FormalPara Overview

  • Networks consist of actors who are connected by relationships and whose connections are made up of different social structures.

  • It is assumed that social networks have an effect on the actors and that actors influence the networks, in turn.

  • There is a distinction between whole networks and ego-centered networks. In the whole network analysis, the respective actors and their relations are considered within predefined limits. In the case of egocentric networks, the interpersonal networking of a particular actor is at the center of the analysis.

  • Methodologically, a distinction can be made between qualitative—often consisting of visual access—and quantitative network research. So far, the focus in health research has been more on quantitative approaches.

  • Tested methods of network analysis in health research do not generally exist. Research must therefore always be adapted to the research issue.

  • The time required to collect network data can be very high, so network surveys should always be tested in pre-tests.

1 What Is a Network?

“Networks” seem to be omnipresent in modern societies (e.g., networking, online social networks such as Facebook and Twitter, or even criminal and terrorist networks), but the term and its meaning in everyday life often remain amorphous. In order to be able to work with the concept in a scientifically adequate way, this chapter introduces the term “social network,” different forms of network analysis, and survey and data evaluation strategies. What follows is a short overview of different methods and important literature references are given, which can be referred to in more detail, if necessary.

The axiom of network research assumes that elements—so-called nodes—can enter into relationships—so-called edges—with other elements. The smallest unit of such social relations is the dyad, the relation between two actors. Dyads, in turn, do not exist solitarily, but join together to form larger networks, where they also form certain structures. However, a uniform definition of (social) networks does not exist. How a network is defined also depends on the object under investigation.

A network can generally be understood as “[...] a set of relevant nodes connected by one or more relations” (Martin & Wellman, 2011, p. 11). This definition can be applied to social as well as non-social, technical, or physical-material elements such as road or electricity networks. Besides this formal definition of networks, there are definitions that focus more on social action and the mutual influence of networks and action. Clyde Mitchell, for example, defines social networks as “[...] as a specific set of linkages among a defined set of persons, with the additional property that the characteristics of these linkages as a whole may be used to interpret the social of the persons involved” (Mitchell, 1969, p. 2).

Networks differ from the sociological concept of groups in that their existence is determined by the drawing of boundaries, not by fundamentally open relations: “(A) fundamental part of the concept of a group is the existence of boundaries” (Borgatti & Halgin, 2011, p. 1169). The concept thus distinguishes between ingroup and outgroup. In some cases, however, groups are also referred to as networks, although social relationships within the group are not explicitly investigated at the dyadic level, but rather presumed. Groups can thus also be understood as a subcategory of particularly dense networks. “(U)nlike networks, [groups] depend upon the merging of social relations within a shared space and with a recognizable culture. Although groups are distinguished from networks through their boundaries, pasts, and identifications, groups are in some regards dense networks” (Fine, 2012, p. 168). In contrast to classical social science methods, network analysis includes not only personal attributes (e.g., gender, age, income) but also relational attributes (e.g., positions of actors in networks). It is thus assumed that the structure of social networks (e.g., support networks) and social outcomes (e.g., health behavior, health) are interdependent and influence each other.

Networks thus consist of so-called nodes (e.g., individuals or collective actors) and relationships, the so-called edges (e.g., kissing, passing on viruses, social pressure), by which the nodes are connected.Footnote 1 The aim of network research is to make causal statements about the effect of relationships on the actors (or vice versa) or to be able to describe the actors and their relationships.

2 Ideal Types of Network Research

Although a network can generally be described as a set of nodes and edges, there are significant differences with regard to the empirical procedure, both in the data collection and in the data evaluation. Ideally, network research can be differentiated along two dimensions (Gamper & Schönhuth, 2020). Along a structural dimension, whole networks and egocentric networks can be distinguished, while along a methodological dimension, quantitative and qualitative procedures of network research can be distinguished. In empirical practice, it is of course possible to deviate from these ideal types. For example, there are research projects that use both qualitative and quantitative methods simultaneously and connect data by triangulation (Dominguez & Hollstein, 2014).

