experiment within the survey 6. Procedures for responding to mandates to report Child maltreatment Elder abuse Intent to harm oneself or others 7. Vulnerable populations: Risks and special protections IRB concerns Children and teenagers Pregnant women Cognitively impaired persons Elderly persons Minorities Students, employees, and institutionalized persons Economically or educationally disadvantaged persons 8. Efforts to minimize refusals to participate in sensitive survey Efforts to induce participation Efforts to evoke answers to sensitive questions 9. Survey procedures for informed consent with special procedures for obtaining consent for respondents under age 18 Regulatory requirements Informed consent as practiced in survey research 10. Effects of signed consent on response rates and other aspects of conducting a survey or other study Regulatory issues Background considerations 11. Researching ethical questions The experiment within a study The focus group Ethnography Meta-communication in the research process 12. Keeping abreast of new developments and answering questions On-line resources Meetings Journals Informal local groups Office of Human Research Protections 13. Briefly annotated bibliography of literature cited Appendix A. State phone numbers for reporting child maltreatment Appendix B. Using focus groups This report identifies best practices to ensure that large sample surveys such as the National Crime Victimization Survey (NCVS) are ethically sound and compliant with the Common Rule. There are three approaches used in this report to identify best practices: *Identify regulations and statutes governing each question, and the position of the Office of Human Research Protections (OHRP) and of major Institutional Review Boards (IRBs) and investigators who oversee large sample survey research. *Recognize that these are general guidelines. Federal regulations recognize that broad areas of investigator and IRB discretion are required to validly investigate sensitive topics. The IRB's mandatory reasoned discussion and decision-making must be based on a broad understanding of the issues, and the IRB must document its decision process in its minutes. *Provide resources that would enable the Bureau of Justice Statistics (BJS) to increase its sophistication and to evolve as issues change with new regulatory requirements and interpretations and with changes in the country's social or political environment. This report addresses the following nine issues: *The risks and benefits of participating in sample surveys on sensitive topics. *Procedures to protect respondent privacy and ensure the confidentiality of data. *Procedures to minimize risks and promote benefits to respondents. *Procedures to respond to requests for help or assistance. *Procedures to respond to revelations of a situation that interviewers must, by statute, report to appropriate authorities. *Dangers and safeguards for vulnerable populations, including children and teenagers, pregnant women, mentally disabled persons, and persons confined to an institution, particularly correctional facilities. *Efforts to minimize refusals to participate in surveys concerned with sensitive topics. *Survey procedures for informed consent; special procedures for obtaining consent for respondents under age 18. *Effects of signed consent forms on response rates and other aspects of conducting a survey or other study 1. Introduction The NCVS, sponsored by the Bureau of Justice Statistics (BJS), is conducted and analyzed by the U.S. Census Bureau which then provides summary statistics and data files to BJS. This survey asks sensitive questions of the same set of respondents every 6 months over 3 years (seven times); respondents include all members of a given household who are at least age 12. Over 40,000 households or living groups are surveyed. This research is exempt from IRB review and from the Common Rule (45 CFR 46, subpart A). However because the NCVS involves complex ethical issues, BJS has requested this review of issues pertinent to the ethics and regulatory requirements of large-scale sample surveys that ask sensitive questions and study vulnerable populations. The federal policy for the protection of human subjects, which formerly pertained only to Health and Human Services research (45 CFR 46, Subpart A), has now been incorporated into the regulatory structure of 17 federal agencies, eight of which have additional human subject protections. ***Footnote 1. These agencies and their relevant regulations are: Housing and Urban Development (24 CFR 60), Justice (28 CFR 46 with additional protections in 28 CFR 512 and 28 CFR Part 22), Transportation (49 CFR 11), Veterans Affairs (38 CFR 16 with additional protections in 38 CFR 17.85, M-3, Part 1, Chapters 9 and 15), Consumer Product Safety (16 CFR 1028), Environmental Protection (40 CFR 26), International Development (11 CFR 225), NASA (14 CFR 1230), NSF (46 CFR 690), Agriculture (7 CFR 16), Commerce (15 CFR 27), Defense (32 CFR 219, plus 12 additional regulatory protections), Education (with extensive additional protections to privacy and confidentiality as noted below), Energy (10 CFR 745), Health and Human Services (45 CFR 46 Subpart A), Social Security (P.I. 103-296), and CIA (Executive Order 12333); the last three agencies also employ Subparts B, C, and D of 45 CFR 46.*** Subpart A, which is now known as the Common Rule, as well as the rest of 45 CFR 46 (Subparts B, C, and D), may be found at under Policy Guidance. Briefly, the Common Rule sets forth the role and operation of the IRB, the required elements of the research protocol and the informed consent, and general criteria for IRB review and approval. The Department of Justice regulation is found at 28 CFR Part 46. Limitations of the Common Rule with respect to survey research The Common Rule poorly defines privacy and confidentiality in survey research. It promotes the prevalent misconception that self-report research is necessarily less risky than experimental or observational research. For example, it exempts anonymous surveys of adults from IRB review (45 CFR 46.101(b)(2)), on the premise that adults can freely protect their privacy by refusing to answer. In fact, self- report questions can induce respondents to reveal far more personal and sensitive aspects of their lives than can be studied ethically by observational or experimental methods (see section 2). Subjects should at least give informed consent based on an accurate understanding of the kinds of questions that will be asked. The Common Rule specifically requires that informed consent include a statement about how the researcher will maintain confidentiality. However it leaves to the IRB and the researcher the subtle matter of understanding what confidentiality is and how it relates to privacy. The Common Rule defines privacy obliquely by reference to private information, as follows: Private information includes information about behavior that occurs in a context in which an individual can reasonably expect that no observation or recording is taking place, and information which has been provided for specific purposes by an individual and which the individual can reasonably expect will not be made public (for example, a medical record). (45 CFR 46.102(f)(2)) This oblique reference to privacy confuses it with confidentiality (an understanding or agreement about the disclosure or nondisclosure of identifiable information to others) and fails to convey the notions of personal privacy (discussed in section 3) that are important to ethical research. It also implies that everyone has the same concerns about others' access to themselves and to identifiable data about themselves, and that researchers and IRBs can accurately assess what others situated differently from themselves would consider as private. Based on this confusing set of definitions, the Common Rule states that: (7) When appropriate, there are adequate provisions to protect the privacy of subjects and to maintain the confidentiality of data. (45 CFR 46.111(a)(7)) Unfortunately, this requirement assumes a level of sophisticated knowledge concerning privacy and confidentiality that most IRBs and researchers do not possess. Given these limitations of the Common Rule, the conscientious survey researcher might look beyond the Common Rule to Subparts B, C, or D of 45 CFR 46 for guidance. Subpart B pertains to biomedical research on fetuses, pregnant women, and human in vitro fertilization, and Subpart C pertains to persons under correctional supervision. Neither of these sections are relevant to the NCVS. However Subpart D pertains to minors and is sensitive to children's personal privacy interests. It requires the child's active assent (when assent would be meaningful) and parental permission, with either party having veto power. It recognizes that there are contexts in which parental permission is not a reasonable requirement to protect the subjects (such as neglected or abused children), and also recognizes a range of circumstances in which parental permission may be waived. The parental permission requirement respects the parents' right to control the conditions of their child's life and their ability to judge the degree of acceptable risk for their child to take. Subpart D requires IRB approval when the research involves minors and anonymous surveys, interviews, and observation of public behavior in which the investigator participates in the activities being observed. It limits a child's exposure to risk, even if there is parental permission and the child's assent. 2. Risks and benefits of participation in sample surveys Risks to privacy and confidentiality have long been deemed the main risks in sample survey research. However the scope and magnitude of this risk has increased as more sensitive topics are studied, and as researchers begin to oversample vulnerable and marginalized populations. As Bersoff and Bersoff (2000) point out, surveys raise issues of privacy that are rarely found in other research. There is the potential for self-report research to glean clinically sensitive data and put it in the hands of non-clinicians. Even if the survey is anonymous, researchers should not ignore the potential for violation of confidentiality and for severe emotional upset to respondents. Standard risk issues, researcher responses, and IRB requirements To prepare a research project for review by an IRB, researchers must provide appropriate discussion of risks and benefits in the protocol and in the informed consent documentation. The IRB is required to evaluate the adequacy of researchers' recognition of potential risk and benefit, their plans for reducing or preventing risk and enhancing benefit, and the appropriateness with which this information is presented in the informed consent. The risks that can arise in survey research are the same basic risks that can arise in other research: *Mere inconvenience when a survey is administered at an inconvenient time or place or simply takes too long to administer. *Emotional or psychological risk when a survey causes upset, or worry, warranted or not, about breach of confidentiality. *Social risk due to stigma or other negative social outcomes of breach of confidentiality. *Physical risk if revelations about others get back to those persons, particularly when researchers study domestic violence, gang activity, or other phenomena concerning violence-prone individuals. *Financial risk if revelations result in loss of employment or insurance coverage. *Legal risk when illegal activities are disclosed Most researchers do not catalog every conceivable risk in their consent document or in their protocol, nor do IRBs expect them to do so. However IRBs are mindful of the risks that researchers may overlook, including those described above. Most of the risks have to do with breach of confidentiality or fear or worry about a possible breach. The IRB is likely to want to know specifically how confidentiality is handled, whether limits to confidentiality are adequately disclosed, and whether promises of confidentiality can actually be guaranteed. IRBs are attuned to issues of "secondary subjects," who are identifiable persons about whom the respondent is asked sensitive questions. These "secondary subjects" are considered subjects of survey research because they are identified and researchers obtain information about them. If feasible, investigators should obtain their informed consent as they too may be placed at risk. If it is deemed impractical or impossible to obtain their informed consent, the waiver of informed consent should meet the requirements specified in the Common Rule, and the IRB must document its discussion and decision in its minutes. Long-term consequences A sensitive matter is usually a troubling one. It may be a matter the respondent fears to discuss lest others learn about their particular situation. Fears of social or economic reprisals can cause respondents to lie, refuse to answer, or answer honestly but worry for a long time afterward. IRBs should recognize the importance of accurately identifying the kinds of unwarranted fears respondents are likely to have, as well as the actual risks. They require investigators to take steps to reduce both kinds of fears, and to ensure that researchers can keep promises of confidentiality. Some matters are so sensitive that the respondent may relive it when discussing it. For example events such as physical assault are so traumatic that the respondent would reexperience old pain by retelling the details yet again. IRBs customarily require that researchers who inquire about sensitive issues such as rape have referral information and even availability of several free therapy sessions for respondents who recount traumatic events and would welcome such assistance in restoring their emotional well being. The benefits to respondents of participation in surveys (not to be confused with financial incentives, which are discussed in section 8) are usually limited to the benefits of an interesting exploration of some topic, an informative and satisfying exchange in the debriefing process, and some written information pertinent to the topics discussed. (See section 4 for other appropriate and feasible additional benefits.) Depending on the particular survey, the IRB will question whether vulnerable populations, non- English speakers, children, and persons whose autonomy is somehow constrained are accorded due protection and respect. These issues are discussed in section 7. Background issues As most skilled interviewers have learned, getting in the door to interview an individual is not as hard as getting out. A well- constructed survey and a skilled interviewer can create powerful motivation to discuss personal matters. This is a source of both benefit and risk, especially with repeated-measure surveys (such as the NCVS) in which a relationship is built up over time. Properly constructed surveys begin with easy, non-threatening questions. Each new topic is initiated with a general, non-threatening question and followed by increasingly specific and sometimes more sensitive questions. This funneling from general to specific makes for easy recall of related information, good comprehension of related questions, and rapid responding. It also makes respondents comfortable with questions they otherwise would refuse to answer. Lonely