“You think that, again, that’s the medication”: reflecting on qualitative methods for interviewing family members of violent and impulsive men in an intervention trial

ABSTRACT Qualitative research in psychology can often maintain standards and assumptions from positivistic and experimental methodologies. Sometimes these issues are well argued against or around logically in literature, often abstractly, but cases of methodological consideration with real cases are rarer. This discussion aims to help methodological reflection and learning by presenting a case of multiple intersecting methodological considerations. The methods included a content analysis completed on interviews with women related to men participating in a violence and intervention trial. This paper presents the limitations and methodological considerations during analysis as a detailed discussion. Considerations arose from assumptions in the research design, interpreter use and leading questions. The discussion describes considerations and solutions depending on the scope required – here working with women in or near contexts of violence. Hopefully, a demonstration of full considerations within such a project encourages similar case-based examinations of research.

• Value free axiology (Freebody, Markauskaite, and Irwin 2011) • Generalisable and representative (Small 2008) • Minimal relationship with participants (Brodsky 2001); disregard power imbalance (Lincoln and Guba 2013) • Sampling at random (Small 2008) • Linear process not iterative (Kvale 1994) • 'What' and 'how many' over 'how' and 'why' (Small 2008) • Rigid Methods (Small 2008) • Calculability (Brinkmann 2018) • Avoid bias or subjectivity (Charmaz 2006) • Cause and effect rather than bidirectional influences (Saldaña 2011) • Discovery versus constructive (Harper 1995) • Generally, hypothetico-deductive, rather than empirically inductive (Fardet, Lebredonchel, and Rock 2023) • Uses unified theory (Kosh, 1993) It has been well argued that such expectations of qualitative paradigms are often inappropriate or problematic (see Big Q versus small q qualitative research in Braun and Clarke 2013;Kidder and Fine 1987). In the training and research for psychology -but also in other fields like health studies -there is evidence that students and researchers view qualitative research as 'less than' the quantitative and experimental counterparts (Devers 1999;Mitchell et al. 2007;Povee and Roberts 2014;Shaw, Dyson, and Peel 2008;Small 2009). That perspective urges the need to resist the notion that 'scientific' qualitative research looks like quantitative and experimental research, meaning we must clarify how qualitative methods can stand independently without emulation. Unfortunately, when attempting to clarify our position, qualitative research has had a history of argument over process and procedure. For example, the schism of Grounded Theory (see Babchuk 1996;Heath and Cowley 2004;Kenny and Fourie 2014) or Braun and Clarke's issues with the consistent use of their 2006 paper for its simplistic 'Thematic Analysis' process despite great strides in Reflexive Thematic Analysis (Braun, Clarke, and Hayfield 2022). Other areas of qualitative inquiry seek to highlight what should be avoided in qualitative analysis; for example, what discourse analysis is not (Antaki et al. 2003;Burman 2004). Such attempts at codifying and standardising qualitative practice can also be met with cogent and important criticisms such as 'proceduralism' and 'methodolatry' Clarke 2021, Chamberlain 2000;King and Brooks 2017). This is paired with the consistent need to situate the change in methodology with wider contexts of working with real people, not just the pursuit of better data or conclusions -whatever that means (Denzin and Lincoln 2018).
Recently, there has been more interest in reflexive and critical reexaminations of qualitative methods such as post-qualitative inquiry (St. Bodén and Gunnarsson 2021;Pierre 2019) as well as a resistance to classical views of methodology altogether (e.g., Nordstrom 2018). Similar reflections have been made for the field of study this article is based on, men's violence (Fraser and Taylor 2022), and qualitative studies on violence intervention have been found to lack rigour while lacking direction for solutions (McGinn et al. 2016). Regardless of analytical or philosophical stance, qualitative researchers are in a position where research needs to be examined on its own terms while being open to examination for new ways to improve its application. Unlike the post-qualitative inquiry examples at the start of this paragraph, not all reflections are critiques of knowledge creation fundamentally -some, like this piece, will be critical reflections on common positivist analytical processes (qualitative content analysis).
Some reflections on methods are abstract approaches to the idea of epistemology, others are granular examinations of the framework process, but there is a missing middle ground -reflection of completed qualitative work. A strength of qualitative research is to examine cases and to understand processes within those cases. That strength can be used here for the assessment and evaluation of qualitative research methods. The suggestion from this discussion, rather than producing evaluation guidelines (e.g., Anderson 2010; Clarke and Braun 2019;Goodman 2017;Kitto, Chesters, and Grbich 2008;Treloar et al. 2000) or highlighting methodological issues at a theoretical or generalised level (Guerin, Leugi, and Thain 2018;Small 2009), is that a lot can be gained by inspecting a case of qualitative research to explain how data collection and analysis as a process produces conclusions with limitations. Most literature of this kind is on specific and perennial issues (e.g., translator use with Culturally and Linguistically Diverse (CALD) communities: Kapborga and Berterö 2002), but there are generalised cases such as Bodén and Gunnarsson's (2021) exploration of writing for post-qualitative inquiry. Here, we seek to highlight and describe all the disparate considerations that arose, often unexpectant, in one case, using a relatively positivistic analysis.
There is also a need for peer-reviewed literature to include this real everyday work of qualitative research. The author, as a new academic and early career researcher in Australia, has been involved in teaching qualitative methodology in psychology at a tertiary level. Such a focus on qualitative research is rare within psychology; only three universities in Australia offer standalone qualitative research courses in 2018-2019 (Fox and Easpaig 2020, 231). Generally, students and peers in psychology have problematised views when it comes to qualitative research (Povee and Roberts 2014). When developing content for qualitative components of psychology research methods courses (largely focused on statistics program training), there was a lack of simple cases of methodological examination or critique appropriate for students only initially grappling with qualitative research. The above cases of post-qualitative inquiry are not appropriate for these new students so a gap, explaining 'issues that arise in real research', became apparent, and this discussion can fill part of that gap for psychology while adding to growing critical reflection on methods in the wider social sciences.
There are also the practical needs of researchers. We can easily acknowledge that qualitative research is heterogeneous and that qualitative researchers often do not know what each other are doing or why they make the choices they do. We will only learn about the diversity and the practicalities of the various processes if we publish when things are changed or reconsidered during the iterative process of qualitative research and when issues arise beyond what fits in a 'limitations section' next to our conclusions.
What we will discuss here, is not for generalisation or uniformity, which is another holdover from positivistic and quantitative emulation. Instead, this article works as a test case for explorations of methodological issues, recorded in the audit trail, from a real and interesting qualitative analysis which was not a break in the mould of positivistic and quantitative-emulating analysis found in small q research. Put succinctly, we are bringing the issues faced in this analysis as a research output to be read by peers and students.

