The development of mixed methods research into a field has led to the development of several typologies, each trying to organise and distinguish the various versions of mixed methods research design (e.g. Moran-Ellis et al 2006; Greene 2007; Teddlie
& Tashakkori 2009; Bazeley & Kemp 2012; Guest 2013). As mentioned earlier, the purpose here is not to discuss the usefulness of such typologies more generally, nor to choose one over the other. Instead, I make use of some of the typologies as tools for describing the design of this study.
Guest argues that the typologies have become too complicated, while at the same time not actually fulfilling their purpose of e.g. aiding the design process and structuring the field of mixed methods research studies (Guest 2013). He instead advocates focus on points of interface, i.e. any point where methods are connected as opposed to a focus on the study as a whole. Thereby, the complexity of a study can be described. Further, he argues for the use of only two dimensions when describing integration in mixed methods research design: the timing and the purpose of the mixing taking place. These two, he argues, are critical for any mixed method research study and at the same time have the necessary descriptive power to distinguish studies. Timing relates to which part of the study the methods are connected within. The purpose of the integration relates to why mixing takes place at the given point of interface and is thus insepa-rable from the purpose of the study as a whole. In this study there are two points of interface, namely 1) the qualitative interviews providing input for the hypotheses of the register data analysis and 2) the conclusion of this extensive summary in which the four papers and their conclusions are drawn together (this point being the primary integration and the one planned from the outset). The purpose of the first point of in-terface is to qualify the grounds for the hypotheses. The purpose of the second point of interface is to combine the knowledge of different aspects of the housing career in order to gain empirical knowledge of different aspects of the housing career. Such a purpose can be characterised as being a purpose of extension (Greene 2007): the two strands of the project describe contiguous phenomena, making it possible to study a larger part of the greater phenomena complex that a housing career is. The integra-tion of the study is based on a theoretical unison (Moran-Ellis et al 2006) through the common backdrop of life course analysis.
The mixed method research design employed in this study is fairly simple, as only two data sources are used and as integration only takes place at two stages. While Guest has a very valid point for more complex studies which are nearly impossible to put in a typological box, simpler studies can be meaningfully categorized within typolo-gies (a point set forth by Guest as well). One such typology is proposed by Teddlie &
Tashakkori (2009). They divide analyses of mixed methods research into six types:
parallel mixed data analysis, conversion mixed data analysis, sequential mixed data analysis, multilevel mixed data analysis, fully integrated mixed data analysis and ap-plication of analytical techniques from one tradition to another. The analysis conduct-ed in this study belongs to the first category: the analysis of the register data and the qualitative analysis took place as two separate, parallel processes. As they focused on different research questions relating to different aspects of the housing career, the two analyses are neither able to confirm nor oppose each other. While the analy-ses are therefore independent by design, they still influence each other in something analogous to them ‘talking to each other’ in Teddlie & Tashakkori’s words (2009).
As both analyses were conducted by the same researcher, they cannot be completely separated, as the researcher, knowingly or otherwise, might have been influenced by findings within one analysis when conducting the other. A specific way in which the two strands ‘talked to each other’ in this study was that the interviews provided input
to the hypotheses tested in the registers. However, as data in this case refers to different aspects of the housing career, the influence was limited and similar to the way in which any prior knowledge as a researcher is never ‘not there’ when analysing new data.
What links the papers, the research questions, and thus the methods is the focus on different aspects of the housing career. Furthermore, all this is understood through the framework of life course analysis. Thus, while the integration in this mixed meth-ods research study is less ambitious in that mixing takes place at a late stage, it is a consequence of the study’s purpose to analyse different aspects of the housing career through different methods, as opposed to analysing the same aspect of the housing career with different methods.
Figure 2: The mixed methods design of the study (inspired by Guest 2013) Figure 2: The mixed methods design of the study (inspired by Guest 2013)
Data collection and data analysis have taken place as two convergent or parallel pro-cesses (Guest 2013) as depicted in figure 2. No explicit integration took place in the data collection process. In the data analysis, the analysis of the interviews added to the formulation of hypotheses for the analyses of home-leaving patterns. Despite the dif-ferent foci, the interviews touched upon and inspired ideas for the analysis of register data. This point of integration was not planned from the onset of the study. However, as the ideas emerged from the interviews, they were integrated in the register data analysis. This shows how any research process, mono or mixed methods, cannot be anticipated in its entirety from the onset. Apart from this, as mentioned above, it is inevitable that the convergent processes will influence each other at a sub-conscious level in the researcher in a combined knowledge process for the study.
As described, the analytical approach to the interviews was inductive, while the ap-proach to the register data was deductive. This, again, is due to the different purposes
of the two strands of the thesis. Furthermore, as Bergman writes, “few research pro-jects are entirely inductive or deductive” (2008:13), whether mono or mixed methods.
This study shows how a potential consequence of working with mixed methods re-search as convergent processes is that the two strands of the study might move in dif-ferent directions, resulting in bigger differences between the foci of the strands than initially planned. The initial plan for the register data strand of the study was not to fo-cus entirely on home-leavers. However, this fofo-cus emerged by following what turned out to be the most interesting in the data and the area where entirely new empirical knowledge could be gained. The interview strand of the study proceeded more in line with the original plans. While this meant that the differences became bigger, I argue that as it has led to greater empirical gains within each of the strands; it was worth it.
This is especially true since from the outset the project was planned as parallel mixed data analysis, with integration limited to the conclusion of the project and with the two strands working together by extension.
As explained in the theoretical section, housing careers are linked to time. That the unit of study is temporal has implications for the unit of analysis as depicted in figure 3. Registers contain data collected for numerous points in time and the interviewees were asked to tell their housing story over time. The knowledge from the two sources differs in the link to time. Registers are retrospective, referring to time that has already passed and where data is collected for a given time at that given time. The interviews primarily refer to the present. While the interviewees were asked about previous hous-ing situations, their knowledge about the past was their current perception and mem-ory of the past and not (necessarily) as it actually was when it happened (Hoerning 1996). In a similar manner, expectations and dreams for the future were seen from the present life situation. Analytically, time is present as well. The Cox regression models analyse the effects of covariates on home-leaving in a prospective sequence in which covariates at a time preceding home-leaving are used to estimate the hazard for home-leaving. The inductive analyses of the interviews analyse the housing careers as seen from the present looking back at the past. Time thus comes into play in different ways in relation to the two data sources. This is an unavoidable consequence of the data sources and data analysis methods chosen. It is a consequence of utilizing the respective strengths of the two sources. This has to be kept in mind when combining the methods: the results of the two strands cannot be integrated completely, but can inform each other and provide potential explanations for the results of the other strand.
Figure 3: The presence of time in the two empirical sources Figure 3: The presence of time in the two empirical sources
Register data Interviews
1985 1986 1987 1988 … 2003 2004 2005 2006 2012‐2013
Register data Interviews
Time of data collection
Analytical time perspective Time frame of knowledge