or troubled respondents are especially likely to welcome the attention of a respectful and skilled interviewer who promises confidentiality. Respondents may find the interview session interesting, and perhaps even a therapeutic opportunity to recall and reflect upon their experiences or to get some things "off their chest." The ethical and well-prepared interviewer is a good listener and keeps private information confidential. In response to a request for help or to signs of problems, or simply as a routine part of debriefing, the interviewer may provide useful feedback, referrals, or reference materials. It would seem apparent that the respondent can readily refuse to answer any question that is too personal. Answering a question seems tantamount to informed consent, hence it is often assumed that there is little risk involved. This assumption of respondent autonomy and ease of refusing to answer questions is somewhat illusory given what is known about the "foot in the door" technique (Freedman and Fraser, 1966). Once persons agree to a small and benign request, they can be gotten to agree to a larger and less benign request which they would ordinarily never agree to, or to which persons who did not receive the first benign request would not agree. Interviewers can lead respondents who would ordinarily refuse to answer highly personal or embarrassing questions about some aspect of their personal life to answer such questions by first asking them to answer a rather tame question on the same topic. The underlying principle is that people want to appear consistent and cooperative and hence will continue to answer questions even when they would otherwise judge it ill-advised to do so (Cialdini, 1993; Orne, 1962). The power of the well-constructed survey to yield answers to sensitive questions is a good thing in the hands of an ethical, sensitive, well-trained interviewer, but it can be dangerous otherwise. There are several possibly serious risks. Details of some of these risks and associated risk prevention strategies are discussed in subsequent sections. Respondents may: *reveal reportable criminal activities, *reveal information damaging to their social or financial standing, employability, or reputation which the researcher fails to treat as confidential, *reveal information that is treated with appropriate confidentiality but which nevertheless causes the respondent to worry about the confidentiality of the disclosure or to feel embarrassed, *reach out for help such as by revealing suicidal ideation to an insensitive interviewer whose inappropriate response confirms the respondent's sense of hopelessness, *reach out for help but then feel betrayed or embarrassed that the interviewer took stronger action than was expected, *recall and ruminate about unresolved issues that the interviewer fails to recognize or respond to appropriately, *reveal information in a family, community, or organizational setting where their privacy or the privacy of others in that group is not accorded due respect, or reveal potentially damaging information about another identified person (the secondary subject), *respond to emotionally charged topics that may cause them to focus long afterward on painful memories such as brutal crime victimization, or *experiment with socially unacceptable behavior in response to an interviewer who uses the "everybody does it" approach to evoking answers about unacceptable behavior. The appropriate safeguard is not to weaken the power of the survey to gather information, but to strengthen the protections offered. The IRB frequently identifies the possible risks, and helps the researcher become sensitive to possible signs of risk and to plan appropriate safeguards. In addition to safeguards to confidentiality, investigators should consider how to conclude the interview. The respondent should have an opportunity to express reactions or ask questions. The conclusion should be a two-way conversation and should not be carried out in a perfunctory manner. It should adequately settle any questions or concerns the respondent may have, and return the respondent to a positive, satisfied state of mind. Many interviewers are unprepared for risks even after they are sensitized to their possible occurrence. Researchers should develop referral and feedback information that might be generally useful to all respondents. They should also locate institutional or outside resources that can respond appropriately to respondents who reach out for help or indicate distress. There should be competent professionals who are available to researchers for consultation when issues arise which the researcher (interviewer) feels unprepared to handle. However the investigators must always respect the wishes of the respondent and keep promised conditions of confidentiality. Giving more help than is wanted is often harmful, not helpful. Moreover the interviewer should not attempt to help beyond providing referrals or other written resource information. These topics are discussed further in sections 3 through 7.3. Procedures to protect privacy and maintain confidentiality The Common Rule and most guidelines for IRBs emphasize the importance of privacy and confidentiality but are neither specific nor detailed in their recommendations. There is a superb literature on approaches to respecting privacy and confidentiality with which survey researchers should be familiar. Standard issues and problems The Common Rule leaves much to the judgment of IRBs with respect to privacy and confidentiality, so that the same degree of caution need not be imposed on all research. 45 CFR 46.111 (a)(1) states that "Risks to subjects are minimized: (i) By using procedures which are consistent with sound research design, and which do not unnecessarily expose subject to risk," and 45 CRF 46.111(a)(7) states that "When appropriate, there are adequate provisions to protect the privacy of subjects and to maintain the confidentiality of data." How do IRBs identify the relevant issues? Invasion of personal privacy is a subjective matter, and IRB members judge invasion of privacy on their own sense of propriety and on the particular circumstances of the study. This can be an inadequate basis for judgment. Confidentiality is an objective but complex matter and involves many possible judgments depending on the research. Confidentiality pertains to data on identifiable persons. In recent times, IRBs have become increasingly concerned about what constitutes an identifiable respondent of survey research. When is a survey truly anonymous? Even when the names of respondents are never attached to their data, there is increasing concern about deductive identification of otherwise anonymous respondents on the basis of such elements of their data as birth date, occupation, zip code, race, and gender. The issue of an "identifiable subject" also arises if the researcher wants access to existing records to identify persons suitable for the proposed study; if the data being sought are sensitive, the IRB may judge that consent of subjects should be obtained for accessing those existing data. If existing school data are sought on youngsters, the Buckley Amendment (the General Education Provisions Act (20 USC 1232)) requires parental permission for release of identifiable information about children in public schools. IRBs do not consider sample survey research anonymous (and hence exempt from IRB review) if identifiers that accompany the data are later stripped from the data. If a unique identifier was attached to it at some point in the process -- for purposes of respondent selection and interviewing or for recontacting selected subjects by a supervisor checking on the work of the interviewer -- it is not an anonymous survey. A survey that involves any identifiable data at any point in the research process is subject to IRB review. If researchers are collecting sensitive survey data about identified individuals, the IRB will inquire whether there are provisions for protecting the confidentiality of the data. Such provisions typically include substituting codes for identifiers and storing the code key elsewhere, removing face sheets (typically containing such information as names, phone numbers, or addresses), destruction of identifying information such as computer sheets, keeping data in locked files, impressing on research assistants the importance of confidentiality, and limiting access to the data by various means. Data from large sample surveys are normally stored electronically for easy access and analysis. Whenever identifiers accompany these data or when deductive identification would be easy, there is major concern about the security of the computer system on which the data are stored, and researchers must satisfy these concerns. A particularly difficult issue has to do with the training and supervision of interviewers and research assistants. Where possible, the research staff should not be persons who might know some of the respondents, though this is difficult to ensure. Researchers who are concerned about the cultural and linguistic matching of interviewers with subject populations should consider hiring and training local people. However, there is the risk that the interviewer will know the respondent. The Common Rule defines and discusses privacy and confidentiality in ways more appropriate to biomedical research than to survey research. It fails to recognize both the aspects of personal privacy that the effective interviewer must respect and the individual subjectivity and diversity of people's sense of privacy. With respect to confidentiality, it gives no hint of the vast technical literature on methods of ensuring confidentiality in survey research. Privacy The difficulty of defining invasion of one's own privacy is evocatively expressed by Melton (1992, p. 66): 'I know it when I feel it.' A gut sense of personal violation may be the tie that binds such disparate events as being subjected to a body search, being the subject of gossip, having one's mail read, being asked one's income, or having one's house entered without permission. It should come as no surprise that such an intensely personal construct is difficult to define. It is difficult to define, understand, and respect the privacy of other persons situated differently from ourselves. Without a useful definition or theory of privacy to guide them, researchers and IRBs must depend on their own culture-bound notions of privacy. They invoke their personal and idiosyncratic definitions, resulting in a capricious standard of protection. The meaning of privacy in survey research inheres in the culture and personal circumstances of the particular subject, the context and nature of the research, and the social and political environment in which the research occurs. A useful definition of privacy that recognizes these manifold elements is borrowed from, and based on, the elegant theory of personal privacy developed by Laufer and Wolfe (1977): Privacy refers to persons and to their interest in controlling the access of others to themselves. This theory of personal privacy recognizes the manifold cultural, developmental, and situational elements by which individuals orchestrate their privacy. It recognizes that people have an interest in (a) controlling the time, place, and nature of the information they give to others, and (b) controlling the information or experiences that are proffered to them. Thus informed consent serves as a control mechanism, provided the prospective subjects of survey research are adequately informed of what it is they will be asked and what they may experience. Laufer and Wolfe's theory would be highly useful to efforts to educate IRBs; to design ethical elements of recruitment, consent, location, circumstances, and content of surveys; and to train interviewers. Laufer and Wolfe's theory embodies four dimensions of privacy: The self-ego dimension refers to the development of autonomy and personal dignity. Young children have an aversion to being alone. By middle childhood, children seek time alone to establish a sense of self and to nurture new ideas, creating a basis for self-esteem, personal strength, and dignity. By age 6 or 7, children have a need and right to privacy not found in infants and younger children. Teenagers are intensely private, as they seek to forge an identity separate from that of their parents. Teenagers would be embarrassed to be interviewed about personal matters in the presence of their parents or others, and in the presence of their parents would most likely refuse to be interviewed or lie in their answers. Adults continue to need time alone and develop many means of protecting that privacy. The environmental dimension includes socio- physical, cultural, and life-cycle dimensions. Socio-physical elements are physical or technological elements that offer privacy; more affluent individuals tend to have more of such barriers to unwanted intrusion. Cultural elements include norms for achieving privacy; for example some cultures permit lying while others permit persons to have private rooms and telephones. Life-cycle elements vary with age, occupation, available technology, and changing socio-cultural patterns. The ways one establishes privacy at one age, under one set of circumstances, may be unsatisfactory or unavailable at a later stage. The interpersonal dimension refers to how interaction and information are managed. One's social setting and its physical characteristics provide options for managing social interaction; physical and social boundaries can be used to control people's access to one another. The control/choice dimension grows out of the other dimensions. Young children have no control over their privacy except through hiding. They learn to use personal, cultural, and physical resources to control their privacy. Events that would threaten one's privacy early in the development of control/choice are later so easy to control that they are no longer considered a threat to privacy. Understanding and respecting the privacy of others How do the researcher and IRB learn about the privacy interests of persons situated differently from themselves? Networks of local researchers, educators and outreach workers such as social workers, farm agents, and public health nurses can share valuable information about the most appropriate ways to approach members of various cultures. Interviewers who are of the same culture and backgrounds are vital to some sensitive research in some kinds of communities. Focus groups and other forms of community consultation are useful ways to learn about a culture, how the individuals within that community perceive the research, and how the research that would be objectionable to them can be made acceptable. The community meetings held by Fisher and Wallace (2000) are a good example of learning the views and suspicions of members of inner-city African Americans about studies of adolescent risk behavior. The community consultation discussed in Melton et al. (1988) is a dramatic example of the explosive acquaintance process of AIDS activists with AZT researchers. While only the Fisher and Wallace account focuses on survey research, there is much that survey researchers could learn from both accounts about the importance of understanding the perspective of one's subjects in field-based applied research. Community consultation