The study was approved by the Scientific Council of the European Center for Higher Education in Osteopathy. Respondents were recruited from first year students of an osteopathy school in France, on a voluntary basis. We paid attention to gender selection bias, respecting the same proportion of women and men among the respondents as in the promotion (Wang 2013)14, (Vlassof 2007)15. During the first week of the course, we conducted a semi-directed questionnaire survey. The first question asked was very open-ended in order to guide the students as little as possible and to be able to glimpse at whether or not his or her conception spontaneously followed from the three domains of the biopsychosocial model: “What does health mean to you?”. At the end of this first question, the students were asked whether they thought that “factors influence health” and then, according to their answers, they were asked to specify how each of the factors they had mentioned influenced health.
The contents of the verbatims were analyzed according to their blocks of meaning (Bardin 2001)16. The meaning blocks were indexed in order to create different categories. A referencing chronology was applied during indexing. The first referencing rank refers to the determinants of health in the Dahlgren and Whitehead model (Dahlgren and Whitehead 2002)17. The second rank refers to the domains of the biopsychosocial model. The third and final rank refers to the factors influencing health, mentioned by students. Some verbatims were indexed in several categories. When some content in the verbatims was not explicit enough to be indexed in one category, it was taken out of the stream. An Excel® file was used to create a database of all the meaning blocks.
Our proposal to model students’ conception of health is based, in its first stage, on a proven methodology (Pizon 2019)10, (Cardot 2010)13 that takes the three domains of the biopsychosocial model on “an equal footing, in a system of complex, multiple and circular causalities” (Pizon 2019)10. As these studies have shown, each of these domains can be recruited by individuals to develop their conception of health. Depending on the circumstances, this recruitment may be cumulative, resulting in interactions between two or even all three domains. These interactions could be schematized in the form of three interlocking circles, materializing seven sectors, which support seven potential categories making up the conception. (Fig. 1)
While this modeling by interconnected circles indirectly illustrates the existence of links between the constituent categories of the conception, it does not allow the strength of the links between the categories or between the indicators of these categories to emerge. To bring out these conception systems, we propose using network analysis and visualization software. These types of software allow us to visualize the global connectivity of content. Their algorithm is based on graph theory, which produces abstract models of network designs linking objects.
From a general point of view, a graph is a set of pairs, triangles, quadrilaterals, etc., made up of vertices (or nodes) and edges (or lines) connecting these vertices. The edges can indicate the direction of the relationship between the vertices. In this case, they are represented by an arrow. The graph is then said to be oriented or directed. We will only describe the so-called multipolar graph shape, which corresponds to the one taken by the network of our data. Multipolar graphs are characterized by two types of edges; those that form the links emanating from the pole to distinctive nodes and those that form the links emanating from one pole to another pole. Poles can take a structured architecture. We have chosen the centralized structure with respect to the three domains of the biopsychosocial model. Consequently, in the graphs presented below, the more frequently one of these domains will have been cited by students, the more central it will be in the graph. A graph is labelled. Each vertex or edge belongs to a set, and therefore bears a label. A label can belong to any set: color set, word set, object set, etc. The label can be used to identify the set. Our graph will, on the one hand, have colored labels to differentiate the three domains of health (biological, psychological and social), and on the other hand, labels indicating the nature of the nodes. In summary, the typology of our graphs is multipolar, oriented, with double labeling, in order to show the structuring poles of the conceptions, the links between categories and their meaning. (Fig. 2)
We chose the Gephi 9.2 network visualization software because of its free access. The Excel-format database has been changed into “csv” format, so that the three categories constituting the conceptions do not appear in the same column, with semicolons as separators, since the verbatims are written in French (CSV separator: semicolon). We have selected the recommended modeling for small networks called “force atlas” (ESIPE 2013)18. The “actors” of the network are the matrices items we defined in the previous paragraph. The force atlas modeling algorithm is based on the principle of the forces of attraction or repulsion of the actors composing the network. The actors, represented by nodes in the graphs, are considered attracted to each other when they are associated in a verbatim sequence. The nodes that represent them are then all the closer to each other, the closer the associations are in the verbatim. Conversely, when actors are not associated in verbatim sequences or are distant from each other in verbatim sequences, they are considered to be repulsed. The size of the nodes corresponds to the number of citations in the verbatims. The thickness of the links corresponds to the quantity of associations between items.
So as to not overload the graph and in order to make it more readable, when a triple association matrix [Determinant - Domain - Factor] was first created in the database, the matrices of the same nature found later in the verbatims were replaced by the only one [Factor] that was placed on the same line of the spreadsheet. This in no way changes the links between determinants and domains and between domains and factors, since the entry by the couple [Determinant - Domain] is located in the first two cells of the spreadsheet row. This keeps the determinants off-centered on the graph, centralizes the domains and illustrates the frequency of the links between the domains and the factors. If we had only populated our database with triple association matrices, all determinants and domains would have been centralized. The links between determinants and domains would have overlapped in number and thickness with the links between determinants and factors, making the graph incomprehensible.
We did not carry out frequency measurements or comparisons of the different categories because our sample was only partially representative of the entire promotion.