Resopnse to OMB

Thank you for your interest and time in providing this feedback on our OMB package. It was very helpful. It has guided us in writing a stronger package that provides more clarity on the goals of this formative research and how we will accomplish them. Below we provide a detailed response to each of your comments and indicate how and where we have made changes to the ICR when appropriate.

1. Mechanical Turk is not representative of the adult U.S. population. The CDC should request the demographic profile of the full MT frame to document the differences to the general population.

The study cannot nor is it intended to generalize findings to the entire U.S. population or U.S. prison population. The respondent universe is a targeted and diverse convenience sample that allows for lower cost data collection from a diverse group of respondents as well as better access to hard to reach small populations (i.e., gay and lesbian, perpetrators not identified by criminal justice system, and IPV perpetrators in general who are a minority of the population as well). Importantly, the sample will also include people who are not currently or have never been in a violent intimate relationship. These individuals are necessary because some will have experienced risk factors for violence (such as history of child abuse or substance use) and yet are not perpetrators of violence. In formative research such as this it is necessary to use previous research to construct a wide net so we can begin to see in a single sample which factors appear to be necessary and sufficient conditions for violent perpetration.

While it is not possible to get the demographic profile of all MT workers (plus we are restricting to U.S. workers) we will have full demographic profile of our actual sample which could be compared to general population. But again, we already know MT is not representative of the general population (e.g., younger, underemployed) and it is not our goal to collect a representative sample.

2. By the same token, the prisoner population is not a random sample of U.S. residents, and those in a work release program are not necessarily representative of the state prison population. I would think that controlling for age, sex, race, education, marital status, and a number of other demographic variables would require far larger samples than those proposed, as would any sort of standardization method.

We are not attempting to obtain a representative sample of prisoners. This is basic formative research for which representative samples are not necessary. The intention of augmenting our convenience sample with incarcerated respondents is to increase the proportion of our sample that has a history of violence and criminality to ensure sufficient variance within our total sample. Given that these individuals will be part of a sample of 2000+ (500+ in subsample analyses) we will have plenty of power. However, we would not be controlling for above mentioned variables as these variables will be part of the latent class indicator models and predictors of class membership.

3. The estimated response rate for prisoners is too high. In 2016, BJS had a response rate of 70% for prisoners in secure state and federal correctional facilities, which was similar to the RR for persons in prison and jail in the 2012 National Inmate Survey. An even more realistic comparison would be the National Former Prisoner Survey in 2008, which surveyed persons on active parole. While these individuals were not returning in the evening to a halfway house/community facility like the ones from which the CDC proposes sampling, they are more like work release prisoners than those in secure facilities, which include prisoners who will never be released. The RR for the 2008 NFPS was 60.6%. So split the difference and assume about a 65% RR at best.

Thank you for these insights. We have changed our response rate to 65%.

4. Some of these questions (in particular #32 (stalking), #51, #53, #54, #55, #56, #58, #62, #64, #66, #82, #83, but also #49 to some extent) could be self-incriminating for the sample of prisoners. Not sure how the CDC's IRB determined that this wasn't human subjects research, but I'd be surprised if self-incrimination wasn't questioned given the protected status of prisoners.

The study was deemed to be research involving human participants. As indicated in Attachment H, CDC determined that this activity is research involving human subjects but that CDC – including employees, visiting scientists, fellows, and on-site contractors (but not off-site contractors) - will NOT obtain data by intervening or interacting with participants and will NOT have access to identifiable (including coded) private data or biological specimens. For these reasons CDC is considered non-engaged. Further, CDC determined that the contractor’s IRB should be the IRB of record because they had the resources to conduct a full IRB review that included thorough review by a prisoner’s advocate. All protocols and consents were reviewed by the contractor’s IRB as well as an outside prisoner advocate. AIR is registered with the Office for Human Research Protections (OHRP) as a research institution (IORG0000260). AIR’s federally registered IRB (IRB00000436; Federalwide Assurance FWA00003952) has conducted expedited and full-board reviews of research involving human subjects for more than 23 years. The study was approved to include prisoners as subjects by the Prisoner Research Coordinator (Julia Gorey, J.D.) at the Office for Human Research Protections of the Department of Health and Human Services (see Attachment I included in this email). No personally identifiable information (PII) will be linked to survey responses and all respondents are free to skip any questions.