can also ameliorate fears among potential respondents who feel a need to avoid strangers. For example, a large-scale survey of the housing needs of the elderly required that interviewers go to the homes of elderly people, who are fearful of scam artists, robbers, and burglars. The interview teams began by making well-advertised visits to senior centers in the neighborhoods where they would be conducting their survey. They explained their project, answered questions, and made certain that the local newspaper carried the story of their presentation and project, along with photographs of them. Their first contact with their prospective respondents was by letter, and it included a copy of the newspaper article with the pictures of the interviewers. The interviewers then phoned to make an appointment for their visit. When they appeared at the door, they carried with them a copy of the newspaper article so that the respondents could see that they were indeed the designated interviewers. Confidentiality The following definition of confidentiality is adapted from that developed by Boruch and Cecil (1979): Confidentiality is an extension of the concept of privacy; it refers to (a) identifiable data (some information about a person that would permit others to identify the specific person, such as a non-anonymous survey, notes or a videotape of the person) and (b) agreements about how those data are to be handled in keeping with respondents' interest in controlling the access of others to information about themselves. The two critical elements of this definition identifiable data and agreement about the handling of the data indicate the critical role of informed consent, which states how the researcher will control access to the data and secures the respondent's agreement to participate under these conditions. This definition further underlines the importance of planning before gathering sensitive data. It is important that researchers make early plans regarding techniques to ensure confidentiality. They should incorporate these plans into the methodology and into any consent agreements with respondents or contractual agreements with subsequent users of the data, including funders who may wish to audit the data. Investigators should include all of these details in the IRB protocol. This definition of confidentiality leads naturally to the literature on procedural, methodological, statistical, and legal approaches to ensuring the confidentiality of survey research data. Methods and procedures of ensuring confidentiality****Footnote: Portions of this section are drawn from an unpublished paper that was commissioned by the National Bioethics Advisory Commission: J. E. Sieber, Privacy and Confidentiality: As Related to Human Research in Social and Behavioral Science, 2000.*** Approaches to ensuring confidentiality of survey research fall into seven categories: * Procedures that eliminate linkage of data to unique identifiers *Intersystem linkage *Statistical strategies *Agreements: Data sharing, secondary analysis, or audit of data *Legal protections *Descriptive statistics and raw data releases *Internet research Procedures that eliminate linkage of data to unique identifiers. Anonymity offers the best insurance that disclosure of subjects' responses will not occur. Researchers have developed dozens of techniques that are responsive both to the need for anonymity and to other research needs. Different kinds of data cross- sectional, longitudinal, and data from multiple sources bring with them different research requirements and different ways of meeting those without using unique identifiers of subjects. The following brief summary is illustrative, not comprehensive. See Boruch and Cecil (1979) for a comprehensive review. Cross-sectional surveys, in their simplest form, require just one data collection session. Anonymity in which even the researcher is at all times ignorant to the identity of subjects protects the respondent from legal prosecution, social embarrassment, and concern that the data may fall into the wrong hands. However it may be desirable to have some form of follow-up to test for sampling validity, response validity, or to do further research on some or all subjects. These refinements are impossible with complete anonymity, but can be achieved through temporarily identified responses with subsequent destruction of identifiers, or through use of brokers to provide anonymous data to the researcher after completing one or more of these refinements. Longitudinal surveys track individual subjects over time. There are many ways in which aliases or arbitrary identifiers can be used as a basis for linking observations over time while preserving the confidentiality of individual responses. For example, subjects may choose an easily remembered alias and use it on repeated occasions. Some approaches are quite complex. For example, research by the American Council on Education (Austin and Boruch, 1970) on political activism among American college students used the following three-file linkage system: 1. Initial data collection: * File A contains each subjects' data and arbitrary account number (X). *File B pairs each subject's name with a second arbitrary account number (Y). File C matches the two sets of account numbers, X and Y. *File C is shipped to a researcher in a foreign country 2. Second data collection: * Second set of identifiable longitudinal data are gathered. * Names are replaced by their Y account number; this file is shipped to the foreign researcher. 3. Data analysis: * Foreign researcher substitutes X account numbers with corresponding Y numbers. * Each set of data files is returned to the data analysts. * Data are organized in longitudinal sequences; the identity of each subject is unknown. * The longitudinal data are analyzed. * Foreign researcher destroys File C; the three files can never be merged to learn subject identities. Thus the data were safe. Conceivably, foreign discovery procedures could be used to obtain some of the identifiable data before File C is destroyed. Therefore a Certificate of Confidentiality (see below) could be obtained to preclude that unlikely event. Intersystem linkage. Intersystem linkage is sometimes necessary to link research records on subjects with other, independently stored records on the same individuals. In the case of highly sensitive data such as psychiatric or police records, a linkage strategy may be needed so that the researcher does not have access to any identified records. One such method is as follows: 1. Researcher wishes to link data on 50 subjects with information from their police records. 2. Subject provides data and an alias (no name) to the researcher. 3. Subject provides to the archive (police) his name and alias. 4. Archive provides the requested police information with the aliases (not the names) attached 5. Researcher analyzes relationship between his research data and the police record data. This brief summary is merely illustrative of some of the many specific procedures for preserving anonymity or confidentiality. The actual literature on this topic is immense. (See Boruch and Cecil (1979), and Campbell, Boruch, Schwartz, and Steinberg (1977). See also current U.S. Census Bureau papers on this topic.) Statistical strategies. Researchers have developed various statistical strategies to eliminate any direct link between the respondent's identity and his true answer. All of these methods involve the injection of a specified amount of random error into the dataset so that researchers cannot ascertain an individual's true identity but can still perform a useful statistical analysis of the data. The statistically elegant randomized response method is used in direct interview to protect privacy and to ensure confidentiality. The following is an oversimplified example: Suppose the researcher wished to ask whether respondents had struck their child in anger this week or cheated on their income tax this year obvious invasions of respondents' privacy. Each respondent is instructed to roll a die in his cupped hands and observe which number came up without showing it to the researcher. If the predetermined number was rolled, the respondent is to respond "yes," irrespective of the true answer. By an algebraic removal of the expected number of false "yes" answers from the data, the researcher can determine the true proportion of "yes" responses. Neither the researcher nor anyone else besides the respondent knows who gave a true "yes" response. The randomized response method has been in use since 1965, and researchers have developed many statistical variations of it. Although it has been tested in many settings, the jury is still out on its usefulness. It tends to produce somewhat more admissions of undesirable behavior than traditional face-to-face interviewing, but brings with it many disadvantages. It is difficult for subjects to understand and believe, and time-consuming to explain. The interviewer has an important role in establishing trust and understanding, and respondents of limited ability have difficulty understanding and trusting the procedure (Landsheer, van der Heijden, and van Gils, 1999). It injects random error necessitating larger samples. Without understanding the reasons why people might refuse to give candid answers, the routine use of a difficult method such as this seems inappropriate. For critical evaluations of the randomized response method, see Linden and Weiss (1994), Umesh and Peterson (1991), and van der Heijden, van Gils, Bouts, and Hox (2000). Methodologists have designed statistical strategies for use with longitudinal and multiple source data; see Boruch and Cecil (1979). Agreements: Data sharing, secondary analysis, and audit of data. Concern for the integrity of data and for extending the analyses of important datasets brings with it the need to do so without risk to privacy or confidentiality. The simplest solution is to render the data anonymous. However anonymity may render the data useless to the secondary user or auditor. To satisfy the needs of secondary users while also protecting the interests of respondents, the researcher can employ procedures that diminish (a) outright breach of confidentiality, (b) likelihood of deductive identification, (c) the sensitivity of the information to which the secondary users have access, or (d) the need for the secondary user to actually take possession of the data (see Boruch and Cecil, 1979). Researchers and IRBs need to develop contractual agreements with funders who require audits and secondary analysts with whom they share data about how they will ensure confidentiality. They should also include information in the informed consent so that potential subjects understand what researchers will do with the data subsequent to the initial project. Legal protections of confidentiality. Statutory protection of research data enables researchers to protect the confidentiality of research records on identifiable individuals from subpoena. Subpoena of social research data is rare. However if vulnerable data could not be protected from subpoena, there would be a chilling effect, especially on criminological and delinquency research. There is growing use of Certificates of Confidentiality. However researchers and IRBs are often unclear about the protections these provide and their limitations. Certificates of Confidentiality. The Public Health Service Act (PHSA) was amended (1970) to authorize researchers to withhold information concerning the identity of participants in research on use and effect of drugs. The Secretary of the Department of Health and Human Services grants this authority by issuing Certificates of Confidentiality. A 1988 amendment broadened its scope to include mental health, biomedical, clinical, behavioral, and social research. Under this amendment, the Secretary of DHHS may authorize persons engaged in biomedical, behavioral, clinical, or other research (including research on mental health, including research on the use and effect of alcohol and other psychoactive drugs), to protect the privacy of individuals who are the subject of such research by withholding from all persons not connected with the conduct of such research the names or other identifying characteristics of such individuals. Persons so authorized to protect the privacy of such individuals may not be compelled in any federal, state, or local civil, criminal, administrative, legislative, or other proceedings to identify such individuals. (42 U.S.C. 242a(b)(1989)) Various institutes within DHHS are authorized to issue certificates. Since 1993 DHHS can grant certificates for research that is not federally funded. DHHS regards a certificate's protection to supercede State law; this position has been challenged and upheld in the New York Court of Appeals (People v. Newman, 32 N.Y.2d 379, 298 N.E.2d 651, 345 N.Y.S.2d 502, 1973) (Boikess, 2000). A certificate does not protect identifiable data of "secondary subjects," a point which researchers may fail to clarify in the informed consent. The certificate only protects researchers against compelled disclosure of subjects' names or other identifiers, coupled with their data. It does not protect subjects who voluntarily consent to disclose their research records, nor preclude a researcher from reporting the identity of subjects who disclose intentions to harm themselves or others. Moreover the language of PHSA is imprecise, which gives rise to uncertainty. It offers protection to "names and other identifying characteristics," but the data of a known subject may not necessarily be protected. Melton (1992, p. 81) provides an example of this possible loophole: [I]n one of my own studies, all of the children in a particular county who are involved in criminal child abuse prosecutions are invited to participate. Knowing that fact, a defense attorney might seek the data of a particular child (not the names of participants) as a fishing expedition for information intended to impeach the child's testimony. A literal interpretation of the statute would suggest that the subpoena might be enforceable if the data could be shown in some way to be relevant to the proceeding. Although it is also possible perhaps even probable that a court would interpret the statute more broadly in keeping with Congressional intent, the uncertainty prevents unequivocal offers of confidentiality to participants and, therefore, should be eliminated by a technical amendment. It is also unclear whether child abuse reporting laws are abrogated by a Certificate of Confidentiality. Is such reporting a "legal proceeding" that cannot be mandated under a certificate? Researchers must request the certificate before each research undertaking. Subpoenas typically occur for reasons unrelated to the study itself and therefore are not reasonably foreseeable by either the subjects or the investigator. Hence the protections offered by a certificate may be unavailable when needed. Researchers sometimes send data to a foreign country, although this does not always guarantee protection. Placing data in a foreign country and laws governing foreign discovery. ***Footnote: am indebted to Dr. Joe Cecil and Jason Gilbert, Federal Judicial Center, for providing me with their detailed summary and analysis of these issues.