The case -women's experience of an intervention for violence and impulsivity
The COVID-19 pandemic spurred on a second pandemic: an increase in domestic and intimate partner violence (Evans et al. 2020;Lyons & Brewer, 2021). In this context, questions about the efficacy of common interventions for men's violence arose as well as critiques about how women's perspectives are rarely the central focus. A qualitative content analysis of interviews of women with sons, fathers, or partners on a pharmacotherapy trial intervention (ReINVEST) (Butler et al. 2021) contained several evaluative and critical methodological questions for the data and processes.
The qualitative analysis of interviews is a sub-study from the 'Reducing impulsive behaviour in male, repeat violent offenders' (ReINVEST) Trial. Sertraline hydrochloride (an antidepressant medication) or placebo was given to men who were repeat offenders who were violent or highly impulsive in the lives of these women. The aim was to see whether the medication reduced violent reoffending (see Butler et al. 2021). The men were screened by clinicians with follow-ups over the 12 months. Relatives of these men could attend these appointments and get to know the clinician if the men allowed it.
Towards the end of the trial, the voices of women in the lives of these men were sought for feedback on the trial's wider contextual implications. The men on the trial were asked to provide details of a woman in their lives willing to undertake the interview -as the process of recruiting these women.
Qualitative content analysis of 26 interviews with family members was undertaken to gain insights into (a) perspectives on the impacts of the ReINVEST trial on their lives and the related ReINVEST trial participant's life; (b) contextual elements relating to continued involvement with ReINVEST; and (c) knowledge, opinions, and feedback regarding the implementation of ReINVEST. Several key methodological or ethical concerns were raised throughout the analysis, which are the core of what we report here. The content analysis itself, without this critical lens, is presented in a separate publication (Thain et al. In preparation).

Research aim
The aim is to provide an example critique, specifically of content analysis, with one-off interviews as a sub-study of a positivist experiment (randomised control trial) to be a guide to improving rigour and remaining critical in future on-the-ground qualitative research.

Interview sample description
Twenty-seven women (interviewees) related to a ReINVEST trial participant were recruited (current partners n = 15; mothers n = 7; ex-partners n = 2; daughters n = 1; and mother-in-law n = 1; 100% women). The interviews were associated with 25 separate ReINVEST trial participants (100% menthree interviewees were related to the same trial participant). Table 1 contains a demographic breakdown of the interviewees. One interview was removed from the dataset (see results), leaving 26 interviewees and 24 ReINVEST trial participants represented. There is significant variability among the interviewees concerning the relationship between the interviewee and the trial participants, knowledge, and expectations of the trial and participant behaviour, understanding of the ReINVEST trial, and beliefs about the trial condition.

Recruitment
A convenience sample of participants' partners and family members was recruited for this study. These interviewees were recruited through ReINVEST clinicians asking their current (trial) participants if their partner or family member (women) would be interested in participating in an  interview. Interviewees were reimbursed with $75 for their time. No qualitative data with the men was made available to the analyst of this dataset.

Procedures
Interviews were conducted by telephone and carried out by two interviewers, one under the guidance of a senior ReINVEST clinician. The clinician completed three initial interviews while observed by a junior interviewer. The clinician observed the junior interviewer for the fourth and fifth interviews before the junior interviewer subsequently completed the following 22 interviews. Interviews began with consent discussions. Interviewees were reassured that the interview content was confidential and would not be disclosed to their partner/family member, nor would they be able to be identified by their answers. They were advised that they did not have to answer any questions that they felt uncomfortable with and were able to stop the interview at any time. They were offered the opportunity to be debriefed by a ReINVEST clinician after the interview or at any time should they feel this necessary. Interviewees were also provided contact details of support services. At the end of the interviews, interviewees were invited to join an all-women zoom group to discuss these topics more and debrief.
The interviews collected data on the impacts, opinions, and other contextual information relevant to the ReINVEST trial from the interviewees' point of view. Though variation was present, these interviews were semi-structured and followed an interview schedule. The interviews lasted, on average, 40 min (range 20-80 min). An independent company professionally transcribed this data in a simple verbatim style (though some 'aside discussions' were not transcribed). The transcripts were then given to members of the Australian Centre for Child Protection (ACCP) to conduct qualitative analyses. All data collection and analysis were carried out with ethics approval from the UNSW human research ethics committee (HC17771).

