Introduction

The field of family assessment measures is very active. As early as 1996, more than 1000 instruments were available in this area [1]. This plethora can be explained by the large number of facets of family functioning and by the fact that family issues interest many disciplines, like psychology, sociology, medicine or educational science. This plethora also points to the fact that we are at the moment far from having a consensual unified theory of the family [2]. Some authors acknowledge positive aspects of this situation—“Growing diversity may be viewed as a sign of health”, but others underline the potential drawbacks “[it] slows the advancement of family psychology as an intervention science” [3].

Even when considering the particular situation of families with an adolescent, the theoretical positions are, indeed, numerous. Gavazzi [4] points to five of them among the most well-known:

1. Family development theory (which focuses on the typical changes experienced by families as they move through their life course, such as marriage, arrival of young children, the moment when the children leave home, and the empty nest);
2. Family system theory (which suggests that families are systems of interconnected subjects, none of whom can be understood in isolation from the whole);
3. Ecological theory (families are an ecosystem, families interact with their environment to form a bigger ecosystem, etc.);
4. Attachment theory (children look for a caregiver when they are distressed, in the hope that they will receive protection and emotional support);
5. Social learning theory (which suggests that children adopt behaviors by the observation of their environment, in particular their family).

In addition to this range of theoretical standpoints, there is also an impressive diversity in methodological, epistemological and even ontological positions implicitly adopted in the research work carried out in the field. Research can use qualitative or quantitative methods, post-positivist or constructivist ontologies. The researcher can claim a neutral position regarding the families studied or, on the contrary, conduct his/her analyses through the lens of his/her own subjectivity.

Ultimately a critical question inevitably arises: in view of this heterogeneity, is communication possible across the field of family studies? This question became a real practical issue in 2008 when the Université Paris-Descartes called for interdisciplinary work on the subject of adolescence. Three research units decided to respond together to the proposal: one team of sociologists (CERLIS: “Research on social bonds”), one team of psychoanalysts (EA4046: “Clinical psychology, psychopathology and psychoanalysis”) and one team of child and adolescent psychiatrists (INSERM U669: “Mental health and public health”). Because of a common interest in family studies, the question of “intra-familial interactions as perceived by adolescents” was chosen as a background for the upcoming research. After a few meetings, it turned out however that it was really difficult to design a common project because of serious methodological and epistemological differences. It was then decided to deal specifically with these differences. Our research question was thus: “beyond our specificities concerning concepts, words, methods and traditions relating to what we call ‘families’, is there a common core, a common perspective that can emerge to address intra-familial interactions as perceived by adolescents?”

Because some researchers in our group favoured qualitative approaches and others quantitative approaches, a mixed method design was adopted:

1. First, we decided on the data needed to address our research question, and it was agreed a priori that the data would be of a qualitative nature.
2. Second, on the basis of three explicit theoretical frameworks and of data available from a pilot study, each team (sociologists, psychoanalysts, child and adolescent psychiatrists) conceived a series of dimensions that could be assessed from a textual corpus collected from a sample of adolescents.
3. Third, multidimensional statistical analyses were used to test whether or not these dimensions had something in common. Some (qualitative) content analyses were also performed, but they are intended to be presented in another paper.

Let us now see this methodology in greater detail.

Method

Results

Description of the study population

Among results in Table 1, we find a mean age of 14,7 years (sd. 2) and a proportion of 51% girls. Seventy-one percent of the adolescents were living with their two parents. There was an average number of 4.3 persons living in the same house and the average size of the sibship was 2 (s.d. 1.3). For about 27 percent of the adolescents in the sample, French and at least one other language were spoken at home.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Description of the study population.

https://doi.org/10.1371/journal.pone.0132153.t001

Reliability of item scoring

The percentages of “0” ratings (which indicate a floor effect), the percentages of “10” ratings (ceiling effect) and the percentage of responses for which a given item was not assessable are presented in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Item analysis.

https://doi.org/10.1371/journal.pone.0132153.t002

Concerning the sociological items, responses relating to belonging and to the affective atmosphere provided a good amount of data with no floor or ceiling effect. Responses relating to freedom and recognition of personal merit were not very informative (small variance). Responses relating to the family forming a whole, to conflict and to injustice showed a floor effect (which was fairly moderate for the "forming a whole" effect). Concerning the psychoanalytic items, responses were fairly complete with neither floor nor ceiling effect. For the psychiatric items, there was a slight floor effect for the items concerning negative relationship with parents, negative relationship with siblings, and for the clinical assessment. The large number of missing responses should also be noted for the two items relating to siblings.