*** Many survey researchers believe that sending confidential data to a foreign country - -such as to a colleague in Canada --protects the data from subpoena. However this is only a deterrent from subpoena, not a guarantee of protection. Federal Rules of Civil Procedure govern the procedures for discovery, including foreign discovery. Rule 26(b) states that parties may obtain discovery of anything that is relevant, not privileged, and admissible or "reasonably calculated to lead to the discovery of admissible evidence." Rule 34 states: (a)Scope. Any party may serve on any other party a request (1) to produce and permit the party making the request, (2) to inspect and copy, any designated documents, or (3) to inspect and copy, test or sample any tangible things which constitute or contain matters within the scope of rule 26(b) and which are in the possession, custody, or control of the party upon whom the request is served. (c)Persons Not Parties. A person not a party to the action may be compelled to produce documents and things or to submit to an inspection. The courts cannot compel a party to produce data if the party does not have "possession, custody, or control" of the documents, but it is unclear what constitutes "control." If a researcher sends data out of the country for the express purpose of preventing subpoena, does this qualify as loss of control in the eyes of a court? Jason Gilbert (2000), a legal intern at the Federal Judicial Center, offers the following analysis of this question: While the courts seem to have settled on defining control as when a party has a legal right to obtain something, questions remain for the researcher seeking to give up control of research data to a foreign colleague in an attempt to protect it from being disclosed. Legal rights to possession can come from a variety of sources, particularly when one is considering intellectual property such as research data. If a researcher were to create a set of data, when exactly would he or she no longer have a legal right to that set of data? What if the researcher gave one part of the data to a colleague? What if the researcher only gave up a small "key" to the data that allowed the individuals who participated in the study to be identified? What if the researcher gave part, or even all, of the data to a colleague but still continued to collaborate with that colleague to perform analysis on the data even though it was not in the researcher's possession? Would that researcher still have a legal right to get back what he or she had surrendered? While the concept of giving away the legal right of possession is relatively straightforward, the mechanics of how exactly a researcher can give away the legal right to possess his own data (particularly if one does not allow for a sale or some type of contract) remains unclear. Gilbert also reminds us of some other implications of "loss of control" of data. (1) Transfer of all data out of the country would mean loss of all electronic or hard copies in the researcher's possession. (2) A researcher must never transfer data after receiving a subpoena. Even if it is done as a safeguard beforehand, the researcher may still be found to have acted not in good faith and be cited for contempt of court. (3) If the research is done under a contract requiring that the researcher maintain control of the data, relinquishing control to a foreign colleague would constitute a breach of that contract. (4) The researcher's professional code of ethics or a future journal editor may require that the researcher maintain control of the data. A researcher who loses control of data by sending it to a foreign colleague places that colleague at risk of receiving a subpoena for the data and of having to seek legal means of protecting confidentiality. However the rules and procedures of foreign discovery are complex, expensive, and time-consuming. If the colleague who controls the subpoenaed information is a foreign national residing outside of the United States, the party seeking the data must follow appropriate procedures for foreign production. The United States has ratified various treaties concerning obtaining of evidence from foreign countries, and each country has its own procedures. Discovery in a foreign country involves sending a formal "letter of request" by the court where the action is pending to a court in the foreign country. This letter requests the foreign court to request documents of the person in possession of the desired information. There are various diplomatic and legal approaches to delivering such a request and accomplishing the discovery. These may make discovery of the information too unattractive to pursue. Descriptive statistics and raw data releases. Statisticians in governmental agencies in the United States, Great Britain, and Sweden have developed practices of adjusting tabular presentations so that deductive identification is not possible. Deductive identification could occur if one knew some facts about an individual, perhaps in conjunction with their zip code. By searching the files for that zip code and locating the individual whose data matched those known facts one could deduce additional information from the other data associated with that individual. For example, if an 84-year-old Hispanic woman from the 01373 zip code area was known to have had quite a few husbands and to have become quite wealthy, it would confirm quite a few suspicions to learn that there was an 84-year-old Hispanic woman at that zip code whose annual interest income was in seven figures and who had had 20 husbands. The most common way to prevent deductive disclosure is to broaden categories so that data from unique individuals or from groups of data containing some unique individuals such as top income earners or persons involved in high- profile criminal victimization are not apparent. Another is error inoculation so that no individual case could be assumed to be correct (as in the random-response method). When the U.S. Census Bureau's Disclosure Review Board reviews NCVS datasets in anticipation of their release to the InterUniversity Consortium for Political and Social Research (ICPSR), the Review Board's primary role is to scrutinize data for extreme cases that might uniquely identify the individuals involved or even subject them to criminal investigation. Thus, instead of showing that a given 84-year-old Hispanic woman had been widowed 20 times and had an interest income of over $1,000,000, it might show that the person had been widowed more than four times and had an annual interest income of over $100,000. Internet research. There is rapidly emerging literature on various kinds of Internet research, associated methods of solving problems of privacy and confidentiality, and uncertainties or vulnerabilities connected with these "solutions." Researchers' insouciant claims that Internet data are anonymous or that confidentiality will be protected are reminiscent of such promises regarding non-web research of several decades ago. This area of research will grow rapidly since it enables researchers to reach far-flung subjects quickly, inexpensively, round-the-clock, and without a research staff. The problems and solutions to issues of privacy and confidentiality will change rapidly as new technologies render old problems and old solutions obsolete. Some of the rapidly evolving issues include: * How to ensure that researchers do not study children under rules that pertain to adults. * How to ensure anonymity of responses given that web page software logs as header lines the IP address of the machine from which the respondent accessed the researcher's web page. * How to store an on-line data file so that unauthorized persons cannot access it. The advice offered by OHRP and generally heeded by IRBs is that the safest practice is to use the Internet for purposes of recruitment and screening. After locating a qualified respondent, a researcher can e-mail the survey and a respondent return it as an attachment. Given the uncertainties, especially with regard to assurances of confidentiality, it is reasonable at this stage to recommend that assurances of confidentiality contain appropriate disclaimers. Confidentiality assurances and their consequences The link between promises of confidentiality and willingness to participate in surveys is tenuous. Researchers' promises of confidentiality are not always effective in producing trust in research participants. Such promises cannot always be kept due to faulty data management practices and other possible compulsory disclosures. Moreover, the relationship between faith in confidentiality promises and participation in survey research is not what most suppose it to be. Singer, Mathiowetz, and Couper (1993) investigated the relationship between concerns about confidentiality and mail returns to the 1990 census. Such concerns have only a very slight effect on survey participation, and this relationship holds even when demographic variables known to be related to concerns and survey participation are controlled. Similarly assurances of confidentiality have unexpected effects. Singer, VonThurn, and Miller (1995) conducted meta-analysis of 30 research reports on the relationship between various forms of confidentiality assurances (anonymity, use of the randomized response method, and verbal assurances). They found that the effect of confidentiality assurances on willingness to respond is small, positive, statistically significant, and robust in the presence of various control variables, but only when sensitive questions are asked. The effect is small and negative when the questions asked are not sensitive ones. Elaborate assurances of confidentiality defeat their purpose when the contents of the survey are not sensitive. Apparently such assurances of confidentiality heighten respondents' perceptions of the sensitivity or threat of the survey or arouse their suspicions (Singer, Hippler, and Schwarz (1992)). Emerging issues of privacy and confidentiality Electronic data collection and storage practices are changing rapidly. Interviewers and their subjects need not meet face-to-face and may even reside in different parts of the world. Emerging issues of confidentiality are more varied and dangerous than current policy makers can easily anticipate. Soon issues of confidentiality will be transformed in ways we cannot imagine today. There are already digital communication networks on a global scale, and hackers with a laptop computer and Internet technology could download any electronic data stored on any server anywhere in the world. There are also emerging technologies for protecting communication and personal identity, and there is a whole new cohort of technology- sophisticated privacy activists. Governments are developing and testing new laws that protect data, and globalization of culture and policy processes is occurring. The American Association for the Advancement of Science, the National Science Foundation, and various Internet research groups are now actively exploring these issues. Scientific societies concerned with the protection of social and behavioral research data the American Statistical Association, American Psychological Association, and American Sociological Association will continue discussing these issues at their annual meetings for years to come. 4. Other procedures for minimizing risks and promoting benefits One cannot begin to minimize risks or promote benefits unless the risks and benefits are first identified. Although conducting a risk and benefit assessment is a fundamental concept in the planning of ethically responsible research, some of the possible risks and benefits are not immediately obvious. Possible risks and benefits are virtually unlimited. Identifying and minimizing risks In section 3, six kinds of risk were identified: inconvenience, physical risk, psychological or emotional risk, social risk, economic risk, and legal risk. They are but one dimension of the risk matrix that is presented here (figure 1). A second dimension is the location within the research process where the risk might occur. A third dimension is the specific vulnerability of the individual subject; for example, public figures are vulnerable, but in a different way than those engaged in illegal activities or those who are institutionalized. The six kinds of risk could be elaborated or conceptualized in various ways. In section 3, we viewed them simply as subsets of risk related to invasion of privacy or breach of confidentiality. They are also issues of personal safety or well being. Some of these risks can come about through misunderstanding, outright deception or concealment in the way the survey is administered, and inequitable treatment by the researchers. Researchers can address these risks in the communication process of informed consent (with informed consent being regarded as an ongoing communication process, not simply a consent form). Issues of ownership of the data and the knowledge are also relevant. If shared, are the data shared responsibly? Is the new knowledge used in a way that benefits the respondents or harms them? What is the relationship of the project to the gatekeepers who facilitated the recruitment process, and to the opinion leaders who will create attitudes toward the project and the respondents? Figure 1. Risk matrix, by part of the research process at which risk occurs and by type of respondent Kind of risk Part of research process Inconve- Psycho- and type of respondent nience Physical logical Social Economic Legal The theory or idea Public figure Lack autonomy Lack resources Stigmatized Criminal Secondary subject Disabled The research process itself Public figure Lack autonomy Lack resources Stigmatized Criminal Secondary subject Disabled The setting in which the research occurs Public figure Lack autonomy Lack resources Stigmatized Criminal Secondary subject Disabled The use or dissemination of the findings Public figure Lack autonomy Lack resources Stigmatized Criminal Secondary subject Disabled Researchers may conceptualize the scientific or research activity as having four parts, each of which has risks (as well as benefits): The idea, theory, or hypotheses that drives the design of the survey, sampling of populations of interest, and analysis and interpretation of the results. A new idea can affect individuals and change cultural values in fundamental ways. The idea may not be correct but it might be falsely confirmed by faulty analysis and interpretation. The idea may be correct, but be disseminated in a way that does harm. The researcher has four ways to reduce the risk of false confirmation or dissemination of damaging ideas: recognize that the null hypothesis could be true; design the research so that each possible theoretical orientation is tested fairly; remember the limitations of the models and measures employed and warn that application and generalization to other populations must be done with caution; and share the documented data with other scientists who want to verify the findings or test alternative hypotheses. The research itself has several stages, each of which could involve risks: the recruitment, the induction, the consent and survey, the debriefing, data analysis, and data sharing. The institutional, community, or group setting of the research. Research is rarely only a matter between researcher and respondent; there are usually third parties involved. Every