Using fsQCA to Illuminate Person Attributes of Music Engagement in Alzheimer's Disease

Abstract

Preserved engagement with music in Alzheimer's disease (AD) is noteworthy given that such persons lack interest and engagement in the activities of daily life. Because music engagement is associated with increased well-being, illuminating personal attributes that facilitate music engagement is an important step towards utilizing music as a therapeutic tool. Here, we use Fuzzy Set Qualitative Comparative Analysis, a systematic approach to case study series analysis, to explore the role of personal attributes such as musical semantic memories, music perceptual abilities, and overall cognitive status in facilitating music engagement in 15 individuals with a diagnosis of probable AD. Nine different solution terms revealed many different pathways to preserved music engagement in AD. Solutions demonstrated the equifinality of music engagement and the usefulness of the qualitative comparative analysis approach. This article is meant to provide both concrete evidence for the role of different person attributes in music engagement in AD and an illustration of the application of qualitative comparative analysis. We discuss our results using the Comprehensive Process Model as a framework and provide suggestions on how to incorporate qualitative comparative analysis in the research workflow.

What allows individuals affected by Alzheimer's disease (AD) to engage with music, despite suffering cognitive impairment in multiple domains and resulting loss of function in daily life? Engagement occurs when an individual is attentive to or occupied with a stimulus and, according to the Comprehensive Process Model (Cohen-Mansfield et al., 2011), is a product of interactions between the person, the stimulus, and the environment. Individuals living with dementia often display a lack of engagement in their daily lives, to the detriment of their quality of life (Cohen-Mansfield et al., 2009). We have argued that the Comprehensive Process Model is a useful theoretical framework for building an account of music engagement in AD (Vanstone & Cuddy, 2020), and some considerations related to the present study are described in the following paragraphs.

Engagement with musical activities is associated with increased well-being in AD (Särkämö et al., 2014). Music-based interventions, which inherently involve music engagement, show promise for managing the behavioral-affective symptoms of dementia (McDermott et al., 2013; Moreno-Morales et al., 2020; Vink et al., 2004). Clinical guidelines recommend that people with dementia be offered activities to promote engagement and well-being, with specific reference to music activities (National Institute for Health and Care Excellence, 2018). It is clear, then, that music engagement does occur in AD, and that maintaining or facilitating it is an issue of clinical importance. However, since engagement is facilitated by a range of factors—broadly subsumed under the categories of person, stimulus, and environment—it is a methodologically complex task to identify what conditions are necessary or sufficient for a person with AD to engage with music.

To sustain attention to a stimulus, to engage in a meaningful way, one must be able to perceive and process that stimulus, at least to some extent. If certain music cognitive abilities are preserved in a listener with AD, then it stands to reason that those preserved abilities could be one person attribute that facilitates engagement with musical stimuli (King et al., 2019). Although music cognitive abilities are not universally immune to AD, some aspects of music processing are more durable than others (Baird & Samson, 2009, 2015; Baird et al., 2020), and numerous case studies have documented instances in which individuals retain musical abilities despite severe impairment in other domains.

Musical semantic memory refers to the memory system for musical knowledge, including melodies—a system analogous to, although functionally dissociable from, the memory systems for verbal or extramusical knowledge (Cuddy, 2018). It has also been referred to as the musical lexicon (Peretz et al., 2009). Neuropsychological evidence suggests that musical semantic memories appear particularly robust in the face of AD, sometimes even in individuals with severe AD (Cuddy et al., 2012). Many listeners with AD also show preserved ability to sing tunes when prompted by the lyrics. Certain aspects of melodic processing, particularly the ability to detect distortions in melodies, also tend to be preserved in AD participants with mild AD as well as in some whose AD is at the moderate stage (Kerer et al., 2013; Vanstone & Cuddy, 2010). Relative to robustly encoded memory for long-familiar melodies, explicit memory for newly presented melodies is more likely to be impaired by AD (Bartlett et al., 1995; Ménard & Belleville, 2009; Vanstone et al., 2012), as is the ability to recall the names of familiar melodies (Kerer et al., 2013). Of the forms of musical memory studied to date, then, musical semantic memories appear most likely to be preserved in AD, which suggests that it has potential to be broadly relevant to music engagement.