Analysis process
The author undertook all analyses. A qualitative content analysis was carried out on the interview transcripts; the analyst had no knowledge of treatment conditions as the trial was a double-blind randomised control trial (blinded). Afterthe analysiss was completed and written,a subsequentt analysis of unblinded data was undertaken in addition to the blinded qualitative reportthese comparisons are found in the content analysis paper published separately (Thain et al. In preparation). Separating blinded and unblinded analysis outcomes maintains rigour. More importantly, the substance of the qualitative analysis, when kept blinded, privileges and focuses on the voices, experiences, and input of the women.
A hybrid approach of using both deductive and inductive codes was used. An initial review of the study documents was completed to develop deductive codes and questions for the analysis. Once ethics approval and access to the transcripts were available, open (initial/eclectic) coding produced inductive codes (see Saldaña 2021). Two coding passes were completed with additional refinement and organisation of the codes at the end of each pass. Open coding allowed for a simultaneous latent (interpreted meaning of the text) and manifest (directly observable from the text) interpretation of the interview content. After coding passes and report drafts, a group discussion about methodological and analytical issues was conducted -these discussions both generated the content for this publication and the motivation that such conversation outcomes should be reported more.
The content analysis included four broad categories to answer the research questions, made up of deductive codes: trial impacts, opinion of the trial, motivation (for ReINVEST trial participant) to enter the trial, and suggestions for improvements to the trial. Each broad category has subcategories made up of inductive codes. See Table 2 for Content analysis categories and codes:

Evaluative note of analysis framework
It is common for methodological and analytical frameworks to appear in qualitative research with one reference to an article or book that mentions or describes that process. For example, the arguably over referencing of Clarke's (2006, 2022) paper on thematic analysis in psychology. Often it is clear to scholars of qualitative methods that when an analytical framework is cited in such a manner that more justification is needed for the use of Sentiment grouping by analyst (Thain). Deductive code from the initial schedule or due to volume. Inductive codes from the data. Fatigue, sleepiness, memory complaints, sexual dysfunction, a lack of sex drive, or other medication concerns.
a framework on the current data (Green et al. 2007). So, below is how the application of Qualitative Content Analysis was evaluated for this case. The aims of the research fell squarely in a positivist approach to science, seeking to treat the women's verbal reports as reflective (Hall 1997) presentations of their real experience of being connected to the trial. Those common experiences were to be distilled and translated to audiences in government, funding, and public political areas, as well as conversations with stakeholders. These external influences determined the choice of single interview analysis by the ReINVEST team for data collection, and subsequently become the context for the analysis that the analyst could undertake. Working within this restricted, though common, framework led to the considerations below.
The content analysis focused on broad data summaries for feedback for the ReINVEST trial. For that purpose, several evaluative references for content analysis were considered to check the applicability of the analysis (Cho and Lee 2014; Devi Prasad 2019; Elo et al. 2014;Erlingsson and Brysiewicz 2017). This literature informed and assisted with the conceptual approach to the qualitative content analysis (esp. Erlingsson and Brysiewicz 2017). This analysis, while sufficient, cannot meet all evaluative criteria because separate teams completed the design, data collection, and analysis. Still, the evaluative points found in the literature were considered where possible.
Peer review, consistent reflexive noting, and value assessment were completed throughout the analysis and write-up to maintain rigour and trustworthiness. Due to ethical constraints and the separation of data collectors and analysts, other common practices like member checking and triangulation with the sample were not possible. This consistent evaluation process is the key area in which notes, and arguments were made for the methodological considerations below.

Results of the content analysis
Part of the content analysis results are published separately (Thain et al. In preparation), while the full content analysis comprised a report internally for the trial. Readers can contrast the other publications as the 'usual reporting' found in the literature. What should be noted is the limits of journal articles in the assessment of the quality of the data despite common reporting conventions (i.e., this article's methodological considerations are longer than the content analysis publication).
The results indicate that the ReINVEST trial was overwhelmingly positive for the trial participants, the women, and their families, directly affecting target behaviours. The trial stood out to women as unique in how it treated them or their families as real people with real problems. The caring work of the clinicians in the trial was a clear strength. Extraneous contextual factors were highlighted when describing the variation of behavioural change for men. Initial apprehension about the trial was due to a lack of understanding, but this reduced over time. The women recalled an array of motivations for the men to enter the trial, but overall, the possible loss of relationships and growing awareness of a need to change were powerful motivators. Suggestions for changing the intervention were sparse alongside the high praise, though earlier and more consistent contact between the trial and the family was suggested. These women were as much on the trial as their partners, sons, or fathers were, and the impacts reached beyond them. Unblinded data analysis completed after the primary qualitative analysis suggests that there were few differences between the treatment arms from the perspective of these women, indicating that contextual factors or other trial elements (e.g., clinical assistance) are powerful mediators of such reported change.

Discussion of methodological considerations and suggestions
The data stand out with the ability to display the women's voices and allow reflection on the trial impacts. The key research questions could be clearly and directly answered by the data provided, which is an achievement sometimes not met by such open methods, especially as a late addition to a larger research program. Similarly, there were a variety of perspectives meaning the recruitment was not unnecessarily narrow. Throughout the research period, there were powerful relationships fostered, which likely allowed for some of the information shared to be available for collection.
While those positives for the data collection stand out, the reflexive practice highlighted several methodological considerations that impact the use of the findings to be transferable beyond the immediate cases: a persistent consideration for qualitative work. Given the generally summarised, often quantified, and volume-focused nature of qualitative content analysis, issues must be considered for any meaningful application of the conclusions. The issues most pressing in the reflexive notes and meetings included sampling bias, the ethical and practical limitations of the interview type, and the direction used in the semi-structured interview questions, in particular an issue labelled as 'inducement'. These considerations, described below, are useful methodologically for researchers taking up this framework or research more in this topic area.
The most important note for this publication is that these considerations, do little to nothing for reducing the importance and the power of these women's experiences being connected to the ReINVEST trial. The words, and their experiences presented in the content analysis stand alone and remain trustworthy, dependable, and rigorous in the scope. Any attempt or insinuation that the below considerations make the content analysis incorrect or the findings irrelevant are not only intellectually flawed, they are ethically iniquitous to the need to value these women's voices above any research assessment of quality.
Below is a Table 3 containing an overview of the issues and their possible solutions.