2.1 Whole Networks and Ego-Centered Networks

The whole network analysis focusses on nodes and their edges within predefined borders. The emphasis is on the internal networking of the actors in this predefined area (e.g., sex partners in a school, transmission of diseases in a village, influence of smoking behavior in an association). In the ideal-typical case, the relations outside these defined limits are not included in the analysis. Thus, the research focus is on a certain number of actors and their very specific relationships. The demarcation of the boundaries should be well justified and described, since every distinction has an impact on the data and results. Boundaries can be determined, for example, on the basis of certain theories or even empirical knowledge. In research, however, there are also pragmatic demarcations that are due to the field of research (Laumann et al., 1983). Usually actors (e.g., pupils) are asked about their connections to other persons (e.g., classmates) in a predefined area (e.g., school class). In addition to predefined lists of names, with the help of which the relevant contact persons only have to be selected, the interviewees can, in some cases, determine the names of the contact partners themselves. However, these contact persons must be part of the predefined set (e.g., school class). In addition to the relationship parameters (e.g., friendship relationships, love relationships), the respondents are asked further questions about themselves (e.g., age, health status, body mass index). Building on this, all relationships and attributes are transferred into a whole network. In other cases, for example, on the Internet, data on relationships (e.g., Twitter, Facebook) are already available in digital form. A rarely used method of data collection is participatory/non-participatory observation (Desmond, 2014). Here, relationships between actors are registered and recorded on the basis of observations, such as the passing of cigarettes to the schoolyard. In many studies, these results are presented or depicted visually.

A prominent example of a whole network analysis from the field of health research is the investigation of romantic and sexual networks in Jefferson High School located in a small town in the USA (Bearman et al., 2004). The study focuses on the risk of infection with sexually transmitted diseases in adolescents and the possibility of prevention. For this purpose, the whole network of about 800 students at Jefferson High School was surveyed. The nodes in this case are the pupils of the school. The edges depict the romantic and sexual relationships between them over the last 18 months. Here, 573 students stated that they had entered into one or more such relationships.

Different ways of establishing relationships with others result in different forms of networks. These range from simple dyads to triads to a large network component with many actors interwoven in different ways (see Fig. 1). By comparing them with randomly generated networks, the researchers were able to determine that the observed structures differ radically from the randomly generated networks. Specifically, we find that real sexual and romantic networks are characterized by much longer contact chains and far fewer cycles (Bearman et al., 2004, p. 44). As a result, many people (here just under 50%) are indirectly connected to each other and thus cannot keep track of the number of sexual relationships in their entirety. An indirect chain of relationships results, for example, when a sick “pupil A” had a relationship with “pupil B” and the latter then enters into another relationship with “pupil C.” If C does not learn about the relationship between A and B, C has no idea that B could transmit the diseases of A to C. Through this kind of networking, a disease can be transmitted quickly and infect a large number of students. In order to avoid infection, it is therefore important to “break up” the large cluster so that the virus can be stopped in its spread. This requires changing the behavior of some students (e.g., by using contraceptives), as the cluster will then break up into individual chains and the infection will be reduced.

Fig. 1
figure 1

Sexual and romantic relationships of female students* within Jefferson High School. Source: Bearman et al. (2004, p. 58)

Full size image

As the example illustrates, the network boundary is the “school grounds” of Jefferson High School. Therefore, “only” the romantic and sexual relationships of the students of this school are analyzed. Sexual and romantic relationships with people outside the school, such as pupils in another school, are not considered here. In addition, other types of relationships (e.g., friendships), beyond sexual and romantic relationships, are not included in the analysis.

The egocentric network research is subject to a slightly different logic. Here, the interpersonal networking of a specific actor, the ego, is the focus of attention. From the point of view of the respondent (= ego), certain persons and their relationships to one another are questioned (Burt, 1980; McCallister & Fischer, 1978; Wellman, 1979). The ego-centered network consists of relationships of the respondent actor (ego) to other actors in their network, the so-called alters, with whom they are directly linked. In some studies, ego is also asked about relations between the alters.

First, ego is interviewed about their subjective view on their relationships and has to name persons with whom they have certain relationships (e.g., smoking together, sexual relationship, exchange of syringes), usually predefined by the researcher. These questions are also called actor generators (these include, for example, name generators, resource generators, position generators), since these generate network actors. The best known are the name generators, which can be divided into interaction-approach (e.g., with whom have you interacted […]), role-relation-approach (e.g., three best friends), affective-approach (e.g., actors you feel close to), and exchange-approach (e.g., who helped you) (Bidart & Charbonneau, 2011; Marin & Hampton, 2007).Footnote 2 There is no predefined list of names, as in the case of the whole network analysis. The researcher does not know the names of the contact persons in advance and there is no clear border in the whole of network research (e.g., Suitor & Pillemer, 1993). Therefore, the researcher has to decide how many alters should be collected and how the border of the ego-network is defined (McCarty et al., 2007). This is an important process, because it does affect the duration of the survey, the effort of the interviewee to answer the survey questions, and the time to assess the alter-alter relation. An overview of possibilities is discussed in the article by Perry and Roth (2021).