It is likely that musical semantic memories have a mechanistic role in explaining music engagement in persons with AD also because these memories are implicated in a number of the other mechanisms by which music engagement is known to occur in the general population. First, listeners tend to have more intense and pleasant emotional responses to familiar music relative to unfamiliar music (Ali & Peynircioǧlu, 2010; Reschke-Hernández et al., 2020). To the extent that preserved memory enhances the feeling of familiarity, it would be expected to enhance emotional response. Second, music-evoked autobiographical memories are more likely to be elicited by familiar music (Janata et al., 2007). This phenomenon extends to listeners with AD (El Haj et al., 2012), for whom it tends to be a positive experience (Cuddy et al., 2017). Furthermore, long-familiar music is often part of a repertoire of music that is widely known within a culture, thus creating potential for social connection through shared musical experience. Individuals are able to participate more meaningfully in a group sing-along, for example, if they remember the melodies being sung.

However, it is unlikely that preserved musical semantic memories would be the single explanatory factor for music engagement in AD. Music engagement is manifest in a variety of forms, some of which involve other higher-order functional abilities. For example, an avid record collector would show decreased music engagement if they lost the physical or cognitive ability to operate a record player. Indeed, musical abilities and practices evolve across the lifespan (Trehub et al., 2019) and are shaped by individual differences within the person (Delsing et al., 2008; Kreutz et al., 2008; Rentfrow et al., 2011) and by the social and cultural environment (Jacoby & McDermott, 2017; McDermott et al., 2016). Musical semantic memories would be predicted to play a role in facilitating music engagement for people with AD, but this role needs to be considered in the context of other factors. Framed in terms of the Comprehensive Process Model, we ask: What combination of characteristics within the person, the stimulus, and the environment are likely to give rise to engagement with musical stimuli in people with AD?

This question poses a few methodological challenges. It involves a novel application of the Comprehensive Process Model to an exploratory question in the psychology of music and dementia. Given the complexity of how individuals develop their patterns of music engagement, we also expect there to be multiple “pathways,” or combinations of favorable conditions, that facilitate a person with AD to engage with music; in other words, the relationship between musical semantic memory and music engagement is unlikely to be linear. The sample sizes required to address this question with a regression analysis are prohibitive given that it involves a clinical population and an exploratory research question. Furthermore, regression-based approaches risk overlooking the expected relationships and multiple distinct patterns of association between musical semantic memory, other variables, and music engagement.

To address these challenges, we turned to Fuzzy Set Qualitative Comparative Analysis (fsQCA) to make systematic comparisons across a series of 15 individuals with AD. Fundamentally, fsQCA is a method for making systematic comparisons across a series of cases, identifying the combinations of conditions that are necessary or sufficient for a given outcome to occur (Ragin, 2009).

QCA was first adopted as a case comparative method in sociology and political science. Although QCA has often been applied to psychological data in the context of management research (Pappas & Woodside, 2021), its application in psychology has developed more slowly. However, in more recent years, it has been applied as a complementary approach to structural equation modelling in larger datasets (e.g., Navarro-Mateu et al., 2020), as a standalone approach to analyzing treatment outcome data (e.g., Haynes et al., 2017), and as a technique for evidence synthesis across treatment outcome studies (Batho et al., 2021). Despite the importance of case series research in cognitive neuropsychology (Schwartz & Dell, 2010), we have not identified any published research in which fsQCA is used to analyze a case series involving quantitative data from cognitive tasks.

For the present exposition, we adopt fsQCA terminology. Each participant is referred to as a case. Each case is assessed on several variables, called the conditions and the outcome. In the present study, conditions are variables such as dementia severity and level of general education (a proxy for premorbid socio-cognitive status) and the outcome is level of music engagement. FsQCA generates solution terms that indicate how different conditions may be combined to account for the outcome.