The Kappa coefficient was used to assess inter-rater reliability. According to Fleiss [18], kappas over .75 are considered as “excellent”, from .40 to .75 as “fair to good”, and below .40 as “poor”. In Table 2, all kappas related to sociological items are “fair to good” or “excellent”. Concerning the psychoanalytical and psychiatric items, except for the item about intergenerational boundaries (for which the inter-rater agreement was “poor”), all inter-rater agreements on the scores themselves are “fair to good” or “excellent”. Some inter-rater agreements about missing data for an item are only “poor”. For instance, concerning the item “negative relationships with siblings”, if no mention of this point was available in a text, rater 1 considered that it was not possible to rate the item while rater 2 considered that the rating was “0” (data not shown).

Unidimensionality of the set of items

The screeplot (Fig 1) is in favor of unidimensionality (the first eigenvalue is clearly above the others, it is the only value obviously above what could be expected by chance, and it represents 50.4% of the total variance). A solution with 2 factors can however also be envisaged because the second eigenvalue, even if it is negligible compared to the first, appears to be above that could be expected by chance in the the simulated screeplots. Finally, if one follows the classic but overly lax “Keiser rule” (eigenvalues above 1), a solution with 4 factors is of potential interest.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Screeplot.

Screeplot of the eigen values for the item correlation matrix, parallel analysis. In addition to the scree plot of the study data, 20 random data sets with the same number of subjects and items were generated and the corresponding scree plot represented. This can be used to determine eigen values above “what could be expected by chance”.

https://doi.org/10.1371/journal.pone.0132153.g001

Discussion

Numerous family assessment measures already exist and our intention was not to develop and promote a new one. Instead, this paper aimed to look with a statistical perspective and across theoretical frameworks to the level of diversity in the manners intra-familial interactions are perceived by adolescents. More precisely, when disciplines as different as sociology, psychoanalysis and child and adolescent psychiatry are asked to characterize intra-familial interactions, do they catch authentic different facets or do they apprehend mostly the same dimension? The empirical study designed to answer this question used a mixed method design: an open question asked of adolescents and a rating of the answers collected on the basis of 18 dimensions related to 3 different theoretical frameworks.

The main result of this study is the existence of a strong underlying factor that explains a large amount of variance in all the 18 ratings coming from the 3 disciplines. This factor is bipolar, it reflects the level of positivity/negativity of the adolescent’s point of view concerning his/her intra-familial relationships. This result can however be refined. Beyond this single common factor two factors can be considered. One is related to positive feelings and the other to negative feelings. Of course, these two factors are highly, negatively correlated (r = -0.65). But there is here something that is well known by clinicians [15] and psychometricians [19]: negative feelings are not merely the opposite of positive feelings. Ambivalence is common, and this appears to be true in particular concerning intra-familial relationships as considered by adolescents. Two other factors can also be estimated, they are however at the very limits of statistical relevance. One concerns individuation and the other relationships with siblings.

Of course, these results need to be moderated because of obvious limitations in the study design. By construction, we are restricted here to intra-familial relationships as viewed by adolescents. These relationships are described in a short written text which can be sometimes insufficiently informative (as shown by a non-negligible percentage of missing answers to some items). In line with this methodological limitation, the theoretical frameworks proposed by the 3 teams of researchers had to be compatible with the elements available in the collected corpus. This perhaps led to a restriction in the level of complexity of the concepts that were tackled, which possibly induced an artificially simple factor structure for the dimensions assessed. Moreover, the statistical approach used here (i.e. unidimensionality conceived according to a psychometric standpoint) is not indifferent. The property of local independence which underpins the notion of psychometric unidimensionality does not equate with semantic equivalence. Local independence of a set of items corresponds to a situation where the responses to these items are independent conditional to a measured or unmeasured variable. The items have thus “only one thing in common” and the importance of this common core can be captured by the amount of variance explained by the common factor. Semantic equivalence is a notion that is much more subtle and delicate to define and to assess. Finally, the results have only a limited level of generalizability because of the sampling process, restricted to a given country at a particular moment. The sample is stratified according different categories of age and localization (rural versus urban), but obviously there is no ambition of representativeness there.