Based on this concept, ego is asked to provide further information about the named alters and the relations between ego and alters (so-called actor interpreters). At the end, information about ego will be asked. This could be, for example, socio-demographic information, smoking behavior, or health status. Many studies also ask ego questions about the relationships between the alters, for example, to what extent the alters are in contact with each other. This is not absolutely necessary if certain statistical measures or questions are not considered essential for one’s own question (McCarty et al., 2016). In contrast to the whole network analysis, where the contact persons are specified by some kind of border, the interviewee is free to name them. In addition, the information about the alters (e.g., gender, health status) comes from ego and not from the alters themselves.

An example is the longitudinal study by Perry and Pescosolido (2015). The researchers asked about 171 persons (egos), who the egos contacted in the case psychological illness. The research interest was focused on the activation of support services and the kind of networks that were used for the health issues of egos. The sample consisted of a group of patients with severe mental illness and a group with less severe disorders who were receiving psychological treatment for the first time. The following actor generator (here specifically a name generator) was used in the study: “I’m interested in who, among all of the people in your life, you talk to about health problems when they come up. Who are the people that you discuss your health with or you can really count on when you have physical or emotional problems?” (Perry & Pescosolido, 2015, p. 119). In contrast to the study by Bearman et al. (2004), the focus here is not on the connection of the actors within a certain boundary and between these persons, but on the effect of persons on the well-being of the egos against the background of their personal networks. In other words, they were trying to determine which networks can be helpful for ego to feel more comfortable. The aim was to make general statements. As the study shows, networks play an important role especially against the background of emotional support and information: “Social networks have the potential to serve as conduits of general emotional support and information. However, according to our findings, it is not these general support processes that drive recovery outcomes. Rather, the key factor appears to be activation of particular kinds of people for health discussion. This indicates that achieving a state of recovery may be facilitated by cultivating a social safety net that can provide targeted, health-related advice, affirmation, and instrumental aid that buoys the treatment process and permits gains in self-sufficiency and productivity” (Perry & Pescosolido, 2015, p. 126).

In quantitative ego-centered network research, visualizations are usually dispensed with, since here several individual networks (in this example, 171 individual networks) would have to be visualized and the added value could be considered rather low. But there are two are instances where it can be useful to visualize an ideal type of an ego-centered network. First, it can be helpful to convey a theoretical concept with the help of visualization. For instance, the egocentric network studies conducted by Bott (1955), Cornwell (2012), and Perry and Pescosolido (2012) are just some examples that depict ideal-type network visuals to emphasize their theoretical concepts. However, visualization can also be useful when several egocentric networks, that were collected and analyzed by the researcher, can be reduced to a few ideal types of networks and then presented visually. For example, Wellman (1988) formed an ideal type of an ego-centered network based on his network-collected data and several egocentric networks. The situation is different in visual or qualitative egocentric network research, which will be discussed later (see Sect. 2.2).

Therefore, whole and ego-centered network analyses differ. Although it is possible to isolate individual ego-centered networks from whole networks, these are always subnetworks from a predefined area defined by the researcher. Conversely, the transformation from ego networks to whole networks is very difficult or even impossible. In the research process, researchers should therefore choose one of the two methods. This decision is essential, since both methods differ in terms of their respective data collection and, in some cases, in data analysis. This will be discussed in more detail later (see Sect. 2.2). The choice for one of the two procedures should be strongly oriented toward the research question and also take into account for access to the field.

If the research question is aimed at the internal networking of actors, such as the passing on of cigarettes by pupils in a school, the whole network analysis is the more appropriate tool. In the case of whole networks, the focus is on a group that can be easily isolated and its internal networking. If the focus is on the influence of friends on the drug use of homeless people, the egocentric network analysis would be more suitable. In this case, the “social border” is not clear, and not only can the internal cross-linking be interesting for the research, but also the relationships outside the group of homeless people. There are also differences in evaluation procedures. For example, not all statistical measurement methods are applicable to egocentric network analysis (see Sect. 2.2).