The “fuzziness” of fsQCA refers to the fact that assessment is not binary and, in this respect, is unlike assessment for crisp set QCA. In fsQCA, cases are assigned fuzzy set scores between 0 and 1 for each condition. 0 indicates that the condition is not present in the case, 1 indicates that it is fully present, and 0.5 indicates maximal ambiguity regarding whether it is present or not.

Although conceptually similar to cluster analysis, which answers questions about which cases are similar to each other, fsQCA can identify the different ways in which predictor variables can be configured to provide the conditions in which the outcome of interest occurs (Greckhamer et al., 2018; Ordanini et al., 2014; Pappas & Woodside, 2021). It is thus particularly well suited to explore the present research question of how different predictors may lead to music engagement in AD patients. This paper aims to address this question as well as provide an illustrative example of using fsQCA in a psychology case series. Here, we describe the data processing and data analysis steps in some detail, and for a more in-depth introduction to fsQCA we refer to the comprehensive text by Schneider and Wagemann (Schneider & Wagemann, 2012).

The strength of fsQCA for our current dataset may also be highlighted by considering other, more standard approaches, such as running a multiple regression. Those techniques are particularly suitable to larger datasets in which the relationships between the predictor variables and dependent variable are relatively consistent across participants. In contrast, fsQCA is better suited where there are different pathways to the outcome of interest, different combinations (configurations) of conditions that are necessary or sufficient for the outcome to occur. For example, a variety of different conditions may account for the outcome “being at concerts,” such as “enjoying the band” or “being a stagehand.”

FsQCA might then provide several solution terms for “being at concerts”: one might be “enjoying the band” while not “being a stagehand”; a second might be “being a stagehand” but not “enjoying the band,” a third might be “enjoying the band” while also “being a stagehand.” Given that different people are described by these three solution terms, a regression analysis would not be able to generate a suitable regression term, as the predictors “enjoying the band” and “being a stagehand” can be either positive or negative while leading to the same outcome, namely “being at concerts.” The strength of fsQCA is thus that it can provide insights that include the logical operator “or” as well as “and.”

Methods

Participants

Fifteen participants with a diagnosis of probable AD (5 female, 10 male) were recruited. We refer to this group as AD participants forthwith. All were 65 years of age or older, M = 76.07 years, SD = 6.87 years of age, had normal or corrected-to-normal hearing, with no history of stroke, other acquired brain injury, or substance dependence. Additionally, each participant was required to have a caregiver who had regular and ongoing contact with the participant and who was willing to provide information regarding the participant's background and current functioning.

A separate group of n = 29 older adult control participants (17 female, 12 male) was recruited who fulfilled the same criteria as our AD participants, M = 71.03 years, SD = 6.04 years of age. These participants underwent cognitive screening to rule out undiagnosed dementia or mild cognitive impairment.

All participants provided written informed consent for participation in the study and for their anonymized data to be used in disseminating study findings. For participants with AD who lacked capacity to provide consent, written consent was obtained from a caregiver instead.

Measures

Standardized tests from the test repertoire and the literature were used to assess present cognitive status, musical semantic memory, music perceptual abilities, and musical engagement. Demographic data regarding level of music education and general education served as a proxy for premorbid socio-cognitive status.

Cognitive Screening: Mini-mental Status Exam (MMSE) and Montreal Cognitive Assessment (MoCA)

The Mini-Mental Status Exam (MMSE; Folstein et al., 1975) is a brief test used to evaluate global cognitive function in orientation, memory, attention, and language. The MMSE was used as a marker of dementia severity in the AD group and was completed by control participants to facilitate comparison between groups. To establish that control participants were unlikely to be living with undiagnosed dementia or mild cognitive impairment, they also completed the Montreal Cognitive Assessment (MoCA; Ziad et al., 2005), which is well-established as a more sensitive screening instrument.

