1 Introduction and Overview *

As both the variety and the specialization of methods increases steadily, combining methods is a common thread in actual research and self-reflexive scientific discourses. There are different catch-phrases for these discussions like "mixed methods" (Tashakkori and Teddlie 1998, Tashakkori and Teddlie 2003), "integrative social sciences" (Seipel and Rieker 2003), or "triangulation" (Flick 2011) - the latter gained arguably the most prominence throughout the last years. As the different catch-phrases or labels indicate, these debates both attract quite some attention and take place in various scholarly communities that each by and large focus on one aspect, be it a critical discussion on whether and how to combine qualitative (small-N) and quantitative (large-N) studies (see, for instance, "integrative social sciences") or be it an analysis of various mechanisms of aggregating data, which is the focus of, for example, the "triangulation" camp. As fruitful as these debates were for advancing the methodological reflections and in fostering a pluralistic thinking in social sciences, they come with a certain baggage: firstly, there is a tendency of being either caught up in some ideological debates (a strive for pluralism and methodological non-conformism) or focused overly on a specific subfield (data aggregation in triangulation) losing sight of the overall picture of combining methods. Secondly, terms like triangulation and phrases like mixing methods have become ubiquitous in articles, conference papers, and grant proposals oftentimes not adhering to the rigor of the original concepts and thus making them more and more devoid of any methodological substance. Thirdly, there are new advances in the methodological literature that have not received attention by scholars devoted to method combinations. A case in point is the role of temporality, most notably the discussions on causal process tracing or causal chains in case study designs (see, for instance, Blatter and Haverland forthcoming).

Hence, this article attempts to map a more comprehensive picture of combining methods, introduces a typology that focuses on the basic standards of each subtype, and exemplifies one specific subtype (causal chains) by investigating a theoretical framework, which explains humanitarian interventions as a multilevel process.

At the core of the typology is both a distinction between method triangulation and method parallelization and corresponding subtypes differentiated between the combination of data-generation and data-analysis techniques. While scholars often refer to triangulation when in fact undertaking parallelization, the two should be clearly distinguished: method triangulation is based on a vertical logic of combining methods to aggregate data or data-analyses for the score of one explanatory factor, method parallelization follows a horizontal logic of combining methods according to their conceptual linkages.

In the first part, the article introduces this typology and distinguishes between two subtypes of method triangulation (data generation and data analysis triangulation) and three subtypes of method parallelization: multivariate designs, research programs, and causal chains, whereas the latter links to current debates in methodological literature.

To illustrate the differences and applications, the article investigates international decision-making in the field of humanitarian interventions. To that end, humanitarian interventions are modeled as a causal process, which is analyzed by falling back on different methods at different steps of the process but sometimes triangulating methods to explore one condition at one step of the process.

2 Triangulating and Parallelizing Methods - Mapping Method Combinations

Combining methods in one research endeavor is nowadays quite common. However, mono-method designs dominated the scene before the 1990s (Tashakkori and Teddlie 1998), even though the methodological reflection of combining various methods had already advanced quite considerably at that point of time: as early as 1959, Campbell and Fiske called for combinations of various operationalizations of one explanatory factor to refine the measurement and to make the results more reliable and valid. Therefore, they introduced a now seminal technique of forcing researchers to think of alternative methods in one analysis, the multimethod multitrait matrix (Campbell and Fiske 1959).

While Campbell and Fiske focused on quantitative methods in psychological research, much of the methodological debates in the subsequent decades were consumed by the partially ideologically driven conflict between camps that were most commonly referred to as positivist-quantitative and interpretative-qualitative camps. The debate of combining methods was thus driven by an often pragmatic stance (Morgan 2007) against these methodological divides: proponents of "integrative methods" or "mixed methods" elaborated on various ways of integrating quantitative and qualitative methods in operationalizing one factor or throughout the research process. Tashakkori and Teddlie provide an in-depth overview of this development (Tashakkori and Teddlie 1998).

Combining elements of both quantitative and qualitative approaches was the common thread of most mixed method, triangulation and integrative method debates. Scholars engaged in them reflected on their philosophical assumptions, inquiry logics, guidelines and good standards of research practice, and even on the sociopolitical commitments of the researchers themselves (Creswell et al. 2003, Greene 2006).

In the following, a typology of method combinations will be presented that goes beyond these debate in three ways: firstly, it is truly integrative leaving the old method divides aside. Hence, it discusses combinations of methods no matter if they are qualitative or quantitative by nature. Secondly, it maps the complete field of method combinations by looking at and delineating both triangulating and parallelizing methods - and not either or. Thirdly, it puts a certain emphasis on the inclusion of developments that have gained considerable prominence throughout the last decade: temporality and causal processes.

2.1 Data Generation, Data Analysis, and Two Logics of Method Combinations

The term "method" is used for both techniques to generate data and for techniques to analyze data. Data generation methods provide techniques of standardization for tapping into new data sources and transforming raw data into analyzable forms. Narrative interviews or surveys are cases in point. Methods of data analysis refer to what comes after: by using analyzable data (even from the literature) they provide techniques to structure them and to apply them to some kind of causal inferences. The proof of causal relationships can take different ways: for instance, either a more positivistic one by employing experimental or quasi-experimental methods to draw conclusions from empirical data in a logical and controlled way (for instance, through systematic and covariate cross- case comparisons), a more inductive-naturalist one by focusing on the richness of the data and finding "smoking-gun" observations that can hardly be disputed by common sense (see, for instance, George and Bennett 2005) or rather theory driven in the form of a congruence analysis (see Blatter and Blume 2008, Blatter and Haverland forthcoming). When combining methods, one should keep in mind whether one refers to data generation methods or data analysis methods. In some cases this leads to quite different challenges.

In general, two basic logics of combining methods should be clearly distinguished. First, the vertical logic, which refers to combining several methods in operationalizing and measuring one explanatory factor. Hence, one is mixing different data generation or data analysis techniques to measure the score of one explanatory factor. The challenge here is to identify the rules according to which data is aggregated. This is the logic behind method triangulation. Second, the horizontal logic, according to which methods are combined along a conceptual sequence in one research design or within a larger research process - method parallelization. The challenge is thus to be found in the conceptual depth of the theoretical framework.

In the following these two basic types of combining methods will be introduced in more detail and subtypes distinguished by falling back on the differentiation between data analysis and data generation methods. Figure 1 summarizes the typology.

[FIGURE 1 OMITTED]

2.2 Method Triangulation

Combining methods according to the first logic is called triangulation. Two or more methods converge in the measurement of the score of one explanatory factor graphically forming a triangle (see figure 1). Triangulation is not a new term and the concept has been methodologically refined since the 1970s. However, scholars sometimes use it as a synonym for every kind of method combination by and large neglecting the vertical logic behind it.

Conceptually, it was coined by Denzin, who defines it broadly as "the combination of methodologies in the study of the same phenomena" (Denzin 1970: 297). But even before, as Flick (2011: 7-10) points out, triangulation has been used frequently or even highlighted as a research strategy (Strauss et al. 1964: 36) without referring to the yet relatively unknown terminology. Denzin (1970) distinguishes between data triangulation (the use of multiple sources of data), methodological triangulation (the use of multiple methods), investigator triangulation (the use of multiple interviewees, for instance), and theory triangulation (the use of multiple theoretical perspectives).

Discussing methods, however, this article focuses rather on methodological triangulation, relating to it as method triangulation. Denzin differentiates two subtypes: within- and betweenmethod...