Thus, whenever the research interest is directed at the internal structure of a network and the connections between a predetermined number of actors are known or of interest, the whole network analysis is particularly well suited. Here, the researcher determines who belongs in the sample and who does not. The ego-centered network research is used when the relationships are not only to be analyzed between actors in a certain predefined space, but also when the interest goes beyond that. In this case, the focal ego is selected by a sampling procedure, but the persons (alters) of ego are not specified. The procedure is particularly suitable if one focuses on a specific group and wants to consider its general embedding in the social environment without having defined it beforehand.

2.2 Quantitative and Qualitative Network Analysis

In addition to the distinction between ego-centered and whole network analyses, a differentiation can also be made on the continuum between open/qualitative and standard/quantitative research. While social network analysis of the last 40 years was predominantly standardized and quantitative, less standardized research approaches for social network research (Freeman, 2004; Gamper, 2015) are now (again) more frequently discussed, and concepts of network analysis as method combinations of open and standardized approaches are presented (Dominguez & Hollstein, 2014; Gamper et al., 2012).

In standardized network research, the focus of interest is on so-called statistical structural descriptions or causal relationships, which include distribution properties of features, the testing of hypotheses and explanatory models, the discovery of correlations, and the development of alternative hypotheses and explanatory patterns. In contrast to classical research, in which attributes (e.g., age, gender) and their interrelationships are examined, here relationship aspects are also included in the analysis or are even at the center of the research. Using structured and standardized data, structural measures such as network size, centrality, heterogeneity, and density are calculated (Wasserman & Faust, 1994). For better understanding, below I will briefly discuss some of the measures. A mathematical derivation is not given in this introduction. There are introductory books by Wasserman and Faust (1994), Knoke and Song (2019), or Newman (2018) that give a very good theoretical and statistical overview.

First of all, a distinction can be made between network parameters, that is, aspects that cover the entire network, and measures that affect the actors of a network, so-called actor parameters. For network-related measures, for example, the network size, density, or clique calculations can be given.Footnote 3 The network size is probably the simplest measure. Here, the actors in a network are summed up. Density is the degree of connectivity of the network, which results from the connections of the individual actors with each other. The maximum densityFootnote 4 is reached at a value of 1, that is, when everyone is connected to everyone else in the network (Seidman, 1983; Wasserman & Faust, 1994). The value 0 is the minimum value and means that no relationships exist in a network. Elisabeth Bott (1953) distinguished between “tightly-knitted” and “loosely-knitted” networks, which refer to the networking of the network members between each other. In a tight network, many actors are interwoven with each other. It is assumed here that a high density can, for example, lead to strong control or that diseases (e.g., caused by viruses) can spread faster. The above-mentioned transmission of sexually transmitted infectious diseases in the Sexual and Romantic Network of Jefferson High School can be cited as an example (Bearman et al., 2004).

In addition to measures that relate to the entire network, there are also measures that relate to individual actors. So-called centrality or centralization measuresFootnote 5 examine the question of relevance of actors within a network. However, no agreement has yet been reached as to how centrality is to be conceptually understood and measured: “There is certainly no unanimity on exactly what centrality is or on its conceptual foundations, and there is little agreement on the proper procedure for its measurement” (Freeman, 1978, p. 217). Consequently, there are different forms and types of calculation of centrality (a critical review can be found in Landherr et al., 2010). Some focus on aspects like control, power, and prestige, while others concentrate on the flow of information and still others on the accessibility of people within a network. The simplest form is degree centrality. In this case, the most central actor is the actor with the most relationships within the network. In the case of betweenness centrality, the most central actor is the one who is on the shortest route between two vertexes in the network. In the case of closeness centrality, the most central node is the one that has the shortest distances to all other nodes within a network. When calculating the eigenvector centrality, all actors are assigned a score on the basis of their respective interconnections in the network. The most central actor is the actor who has a lot of relationships with actors who also unite many relationships and are therefore very central (Wasserman & Faust, 1994).Footnote 6