Instrumental Tunes Test (ITT)

The Instrumental Tunes Test (ITT; Sikka et al., 2015) contains a series of 50 excerpts of instrumental melodies recorded with a synthesized piano timbre. Half of these excerpts are well known to English-speaking, North American listeners. The other half are melodies composed by altering the note order of the familiar tunes. This preserves the musical characteristics of the familiar melodies but prevents recognition. After each melody was presented, the participant was requested to indicate whether the melody was familiar. Performance on the ITT can be viewed as an indicator of musical semantic memory.

Montreal Battery for Evaluation of Amusia (MBEA)

A shortened version of the Montreal Battery for Evaluation of Amusia (MBEA; Peretz et al., 2003) was administered as a general measure of music perceptual abilities. Eighty pairs of short melodies were presented to the participants, who were asked to identify whether each pair was the same or different. In 40 trials, the presented pair is identical, in the remaining trials, a note is changed, affecting the musical interval, the melodic contour, the musical scale, or rhythm of the melody.

Music Engagement Questionnaire (MusEQ)

Lastly, participants or their caregivers completed the Music Engagement Questionnaire (MusEQ; Vanstone et al., 2016), which contains 35 items assessing an individual's current degree of engagement with music activities. However, some items imply additional, non-musical activities of daily living—for example, “listening to music while performing household chores”—and as an individual's dementia progresses, it becomes less likely that they will carry out such activities. Content analysis of the 35 MusEQ items revealed 11 items that do not make reference to higher-order, non-musical activities. These items were used to calculate a “basic MusEq” score that is less influenced by the participant's broader functional impairments, capturing instead the affective and behavioral responses that most closely relate to music itself. Indeed, the difference between basic and full scale scores was negatively correlated with MMSE scores, r(42) = −.42, p = .009; participants with the greatest cognitive impairment were the ones whose MusEQ scores increased most when only the basic scale items were included. Internal reliability of these basic items was high, α = .88, and the basic score correlated highly with the total MusEQ score, r(42) = .88, p < .001. Items for the basic score are listed in Table 1.

Table 1.

MusEQ items—participants or their caregivers were given the following instructions: “Thinking of the person you described in the previous questionnaire, please indicate how much each of the following statements describes him/her during the past month.” Answers were indicated on a five-point scale, with 1 = “Not at all,” and 5 = “Very much.”

MusEQ items
1. Moves his/her body to the beat of music that is playing (e.g., tapping my feet or bobbing his/her head).
2. When by him/herself or with close family/friends, (s)he will hum or sing along with music that is being played.
3. Sings as (s)he goes about daily activities
4. Tells others if (s)he hears a song that (s)he really likes.
5. If others are singing, (s)he joins in.
6. His/her social activities revolve around music.
7. Is usually excited by the chance to hear the music that (s)he likes.
8. People who know her/him describe her/him as a musical person.
9. Relaxes when listening to peaceful music.
10. When (s)he listens to familiar music, (s)he recalls events from her/his past.
11. Becomes cheerful when listening to music (s)he enjoys.

Measures were collected in a quiet room at the participant's home or in the laboratory, according to participant preference. During testing, participants were encouraged to take breaks as necessary or split testing into multiple shorter sessions if fatigued or inattentive.

Data Processing

Test Scoring

ITT and MBEA responses were scored using an index of discriminability, d′, calculated from their hit and false alarm rates (see Stanislaw & Todorov, 1999). MMSE and MOCA responses were scored according to the established scoring rules. Responses on the basic MusEQ items were averaged, such that the lowest possible average of 1.00 denotes an individual “not at all” engaged in musical activities, and the highest possible average of 5.00 denotes an individual who is “very much” engaged in musical activities.

Calibration of Fuzzy Set Scores

Test scores were converted to fuzzy set scores using a log odds transformation to derive values between 0 and 1. (Please see Ragin (2008) for a helpful and detailed discussion of the rationale and process for fuzzy set calibration.) As described in the introduction, a fuzzy set score of 1 denotes full membership in the condition or outcome. A fuzzy set score of 0 denotes full non-membership in the condition or outcome, with 0.5 indicating the maximal level of ambiguity in membership.

