Background
Longitudinal qualitative research (LQR) is said to be that which focuses on experience over time, with change being the key focus of analysis [
1]. Alongside understanding what change has happened, LQR explores how and why change happens within a socio-cultural context [
2]. Practically, LQR can be understood as having two purposes: to collect data about a phenomenon over two or more time periods, or an analysis which involves comparisons of data across time periods [
3]. LQR has a steeped history in the social science arena, for instance in well-known datasets such as the
Timescapes series in the UK [
4]. It is starting to be used in health research and health services research. Within health research, LQR most often takes the form of illuminating illness or recovery trajectories of patients in order to inform future health care priorities [
5]. This most often takes the form of ‘serial interviews’ with the same cohort of patients over a given time period, about a specific disease or condition [
1‐
3,
5‐
9]. The emphasis is on repeated contact with the same participants over time. Descriptions of this particular method are almost to the exclusion of other ways of collecting LQR data. Little methodological work has been published in relation to how LQR can be undertaken in relation to evaluations, intervention assessment or embedded as part of a randomized controlled trial in health research (although some authors working in the social policy space have explored elements of the above [
10,
11]). Simultaneously, there is a dearth of literature which examines LQR in relation to applied health research [
1,
2], as opposed to health research with patients.
The extensive volume of data which LQR can capture, alongside the inherent complexity resident in it, is said to be problematic. Narrative methods require specific attention to detail and therefore may be unsuitable for studies with large numbers of participants [
8] or a large amount of data. Managing large quantities of qualitative data has the potential for the researcher to feel like they are ‘drowning in data’ [
12] with the path to interpretation fraught with complexity [
6]. Management and analysis of large volumes of temporal data is a key consideration of LQR researchers [
4]. Correspondingly, there are few ‘off the shelf’ procedures for analysing LQR leading to researchers being unsure with what to do with their data [
4]. This can lead to research teams having to design their own bespoke analytic methods to meet this need [
5]. Particularly, the analysis of multi-dimensional data is a challenge which is not well described or reported in the literature [
1]. By multi-dimensional, this could mean any study which seeks to involve more than one qualitative method in a longitudinal manner, for instance, multiple instances of interviews and ethnography over time. This challenge of reporting could be because this field is in its relative infancy, with LQR studies involving multiple methods in applied health research only starting to appear in the literature in the past few years [
13‐
17].
It is useful to look at the current state of play with regard to LQR analysis. Authors working in the health LQR sphere who have published their analytic strategy tend to have, across all data sources, undertaken a thematic or constant comparison analysis [
7,
15], a narrative approach [
5,
6,
8] or deductively coded against an existing conceptual framework or taxonomy [
14]. Explicit and careful attention was paid to the analytic process in a LQR study conducted in England about how motivational interviewing for depression after stroke may be effective [
7]. ‘Parallel-serial memoing’ was the resultant technique developed and allowed a consensus to develop across different researchers in the same team. The focus was placed on how different researcher’s interpretations of the same dataset can be coherently brought together over time. The LQR dataset was based on one data source; transcripts of several motivational interviewing sessions. The research team conducted a thematic analysis based on the serial memos they developed in parallel to each other. A study in New Zealand conducted repeated interviews over 24 months with patients who had suffered a traumatic brain injury, and also their family members [
5]. The research team describe using a narrative style analytic approach using “case sets” (one case set per participant) whereby transcripts at the 12 and 24 month time points were coded based on codes developed a priori from the six month time point transcripts in order to capture change or maintenance. The analysis underpinning a LQR study undertaken with first time parents in Austria is one of the few published accounts of how multiple and sometimes differing perspectives on the same topic over time can be analysed in a relatively systematic manner [
18]. However, this articulated analysis relied on just one method – serial interviewing.
Despite the advances in the LQR analysis field described above, concrete descriptions of how research teams coherently and meaningfully
integrated and made sense of the data
over time from different sources are largely absent and elusive. Subsequently, there is minimal practical guidance given to researchers who may want to undertake this task. This risks the researcher approaching the analytic stage of a LQR project with a lack of described techniques in order to concentrate the data into a sufficiently meaningful
focused account. This focused account could take many different forms. For example, it could portray how a team of healthcare ward staff interacted with an intervention over the period of an 18 month study, using data collected from in depth interviews and ethnographic field notes. Equally, it could pertain to how an individual GP utilizes a new software programme, based on think aloud interviews and non-participant observation over a 12 month implementation period. Critically, this focused account should aim to integrate data from all methods used in a LQR study in order to make sense of ‘what happened {to the GP or ward team} during the lifecourse of the study’ (changes over time) but also ‘what happened across the whole dataset at different time points’ (comparisons between GPs or ward teams at any specified point in the research process). It should aim to do this whilst maintaining a distilled version of the richness embodied in the data sources rather than a reductionist, dispersed account. It has been stated that analytic strategies which purport the first stage as coding or sorting text into discrete units of meaning risk stripping contextual richness away [
19] and ‘breaking apart’ a participant’s story [
6].