In addition to the distinction between network parameters and actor parameters, it is also possible to differentiate (ideally) between structure-describing methods and inferential statistics models that examine causal relationships. The structure-describing methods detail the structure of the network or focus on a few parts of the network. The so-called density method has already been presented. In addition, there are measures such as clique analysis, cluster analysis, component analysis, block model analysis, or the triad census. Clique, cluster, and component methods attempt to filter out subgraphs from a network whose internal density is higher than the density of the entire network (Luce, 1950; Moody & White, 2003). In this case, there are different procedures such as n-core, n-clan, and n-clique procedures (Mokken, 1979). For example, n-clique is a maximum subgraph in which the path distance, that is, the number of actors by which all nodes in the network are connected to each other, is not greater than a predetermined “value n” (Bron & Kerbosch, 1973). Thus, groups can be filtered out and distinguished hierarchically according to this calculated distance. The component method is similar. Components are subgraphs, that is, parts of a network consisting of nodes and are interwoven with each other. A strongly connected component is a group of nodes in which all nodes are connected by directed edges (for example, all actors in one part of the network lend each other cigarettes). In addition, there are also weak connected components, where each node is connected by exactly one path. For undirected networks where the direction of a relationship is not given (e.g., “Do you meet person XY occasionally?”), no strong or weak connected component can be calculated. In this case, there are just connected components (De Nooy et al., 2011, p. 77). Another explorative method, which is based on a data-reducing representation of nodes and edges, is the so-called “blockmodel analysis,” where actors and relationships to groups of actors and bundles of relationships are clustered (White et al., 1976) and thus form a reduced image of the network structure. Through clustering, hierarchies, center-periphery groupings, or even cliques can be visually presented and analyzed. A distinction is made here between a posteriori blockmodels, in which actors are grouped based on similar positions in the network, and a priori blockmodels, in which actors are grouped based on characteristics (Wasserman & Anderson, 1987). The statistical procedure of the “triad census” goes back to Heider (1958) (see also chapter “Social Network Theories: An Overview”). It examines how often closed triads—three actors directly connected to each other—occur in a network. In a directed network, 16 different triad types (isomorph classes) can be differentiated depending on the direction and type of relationship (Holland & Leinhardt, 1970). The labeling of the triad types is based on the MAN scheme: Mutual Dyads (i.e., reciprocal relationship), Asymmetric Dyads (i.e., one-sided relationship), and Null Dyads (i.e., no relationship). From the significantly reduced or increased presence of certain triad types, it is possible to draw conclusions about specific microstructural mechanisms in social networks, for example, whether a network is rather hierarchical or flatly structured. Figure 2 shows a complete triad count showing all 16 triad configurations. Here, the MAN scheme is applied. Seven triangle configurations, in which all three nodes are connected by either asymmetric or mutual edges, are shown in black. The weighting factor (wu) for each of the seven triangle configurations is based on the probability that the triangle is transitive, assuming that each individual in a mutual dyad has the same probability of being dominant (a short introduction can be found in Faust, 2007).

Fig. 2
figure 2

Triad census and the MAN scheme (mutual, asymmetric, null). Source: Shizuka and McDonald (2012, p. 934)

Full size image

Two stochastic methods that are implemented include Exponential Random Graph Models (ERGMs) and stochastic actor-oriented models (SAOM). ERGMs are stochastic models of empirical networks (Robins et al., 2007). They are used to test structural relationships with a few local parameters. A multivariate model can be created in which parameters such as reciprocity, transitivity, homophily, and centrality are tested for significance. Dependent variables are the edges, while the independent variables can be attributes (e.g., age, gender) as well as relationships (e.g., strong or weak ties). The basis of ERGMs is a Markov chain Monte Carlo estimation process that generates a sequence of random networks containing stepwise small changes of different parameters (a short introduction can be found in Robins et al., 2007 and van der Pol, 2019).

SAOMs, often carried out through RSIENA, were designed for modeling the dynamics of longitudinal network data (Snijders et al., 2010). In this statistical procedure, influence or/and selection effects (see chapter “Social Network Theories: An Overview”) are investigated and tested, in other words the extent to which attributes have an effect on relations (selection), relations on relations (reciprocity), attributes on attributes (control variable), or relations on attributes (influence). The model is based on four assumptions: Actors have an influence on outgoing relationships, the change of relationships is done in so-called microsteps (actors have the possibility to dissolve, enter, or maintain a relationship), the change will be made in such a way that the change implies an increase in benefit for the actor (rational choice approach), and the benefit function includes a random component in addition to individual effects and their parameters (objective function) (a short nontechnical introduction can be found in Steglich et al., 2006).

It should be noted that not all measures for ego-centered and whole networks are applied in the same way. For example, the different measures of centrality are only applicable to a large number of actors within a network. Ego networks are often too small and do not have enough nodes within. Also, blockmodel analysis, in which clusters are formed against the background of the relationship structure, is not found in the ego-centered network analysis at all.