When converting data to fuzzy set scores, we specify which values in the data correspond to fuzzy set scores of 0.05 and 0.95, rather than 0 and 1, in order to avoid generating fuzzy set scores with infinity values (Ragin, 2008). The values corresponding to fuzzy set scores of 0.05, 0.5, and 0.95 are referred to as the thresholds for coding non-membership, maximally ambiguous membership, and full membership, respectively, in each set. In fsQCA, the coding values are specified based on a theoretical understanding of the underlying data; fuzzy set scores should reflect the meaning of the data in relation to the conditions and outcome, rather than a “one size fits all” statistical transformation of the data. Our choices of coding scheme values are described below.

For the ITT and MBEA, control participants’ scores of these measures were used to calibrate the fuzzy set scores. Full membership (preserved ability) was assigned to AD participants if their scores were greater than or equal to the control group mean. Full non-membership (impaired ability) was coded as less than or equal to two standard deviations below the control group mean. Maximal ambiguity was defined as one standard deviation below the control group mean. Because full membership (preserved ability) is determined by the control group mean—rather than, for example, the maximum possible score on these tests—a high fuzzy set score suggests that the participant's ability has been relatively unaffected by cognitive decline.

MMSE scores denoting mild AD or severe MD were derived from literature on MMSE and dementia severity (Perneczky et al., 2006), and they were coded such that full membership indicates no evidence of cognitive impairment (control participants), while non-membership indicates severe impairment. Full membership and full non-membership for general education were defined as having at least four years of university study and no post-secondary education, respectively. Full membership and full non-membership for music education were defined as at least 10 years of music training and less than one year of music training respectively. Coding values for the latter two variables were identified based on the distributions within the overall dataset, allowing us to distinguish those with the most education from those with the least.

Fuzzy set scores were coded for MusEQ by defining full membership and full non-membership in relation to the Likert scale descriptors used in the MusEQ. Answering consistently that the various descriptions of music engagement very much described a participant yielded a fuzzy set score of 1. Conversely, a fuzzy set score of 0 was used for cases in which participants or their caregivers had consistently answered that the various descriptions of music engagement do not at all describe a participant. The coding scheme is summarized in Table 2. Note that music education was assessed using a categorical scale. Thus, fuzzy set scores are given for each of the answers on the scale.

Table 2.

Coding scheme for conditions and outcome.

	Full non-membership (fuzzy set score = 0.05)	Maximal ambiguity (fuzzy set score = 0.5)	Full membership (fuzzy set score = 0.95)
ITT (d′)	≤ 0.84	1.38	≥ 1.92
MBEA (d′)	≤ 1.01	1.77	≥ 2.52
MMSE	≤ 10	21	30
General education (years)	12	14	≥ 16
Music education (years)	<1	1–3: 0.33 4–7: 0.67 8–10: 0.83	>10
MusEQ (average rating)	1	3	5

To study patterns in which the absence of one condition interacts with what other conditions are present, fuzzy set scores then also need to be negated. To do so, the fuzzy set scores obtained via the coding scheme described above were subtracted from 1 to obtain negated forms of fuzzy set scores. Entering these additional scores to fsQCA allows the exploration of not only which conditions produce high engagement with music in everyday life but also which conditions produce low engagement with music in everyday life

Data Analysis

FsQCA was implemented using MATLAB scripts provided by Koranji (2016) on GitHub. Configuration of conditions were retained for interpretation if they showed consistency greater than 0.80 with the outcome. These configurations are called solution terms. Consistency scores measure the proportion of a solution term's membership that also has membership in the outcome and thus can be used as an indicator for how well a relation between condition and outcome is supported. Membership in a solution term is dependent on the degree to which membership is given to the conditions of the configurations. For example, if all the people who would reply yes to “being a stagehand” would have the same membership level for the outcome “being at concerts,” then a solution term with only this condition would have a consistency of 100%—the condition is consistently associated with the outcome.