setting has its members, structure, culture, and interests. Most settings have gatekeepers who impose rules on the research transaction. All of these elements may impose pressures and risks on both the respondents and the researcher. Because individuals vary in their degree of personal autonomy and institutions in their degree of control and coercive power, the kinds of harm that may result from research in those settings also range greatly. Uses of the research findings. In their enthusiasm to use what is learned, investigators can overlook that many findings are based on measures of dubious reliability or account for so little of the variance that they should not be considered practically useful. Researchers can misuse findings that are useful for one purpose but harmful for other purposes. Respondent vulnerabilities are covered in section 7. The following are some examples of the range of vulnerabilities researchers should consider when engaging in risk assessment. Respondents who are public figures are especially exposed to attention and criticism; for them privacy is a sought-after luxury. Those lacking in autonomy or resources to protect their privacy lack the usual means of preventing intrusions from others; these may include children and the mentally and emotionally disabled. Scapegoats and targets of prejudice, such as homosexuals, persons with HIV infection, and some racial minorities, are especially sensitive to researchers' interpretation of their lives and their data. Institutionalized persons lack the normal degree of autonomy and may also be stigmatized. Persons engaged in illegal activities have many fears concerning disclosure of information about themselves. Persons damaged by the respondents' revelations, the "secondary subjects," are harmed often without any knowledge of the harmful disclosure. The risk matrix serves to sensitize researchers and IRBs to the kinds of risks they may anticipate (figure 1). Once risks are identified the quest for safeguards or protections can begin. Researchers can minimize some of the risks identified by this matrix by use of approaches that respect privacy and ensure confidentiality. Researchers can minimize other risks through mindfulness of the possible harms that may result from the research findings. Identifying and maximizing possible benefits Survey research is typically considered to benefit respondents by giving them an opportunity to discuss a topic relevant to their life, and by providing valuable information to organizations and to society in general. However there are other benefits to the respondents and their community that are typically overlooked (figure 2). Figure 2. Benefit matrix for survey respondents and their community Benefit Respondents Commmunity Relationships Gain rapport with Create ties to the project respectful researcher Knowledge Conduct informative Understand victimization debriefing, having one's issues relevant to questions answered community improvement Material Provide informational Provide brochures, books, resources handouts videotapes, and other communication media Training Develop relationship Teach crime deterrence between some of the respondents and the community-based project Earn esteem Earn praise for program Empowerment Earn prestige from an effective program Cultural anthropologists have pioneered in thinking about ways to benefit the respondent's community partly for practical reasons. Without making the research a win-win situation for themselves, respondents, and the host community, researchers would quickly lose support from the hosts. Anthropologists such as Pertti Pelto (described in Sieber, 1991) have hired selected community members as research assistants, then shared their data with the host community so that it can be used for public policy purposes. Thus they have left behind a cadre of new researchers who can remain in contact with the investigator or can continue to build on and analyze the dataset on behalf of their community. With this model in mind, researchers should consider the potential value in employing and training members of some target communities to assist with some aspect of the NCVS interviews or dissemination of information. Conceivably, once researchers interview a cohort seven times and the project is ready to move on, researchers could distribute information on crime prevention to the particular respondents and to the community, and to assist the community (via the police department or other agencies) in helping its citizens make optimal use of that information. Researchers should use both the risk and benefit matrices early in the research planning process to generate ideas. They can then translate the resultant assessment of possible risks and benefits into actual operations. 5. Procedures for responding to requests for help or assistance Effective interviewers are skilled at establishing trust and rapport. These qualities tend to instill in respondents a sense that the researcher is a responsive and capable professional. When the interview topic is a sensitive one that evokes concerns in a troubled respondent, it is not unusual for that person to reach out to the interviewer for help. The interviewer is often only that -- a paid interviewer, not a helping professional. However because of the sensitivity of such situations -- this may be the first time the individual has dared to reach out for help -- it is important that the interviewer not rebuff or ignore the request, and respond in an appropriate and helpful manner without getting personally involved. Standard procedures When researchers work with vulnerable populations or in situations where they may be asked to provide help or assistance, it is standard for IRBs to require that the researchers prepare a set of referrals. The required response varies with the type of research. In some cases researchers give respondents an information sheet (that may include referrals) to thank them for participating and as part of the debriefing. In other cases, researchers develop various referral sheets, depending on the kind of help that is requested. In research with vulnerable participants who are at high risk, the researchers establish a relationship with specific professionals and arrange for the referred respondents to have several sessions with the professional. In the case of high-risk respondents, IRBs may require appropriate training to help interviewers recognize and interpret responses indicative of need for help. Research institutions that engage in research with high-risk populations usually require their interviewers to handle requests for help as follows: If the request comes at the beginning or during the interview, the interviewer is to ask if the interview can continue and promise to give help or assistance when the interview is concluded. At the end of the interview, the interviewer gives the respondent the appropriate referral information. If the respondent simply cannot wait for the end of the interview, the interviewer provides assistance, thanks the respondent for the information provided so far, and discards the interview responses. Researchers may write the costs of the professional's services into grant support for the project or into the research institute's budget. Background considerations It is important to understand the degree of response that might be appropriate in a given project. When should interviewers actively intervene in the lives of persons who need help and when should they simply provide referral information? It is useful to consider a range of kinds of high-risk respondents. Survey researchers who study social psychological aspects of HIV infection and injection drug use have faced the ultimate challenges of responding appropriately to requests for help. Hence investigators now apply, when appropriate, a range of practices that were developed in survey research on HIV infection and drug treatment to other high-risk settings. Interviewers in the HIV and drug treatment fields receive HIV post-test counseling to acquaint them with the feelings, concerns, and behaviors of persons who have recently learned that they have HIV, who fear that they may have HIV, or who have significant others who have HIV. Contacts for counseling and treatment are set up prior to the interviews so that the referral process can work seamlessly. This includes a police contact. In survey research in the HIV and drug treatment fields there is now a routine that is followed quite strictly. If an interviewer encounters someone who is distressed, or someone the interviewer is worried about, the interviewer has a telephone number and perhaps a cell phone he can use to call for help right away or give to the respondent to use right away. When survey research is conducted on the streets, projects require that a team member accompany the interviewer to the interview, even if there is only one interview scheduled. The team approach is used for the safety of both the respondent and the interviewer. If the interview is held in the project's field office, another team member is always present when a respondent is in the office. For very immediate emergencies the interviewer helps the recruiter or receptionist until assistance arrives. Again, everyone has prearranged telephone numbers. Ultimately, such field survey projects must rely upon the skill, humanity, and insight of their staff to sense incipient problems and to deal with them. This is one reason why researchers seek staff members who have related prior experience, including knowledge of this culture in their own lives and prior work with emergency situations. As one project director commented: "We seek staff among people who live on the edge. There is no substitute for experience." Some of the same precautions are often used with survey research on prostitutes, drug dealers, and other members of street cultures. However most survey research is not conducted "on the edge." Interviewer training and preparedness remain important but the emergencies they may face are often more subtle, less immediate, and less frequent. A "cry for help" may be implicit or explicit. Examples of implicit responses include statements indicating desperation, suicidal ideation, or a high score on a depression scale. Explicit responses might be a statement of extreme duress such as "My spouse has threatened to kill me," accompanied by a request for help or advice. It could be about someone else in danger, for example "My spouse threatened to kill his boss. What should I do? Can you help me?" (There are also requests for help that would trigger mandated reporting to authorities; see section 6.) It may come from a respondent to an anonymous survey, from an identified survey respondent, or face-to-face in an interview. Interviewers have effective and responsible ways to answer such cries for help but must take appropriate precautions. Implicit request The ethical requirement of beneficence (National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research, 1979) and the code of ethics of the American Psychological Association (1992) would seem to endorse the responsibility of researchers to intervene when serious distress is discovered or caused by the research procedure, but may conflict with the principles of respect for autonomy and privacy. If the researcher is to take any steps in response to apparent need, should the respondent be warned in the consent statement that such action would be taken? If this is done, what effect would such a warning have on willingness to participate or to be candid? This set of questions was answered quite effectively in an experiment by Stanton, Burker, and Kershaw (1991) on the effects of researcher follow-up of distressed subjects. Since researchers studying depression often encounter respondents in serious preexisting distress, these investigators sought to study the ethical responsibilities of such researchers. A literature search revealed that researchers' responses to such subjects range from doing nothing to recontacting both the distressed subject and a significant other. Stanton et al. studied whether respondents adjust their reports of distress on three standardized measures of depression as a function of their expectation of experimenter follow up. The content of their informed consent was manipulated to reflect four levels of researcher intervention: (1) no intervention, (2) provision of a list of treatment resources to all subjects, (3) follow- up contact by the researcher with distressed subjects, and (4) follow-up contact by the researcher with the distressed subject and a significant other. Their results were quite dramatic and consistent across all three measures of depression. Respondents reported the highest average level of distress in condition 2, in which all subjects were to receive a list of treatment resources. The next highest average level of distress was reported in condition 1 (no intervention). Mean distress measures were markedly lower when subjects expected the researcher to recontact distressed subjects (condition 3), and much lower when they expected the researcher to recontact them and their significant other. Some of these effects were slightly stronger for men than for women, but the gender differences were minor. The implications for survey researchers are quite clear: Respondents do not want to receive personal help unless they ask for it. However they appreciate receiving information that would enable them to better understand their situation or to seek help for themselves. The expectation that the researcher might intervene personally is so distasteful to most that respondents will falsify their answers to avoid this. The expectation that the researcher might contact family members or significant others is even more objectionable. Obviously, the simple provision of information to all subjects both provides the desired kind of assistance and respects privacy and autonomy. Moreover it is easy to administer and the interviewer will not need any specialized training. The anonymous survey that may reveal serious distress is a related issue. Researchers may administer anonymously surveys of topics such as mental health, stress and coping, high-risk behaviors, health concerns, and experience of abuse or violence due to the unwillingness of many respondents to reveal personal or embarrassing details about their lives. How does one intervene to help anonymous persons in serious difficulty? The Stanton et al. study emphasizes that personal or clinical intervention by the researcher is unwelcome if not requested. Researchers have used the following approaches. The interviewer may provide: a list of treatment resources, self-help guides, or useful factual information to all respondents, ample opportunity for debriefing in which the respondent can express any concerns or interests and receive a referral slip (with no name or other identifier) for a nearby clinic if the respondent so desires, an admission ticket to a nearby workshop or panel on the topic, or web addresses to constructive web discussion groups on the topic. If respondents come from identifiable groups such as classrooms or workplaces the researcher may arrange for the organization to provide an appropriate educational program to the whole group. Explicit request The interviewer can respond less awkwardly to an explicit request for help because there