As an example of a quantitative network study, the Framingham Heart Study will be cited here. Starting in 1948, data were regularly collected in the city of Framingham in the USA to determine, for example, the causes and risks of heart disease and arteriosclerosis. Since 1983, network data have also been collected, which Fowler and Christakis (2008) have used to investigate the connection between “being happy” and being embedded in networks with the help of regression. In the process, 4739 persons were medically accompanied from 1983 to 2003. The results show that happy people are particularly at the center of the network, that is, they are central and form clusters (see Fig. 3): If you are surrounded by many happy people, you are very likely to be happy. More unhappy actors tend to be located on the periphery of the network. It is also shown that happy people in the network have a great influence on the feeling of happiness of ego and that this influence can spread over three edges (Fowler & Christakis, 2008).Footnote 7

Fig. 3
figure 3

The clustering of happy and less happy people in the city of Framingham. The lines between the nodes indicate the relationship (black for siblings and red for friends and spouses). The color of the nodes indicates the happiness of the ego, with blue shades meaning least happy, green a little happy, and yellow shades the happiest. Source: Fowler and Christakis (2008, p. 3)

Full size image

As a second example, a SIENA model is given here. Using longitudinal data, the smoking behavior of students in Finland was investigated. Mercken et al. (2010) investigated selection and influencing factors between pupils. They investigated to what extent smoking among friends is “socially contagious” (influence) or whether friendships develop due to interest in smoking (selection). They were also asked whether their own parents or siblings also smoke. It was found that both students chose their friends based on their smoking behavior. For girls, on the other hand, the smoking behavior in the “clique” also showed an influence factor. It was also evident that the smoking behavior of the parents had a significant influence. In contrast to the study by Fowler and Christakis (2008), no visualization was used here.

Qualitative network methods are rather underrepresented within network analysis in general and in health research in particular. In contrast to the quantitative approaches, the focus here is on understanding relationships or mechanisms and the subjective view of the actors’ networks. Mixed-method (Small, 2017) and qualitative SNA approaches have proved to be fruitful, especially by bridging personal and structural dimensions (Bernardi, 2011), exploring the contextualized nature of social relations (Bellotti, 2016; Molina et al., 2014), offering elaborated and nuanced differentiation of social relations to overcome categorizations (Sommer & Gamper, 2018), and detecting dynamics and temporal changes (Ryan et al., 2014). The interest lies in the stories behind the relationships, since according to White (1992) networks do not represent given realities, but are phenomenological constructs that are given meaning by the actors (see also chapter “Social Network Theories: An Overview”). The so-called “stories” (descriptions or interpretations of meaning), which make it possible to structure events in such a way that they function as part of a relationship history, which contains the subjective-social “meaning” of the relationship, are regarded here as the substrate of social networks. Thus, in order to be able to construct the emergence of networks or the dynamic change of networks, the stories of the persons and the possibilities for action in the respective context must be understood (White, 1992).

The theoretical discussion about a cultural or constructivist opening of network research (e.g., Emirbayer & Goodwin, 1994; White, 1992) goes hand in hand with the need for less standardized or qualitative-methodological approaches (Hollstein & Straus, 2006). The so-called visual network research (Gamper & Kronenwett, 2012; Gamper & Schönhuth, 2020), which is dominant in qualitative network research, is presented here. Since the 1980s, so-called network maps and network drawings (Gamper & Schönhuth, 2020) have been used in data collection, which are used to collect subjective experiences and attitudes of the actors. The most open form of visual network research is the network drawing. Using a narrative stimulus, the respondent draws his or her individual network on a non-structured sheet of paper or reconstructs it with the help of a software program (e.g., VennMaker). In this way, internal network images are made visible without any concrete specifications by the researcher. The researcher carries out subjective attribution of meaning, whereby the evaluation takes place within the framework of a communicative validation. Through the interviews conducted, the statements and interpretations flow into the analysis (Herz et al., 2014; Molina et al., 2014).

