You are on page 1of 24

Journal of Mixed Methods Research

4(2) 144–167
Enlightened Eclecticism or ªThe Author(s) 2010
Reprints and permission: http://www.
Hazardous Hotchpotch? sagepub.com/journalsPermissions.nav
DOI: 10.1177/1558689810364987
Mixed Methods and http://jmmr.sagepub.com

Triangulation Strategies in
Comparative Public Policy
Research

Frieder Wolf1

Abstract
This article discusses the trend toward triangulation and mixing methods in comparative public
policy research and proposes criteria for further such endeavors, especially regarding the
sophistication of the respective (qualitative and quantitative) parts, the way the different analy-
ses are combined, and the aspects of the social world as well as the statements about them
political researchers are interested in. It draws on both the current methodological debates
and a review of relatively recent contributions to the comparative public policy literature em-
ploying several methods. The final recommendation is to consider thoroughly whether to
engage in triangulation, and if doing so, to use tailor-made triangulation strategies fitted to
the research questions and interests.

Keywords
combining qualitative and quantitative methods, comparative public policy, nested analysis,
research design, triangulation

The Trend Toward Triangulation in Comparative Public Policy*


The academic subdiscipline of comparative public policy is mainly concerned with ‘‘what govern-
ments do, why they do it, and what difference it makes’’ (Dye, 1976). Originally, the policies most
closely related with economic performance and social well-being were positioned at the center of
its attention, whereas in more recent years further fields, such as environmental or education pol-
icy, have gained prominence, and the interplay between public and private problem-solving

1
University of Heidelberg, Heidelberg, Germany

Corresponding Author:
Frieder Wolf, Institut für Politische Wissenschaft, University of Heidelberg, Bergheimer Straße 58, 69115 Heidelberg,
Germany
Email: wolf@uni-hd.de
Wolf 145

activities has come under more thorough investigation. Regarding methods and methodology, after
several cycles of (more or less fruitful) qualitative–quantitative confrontation (which also had an
European vs. American component to it, cf. Berg-Schlosser, 2006, p. 169; Moses, Rihoux, & Kit-
tel, 2005; Sartori, 2004; Topper, 2005), there has now been a trend toward triangulation and mixing
methods in comparative public policy research. About a decade ago, it was ‘‘difficult to defend the
decision not to use panel data’’ (Kittel & Winner, 2002, p. 5; see also Hall, 2003, p. 381) in many
fields of political research but especially in comparative public policy. Nowadays, we have come
one step further: Scholars are regularly expected to demonstrate both familiarity with the latest
fashions in quantitative modeling and a critical awareness of their methodological perils, and to
combine macroquantitative methods with in-depth case studies (based on meticulous process trac-
ing, numerous expert interviews, and the like) at the same time.2 By now comparative public pol-
icy, and especially the study of the welfare state, has been hailed for the fact that ‘‘the willingness
and ability of individual scholars in this literature to combine methodological approaches, some-
times in a single work, is remarkable’’ (Amenta, 2003, p. 112). According to a recent review of
three leading journals, studies concerned with policies are an underrepresented subfield of compar-
ative politics (Munck & Snyder, 2007, pp. 9, 12), yet this subfield features a higher-than-average
share of mixed methods designs (p. 17).3
con entusiasmo This article tries to answer the question whether this is a development worth welcoming

wholeheartedly and pursuing further. In a first step (Section ‘‘The Promise of Combining Differ-
advertencias
ent Methods—and Three Crucial Caveats’’), different arguments raised in the methodological
debate in favor of and opposing the trend toward triangulation and mixing methods are discussed,
and criteria for promising research of this kind, or, in other words, for its evaluation, are iden-
tified. Then (in Section ‘‘A Review of Recent Mixed Methods Works in Comparative Public Pol-
solvencia
icy’’), the soundness of several eclectic4 research designs is assessed, and some suggestions
que intenta regarding the improvement of future triangulative and mixed methods research are offered.
reunir This second step mainly draws on a review of relatively recent contributions to the comparative
procurando
public policy literature employing several methods, highlighting two especially instructive
conciliar sistemas
works. In the conclusion (the final section), it is stated that although triangulation and mixing
methods are promising in many research contexts, they are certainly no panacea. My final rec-
ommendation is to consider thoroughly whether to engage in them, and if doing so, to use tailor-
made strategies fitted to the research questions and interests. hecho a la medida
A couple of further clarifications are necessary at the outset: This article has no missionary
intention directed at those scholars who oppose either quantitative or qualitative methods in
general. (Of course, the usual disclaimer about the terms qualitative and quantitative being
somewhat problematic labels also applies. By ‘‘quantitative methods’’ I mean methods that
use mathematical tools to identify patterns in larger collections of mostly numerical data, and
by ‘‘qualitative methods’’ I mean methods that extract patterns or meaning from all sorts of sour-
ces in any other systematic way.) It is merely a contribution to the discussion amongst those who
are open-minded toward mixing methods in general, but disagree about the degree of feasibility
and the merits of different ways of doing so. factibilidad