We have devised an analytic process which speaks to the above issue. It is called a
pen portrait and has been used by the authors of this paper to successfully concentrate a large amount of longitudinal qualitative data into a focused account, in a previous empirical study [
17]. The aim of this paper is to describe and explicate the process of creating and using pen portraits to conduct an analysis of LQR data.
Discussion
In this paper, we have outlined the process of constructing a pen portrait with the intent that researchers may use this process in their own analyses of LQR data. We note four distinct stages: understanding and defining the core focus, designing the basic pen portrait structure, populating the content and, finally, interpretation. We give a large amount of instructive and - what we hope is helpful - detail in the first three stages but would encourage researchers to read more widely around issues of interpretation. Throughout our account, we provide pertinent examples of how we personally employed the stages described through reflections based on the dataset for which the pen portrait process was originally devised.
Braun and Clarke, in their 2006 classic text [
22], state that a previous absence of clear and concise guidelines around thematic analysis may have led to an ‘anything goes’ critique of qualitative research. That is, by not discussing the ‘how to’ of analysis, techniques are therefore kept mysterious and elitist. Concrete advice on how to perform an analysis (of any kind) works towards making the analytic method accessible and democratic. We devised this bespoke analytic process because a search of the methodological literature provided no guidance whatsoever as to how an applied health researcher should go about the task of integrating large amounts of qualitative data from multiple sources over time, in a focused manner. LQR methods in the social sciences are seemingly well rehearsed [
4] but their analytic strategies – where explicated and published - offer little assistance as they tend to focus on serial interviewing of the same participants over a period of years. In contrast, our project saw us collect qualitative data from 17
teams of people, using three distinct methods over an 18 month period. We needed an analytic method which was less about exploration and significantly more about answering specific research questions which were formulated a priori.
Several authors have noticed the above lack of instruction in the LQR methodological literature and have issued pleas for health LQR researchers to publish their methodological reflections in order to move the method forward [
1,
2]. Calman et al. (2013) [
1] have noted that the published literature relating to LQR is “limited in highlighting debates about LQR, focusing on the reporting on findings rather than developing debate about this emerging methodology”. We hope that by demonstrating the stages of the pen portrait method, and using a worked example to illustrate context, that we have answered this call and provided clear and concise guidelines. We believe that our specific contribution to moving LQR analysis forward is the novelty of proposing a technique which explicitly looks to integrate different methods over time. Some literature already exists with regard to researchers being able to make meaningful sense of change over time based on one method (such a serial interviewing of the same patients). Bringing data together from different qualitative methods, captured over time, is largely non-existent. This matters because applied health researchers are increasingly making us of multiple methods within the same study [
13‐
17] but have no analytic instruction available to them. More important to us than bridging a gap in the methodological texts, our intention is that researchers are able to use the stages of the pen portrait as described in this paper practically, to develop a
focused understanding of what their LQR data is telling them.
Of great importance to us as developers of this technique, is the notion of adaptability and flexibility in its use going forward. To provide an analogy, we expect that we have given people the overall recipe for the dish but we expect that elements of the ingredients and their ratios will change over time, potentially leading to improvements in the flavour. We propose that the potential scope for the pen portrait technique is far-reaching and diverse. We see few restrictions on the ‘unit of analysis’ to which this could apply - in our case this was a ward, but it could equally be applied to an individual (following a health professional or a patient over time). In our case, we chose ‘engagement’ as our focus but we could have chosen other factors such as staff attitudes or perceptions. Outside of the realm of interventions, other foci could include patient experiences (e.g. disease symptoms or satisfaction). Finally, we believe the number of analytical units to also be flexible. In our case, we analysed the engagement trajectories of 17 intervention wards. We see no reason why the technique could not be applied to just one unit - e.g. one person or one ward - if the research question was not concerned with comparison between units but about a particular unit’s trajectory. A potential limitation is the number of units of analysis - and indeed the volume of data - that can be included which will be limited by the need for a largely consistent approach to the pen portrait steps. This issue may be hard to control in a very large study involving more than a small group of qualitative researchers.
We have already encountered a natural experiment in this regard as colleagues in our applied health research team have started to use the pen portrait technique, in the absence of any other structured manner of integrating multiple qualitative sources over time. Louch et al. (2018) [
16] are the first to publish their findings (aside from our previous work for which the method was developed [
17]). It is interesting to see how Louch et al. adapted our original premise by adding to and subtracting from elements of our approach which did not directly fit their analytic need. This demonstrates how the pen portrait technique has been taken forward as a
concept rather than rigid proforma. Louch et al. go further than we did in developing distinct parts of the pen portrait which intuitively spoke to the niche needs of their analytic project. We hope others will adapt the technique for their own purposes.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.