Justification for removed data
Data can often be removed with a small note on why, but often the rationale is not directly present for others to learn from, here is that rationale. One interview of 27 was removed from the analysis because an ad hoc unforeseen interpreter was used. This interpreter was the interviewee and trial participant's male family member (not on the trial). While, on the surface, it appeared that this data is useable, a simplistic literature review with the search phrase 'interpreter use in qualitative research' brought into question the dependability of this interview.
Several papers addressed similar issues. Kapborga and Berterö (2002), a first-time use of interpreters for the authors, supported previous literature arguing that interpreters must be well-trained in the research methodology, not just the language. They suggest an extended period with CALD participants to affirm understanding and knowledge of the community and culture in question. Those two key elements are absent for the removed interview.
Similarly, Edwards (1998), in a comprehensive critical examination of interpreters in qualitative research, raises several points. They focus on possible barriers to open communication from the participants. For example, 'who an interpreter is' in context for the community is key; professionals may be a hindrance given their social power compared to community-based volunteers or close relations for participants (e.g., family). In that latter and more relevant case, matching interviewees' and interpreters' social characteristics are preferred, as Wallin and Ahlström (2006) and Sawrikar (2015) argued. Correspondence in identity was not met, so more quality issues exist for the removed interview. Consistently in the literature are standard issues of time, training, and translation accuracy. Edwards (1998) highlighted the innate mismatch between languages that interrupts understanding and contextual differences and that mismatch leads to conclusions seeming stronger due to translation rather than the reverse (also in Kapborga and Berterö 2002). Again, none of these issues can be overcome for this interpreter use.
Given that the accuracy is limited, the training was non-existent, and the mismatch of gender and position was so stark, the translation's merit is seriously questioned. A key requirement is direct knowledge of the interviewees' culture which is absent (Kapborga and Berterö 2002). Similarly, the possibilities for monitoring and reciprocity contexts in this situation made the consequation for the interpreter highly problematic (Guerin 2016). These aspects led to the simple answers that, in the view of critical examination, this use of interpretation, while possibly better than a professional, is at best an unknown or at worst lacking dependability or credibility.
More time, purposeful consideration, and appropriate CALD community connections need to be present for valid interpretation. Improving interpretation, however, does not allay perennial issues such as the privileging of 'accuracy translation' as if it improves the truth value of analysis as positivist 'fact' nor does it remove the researcher's likely ignorance of discursivecontextual-interpretation when answers are translated (Vara and Patel 2012). Such 'accuracy' and 'de-contextualised interpretation' issues are present even without lingual translation and can only be solved with the interviewer and interpreter/informants working -spending more time -to gain a better understanding of the participants experiences and contexts.
In some cases, interviews cannot be used, in other cases interviews are justified in their use despite flaws or special considerations. Often in journals there is no room to consider these decisions, but we as researchers would be better off with our rationale being presented in full when these choices are made. These issues are not unique to qualitative methods. Quantitative methods remove people in social science data as outliers or data errors. Outliers 'deviates so much from other observations as to arouse suspicions that it was generated by a different mechanism' (Hawkins 1980, 1). Committed qualitative and mixed methods researchers should be interested in producing methods to find out those mechanisms when these outliers are people discounted from research. This first issue stands out as descriptive not interpretive data and analysis (Morse 2018), and an assumption we are discovering versus constructing data (Harper 1995).

Sampling
For recruitment there was an assumption of sample universe homogeneity (see Robinson, 2014 for more on this term) -allowing self-selection through partner-based recruiting. The analysis, based on that approach, was a positivist content analysis guided by seeking the general answer to the research questions, questions which also assume homogeneity as a marker of validation. This on reflection is a useable but not entirely sound epistemological choice. Issues arise that challenge homogeneity or the semblance of it in the data (below). The focus on 'what' is said and 'how many' participants say it belies 'how' and 'why' participants were recruited and how and why they say what they say (Small 2008).

Variation in relationship
Table 1 's demographic factors of participants show these women had different relationships, experiences, and opinions of the ReINVEST trial participants. These different perspectives on the trial and the participants are a strength for the analysis; qualitative research excels at utilising heterogeneity to better explain processes even when similarity is present or sought. That said, content analysis displays answers as uniform and homogeneous. So, variation becomes a serious and difficult consideration for each interviewee's credibility in supplying reflective or 'objective' verbal reports of complex phenomena (Hall 1997;or 'type 1' in;Guerin, Leugi, and Thain 2018). Arguably in the analysis, authors would comment on these considerations for each instance of participants data, but this would quickly become unruly.
An example of such a variation is the type of relationships between these men and women. Some interviewees and trial participants lived together, others apart, some shared daily activities while others did not, and some reported that they were consistently involved in each other's lives while others were rarely involved. Otherwise, the family structure by children, kin, and support networks, were vastly different. These contexts greatly altered the amount of knowledge, understanding, and expectations the interviewees had of the trial participants' behaviour -also referred to as opportunities for monitoring in contextual research (Guerin 2016). There is an expectation, using reflective assumptions for verbal reports (Hall 1997), that reporting means that the stated event has occurred, but there is no good reason to assume that language is used to just report things that happened (Guerin, Leugi, and Thain 2018). In the future, we can learn from this variation and the effects described below. Primarily, we resolve this with a more justified and directly purposeful sampling in an iterative matter, informed by analysis, to gather the range of knowledge or monitoring behaviour by the interviewees. A similar variation was the level of connection between clinical trial staff and the interviewee. Potentially this variation also explains differences in our understanding of the trial. Some interviewees had impactful relationships with these staff, while others seemed unaware of them. Family members were able to attend clinical appointments with the participant, but the participantinterviewee relational contexts would vary how interviewees could interact with clinicians. This is potentially why interviewees' knowledge varied drastically when asked about their understanding of the trial. Similar interventions should address this unforeseen variation so by the end of a trial, this knowledge is uniform for families impacted. For research we should explore how understanding changes through intervention and seek this process out for its relation to reporting impacts and opinions.