The “reverse” score, the proportion of an outcome's membership that also has membership in a solution term is called coverage and indicates how common a particular combination of conditions and outcome is. The combinations of condition and outcome of interest to us are cases with high consistency and coverage, and cases with high consistency and low coverage. For our example above, “being a stagehand” may have high consistency with an outcome “being at concerts,” but have low coverage given that they would make up a small proportion of the concertgoers. While the condition is well related to the outcome, this relationship is perhaps less common but nonetheless valuable for interpretation.

In other words, consistency tells us how reliably an outcome is observed given a combination of conditions, and coverage tells us how many cases are described by this combination of conditions. For a more complete description of how consistency and coverage are evaluated, please see Ragin (2006). Mathematically, consistency and coverage are defined as following, where X represents membership to a solution term, and Y represents membership to the outcome, and i indicates different cases. Min(X_i,Y_i) refers to the lesser value of X_i and Y_i.

$$Consistency(X_i\le Y_i)={\displaystyle \frac{\sum min(X_i,Y_i)}{\sum X_i}}$$

$$Coverage(X_i\le Y_i)={\displaystyle \frac{\sum min(X_i,Y_i)}{\sum Y_i}}$$

Results

There was a wide range of MusEQ scores (control participants: 1.36–4.09; AD participants: 1.64–5.00), indicating a dataset with sufficient spread to investigate which conditions are necessary or sufficient for music engagement in the everyday life of patients with AD. Interpreting fsQCA involves balancing complexity and parsimony—a solution that included n solution terms (the highest possible complexity) would offer little insight if the objective is to compare n cases. By combining some of these terms together through a process known as minimization, researchers can identify a parsimonious solution with fewer terms, allowing for greater clarity in the interpretation. However, an overly parsimonious solution carries the risk of ignoring theoretically interesting case data. Our analysis is based on an intermediate solution generated by the fsQCA software, which balances consistency and coverage to achieve a solution that allows for meaningful comparison across cases while also capturing important variation between cases.

The nine intermediate solution terms generated by the fsQCA are listed in Table 3 along with their consistency and coverage values. As stated above, only solution terms with consistency greater than 80% are considered. Coverage ranged from 8% to 23%. Five of the seven solution terms for full membership in MusEQ included preserved ITT, and five solution terms included preserved MBEA. Two solution terms included high values on the MMSE, two included high values for music education (abbreviated as MusEd), and two included general education (abbreviated as GenEd). Two of the solution terms were generated for full non-membership in MusEQ. In both of these terms, participants had mild cognitive impairment but no music and no post-secondary education.

Table 3.

Seven fsQCA solution terms for full membership in MusEQ and two solution terms for full non-membership in MusEQ with + indicating membership, − indicating non-membership, and = indicating an ambiguous membership for the different conditions.

Solution term	ITT	MBEA	MMSE	MusEd	GenEd	consistency	coverage
Full membership in MusEQ
1)	+	+	+		=	100%	8%
2)	+	+	−		−	86%	23%
3)	+	+	−	+	−	100%	10%
4)	+	−	−		+	98%	19%
5)	+	−	−		−	87%	14%
6)	−	+	−		−	100%	18%
7)	−	+	+	+	+	100%	10%
Full non-membership in MusEQ
8)	+	−	+	−	−	100%	17%
9)	−	+	+	−	−	93%	23%

Discussion

The many different patterns of conditions highlighted in the solution terms demonstrate the value of employing fsQCA in a small-n study where there are complex or non-linear relationships between variables. Note, in particular, the solution terms that stand in apparent contrast to each other—for example, solution terms 4 and 5 compared with solution terms 6 and 7. Despite differing in some important characteristics, these terms lead to the same outcome, namely, music engagement in daily life. The data demonstrate the equifinality of music engagement, a finding that would be overlooked if the data were analyzed with a correlational approach.

As illustrated in Table 3, each solution term included either ITT, MBEA, or both as conditions for music engagement. FsQCA shows here that preserved musical semantic memory and/or preserved music perceptual abilities are conditions to music engagement. A correlational approach would never be able to capture this information because it cannot provide insights that include the logical operator “or.”