is less chance of offending the individual by proffering assistance. The interviewer can offer assistance in proportion to the nature of the request. In extreme situations, the interviewer may step in, provide a therapeutic debriefing, and arrange for immediate referral. However interviewers must avoid getting drawn into the dual role of therapist/helper and interviewer. They must be prepared to explain to the respondent that they are not qualified to help but can provide the services of a competent caring person who can. A specific recommendation: An experiment within the survey The U.S. Census Bureau faces an unusual dilemma regarding response to requests for assistance because its interviewers return to the same group seven times to study its members' experience of crime victimization. By offering referral information on any of the first six interviews, Census may alter the phenomenon that it seeks to study. Census interviewers currently respond to requests for help only after the last interview unless the individual is in truly dire circumstances. In those extreme cases the interviewer responds earlier in the sequence of interviews, but then drops that household from the survey, replacing it with a demographically similar household. However the current Census practice raises several questions. How much difference does a referral actually make in the life of a respondent who asks for help? Is it helpful? Harmful? Does the person who asked for help actually use the referral? What if researchers provided a list of referrals to all respondents for all of the sorts of crime victimization surveyed? Does the making of a referral, requested or otherwise, impact the nature of the data subsequently gathered from that household? With what frequency are requests made for assistance? There are several implicit concerns in this discussion: (a) concerns for statistical accuracy of estimates of amount of crime victimization, (b) concerns to learn whether providing a referral actually changes the course of events in a victim's life (other than impacting the kind of data Census seeks to gather), (c) concern to help someone who reaches out, (d) concern to provide possibly useful referral information to all respondents in return for their participation, (e) concern about the frequency and nature of requests for assistance, and (f) concern that Census avoids criticism for failure to help. If interviewers do not already record the incidence and nature of requests for assistance, they should do so. What is the relative frequency of the various kinds of requests? What are the demographic characteristics of respondents who make each kind of request? How do requests change in relation to changes in crime rates? These baseline data would provide a useful backdrop for the following experiments: Experiment 1. Effects of referral sheets given to all respondents. Researchers would draw a stratified random sample of respondents and divide them into experimental group members and matched controls. After the first interview, researchers would give all experimental group members a referral sheet recommending local sources of assistance for each of the kinds of crime victimization surveyed. Control group members would receive no referral sheet. At the end of the 3 years and seven interviews, interviewers would reinterview members of both groups to learn whether there were kinds of relevant assistance they had desired, whether they sought assistance, how they went about it, and with what result. Interviewers would note whether the experimental (referral) group members mentioned the referral sheet, used it, were more likely to seek referral of any kind, and whether their data reflected a lowered rate of victimization than matched controls. Experiment 2. Effects of appropriate referrals provided in response to requests for help. When interviewers are asked for help they would use a randomizing procedure to decide whether to offer a referral on odd dates give referral, on even dates do not. Interviewers would use this random assignment procedure in all but the most extreme cases where failure to make a referral would place the respondent in serious jeopardy. Interviewers would note who asked for help and other pertinent details, including whether the respondent was assigned to the experimental or control group, the kind of help requested, when in the sequence of interviews the request was made, and other relevant details. At the end of the seven interviews, interviewers would reinterview all members of the experimental and control groups to learn whether they had sought and obtained assistance, how they went about it, and with what results. Researchers would compare the experimental group's subsequent rate of reported victimization with that of the control group. A conservative version of experiment 2 would employ the replacement sampling procedure Census currently uses. Census would provide referrals, drop those household from the survey, and replace it with a matched household. However Census would continue to interview the dropped households for the entire 3-year sequence so that Census could validly compare their data with that of the remaining households. The postexperiment reinterview would be the same as that described above. By including small-scale experiments such as these into its survey procedure, Census can learn what happens when a referral is made, whether it is helpful, if it can be made more helpful, and whether referrals significantly impact subsequently reported victimization. If referral information is helpful to persons in avoiding or dealing with victimization, but impacts statistical results, researchers might provide referral information only at the end of the 3-year sequence or in response to those kinds of requests for which referrals were found to be helpful. 6. Procedures for responding to mandates to report There are various statutes that require researchers to report revelations of certain kinds of situations to appropriate authorities. These vary by State in terms of what must be reported and who must report. Because of the State-by-State variability of mandated reporting, national surveys that cross State boundaries must either modify their procedures for each State or adopt the most stringent reporting standard for the entire survey. The kinds of revelations that call for reporting to appropriate authorities may include child or elder abuse or neglect and intention to harm others or oneself. It is assumed that the NCVS is exempt from mandates to report because the survey is conducted under Title 13, the U.S. Census Bureau confidentiality statute. Probably, NCVS interviewers would only report explicit verbal mention of abuse, which would occur rarely. In response to such mention, the interviewer could encourage the individual to self-report. The interviewers might have with them the 800 number of a nearby appropriate agency in that State. (See Appendix A for State phone numbers to obtain local agency information and phone numbers.) This would satisfy both the moral duty to respond and the legal duty to abide by conditions of Title 13. Researchers should anticipate (rather than respond to) possible revelations. In the interest of confidentiality, researchers should include a warning of the limits of confidentiality in the informed consent for surveys or interviews that might conceivably reveal evidence of reportable situations such as child abuse. While this might distort the random sampling scheme and jeopardize generalizability by eliminating those who decline to participate, it provides a higher level of confidence in those who choose to participate and eliminates the legal and ethical horrors of having to consider breaching confidentiality to comply with the law. If this procedure raises significant generalizability concerns, the researcher may consider conducting a parallel study of parents who have been convicted of child abuse. Child maltreatment The Federal Child Abuse Prevention and Treatment Act of 1974 required that each State establish child protective services and develop its own mandated reporting laws. By 1978, State reporting laws were in place; Levine and Levine (1983) include the history of these laws. State laws mandate the kinds of situations that must be reported and the kinds of persons who must report helping professionals only or anyone. There is considerable State-to-State variability of laws and unpredictability of court decisions. Because most large sample surveys cross State boundaries, IRBs and survey researchers must know the relevant State laws and trends in court decisions, and should have policies and procedures in place for interpreting those laws correctly. Specifically, researchers should consider State law in the formulation of the informed consent statement about the limits of confidentiality, in the education the IRB provides concerning kinds of evidence that require reporting, and in the plans and consultation arrangements that the researcher and IRB develop for handling such cases. See Kalichman (1999, pp. 14-23) for the definitions of abuse and requirements to report that are excerpted from the reporting statutes of each State. All States require reporting by certain helping professionals such as physicians, psychiatrists, clinical psychologists, counselors, teachers, nurses, and social workers. Some also require reporting by pharmacists and religious healers. In their efforts to locate makers of child pornographic films, Colorado and Illinois require reporting by commercial film developers. Anyone who has reason to suspect child maltreatment must report in the following nine States: Florida, Indiana, Kentucky, Minnesota, Nebraska, New Hampshire, New Jersey, New Mexico, and North Carolina. Obviously, interviewers working in these States are operating under a different mandate than those working in the remaining States. However the likelihood of their discovering reportable cases is largely limited to situations in which they are explicitly told about the maltreatment, an event which would be exceedingly rare. In evaluating a given protocol the IRB must consider whether there is a chance that the researcher will find reasonable evidence of a reportable situation. Reporting laws vary among States with respect to how one learns about the suspected abuse. In some States, a report is required even if the reporting individual learns of it through a third party. Most statutes require the reporting individual to testify in court proceedings, include a criminal penalty for failure to report, and permit civil action against a professional whose failure to report resulted in injury to the child. However all statutes provide immunity from a suit when a report made in good faith turns out to be unfounded (Levine, 1982). If the IRB believes that a reportable revelation might occur, it will require that the informed consent statement include a warning about the limits of confidentiality. A statement adapted from one developed by David Ruja (Gil, 1982) is: What is discussed during our session will be kept confidential with two exceptions: I am compelled by law to inform an appropriate other person if I hear and believe that you are in danger of hurting yourself or someone else, or if there is reasonable suspicion that a child, elder or dependent adult has been abused. The same sort of warning must appear in the parental permission for research on a child. The problems of coping with mandatory reporting have resulted in creative approaches to recruitment and consent procedures. One researcher who wished to study the responses of non-abused children to anatomically correct dolls resorted to the following screening procedure. She told parents that children could not participate who: (a) were under treatment by a mental health worker, (b) did not speak English, (c) had been sexually abused, or (d) had an infectious disease. Thus she could rule out children who had been abused without knowing who they were. The major problem faced by IRBs and investigators is not the State-by-State variability of laws but the lack of clarity about their interpretation. It is not clear whether "reason to believe" refers to a clinical hunch or to firm evidence, nor do these laws define what constitutes abuse. This leaves researchers to consider cultural differences and to weigh these against the possibility that the legal bureaucracy may be more harmful to the victim than are their seemingly abusive relatives. The difficulties of defining abuse are many. Estimates of the amount of child abuse vary from 1% to 30% of all children depending on one's definition (Weis, 1989). How is the act perceived by the child to teach an important lesson (Corbin, 1987), to cure a disease (Gray and Cosgrove, 1985), or out of malice? Thus, added to the costs of breaching confidentiality is the possibility that reporting will harm both the "victim" and the "perpetrators." What is ethically responsible research behavior with respect to reporting? Should the researcher stop and warn the subject who starts to mention abuse that it is reportable information? Should the researcher actively seek and report evidence of abuse and neglect? BJS and U.S. Census Bureau researchers do not have to solve this dilemma. Census interviewers can respond to mention of abuse by encouraging the respondent to self-report and by providing an appropriate phone number. For investigators who do not have statutory protection such as Title 13, this is an area in which IRBs and researchers need consultation. They should establish relationships with social workers, pediatric nurses and physicians, or clinical psychologists who are competent to judge evidence that may trigger reporting. Some IRBs may not recognize when there is risk of uncovering evidence of reportable activity. If risk is recognized, the ambiguity of State laws concerning reporting can lead to extreme IRB decisions such as rejecting the entire protocol or suggesting poor solutions. If the IRB does not have a knowledgeable clinician among its members, it should call upon such a person for advice as needed. Clinically trained practitioners know how to interpret verbal or behavioral communications and are able to determine the appropriate action. They probably are acquainted with the Child Protective Services agency in their area and with the strengths and weaknesses of its professional staff. They will know how to report suspected abuse in a way that maximizes the likelihood of a beneficial outcome. IRBs that frequently review protocols for research that might happen upon evidence of abuse should arrange permanent institutional resources to advise and support researchers in this area. Without a trained clinician to advise on what constitutes "reasonable evidence," a risk-averse researcher or IRB may overreport to protect themselves from possible prosecution. Their duty is to make a considered decision in consultation with others qualified to advise, not to jump to conclusions and report without consultation or sound advice. IRBs should develop guidelines for reporting that are tailored to the specific State and local situation and to the resources available to them for consultation. It is important that investigators at risk of discovering reportable abuse understand the significance of warning respondents of their duty to report. Federal regulations regarding confidentiality