Loss of relationships
More critically, the trial participants become part of the trial by either performing violent or impulsive acts, which in the context of partnership and family relationships may involve abuse and control. For example, Like even with phone calls, with text messages, you know he'll read over my shoulder, and I've only recently had to put him in his place over that. He does deny it, but when I have my own family come to me now and say, hey look, I don't know if you're aware, but he's been doing it again, that's been a big no, no for me, but I basically got to the point where I said to [participant] start accepting the help now, or I'm going to walk away (Interview 2) Do you feel unsafe right now? Yes. (Interview 15) Interviewee: Like even with phone calls, with text messages, you know he'll read over my shoulder, and I've only recently had to put him in his place over that. He does deny it, but when I have my own family come to me now and say, hey look, I don't know if you're aware, but he's been doing it again, that's been a big no, no for me, but I basically got to the point where I said to [participant] start accepting the help now, or I'm going to walk away (Interview 2) Facilitator: Do you feel unsafe right now?
This was not uniform amongst participants but tells us who the men could recruit for interviews. Given that relationships were key for each category of findings, participants who have not significantly changed their behaviour may have lost the relationships that would be their interviewee representative.
Facilitator: So, from where you are now with him, do you feel like you would have coped had he not gotten on the trial?
Interviewee: I don't think so, I still think back on it now after eight months, I'm not sure whether or not -yeah, I'm pretty sure, I don't -I think if he wasn't on that medication, I don't think that we would have lasted these last eight months. I don't think that we would have made it (Interview 9).
More particularly, the possible positive skew, the sample, and indeed population have an in-context survivorship bias for interviewee-participant relationships. Women no longer in relationships with these men any more are not likely to participate. Examining deviation, or unique cases, is a strength of qualitative analyses and allows us to contrast different outcomes for processes and unique contexts. Two cases deviate from positive relationships: one case of worsened behaviour and one ex-partner who was no longer in close contact with the participant. In the future, however, an interesting question to explore is 'Why did relationships break down?' and therefore become unavailable for interview. Information on why trial adjacent relationships break down, who these other relationships are (who are no longer available for interview), and the processes of relationship break down, are questions to answer. Interviews and recruitment did not have the scope to ask the two deviant cases why break down occurred. Purposive sampling of cases of relationships that break down or still have problematic behaviour are needed to answer these questions. A more interpretative constructive approach allows this to happen organically and iteratively in the methods -but opposes the linearity of experimental and quantitative designs.

Expectation of legal ramifications
One power imbalance for this study, largely unavoidable, was the participants or interviewees potentially believing they need to report positive impacts, under erroneous assumptions that poor performance on the trial harms participants 'commitment to change' from a legal point of view. Such ramifications were not an element or mechanism of the trial, but the notion was present in the narrative components of some interviewees' interviews: Interviewee: Then he came back, and he told me, 'Look, I have to do this trial, it's going to help me stay out', and he goes, 'it's going to make me tired for a while but they're going to help me. They're going to help me with . . . ' -because he couldn't -he didn't have the money to do the road traffic offenders, whatever it's called. You helped him, and which was good, you helped him stay out of jail. He's been doing the right thing. He doesn't drive the car no more. (Interview 4)

The women's reputations as responsible family members to these men
Finally, the reputation and social status of the women are intertwined with the participants' behaviour, meaning we should expect the women to take a defensive position, discursively distance, and make comments to protect their reputation in a society that is often judgmental of them (Guerin 2016;Maher et al. 2021). For example, discursive strategies for presenting the men's behaviour were noted including stake inoculation, women distancing oneself from their stake in their claims 'I don't know if it is because I am his mother . . . ' (Interview 11 and 17) to allow comments that could be seen as biased; such biased comments were sometimes extreme case formulations: 'he/it has completely changed' (Interviews 16 and 21), while the same interviewees stated that some things have not changed (see Wiggins 2017 for further explanation on such discursive strategies). Such insights on data are particularly at odds with the choice for qualitative content analysis and speak to the benefit of multiple analytical frameworks being utilised by analysts working with qualitative data (e.g., Spiers and Riley 2019) which would not agree to the rigid linear analysis sometimes suggested.

How many interviewees?
A sufficient participant sample is always a question for qualitative research (Baker and Edwards 2012;Small 2009). This study has 'enough' participants when compared to other studies like it (McGinn et al. 2016;McGinn, Taylor, and McColgan 2021), but a critical inspection of quality rather than quantity determines sample size adequacy.
A possible pool of ~160 ReINVEST participants was available, with 24 participants represented (3 interviews relating to one participant). The much larger 'family member' or 'impacted women' population is represented by 26 interviewees. It is not known how many trial participants were asked to provide a contact, nor how many trial participants had women in their life suitable to be interviewed. The final sample represents a small number of the participant population and a smaller portion of the possible family member population, so these findings' transferability is limited.
Again, samples of this size are extremely common for qualitative studies but are a point for all to improve on. Comparatively, this study has a lot more applications than most. The conclusions are not generalised beyond 160 families on this trial, while other studies claim findings for broader populations with the same sample size. Purposeful sampling, perhaps directed by clinicians with insights into these relationships, could yield more complex cases to improve the quality of the sample without necessarily increasing the number of interviews.
A consideration that is not an issue, but something to contemplate, is the choice only to explore women's experience in the lives of these men. This is a boon to the analysis when framed as exploring women's experiences. The data from these women, however, highlighted the impacts of the intervention beyond themselves and onto other social groups or their whole families and even strangers in context (see Thain et al. In preparation for more). Perspectives of these other close relations, including children, may reveal other impacts, opinions, and motivations. For example, are participants open to other men about motivations or impacts? Are other men a hindrance or support for positive behavioural changes? What is the influence of children in this process? Is the presence of children solely a risk or a benefit in behaviour change? A fuller investigation of the suggested systemic relational changes would benefit the trial and research as holistic. The women in these relationships could potentially be elevated to research partners given their observational position and to avoid scrutiny of their role, or lack thereof, in the intervention's outcome. Finally, and most importantly, a constant reflexive point was that women are often treated, and here treated, as 'the women of men', rather than research participants in their own right as they should be. Power differentials, relationships with participants, and iterative process -against the positivistic or quantitative norm -again would benefit this study.