The results of an fsQCA are interpreted (in the first instance) at the level of its solution terms, allowing for an analysis that simultaneously considers the data from the standpoint of its conditions as well as its cases. In our analysis of fsQCA solution terms, preserved musical semantic memory stands out as particularly relevant to understanding music engagement in people with dementia. Of the seven different combinations of conditions shown in our data to be associated with music engagement, solution terms 1–5 included the condition of preserved musical semantic memory. However, this preserved ability has more explanatory power when examined alongside other factors and, conversely, in light of the sets of cases that share common characteristics. These sets of cases are described in the following paragraphs.

Solution term 1 is perhaps the least surprising; in these cases, music engagement occurs in the context of relatively mild dementia and with intact music perceptual abilities. Although this pattern has relatively small coverage within this case series, it is likely relevant to the many individuals with mild-stage dementia who continue to enjoy music activities in their daily lives. In people with dementia who have these characteristics, we would expect levels of music engagement to mirror those of the general population. Quite simply, these cases are ones where dementia has had less effect on all aspects of cognitive and functional ability and, accordingly, on music engagement.

Music engagement was also observed in cases where musical semantic memory and music perceptual ability were both preserved, despite more advanced cognitive impairment (solution terms 2 and 3). This configuration of conditions was observed whether or not music education was included in the solution term. These cases seem to lend support to the notion that preserved music cognitive abilities are a supportive factor in allowing individuals with dementia to engage with music in daily life.

However, high levels of music engagement were also observed in cases of preserved musical semantic memory (solution terms 4 & 5) and impaired music perceptual ability. These cases illustrate that not all aspects of music cognition must be fully intact in order for an individual to engage with music in daily life. The contribution of musical semantic memory stands out in these cases, where despite a deterioration in the ability to discriminate features of novel melodies, the person is able to rely on a store of long-familiar melodies to scaffold their engagement with real-life musical stimuli.

The converse pattern was observed in solution terms 6 and 7, which were characterized by impaired musical semantic memory and impaired music perceptual abilities. Although the melodies presented in our melody recognition task were shown to be highly familiar in the study population, it cannot be ruled out that the cases described by these solution terms had more idiosyncratic musical experiences earlier in life and, as a result, were less familiar with the melodies. However, there remains the additional possibility that, for some listeners, music could be an engaging stimulus even if their musical memory and perceptual abilities are somewhat impaired. Indeed, most listeners are able—at least at some points in their lives—to engage in positive ways with novel music. These data raise the possibility that this ability could persist for some people with dementia.

We found only two solution terms where music engagement was very low. However, the case data did evidence situations where either musical semantic memory (solution term 8) or music perceptual ability (solution term 9) was preserved even though music engagement was low. In both of these solution terms, cognitive impairment was relatively mild. These case observations could reflect the reality that not all individuals engage with music. Low music engagement could be explained by a host of unobserved factors ranging from musical anhedonia—observed in some case studies of people with dementia (Fletcher et al., 2015)—to individual preferences and lifestyle. The cognitive mechanisms underlying low music engagement could be elucidated in future research by assessing musical anhedonia alongside low music engagement and by collecting more detailed information on participants’ premorbid musical practices.

However, given musical semantic memory's prominence in the solution terms, the present case study suggest that it is one of the abilities that allows the person to relate in a meaningful way to musical stimuli. Furthermore, the influence of musical semantic memory can operate differently in different cases, depending on the presence or absence of the other conditions we examined. Stated from the perspective of the Comprehensive Process Model, our case data suggest that preserved musical semantic memory is a person attribute that, alongside other person attributes, plays an important role in facilitating engagement with musical stimuli. Further, it suggests that familiarity with the musical stimulus is a stimulus attribute that allows the person to engage with music.