require that researchers warn subjects of mandatory reporting requirements, and researchers must be ready to respond appropriately to signs of abuse. Realistically, however, this requirement protects researchers and not victims. Moreover it does not apply to the U.S. Census Bureau. The most authoritative, comprehensive and helpful source of information on mandated reporting of child abuse is Kalichman (1999), who provides comprehensive reporting requirements. Elder abuse Elder abuse is a significant problem, and elderly respondents may self-report to NCVS interviewers. In 1996 a national incidence study found that over 500,000 elders experienced abuse that year, 75% of which was unreported. In 90% of cases, the perpetrator was a family member, two-thirds of whom were adult children or spouses. The generally accepted definitions of elder abuse include: Physical abuse which is the willful infliction of physical pain or injury, including slapping, bruising, sexually molesting, or restraining Sexual abuse which is non-consensual sexual contact of any kind. Financial exploitation which is using the resources of an older person without their consent for someone else's benefit. Neglect which is failure of caretaker to provide goods or services necessary to avoid physical harm, mental anguish, or mental illness. The Administration on Aging (AoA), Department of Health and Human Services, is the only federal agency dedicated to policy development, planning, and delivery of supportive services to elders. There are also State elder abuse prevention programs. There is now federal legislation requiring that States develop legislation similar to that for child maltreatment. However this legislation is relatively new and the mandated programs are not fully developed. AoA funds the National Center on Elder Abuse, which is located at 1225 I Street, N.W., Suite 725, Washington, D.C.; phone 202-898-2683; e- mail NCEA@nasua.org. Their web site, http://www.elderabusecenter.org, includes a State-by-State listing of toll-free phone numbers for reporting elder abuse. Additionally, this web site contains comprehensive information on elder abuse. After reporting, the agency screens calls for potential seriousness, keeping the information confidential. If the agency decides there is violation of State elder abuse laws, the agency assigns a case worker (in emergencies, usually within 24 hours). If the victim needs crisis intervention, services are available. If no abuse is substantiated, most agencies will work as necessary with other community agencies to obtain any needed social or health services for the elder. The elder has the right to refuse services offered. As with child maltreatment, NCVS interviewers could respond to evidence of elder abuse appropriately by urging self-reporting and providing the appropriate State hotline numbers. Intent to harm oneself or others The intent to harm oneself or others are issues that primarily impact mental health clinicians who probe into the motives and intentions of their clients. However it is conceivable that feelings of despair or anger evoked in an interview on crime victimization could give rise to statements about harming oneself or others. Intent to harm oneself is discussed in section 5. An appropriate response is to urge the individual to seek help and to give them an appropriate referral (phone number). Intent to harm another is an issue that rarely arises in either psychotherapy or social/behavioral research. However there are a few State laws governing the duty of psychotherapists to respond to such a threat by warning the intended victim. Consequently psychotherapists have had to include in their informed consent a warning of their duty to report such events. The disposition of courts to find that there is a duty to warn an intended victim has spread throughout the country. Despite the rarity of statements about intent to harm another, the cases in which actual harm has arisen resulted in some surprising court decisions and tremendous amounts of publicity. Under the protection of Title 13, the U.S. Census Bureau need not warn of a duty to report and may not report. However the U.S. Census Bureau may want to consider the kinds of referral information that it could provide to individuals who are planning to take justice into their own hands. This might include a range of alternatives that include counseling, mediation, and referral to appropriate law enforcement agencies. The U.S. Census Bureau should stay current about the development and spread of legal mandates to warn intended victims. The case of Tarasoff v. University of California Regents is instructive. A University of California Berkeley graduate student, Prosenjit Poddar, revealed to a campus psychologist his pathological intent to kill Tatiana Tarasoff, who had spurned his affections. The psychiatrist notified the police, who found the man rational. Poddar understandably did not return to therapy, and stabbed Tarasoff to death. Through a series of appeals, the Tarasoff family persuaded the California Supreme Court (1976) that mental health professionals have a duty to intervene in such cases. Depending on the case, intervention can include warning the intended victim, notifying authorities, or securing an involuntary commitment. Although some therapists and researchers consider this an unacceptable infringement on their duty to hold professional information confidential, the Tarasoff law mandates that professionals must intervene effectively when the client in therapy, including those in research on the therapeutic process, reveals an intent to harm another. Other States have copied the Tarasoff law in various forms. Even in States that do not have "Tarasoff" law, it is reasonable to consider whether victims or their families might seek to apply the Tarasoff principle if a person indicates intent to harm and then commits a violent act. According to existing law, the Tarasoff principle applies only to therapists. However many IRBs require that researchers report to appropriate authorities any disclosure of intent to seriously harm another, irrespective of whether the researcher is a therapist. Some researchers seek to protect themselves from lawsuits for breach of confidentiality by including in the informed consent a statement such as "Your data will be kept confidential within the limits of the law." This is unadvisable. Recent court cases have upheld respondents' rights to an informed consent that they can understand. IRBs are now increasingly careful about how they state their obligation to report. On their IRB's web site, the University of Chicago provides the following example of a possible statement for inclusion in a consent document regarding a survey of persons with depressive symptoms: Some of the questions in the written forms and the interview ask about how you are feeling now. If your answers make us think that you might harm yourself or others, we are required to notify the proper authorities of this risk. The Census Bureau and BJS should prepare policy regarding a response to evidence of child or elder maltreatment, or threats of harm to oneself or others with appropriate referrals, and be sensitive to the requirements of investigators who do not operate under Title 13. 7. Vulnerable populations: Risks and special protections Vulnerable populations normally include children, teenagers, pregnant women, cognitively impaired persons, and institutionalized or imprisoned persons. For purposes of the NCVS and most other social surveys, the vulnerable populations requiring special precautions to minimize risks probably also include the elderly, minorities, economically or educationally disadvantaged persons, and other "institutionalized" groups such as students recruited as research subjects by their teachers and employees recruited as research subjects by their employer/supervisor. IRB concerns In the case of prisoners and children, Subparts C and D of 45 CFR 46 permit IRBs to approve research that involves no more than minimal risk, or that may involve somewhat more risk but will benefit the participants. Subpart C is pertinent to survey research in that it provides extensive safeguards to confidentiality and against coercion so that prisoners are afforded autonomy in their role as research subject. Subpart D is pertinent to survey research in that it requires both parental approval and the child's assent, except as noted in section 9. In the case of each of these populations, IRBs are sensitive to whether it is appropriate to include the given population, consent is informed and autonomous, participation is free of coercion, and the language and presentation are appropriate to the needs of the particular population. Privacy and confidentiality have special significance for each population and researchers should safeguard them appropriately. Specific fears, not necessarily warranted, are germane to each group and investigators should learn and dispel them. Each group has its own culture and language characteristics, which researchers need to understand and respect. Children and teenagers The privacy needs of children and teenagers change markedly as they grow from early childhood through the teen years. Thompson (1982) presents an excellent review of these changes, showing that popular ideas about vulnerability decreasing linearly with age are inaccurate. Older children and teenagers are more easily embarrassed, more concerned about personal and informational privacy, and more likely to feel upset if they reveal more than they intended. As they approach their teens children become sensitive and skeptical about situations in which an interviewer might conceal or withhold information and upset at the prospect of being deceived. In their efforts to forge their own sense of identity, youngsters approaching and in their teen years would not want to be interviewed on a sensitive topic in front of their parents. They might even offer to disclose information only on the condition that the interviewer not reveal it to their parents. Because NCVS interviewers are exempt from mandatory reporting under Title 13, this is less of a dilemma for them than for other researchers. If appropriate, the NCVS interviewer might provide referral information to a troubled teen or urge that they confide in their parents, but need not take the matter farther. Older children and teenagers are developing means of controlling their privacy. In middle childhood, children begin to want to have private places of their own, and are likely to post a sign on the door to their room saying "KEEP OUT." By the teen years, youngsters are intensely private, easily embarrassed, and interested in issues of informational privacy. Some will refuse to provide information if they feel intruded upon. Having a private place to be alone, or to communicate with friends in privacy becomes extremely important. For the teenager, promises of confidentiality are extremely important. In contrast, young children, up to about age 7, are not easily embarrassed and have an insouciant charm and candor that comes from lack of self-awareness. Their means of protecting themselves from unwanted intrusion is their parent's protection of their privacy. Younger children's sense of privacy is enhanced when their parent is present during an interview. By the same token, the requirement of a parent's permission to have the child interviewed serves to protect the overly shy or emotionally unstable child from the researcher's intrusion. The younger the child, the more important it is for investigators to pilot test consent (assent) information and surveys to determine whether children can understand the language. Young children, under age 6, usually do not understand statements dealing with hypothetical situations, and have difficulty with quantifiers and relational concepts. As children mature, they develop increased understanding of the hypothetical or conditional. The use of long sentences that would be understandable to an adult is problematic with younger respondents. Some guidelines for communicating with children in survey research follow: Use short sentences with easy words to improve comprehension. Avoid unnecessary clauses that complicate the question. Avoid use of the passive voice. Examine the child's understanding of words. The interviewer might ask "Tell me what a ..is?" Make sure the child understands the context of the question, perhaps by prefacing a question with a statement like: "Remember the time when (such and such) happened." Avoid asking a young child "why" they did something, which the child will likely perceive as critical and make the child feel defensive. It may work better to rephrase the question as a "what" question such as "What made you feel that way?" Rephrase questions to check for comprehension when the child seems to lack comprehension. Pregnant women Since it is conceivable that pregnant women's experience of crime victimization may differ from that of others, there is clear justification for their inclusion. The only regulatory issue pertaining to their inclusion is whether their participation in any way jeopardizes their health or that of their fetus. This is unlikely to happen in survey research. Cognitively impaired persons Although there are no regulations pertaining to research on the cognitively disabled, there has been considerable discussion and concern (see at http://ohrp.osophs.dhhs.gov/, go to IRB Handbook) about interpretation of the regulations in their case. The primary concern is that persons with psychiatric, cognitive, or developmental disorders, or persons who are substance abusers may lack capacity to understand what interviewers are asking them to do or to make a reasoned decision about participation. Many in this population are institutionalized, which further compromises their ability to exercise free choice. Institutionalized persons may want to participate in research to appear "rational" and "cooperative" to those who make decisions about their release. There is a growing consensus among ethicists that researches should select cognitively disabled persons as subjects only when the research bears some relationship to their situation. In the case of the NCVS, the choice of those cognitively disabled persons who are competent to respond to the survey may be highly appropriate since they may be particularly vulnerable to certain forms of victimization. However from a researcher's point of view the issue is whether they are competent to respond accurately to the survey. This underlines the more fundamental issue that the interviewers, the investigators, and the IRB need to work closely with persons who are knowledgeable about and experienced in working with cognitively impaired populations to decide whether and how to include them in the research. Most that has been written on this topic pertains to biomedical and pharmaceutical research, and it is not easy to apply the same ideas to survey research such as the NCVS. OHRP's primary concern is that incompetent subjects with court-appointed legal guardians provide consent in a responsible fashion. Persons deemed incompetent to decide about participation probably would also be incompetent to participate in survey research. Conceivably, family members could provide a more accurate account of the crime victimization of cognitively impaired persons. However