Realities of research processes
In this retrospective view, the issues are clear, but so are the constraints of the sampling source, from the ethical and practical restrictions. Ethically, it is impossible to identify and reapproach these women given it would be known to their partners due to secure communication channels not being available (see below for more). The ReINVEST trial was the primary research program, with the men involved in the trial being the primary research participants. Given this concentration, the interviewed women were not the focus population of ethical applications and considerations given the trial's scope, therefore applying, getting approval for, and developing secure means of communication was not a practical possibility without major changes to the primary research program. Nonetheless, we cannot avoid the possible positive skew in the data. Such a skew is not a fault of the research team or methods but of a common self-selection issue that requires whole research program shifts to attempt to avoid, potentially still unsuccessfully. With the skew in mind, it is impressive that not all reports were positive, suggesting the method of recruitment is not the sole means of this data quality consideration. Equally, the absence of more negative reports (which may or may not exist) does not by themselves eliminate or even reduce the reported experiences of the women interviewed in the dataset -which are overwhelmingly positive.
A potential solution to the survivorship bias is for interviewers to maintain consistent relationships, ethically, with this population (partners and family members of participants) from the beginning of an intervention. Consistent research relationships will allow process data on behaviour changes and highlight possible lead-up issues that may contribute to attrition. That alteration would include better 'negative' results that are otherwise unlikely to be uncovered with the current sampling method. An ethical and procedural review of the researcherinterviewee relationship also has strong possible ethical and data quality outcomes. Interviewees and the content analysis agree on the suggestion that interventions should be more holistic as well as forge 'more-than-stranger' research connections (see Guerin, Leugi, and Thain 2018;and action research otherwise and Thain et al. In preparation for the women's suggestions).

Interview type
The use of phone interviews was largely a consideration of COVID-19 restrictions and the geographic spread of trial participation. While a good choice in that regard, few methodological and ethical issues are present.

Safety on the phone
Given the trial participants (men's) violent or impulsive inclusion criteria, phone interviews run contrary to some recommendations for research on possibly abusive and violent relationships. The World Health Organization's ethical and safety recommendations for intervention research on violence against women suggest caution around phone use (2016,13). This type of consideration is present in other intimate partner violence (IPV) and domestic violence (DV) literature (e.g., Slakoff, Aujla, and PenzeyMoog 2020). Phone data collection can be problematic without direct and personal knowledge around confidentiality and safety, but in this case COVID-19 related research requirements limited the capacity for alternatives, and it must be noted the staff of ReINVEST acted on advice, from stakeholders, and peer groups, that in-person interviews with clinicians would not be appropriate in this case. Problematic for this point is that the COVID-19 pandemic has seen a sharp rise in partner and domestic abuse (Evans et al., 2020;Lyons & Brewer, 2021).
In this sample, however, one interviewee was in an actively dangerous situation in which they were not alone on the call and requested help which was appropriately provided and followed-up (Interview 15). In three cases, the interviewee was not alone during the interview and was, in fact, with the participants (men,Interviews 5,15,20). In two cases, the participant of the trial speaks in the interview (Interviews 15 and 20). Furthermore, abusive behaviours that put women at risk as warned in the above guiding literature occurred (see previous quote on phone monitoring). Such context brings confidentiality, credibility, and safety concerns forward. Following the literature informed recommendations improve safety, ethical practice, and data credibility. Potentially, these elements arise from experimental processes of the whole project 'avoiding' bias or subjectivity in collection and analysis, which involves better understanding the contexts of interviewees (Charmaz 2006).

Training of interviewers
The use of an interviewer with limited experience connects to treating research objects as immutable and unchanging (Guba and Lincoln 1989) and focusing on rigid methods (Small 2008) over building relationships with participants (Brodsky 2001); that if the interview follows the same general method the same data is produced.
Interviewing, however, is a complex skill, and largely there is no training for qualitative interviewing apart from doing them. The interviewer trained during the data collection, who completed most of the interviews, displayed limited interviewing, trauma-informed, or counselling skills. This need for training and consistent reflection on the suitability of interviewers harkens back to considerations for interpreter use but also on thoughts for ethical and safe work for the interviewer themselves (Dickson-Swift et al. 2009). Given the population and issues that arose, more training may be needed, and high levels of interview and trauma-informed practice should be present in the future. In this research, potentially the clinicians, as suggested previously, should carry out such data collection to have these skills trained and certified. More consideration on the counselling or interpersonal skills benchmarks for qualitative researchers should be made but is out of the scope of this reflection.

Time with participants
Another constant issue for qualitative research is the length of time with interviewees. The current interviews averaged 40 min with generally 5-10 min of ethics, consent and rapport building and 5-10 min of exiting ethics, bringing the content average closer to 20-25 min. This time allocation is completely normal for qualitative interview studies. The quantification alone is also not a problem but given the scheduled questions, many answers were only affirmations or refutations, meaning large sections of time are primarily interviewer speech. In shorter interviews, limited qualitative experience data are collected, and, in these cases, quantitative survey methods would potentially reveal more in less time. These are aligned to the positivistic assumption that generalisable and representative samples present data quickly and more of the same validates a claim (Small 2008). This assumption is easily challenged with iterative and processes to uncover variation and divergence in context.