Given that the individuals are incarcerated we cannot promise that their participation in the survey will remain unknown to staff. Officers must be involved in transporting the prisoners to and from areas (the exception are the work release where there is not this level of security). However, we can promise that their answers are confidential. Prison staff will not be made aware of the contents of the survey or the inmates’ responses to questions. The surveys will be administered and recorded on mini iPads for inmates who are literate. In the medium security facility where guards may need to maintain observation of inmates during administration, the survey will only be administered to inmates who are literate and do not require the survey to be read aloud to them.

Given that the focus of this study is on IPV perpetration and potential latent typologies of perpetrators, it is essential that we assess IPV perpetration which includes stalking and sexual violence, in addition to physical violence and psychological aggression. If we do not ask these questions pertaining to IPV perpetration we will not be able to distinguish perpetrators from non-perpetrators or perpetrators of different typologies. These acts of violence are also frequently assessed in psychology and criminal justice research involving incarcerated populations. We have taken steps to ensure the confidentiality of the prisoners’ responses to all questions precluding potential for self-incrimination. We worked with the prison advocate who reviewed all survey questions to make sure the process was not coercive or put the prisoners at risk. All survey questions were approved by the IRB including prisoners advocate and the Prisoner Research Coordinator (Julia Gorey, J.D.) at the Office for Human Research Protections of the Department of Health and Human Services.

5. Along the same lines, I would be wary of asking these questions of prisoners. In our National Inmate Survey, which asks about sexual victimization in federal, state, and local correctional facilities, 10% of the sample receives a survey that includes no questions on sexual victimization at all, and all surveys are time-limited. This is done to protect the inmates from the correctional officers - everyone can claim that they got the survey without sexual victimization questions, and people with no sexual victimization to report take exactly as long on the survey as people with 10 instances in the past week. While the CDC survey does not specifically ask about victimization by correctional officers, I am still not comfortable with the power arrangement that exists for this survey. Officers will know who took the survey, and could pressure these individuals because of participation.

These are very valid concerns that we considered when developing this study. The survey is presented as “Understanding Relationship Dynamics.” The content of the survey will NOT be shared with correction officers or any facility staff; nor will prisoners’ responses to the survey be shared with anyone. For prisoners with sufficient literacy, the survey will be administered and recorded on a mini iPad to which prison staff will never have access. For inmates with a low literacy score, the survey will be read to the inmate and the interviewer will record their responses on the iPad. However, because in medium security facilities correctional officers may be required to remain in the room with the inmate, we will not administer surveys to inmates with low literacy scores to prevent correctional staff from learning the survey content and inmate responses. We acknowledge that there may always be some potential for staff to discover some content on the survey. However, we have taken significant precautions to ensure responses to the survey questions will be confidential and inmates will be protected. These procedures have been vetted and approved by a prisoner’s advocate, the Institutional Review Board, and the director of prisoner research at the Office of Human Research Protections at DHHS (see attachment I).

6. In their construction of typologies, are the researchers going to consider the issue of multicollinearity between the measures/concept instruments? Each of these sets of questions has significant overlap that could cause spurious results of modelling.

Multi-collinearity is not applicable to the latent variable modeling we will conduct. In regression you worry about collinearity among the predictors (independent variables), but in LPA your variables are outcomes (dependent variables) so that issue isn't involved. In fact, to conduct factor analysis, you need high correlations among indicators to ensure they are measuring the underlying latent construct.

7. In Table 2, providing actual statistics on each instrument would better justify their inclusion.

Survey items were selected by modifying and aggregating questions from extant measures where possible. When necessary, questions were developed specifically for the purposes of this data collection. As such, there would be no statistics to report on these items. All questions were vetted by numerous subject matter experts in IPV research and measurement design. See list of experts in section A8 of the SSA.