persons having minor cognitive impairment who are not institutionalized might be deemed competent to participate in the NCVS. They might be included, with advice and guidance from persons who are knowledgeable and experienced in administering similar instruments to cognitively impaired individuals. Elderly persons The elderly are a heterogeneous group, not usually in need of special protections, except under circumstances of cognitive impairment and institutionalization. However investigators sometimes fail to include them in research because of difficulties in recruiting them. Older persons tend to avoid events that interrupt their daily routine, are inconvenient, or do not directly benefit them (Sachs and Cassell, 1990, p. 236). They are more likely than younger persons to drop out of studies so that investigators may need to recruit more elderly to account for this. Elderly persons who live alone tend to have warranted fears of scam artists and others who would harm them. The use of gatekeepers such as directors of senior centers may provide useful assistance in identifying the investigator as someone it is safe to associate with. Elderly persons who live with younger relatives have a different problem keeping parts of their life private from younger members of their family. Researchers need to recognize this problem. The interviewer should ensure that the interview takes place in a private setting and that they can conduct a phone interview under conditions of privacy at the elder's end of the line. Even those elders who are well educated and have kept abreast of news and changes in our culture may use a somewhat different vocabulary than younger cohorts and may be unfamiliar with some new concepts. The interviewer needs to be sensitive to these possibilities and be prepared to rephrase questions. Elders may have hearing and vision problems that interfere with responding to a survey. Hence elders may require more time, large-print material, and patient explanation until the hard of hearing are able to accurately understand. The very elderly, on average, are far less educated than younger cohorts. Those who have had little formal schooling are not test wise. Some of the formats of written surveys such as item ranking, Likert scales, and multiple choice items are foreign to them, hence difficult for them to use (Levine, 1982). Generally, education, health status, and inadequate communication about the research, rather than age, are responsible for lack of comprehension and recall (Sachs and Cassel, 1990, pp. 235- 236). When interviewing poorly educated elderly, interviewers should use a more conversational approach to yield more satisfactory responding. The wide-ranging demographic differences and corresponding response differences underline the importance of adequate stratified sampling, with oversampling of those segments of special scientific interest to the project. Minorities The inclusion of racial and ethnic minorities in national surveys is vital to the generalizability of the results. However minority groups frequently have a negative perception of research, and it can be difficult to recruit minorities because they fear lack of confidentiality. Focus groups within minority communities can provide useful communication between investigators and community members. In that context, researchers can explain the aims and process of the research and the safeguards to confidentiality. Investigators can learn the perceptions and fears of community members, correct erroneous perceptions, and allay fears. A second reason to do pilot work within the relevant minority communities is to learn about possible linguistic problems. Often it is not simply a matter of obtaining an accurate translation from standard English to another standard language. The combination of dialects and variants can make standard translations inadequate for accurate communication. For example, immigrants from Mexico may speak a variant of Spanish according to their home region, and the Spanish of Mexico itself differs from the Spanish of Central America. Members of racial or ethnic minorities in the United States are relatively more likely to have problems of transportation and child care to participate in studies, especially when repeated interviews are required. Interviewers should be sensitive to these problems and be prepared to offer solutions. Interviewers also need to be sensitive to cultural norms of the respondents. Researchers should include members of the given cultural group on the research team to help ensure the survey's success. Investigators should consult with other researchers, clinicians, or outreach workers who have worked with the given cultural group in a given location. Federal Regulations recommend that IRBs reviewing research involving racial or cultural minorities have a representative of that group on the IRB or invite a consultant from that group to assist with review of the protocol. Students, employees, and institutionalized persons Issues of coercion and confidentiality arise with respect to these groups of persons. Will their gatekeeper, including their employer, teacher, physician, or jailer, pressure them to participate? Will researchers keep their responses confidential and not disclose them to the gatekeeper or others? Do the respondents perceive that they will win favor within their institution by participating in the research? Is recruitment truly a voluntary process? As a safeguard, are respondents given every opportunity to refuse to participate during the consent procedure, and to discontinue their participation at any point in the procedure? If they do not complete their participation in the research, is this fact kept confidential from their gatekeepers? Economically or educationally disadvantaged persons Economic and educational disadvantage tend to go together, but create different problems. The economically disadvantaged person is easily recruited when large financial incentives are offered, and their advocates would argue that large financial incentives may be appropriate. However IRBs are wary of the incentive that is so large that it is coercive. Moreover, if respondents perceive that the financial incentive is contingent on completion of the interview, there is the double threat of coercion and lack of candor. Noncontingent financial incentives are more beneficial in all respects, and they also are more effective in recruitment (Singer, Groves, and Corning, 1999). Economically disadvantaged persons may require financial incentives that are higher than those for economically secure individuals. These incentives are likely to include costs of child care and transportation. Middle-class interviewers may be ineffective at establishing rapport and communicating effectively with poor people. Some researchers employ and train as interviewers persons who are demographically similar to the respondents. For example, Kim (1997) used poor women to survey poor women in an evaluation of federally funded job training programs to determine whether job placement rates and earnings of program participants are higher than those of nonparticipants. Kim incorporated feminist approaches, including asking open-ended questions rather than multiple choice, that resulted in a somewhat more qualitative set of responses. The richness of detail provided perspectives that would otherwise have been overlooked. Kim reports that the approaches she used reduced interviewer bias, improved response rates, and facilitated trust in answering sensitive questions. The poor women trained as interviewers learned about scientific inquiry and earned money. While educational disadvantage tends to be associated with economic disadvantage, that association is not always present. The respondent who seems not to understand questions or instructions and who cannot respond appropriately in test-like formats may simply be inexperienced with school-type activities. The respondent may be perfectly capable of describing the relevant experiences and attitudes, but not in the formats presented. More seriously, the respondent may be illiterate. Many apparently middle-class people are illiterate but have developed considerable skill at faking literacy. The interviewer needs to be sensitive to this possibility and be prepared to switch to an oral format that allows the respondent to answer in a more conversational and open-ended way. 8. Efforts to minimize refusals to participate in sensitive surveys A major ethical value underlying survey research is that it yield valid and useful knowledge a goal that is unattainable unless respondents participate candidly. However most of the human subject issues raised by IRBs and the Common Rule concern the protection of human subjects, not their willingness to participate and be candid. Consequently IRBs sometime reject approaches that seem to provide "too much" motivation to participate. For example, although it is well established that street people, particularly drug addicts, will not participate in research unless they are paid in cash, some IRBs do not want to approve this practice because such respondents will use the money to buy more drugs. IRBs do not want the monetary incentives to be more than the individual would ordinarily earn for that amount of work time. Thus airline pilots who are subjects could be paid far more to participate in a survey than a poor person, a notion to which many poor subjects and their advocates object. There are two kinds of efforts: efforts to induce people to agree to participate in sensitive surveys, and efforts to induce them to give honest answers to sensitive questions. Efforts to induce participation There are two factors that consistently and substantially increase participation: payment of monetary incentives and number of contacts. Regarding number of contacts, the standard practice is to recontact nonrespondents up to three times. By the third contact, investigators get response rates that are usually at least 30% higher than they would have gotten with only one contact. This procedure is viable even with mailed anonymous surveys. Researchers include a postcard containing the respondent's name with the survey that the respondent is to drop in the mail at the same time the survey is mailed. Investigators send repeated mailings only to those who have not mailed their postcard. The postcard could also be used by the respondent to indicate which incentive they want in return for participation. The effects of monetary incentives are more complicated and interesting. Church (1993) performed a meta-analysis of the experimental literature (38 studies) on the effects of incentives in mail surveys. He classified incentives into two categories: whether they were monetary or non-monetary, and whether they were offered with the initial mailing or made contingent on the return of the questionnaire. His conclusions were that: Prepaid incentives are more effective than promised incentives, such as incentives that are contingent on completion of the survey. Prepaid monetary incentives are more effective than gifts offered with the initial mailing. Response rates increase with increasing amounts of money The offer of contingent (promised) money and gifts does not significantly increase response rates. Church's study leaves unanswered the question of whether the effects of incentives on response rates vary by mode of interviewing. Could Church's findings be replicated with face-to- face or phone interviews? Singer et al. (1999) performed a meta-analysis of 39 experiments and used the following independent variables: amount of incentive, type (gift or money), timing (before the interview or promised and given afterward), and burden (whether the interview was longer than an hour or if it included a diary, test, sensitive questions, or panel study). In addition to their main dependent variable, response rate, they also examined quality of data (nonresponse and number of words in response to open-ended questions), cost per interview, and relationship of response to incentives with demographic variables. Singer et al. essentially replicated Church's mail survey findings with telephone and face-to- face survey findings. Moreover they showed that incentives are effective in increasing response rates in both low- and high-burden studies, and for fresh respondents, panel respondents, and those who have refused to respond previously. However they found that the effects of incentives are relatively modest. Some of the details are pertinent: Respondents preferred prepayment to promised payment even in face-to-face interviews. The effects of incentives are linear within the limits of the monetary incentives offered. The money offered varied from $1 to $100; 60% of incentives were less than $10; mean value was $11.39. Gifts were significantly less effective, even controlling for the cash value of the incentive. It was expected that high-burden surveys -- surveys lasting more than 1 hour; surveys with ancillary tasks such as a test or diary, surveys with repeated or panel testing --would benefit more from high incentives than would low-burden surveys, but only an insignificant interaction was found. When low-burden phone or face-to-face interview studies were analyzed separately, they found that incentives had the same effects. When studies involving a diary, test, or other self-administered instrument were excluded, the effects of incentive remained significant. The higher the response rate to the no-incentive condition, the smaller the effect of an incentive. About one-third of the studies analyzed were unpublished, and Singer et al. compared their findings with those of the published studies. They found no differences. Incentives tend to yield better quality responses. Paying an incentive tends to produce higher numbers of respondents in demographic categories such as low income or nonwhite respondents that might otherwise be underrepresented in sample surveys. Singer et al. note that an important question that remains unanswered is whether the costs of surveys with incentives is cost effective in relation to the savings, including fewer calls and mailings to respondents. This elegant analysis also left unanswered the question of whether respondents would regard it as inequitable to offer a monetary incentive to non-respondents. Subsequently, Singer, Groves, and Corning (1999) studied whether cooperative respondents would perceive the use of incentives to convert refusals as inequitable, and whether those who learn of the practice will be less willing to participate in future surveys. Singer et al. found that survey respondents are sensitive to issues of fairness in the distribution of incentives, but these issues are not especially salient and are not among the factors that motivate survey participation. Efforts to evoke answers to sensitive questions A central concept for understanding how to evoke answers to sensitive questions is the "foot-in- the-door" technique (discussed in section 2). Respondents who are first asked innocuous, then mildly sensitive, and then very sensitive questions will cooperate and answer such a sequence of questions. However they would refuse to answer the most sensitive questions if such questions were asked in isolation or near the beginning. A second major concept has to do with obtaining honest answers to sensitive questions.