By allowing researchers to consider their data from the standpoint of both cases and conditions, fsQCA is well-situated to building theory or hypotheses about complex phenomena. The present study was exploratory in nature, and a large sample size would not have been feasible. Given the data available, a purely correlation-based analysis would have left more questions than answers. Since we hypothesized that a variety of different combinations of conditions could lead to the outcome of being highly engaged with music (i.e., the equifinality of music engagement), many associations of interest could be washed out as error in a regression analysis, even with a larger sample. Instead, by examining patterns of overlap between sets of cases, we are able to generate empirically grounded hypotheses that are informed by a theoretical model of music engagement.

Single-case studies (Medina & Fischer-Baum, 2017) and case series analysis (Schwartz & Dell, 2010) are important to the methodology of cognitive neuropsychology. Case comparative approaches in neuropsychology have tended to rely on sharp contrasts, or dissociation, in patterns of impairment between cases, from which inferences are drawn regarding the functional architecture of normal cognition (Coltheart, 2017). However, there are challenges to making systematic comparisons across larger sets of cases. In the present study, fsQCA offered a useful approach to identifying commonalities and differences across cases. Furthermore, by drawing on the flexibility of fuzzy set calibration, we were able to map preserved or impaired cognitive functions (musical semantic memory and music perceptual abilities) against a complex, higher-order construct (music engagement). As a result, this study allows the well-established literature on the functional organization of music cognition (Peretz & Coltheart, 2003) to inform questions concerning the real-life musical functioning of people with AD.

Our case data lend support to the assertion that musical semantic memory plays a role in facilitating music engagement for people with AD. The fsQCA findings point towards two complementary avenues for further research: first, bringing more nuance to the theoretical framework of music engagement through additional small- or medium-n mixed methods studies, and second, establishing the generalizability of the framework in larger-n quantitative studies.

For example, in the former approach, qualitative interviews could elicit rich data on participants’ subjective musical experience and the lifespan trajectory of their musical development, thereby providing insights into environment attributes contributing to engagement, while their profiles of music cognitive abilities could be assessed through a battery of laboratory tasks. Cross-referencing case membership in solution terms against thematic analysis of interview data would permit a more detailed account of the case configurations associated with music engagement. Findings from such a study could, in turn, allow for theoretically motivated larger-n quantitative research, in which regression analysis could be used alongside fsQCA to establish the generalizability of findings while ensuring that the equifinality of music engagement is not dismissed as error.

More generally then, fsQCA is particularly well suited in the early stages of the research workflow. The systematic approach to case study series analysis can provide concrete findings from smaller samples. These findings can then be used to specify hypotheses which can be tested in larger samples later. That is not to say that fsQCA is unsuited for the analyses of larger-n samples, and as we have discussed above, fsQCA offers complementary insights to regression-based approaches for research questions where the equifinality of the outcome is suspected.

While the present study focused on a question related to a basic mechanism of music engagement, fsQCA could also lead to valuable insights in applied music therapy research. The move towards Evidence-Based Practice has raised some concerns about whether paradigms grounded in randomized controlled trials are suitable for addressing some of the core concerns of the discipline (Aigen, 2015). More generally, outcomes research in psychological therapy faces the challenge of understanding individual differences in what factors contribute to effective interventions—“what works for whom?” (Norcross & Wampold, 2011). This question implies equifinality in therapeutic outcomes; fsQCA could help therapy researchers to embrace the complexity of individual differences in treatment response and develop interventions more precisely targeted to clinical needs.

We set out to explore how previous findings of preserved musical semantic memory might help us understand music engagement in people with dementia. However, music engagement is a complex, naturalistic outcome and likely underpinned by mechanisms that vary across individuals; we were faced with the challenge of addressing equifinality in a small exploratory study. FsQCA allowed us to draw on the benefits of a case series design to capture complexity within individual cases, while also taking a systematic approach to making comparisons across cases. The resulting analysis provides support for the claim that musical semantic memory should remain a consideration for future research on music engagement. This novel implementation of fsQCA demonstrates its potential as a methodological tool for other researchers seeking to understand the cognitive architecture of musical experience.