8. Have the researchers given thought to doing a random ordering of their questions? I would expect significant fatigue and response drop-off over a 46-page survey (particularly among the sampled prisoners who are only receiving a piece of fruit, that they could easily obtain while out in the community without answering pages of sensitive questions). I would suggest randomizing the order of the measures so that you have some responses to all of them.

Yes the plan has always been to randomize questions where possible (i.e., some questions have skip patterns/follow up questions that must occur in sequence). We have added this to section B.3 of the SSB.

9. What is the reasoning behind the oversampling of gay/lesbian individuals? Is there evidence to support the idea that they are at a higher risk for IPV?

There is some evidence from national surveillance (NISVS) that gays and lesbians are at higher risk for IPV than heterosexuals (Walters, Chen, & Breiding, 2013, found at: https://www.cdc.gov/violenceprevention/pdf/nisvs_sofindings.pdf). Also, there is some evidence to suggest that same sex dyads have rates of IPV equal to or higher than heterosexual couples (see citations below). However, there is a dearth of etiological IPV research conducted with gay and lesbian populations and as such much less is known about risk and protective factors, correlates, and the relationship dynamics associated with violence in these relationships. Oversampling of gay and lesbian respondents is essential to obtain sufficient sample sizes in these populations because approximately 97% of the U.S. population identifies as heterosexual (Walters, Chen, & Breiding, 2013).

Edwards, K. M., Sylaska, K. M., & Neal, A. M. (2015). Intimate partner violence among sexual minority populations: A critical review of the literature and agenda for future research. Psychology of Violence, 5(2), 112.

Graham, L. M., Jensen, T. M., Givens, A. D., Bowen, G. L., & Rizo, C. F. (2016). Intimate partner violence among same-sex couples in college: a propensity score analysis. Journal of interpersonal violence. DOI:0886260516651628.

Stiles-Shields, C., & Carroll, R. A. (2015). Same-sex domestic violence: Prevalence, unique aspects, and clinical implications. Journal of sex & marital therapy, 41(6), 636-648.

10. How are the researchers determining a criminal history of IPV? Obtaining RAP sheets or sentencing information from the current instance of imprisonment? Self-report? Have they actually looked at the administrative records (RAP sheets, department of corrections (DOC) records) to ensure that information on the victims is included? The offense information I've dealt with in state prisoner records include the state statute(s) for which the inmate was sentenced - so it may be "agg. assault", with no mention of who was assaulted. In addition, it's important that the researchers understand that the sentenced offense is often the result of a plea bargain, so they should pay attention to people convicted of simple assault as well (they may have been pled down from aggravated assault).

Thank you for these insights. For this study, those individuals who have been charged with “Domestic Battery” will be considered to have an IPV-related offense. These charges have their own specific code which is assigned by justice system to indicate which individuals have been convicted of these charges. This is both publically available data and is also administrative data kept by the prison facilities. The prison’s administrative staff (not correctional staff) will provide the sampling frames for individuals with or without current IPV related charges so that AIR will not have access to PII. It is likely and expected that individuals convicted of an IPV related offense will have a history of non-IPV offenses and prisoners incarcerated for non-IPV offenses will have a history of IPV offenses. However, only the current offense will be used for the sampling frames. The purpose of including the stratified forensic populations is to ensure diversity of offenders and increase the number of persons incarcerated with an IPV offense.

11. If they are going to be pulling RAP sheets, it gets even murkier. First, I would question the ethics of looking back in someone's criminal history to identify potential subjects for a survey, but I guess if the state(s) they are working in are open-record states, it would be legal. If they try to do an Interstate Identification Index (Triple-I) search, they would likely only get arrest records for the current state, since whoever is pulling the record would have to use code "R" for research (instead of code "C" for criminal justice). Not all states reply to code R requests for data. In addition, RAP sheets are state-specific, hand-written, and do not use standard terminology for offenses even within a state or law enforcement agency, so they would need to be parsed and standardized to ensure comparability.

We are not pulling RAP sheets. Facility staff will use their administrative file to identify the current convicted offense to generate sampling frames.