These weaknesses are however counterbalanced by authentic strengths. First, the size of the sample studied (close to 200) is substantial and compatible with the statistical routines that have been used [20]. In addition, all ratings were performed blind by two independent raters from each discipline. It can be noted that inter-rater reliability is globally good according to usual standards [18]. This was a real challenge, because the content of many items is subtle and because most raters were not used to quantitative evaluations. This is indeed encouraging as regards the feasibility of multi/inter-disciplinary research grouping together specialists from distinct epistemologies and using very different methodologies. One strength of the present study is the nature of data collected. A real bottom-up process, which stems from the qualitative research paradigm, guided the study. The dimensions were designed from the analysis of texts written by adolescents during a pilot study and the scores allocated for the pivotal study are derived from similar free texts. The ratings are thus obtained from the patient perspective itself and not from guided answers to oriented questions. The main strength of the paper comes surely from its clear-cut main result: a mostly dominant first principal component explains more than 50% of variance, while the second component explains less than 10% and is at the limit of “what could be expected by chance”. The structure of variance of the 18 items is thus compatible with the existence of a unidimensional common core. And this was not expected at the beginning of the study when we defined the three theoretical frameworks. One is based on recognition (being recognized by the other members of the family), one on the psychodynamic constitution of subjectivity and the last on the phenomenological and semiological perception of tension or distress. This means we have here three different ways of considering and gauging intra-familial relationships and, curiously, only one main common core ultimately remains.

Now, on a more conceptual viewpoint, it can be shocking to find only one dimension behind these three different and sophisticated theoretical frameworks. Indeed, how is it possible that social and cultural context (for the sociologists), inner life and development of psychic structures (for the analysts), and clinical symptoms and syndromes (for the psychiatrists) can be summarized without any substantial loss in a unique dimension (positive versus negative feelings about intra-familial relationships)? Does this mean that these theories are somewhat spurious, that they introduce elements of highbrow complexity where reality is in fact straightforward? Indeed this paper cannot have the pretention to deal with this issue. Our search for a common core is based on psychometric paradigm and we have to remember that statistical unidimensionality is not synonymous with semantic equivalence. Certain variables may account for only a small amount of the variance of a phenomenon while they are nevertheless of crucial importance to explain it. It would be in particular the case for a necessary condition that is ubiquitous: it explains 0% of variance of the phenomenon under study, but it is however indispensable to understand it. More important, this study relies on a four-stage reduction process that necessarily weakened the scope of its conclusions: 1/ by construction, the corpus analyzed is limited; 2/ the sociological, psychoanalytical and psychiatric perspectives are considered through a narrow and limited angle; 3/ they are reduced then to a list of 18 quantitative items; 4/ the covariance matrix of the responses to these items is reduced with the use of factor analysis. But reduction is inherent in all quantitative works. No doubt that some qualitative content analyses of the present corpus would cast a different light on the question of the existence of a common core. Obviously, this should be done and presented in another paper.

A final question has thus to be raised at this point. Considering the limitations presented above, are there any practical implications to the present study? Yes, and not only because this work proves that the complexity and diversity of family functioning is not necessarily a curse. There is some room for working together and for common constructs. Indeed, the main implication of this paper concerns evaluation studies. When performing a randomized controlled trial or a cohort study to evaluate a family therapy or a family intervention, there is a need for a primary endpoint, and preferably for a single main primary endpoint to preserve the type one error of the statistical test of hypothesis that will determine the conclusion of the trial [21]. If the phenomenon under study is highly multidimensional, it is difficult to propose a main primary endpoint that will satisfy all viewpoints and the conclusion of the trial will thus be open to criticism. This happens frequently in medicine, for instance in the evaluation of cancer treatments where a small improvement in survival can be criticized because it is obtained at the cost of a deterioration in quality of life. We have proved here that when considering the adolescent’s point of view concerning his/her intra-familial relationships, there is, formally, one single way to quantify the positivity/negativity of this viewpoint, whatever the theoretical framework used to structure it. This is a good news for all those who work for the evaluation and the promotion of family interventions.

Acknowledgments

The authors warmly thank the students and the teachers for their essential participation in the validation study. They also thank Angela Verdier for her edition of the English wording of the paper, and a referee for pointing an important issue.