Realities of research processes, again
Interviews are appropriate to allow women to speak openly. The choice of interviews may be sound, yet the execution can be improved for better data collection. Trends in qualitative research suggest that in-person interviews by interviewers known to the interviewee can benefit the data outcome with appropriate reflexivity practices; furthermore, longer interviews can reveal and allow more narrative and personalised details than shorter interviews (Rubin 2000). For example, when interviewees were asked about what suggestions they had for the intervention, many reported that they had none, followed by reporting some later in the interview. One possibility for this is that the relationships between the interviewer and interviewee did not allow a frank discussion of issues or suggestions (see Thain et al. In preparation for more). The benefits of a more open, lengthy, or more importantly iterative approach to collecting data and understanding women's experiences have been advocated for (Fraser and Taylor 2022). However, intervention staff again acted on advice from stakeholders and a women's advisory group that interviews conducted by clinicians, in this instance, would be inappropriate due to bias (though how this was conceptualised is unknown).
Possible improvements to this research, however, come when we reflect on the purpose and use for this information to be collected. If collection is for intervention evaluation, independent, highly trained interviewers are needed. If collection is for service provision to the interviewees, clinicians with better relationships with the interviewees and hopefully clearer understanding of context are likely the best candidates to collect these data and provide support. If collection is for triangulation with other data in the trial, independent interviewers with more structured interview skills are likely needed. In all cases, and what seems to be missing for the junior interviewer in this data collection is some prior experience, particularly with trauma and possibly violence-related training to better observe for and respond to possible markers for DV and IPV, as is suggested in ethical guidelines for this type of research and from qualitative research on this issue (World Health Organization [WHO] 2016;Fraga, 2016, p. 79).

Interview questions
Interview schedules can be improved by seeking input from peers, as ReINVEST did from stakeholders and a women's advisory group which included domestic violence victim survivors. Though an interview schedule was generally attended to for the structure of these interviews, several unscheduled issues arose in the analysis. These include missing questions, question framing and assumptions, and leading questions (inducement). These suggest consistent audit trailing and iteration on interviews by all involved would be beneficial.

Adherence to interview schedules
The assumption that we should stick to schedules is a holdover from positivistic thinking for rigid methods (Small 2008), this assumption is baked into the content analysis. Issues are present, first, by the haphazard and nonlinear nature of interviews, and the short time of interview, not all questions were asked of all participants. Notably, not all demographic questions were asked (see Table 1). There also seems to be additional questions included late in interviews, such as asking about the interviewee's gender identification, while the absence in previous interviews suggests the expectation was that all interviewees were women and therefore asking was unnecessary. Several pieces of information that could inform further analysis were the length of the relationship, qualitative living status, and the presence and care of children from the perspective of the women, but these were not asked consistently. For example, some women spoke at length about their care work with children relative to the participant, while others were not clear on if care work was part of the interviewee's, or the participant's, experience during the intervention. Such care work is known to be important for this topic; for example, in reconciliation of relationships in a legal context (Groggel 2022). When asked, the questions lacked uniformity in timing and phrasing, which altered some interviewees' understanding. Uniformity is not required in the strictest sense, though consistency in information collected is needed for contrast and comparison.
A conversational and open approach to interviewing and data collection is often commendable when aiming for a full exploration with long repeated interviews. In this case, however, it seems misaligned given the single short interviews and the directed nature of the research questions to be answered, rather than being open-ended research aims ('what were the impacts of the intervention?' versus 'To explore the experience of being a woman related to a participant of a violence intervention'). Again, iteration and reflection reflexively on these questions and ideas in research are beneficial here.

Assumptions in questions
Some questions, especially follow-ups, were not answerable by interviews. A lack of knowledge does not mean that interviewees will not answer these questions but rather that answers lack credibility. For example, questions about the trial participants', but not the interviewee's, mindstate, or behaviours outside the interviewee's observation were asked: 'How did he feel about that?'. Otherwise, questions included an assumption that the interviewee can recall fine retrospective detail or that interviewees have the ability, desire, and accuracy to state causal aspects of behaviour. These issues consistently arise in the reflective treatment of language (Hall 1997) and are constantly at play for qualitative and quantitative research (Guerin, Leugi, and Thain 2018). While most interviewees just answered these unanswerable questions, other interviewees resisted with excessive hedging when they wished to remain undecided.
Facilitator: If he did change then hypothetically do you think he would have reverted back, or do you feel like these changes are something that will maintain if he did go off the medication?
Interviewee: I don't think -I'm not 100 per cent about that one. I will sit -can I sit on the fence about that?
Facilitator: Yeah, of course. That's your honest answer, so I can't ask for anything else. What about . . . Interviewee: I feel like he could maintain it. But, yeah, like I said, there's other factors too. If you take one away it just takes another to fall apart, and then we could all be back at square one or beforehand.  A similar issue is that most interviewer questions were framed positively, or the interviewers avoided negatives. Some of these questions are acceptable, for example, 'What changes have you noticed?' but could be less steering; 'Tell me about life while he has been on the trial'. Other times interviewers were overly positive when responding to hearing positive things, with follow-ups by the interviewer reflecting positive content with extra praise. Without addressing these occurrences, we fall into the assumptions of cause and effect from experimentation (the intervention made these experiences) rather than bidirectional constructive influences (the interview itself influences what is said -Saldaña 2011).

Interviewer identity with research object
Such a steering becomes problematic when the interviewer collectively describes themselves with the intervention, e.g., I'm really glad that our trial is making such a large impact on his life.' (Interview 7 -solidarity highlighted). Potentially interviewees' nonresponses to questions, for example, when asking for suggestions or attempting to say, 'I don't know', are avoiding harming the interviewer as a target of that criticism. Identifying the interviewer with the intervention team leads to a further problem termed 'inducement' in the analysis. This sort of issue is only solved by attending to power imbalances in relationships which is sometimes ignored in the experimental or positivistic frame (Lincoln and Guba 2013).