Due to the openness of the network drawing and the interviews, a quantitative evaluation is not possible. On the other hand, network maps can be described as maps of social relationships, which individuals use to visualize their social networks. In contrast to network drawings, these contain structuring (e.g., the positioning of ego as well as age, or other attributes such as age and gender) and standardization (unification through value assignment). These attributes are more or less predetermined by the researchers and the interviewer’s freedom is restricted. For this purpose, the network maps are partly or fully structured as well as semi-standardized or standardized by the researcher (Gamper & Schönhuth, 2020; Hollstein & Pfeffer, 2010). The pre-structuring can turn out differently. Popular forms include concentric circles (see Fig. 4) or sectors. If the specifications such as the concentric circles around the ego or the sectors are not assigned any discrete attribute values, this is structuring but not standardization (e.g., the social convoy model: Kahn & Antonucci, 1980). In the case of partial or full standardization, these visual items are assigned characteristic values in part or in their total number. The concentric circles that structure the proximity (which can be defined as importance, accessibility, etc.) to the ego can therefore be assigned the values “very important,” “important,” and “less important” to reflect the importance of the alters for the ego. The increase in structuring and standardization is accompanied by the loss of subjective assignment by the interviewee. However, the standardized data obtained can be evaluated using quantitative methods (Gamper & Schönhuth, 2020). The visual survey can be carried out using a paper-and-pencil method; paper, pens, and building block methods; or even a computer program (e.g., VennMaker), each of which has different advantages and disadvantages (Gamper et al., 2012). In addition, the visual survey method can be used in group or individual interviews. With regard to the qualitative evaluation, the focus can be on both the interviews conducted and the results of the different network maps or even drawings. Here, statements from the interviews can be related to the visualizations by first analyzing the interviews and then, in a second step, by investigating and relating the visualizations. Another possibility is to start from the maps or drawings and only then to use the interviews for the analysis. Which of the two approaches is chosen depends strongly on the research question and the data material, such as the focus of the survey (more visual or interview-based) and therefore cannot be answered in a generalized way. Qualitative methods of network research thus focus on mechanisms, behavior, or even individual interpretation and thus reveal, for example, action and thought processes.

Fig. 4
figure 4

Caregiver’s Social Network Transformation. Source: Carpentier and Ducharme (2005, p. 297)

Full size image

Two examples, a mixed-method and a qualitative research study, will be presented. The first study deals with support networks for caregivers of persons with dementia. The mixed-methods approach combines the social network analysis with narrative analysis. The name generator is based on eight questions. These questions are providing the caregiver an emotional, informational, instrumental, and social support network (see Fig. 4).

The caregiver is asked to provide up to five names for each question. These actors can be family members, friends, neighbors, coworkers, volunteers, or professionals (Carpentier & Ducharme, 2005, p. 294). With the aim of the quantitative network analysis, differences between T0 and T1 networks are analyzed in relation to size, density, and homophily, since the caregiver began his or her career. With the help of the narrative method, four goals were addressed. The first involves identifying the actors named in the narration, including support actors, identified by the name generator, but also other actors who have participated in decisions or influenced the course of social relationships. A second phase involves identifying events of the narration. The goal was to produce a collection incorporating the events deemed essential to understand the network transformation process. Third, a temporal map incorporating the actors and events was produced. A diagram was used to analyze when the support relationship started or ended. With the interpretation, the last step, the mechanisms linked to motivations, intentions, and actions over time should be interpreted. “Social policy intended to maintain older persons in the community is based on the establishment of support ties with various resources providing assistance, although very little information is currently available regarding the processes that create and maintain support ties for caregivers” (Carpentier & Ducharme, 2005, p. 308).

The second example is a qualitative psychology or psychotherapy research project in the field of intervention studies. This research deals with the effects of network relationships on mental well-being in Germany. Using three case studies, Silvia Weigl (2016) shows how network maps are used to visualize and reflect on the effects of relationships on the well-being of the subjects. Figure 5 shows a network map in which a respondent has presented his or her own relationships and rated them as positive, negative, or ambivalent (for the meaning of negative relationships, see chapter “Negative Ties and Inequalities in Health”). In addition to the drawing, the clients are also asked about their relationships, which are visualized in network maps. In the therapy sessions, these relationships are discussed, placed in the life phase context, and their influence on well-being is reflected.

Fig. 5
figure 5

Network map of a patient at the beginning of therapy. Source: Weigl (2016, p. 238) (translated in English)

Full size image

The results of the network maps and the subsequent networking of the people to be advised are assessed as positive. The use of network maps makes clear the life situation of the respondents and the importance of the own person in the social network. This increases the self-esteem as well as the perceived self-efficacy. Furthermore, a stabilization of the own position is achieved. There are also different forms of networking by the persons concerned themselves, in which relationships in the network are worked on in a targeted and active way, as well as a relativization of idealized and derogatory views of social relationships in the past. As the example shows, networks thus not only serve as scientific analysis tools, but also offer instruments for intervention in the health sector under the keyword “network work.”

3 Conclusion

This chapter defined the concept of the network and presented the different approaches and procedures of network research and analysis. The method to be used for a particular question was shown by using examples from the field of health research. Networks are associations of persons, institutions, and collective actors—the so-called nodes—which are interwoven by relationships (e.g., sexual relationship, love)—referred to as the edges. It is assumed that the embedding of the actors has consequences for them or that certain actions of the actors affect the relationships within a network.

During the analysis, two main distinctions were identified. Against the background of the structure of the network, a distinction can be made between whole networks and ego-centered networks. The whole network analysis focuses on the internal structure of persons within a predefined area. This can include, for example, pupils in a school class or people in a city. Here, only the relationships among each other are recorded. In ego-centered network research, the interest lies in the embedding of the individual in his or her social environment. Here, ego is asked about their relationships and the persons (alters) and their attributes in their personal network. If the research interest is focused on the internal structure of a group and the boundaries are clearly given from the inside, then a whole network analysis is particularly well suited. The egocentric network research has its advantages when persons of a certain group (e.g., drug addicts, the elderly) and their general embedding, also in comparison with other groups, are to be examined. Besides the methodological dimension, a distinction can also be made between quantitative and qualitative network research. Standardized network research can be differentiated between structure-describing methods and methods for analyzing causal relationships. The structure-describing methods describe the network, including, for example, the size of a network or its density. Stochastic methods, such as exponential random graph models (ERGMs) and actor-oriented models (SAOMs), try to uncover random relationships. Both alignments can cover the entire network (e.g., density, network size) or individual nodes or edges (e.g., centrality measures).

Regarding qualitative network research, different visual methods were presented. A distinction was made between network drawings and network maps. Network drawings are free visualizations, which do not include any pre-structuring by the researcher. With network maps, specifications such as concentric circles are made. Structures can also be standardized by assigning values. This also makes it possible to evaluate the data quantitatively, whereas this is not possible with network drawings. For the coupling of interviews and the visualizations, there are hardly any scientific handouts or standard works. Qualitative research should be resorted to if the focus is more on idiographic constructs such as patterns of interpretation, structures of meaning, or subjective perceptions of networks and relationships. Even with phenomena that are unknown, little known, or researched, qualitative instruments are more suitable because of their thematic openness. Quite often, hypotheses for quantitative network research are generated in qualitative studies. For causal connections or when representative statements are to be made, the different quantitative methods are suitable. It is important here that the survey methods must be adapted to the research field and the research question. Particularly in health research, there is no uniform procedure for name generators, for example, and thus, there is still a lot of room for ideas.

Due to the few studies that exist in the area of network research and health inequalities, tested actor generators or other preliminary work, including in the qualitative area, are very rare. Therefore, research questions have to be constructed and tested. This makes it necessary, for example, to develop one’s own actor generators or to adapt already tested questions to one’s own research. It is central to adapt one’s own questions, both qualitatively and quantitatively, to the theoretical concepts (see also chapter “Social Network Theories: An Overview”). Paradigms such as social support, diffusion research, and social capital, which can be well combined with health issues and networks, are particularly suitable here. Limits are generally apparent with regard to the duration of the surveys. Qualitative and quantitative methods of network research are very time-consuming and take up a lot of space in the survey. Therefore, it should be considered in advance what role networks play in answering the research question on health inequalities. Building on this, the part of the network analysis should be tested in order to be able to estimate the duration of the survey. As the few studies show, a connection between health and networks is of great importance in many fields (e.g., transmission of diseases, health behavior) and should be considered much more strongly, but also methodologically.

Reading Recommendations

  • Carrington, P. J., Scott, J., & Wasserman, S. (Eds.). (2005). Models and methods in social network analysis. Cambridge University Press. Introduction to the methods of network analysis with a focus on quantitative methods.

  • McCarty, C., Lubbers, M. J., Vacca, R., & Molina, J. L. (2019). Conducting personal network research: A practical guide. Guilford Publications.

  • van der Pol, J. (2019). Introduction to network modeling using exponential random graph models (ERGM): Theory and an application using R-Project. Computational Economics, 54(3). 845–875.

  • Gamper, M. & Schönhuth, M. (2020). Visual network research (VNR) – a theoretical and methodological appraisal of an evolving field. Visual Studies, 35(4), 374–393.

  • Domínguez, S., & Hollstein, B. (Eds.). (2014). Mixed methods social networks research: Design and applications. Cambridge University Press. English anthology on the combination of qualitative and quantitative methods in network research.