Inducement
Inducement refers to statements, usually follow-up content reflection, asking interviewees if the trial medication was specifically the causal factor for positive impacts. An important note here is that when participants are cautious about saying medication is the causal factor, they display careful thought. That critical thought contrasts with the reported positives when attributed to the medication, particularly when they are asked in a directed way, 'Oh, that's massive, and you think that again that that's the medication that's made that difference in him?' (Interviewer in Interview 2 -Attribution Reflection Emphasised).
The attributing of positive impacts to medication was done by interviewers, yet interviewees agreed to this -potentially interactive with other issues above like power and identity. The agreement, that medication was the key, is at odds with the finding that 'Contextual Factors' were more influential, positively and negatively, for the men's progress (including relationship changes, drug and alcohol use, and economic shifts see Thain et al. In preparation for more). Occurring frequently throughout all but two negative/divergent cases, these inducements by the interviewer are more often seen near positive impactssee Table 4 for occurrences near impacts.
The likelihood of inducement near positives may be due to the sheer number of positives relative to other sentiment coded impacts. That said, inducement is an action by the interviewer, and therefore, should be present consistently. This should be especially the case for negatives as these are mostly side effects but confirmation that 'it was the medication' was less likely near these than some tangential positive impacts. This issue can be avoided by acknowledging the co-constructive nature of qualitative data (Harper 1995); a more general 'What made these changes occur?' is more suitable for co-construction.

Self-reports -motivations for sharing information
Finally, another perennial issue in all psychology research is the reliance on self-reports, as if they are hard data on the real world over interpretations in context (Guerin, Leugi, and Thain 2018;Morse 2018). Self-reports are contingent on the accuracy, ability, and honesty of those asked to give answers. Unfortunately, for several elements of the negative results, interviewees indicated that the intervention participants had somehow been secretive to the clinical staff (and potentially other self-report measures).
For example, being secretive (Guerin 2016) occurred about potential side effects. These cases seemed to also occur where the women and clinician did not have a relationship that was a source of support despite it being often stated as a positive impact by others (Thain et al. In preparation)   While such admissions show that interviews can gain information that the other methods did not, we are unable to know what is also undisclosed, or even lied about, in interviews without following these experiences iteratively.
Limitations for the measurability of violence are an important consideration in violence interventions -one that has been raised by other authors in this space (Strover, 2005, pp. 451-452;Westmarland and Kelly 2012). Women in the lives of violent men, as most often the survivors of the violence the men perpetrate, are therefore the key observer of target behaviour, but difficulties, distrust, or not wanting to involve formal reporting may limit the data on violence or increase secrecy behaviours generally. Assumed judgement of the men legally was highlighted previously as a potential reason for a positive skew. Importantly, if violent relationships ended or were also subject to controlling abuse, it is unlikely, or not possible, for these women to be research participants or report violence truthfully on a monitored phone call. Again, this highlights the need for examining the social context for interventionmainly relationships between researcher and interview. Here family could be considered the target population, potentially with women as research partners alongside clinical support. For critical methods, secrecy and monitoring are key and help with understanding what a participant can report (Guerin 2016; also see action research). Closer consideration of those contexts can improve outcomes and improve data quality and control.

And finally, the realities of research processes
Altogether it is important to highlight the above 'issues' should not be 'solved'. These behaviours in interviews by both interviewee and interviewer are tools of an interview that are functional. To remain on topic, leading and steering questions and statements must be used by an interviewer. In some cases, the type of question needs to change. In others, it is a reflexive practice of the researcher's relation to the interviewee. That relating point is especially problematic when the interviewer is part of the group being assessed by the interviewees.
Similarly, reflecting speech content to someone can be used to maintain attention or encourage correction. What is difficult is avoiding assumptions in reflection; in this case, any praise of the participants by interviewees was reflected with the assumption that the men's behaviour was worse before the trial. Similarly, assumptions can deliberatively be used if the relationship with participants allows challenging -the participant will resist that assumption. More training is the only way to avoid this incident. Finally, the most consistent possible improvement is to have interviewees elevated beyond informants to researchers or partners of the interventions (if willing and ethical). These women, and family and friends like them, are the key observers of change and context as demonstrated by Interviewee 14 who described that their partner was on the active medication prior to random assignment to the placebo condition based on observation of his behaviour. Including family as a primary motivator for such interventions appropriately accepts these women's position and stake in the interventions (see Participatory Action Research, e.g., Sullivan et al. 2005).

Conclusions
Throughout the consideration common qualities of quantitative, positivistic, and experimental assumptions are impregnated into the process of this research. In most cases raising serious concerns that would be avoided with more nonpositivistic, non-experimental, or qualitative understanding in design. Remembering that overall, the content analysis produced did well to privilege and focus on the voices of the women connected to this trial and provide their assessment of the intervention, we can always be critical in reflecting on successful research and analysis. The considerations presented here move towards a more contextual and holistic approach to research, as well as understanding the intervention beyond the positivist scope of 'women's experiences as reflected in verbal reports' and towards a family-not-individual intervention with important contextual and discursive elements. The suggested improvements could render higher data quality, potentially better outcomes, and better ethical practice for research like this but also more widely for assessing the use of qualitative content analysis and positivist assumptions in research programs.
The analyst here acknowledges that such improvements are always wanted in research but are ultimately subject to funding and timelines set out of the control of a research team -that such limitations are used by some as a reason for reduced or denied funding is a paradoxical issue when highlighting these points for improvement in research. At the very least, I hope that the time spent critically reflecting on my work on this project can help some people think about the myriad of different implications we consider when we do qualitative research.

Data availability statement
The data are not available for sharing but further presentation of it is found in referenced work found throughout the document. Given the sensitive nature of the participation, further information is not ethically permitted on these participants.