Regarding terminology, I use the terms mixed methods research and triangulation inter-
changeably, standing for approaches that involve quantitative as well as qualitative analyses.
Admittedly, a number of arguments against this conflation can be found in the literature. First
of all, there is a general definition of mixed methods research that ought to be acceptable for
most researchers fusión

Mixed methods research is the type of research in which a researcher or team of research-
ers combines elements of qualitative and quantitative research approaches [. . .] for the
146 Journal of Mixed Methods Research 4(2)
amplitud

broad purposes of breadth and depth of understanding and corroboration.’’ (Johnson,


Onwuegbuzie, & Turner, 2007, p. 123)5
However, there is considerable disagreement on the meaning of the term triangulation (for an
elaborate discussion, see Hammersley, 2008; see also Johnson et al., 2007, pp. 113-115; Tashak-
kori & Teddlie, 1998/2008, p. 21, and Jick, 1979/2008). Some reserve it for measuring proce-
dures only (e.g., Brewer & Hunter, 2006, p. 4),6 or for the evaluation of independent studies
of the same topic. Others argue that triangulation is different from mixing methods in that it seeks
convergence, whereas (only) mixed methods research is said to be open for divergence. Further-
more, Hammersley (2008, p. 32) cautions ‘‘against conflating triangulation with the mixing of
quantitative and qualitative approaches’’ because triangulation can occur both within as well
as across methodological borders. (Obviously, I am referring to the latter variant.) So for the pur-
pose of the present article (and in the context of the comparative public policy literature), trian-
gulation is simply meant to refer to the combination of different research modes, or looking at the
same topic from different angles or through different lenses, possibly on very different stages of
the research process, and to serve ‘‘as validity checking’’ (Hammersley, 2008, p. 23) and/or ‘‘as
seeking complementary information’’ (p. 27).7 This also conforms to the common usage of the
term within the comparative public policy literature (for a recent textbook chapter on triangula-
tion, see Pickel, 2009).
Finally, my main concern is with the typical research situations in comparative public policy
with a number of relevant cases (i.e., states) between about 15 and 30 and time spans of a similar
number of years, yet most of my arguments are not strictly limited to these.

The Promise of Combining Different Methods—and


Three Crucial Caveats
This section identifies evaluation criteria for mixed methods research in comparative public pol-
icy based on a discussion of arguments in favor of and against different methods and their com-
bination. It addresses the need of researchers for ‘‘a repertoire of strategies for establishing rigor
within their mixed methods studies’’ (Creswell & Plano Clark, 2007, p. 190; see also Teddlie &
Tashakkori, 2003, pp. 38-43, focusing on ‘‘design quality’’ and ‘‘interpretive rigor’’). Because
‘‘we are still very much in the infant stages of understanding how to judge the quality of mixed
methods practice’’ (Greene, 2008, p. 18), this exercise, while being primarily directed at a com-
parative policy audience and its discipline-based problems,8 should also be of some value for
readers from other disciplines, and help bridge the gap between abstract methodological debates
and ‘‘triangulation in action’’ (a phrase coined by Jick, 1979/2008). Studies that fulfill the cri-
teria advanced in the following should be more ‘‘defensible,’’ or more ‘‘legitimate,’’ to the rel-
evant ‘‘research stakeholders’’ (Onwuegbuzie & Johnson, 2006/2008, p. 274) than those which
do not.
Quantifying analyses, and especially multivariate regression (MR), are ascribed the following
advantages: They allow for processing large numbers of observations (distributed over time and
space) and for doing this in a manner that enables the academic audience to reproduce their
results. Thereby, the conclusions drawn are democratized and made accessible for critique
and enhancements.9 Furthermore, they are often said to be superior when it comes to drawing
causal inferences, to deciding between rival theories or hypotheses and to making prognoses
about future developments. With the advent of advanced computational facilities, easily acces-
sible statistical software packages, and also a certain kind of instructional literature, MR of
pooled time-series (or rather, in the case of comparative public policy, pooled cross-sections)
became something like a standard tool in the second half of the 1990s. Yet soon a growing
Wolf 147
trampas

awareness of the pitfalls and deficiencies associated with pooled MR led to calls for a ‘‘triangu-
lation of different methods on the same problem’’ (Tarrow, 1995, p. 473). Comparative case
studies have been hailed as ideal methodological counterpoints of MR, because they can enable
aclamada
researchers to trace the crucial steps of causal processes and their sequence,10 to identify the con-
ditions under which explanatory variables function and interact in which ways, and to link struc-
tural factors with personal agency and the role of ideas. They are also useful for deriving and,
under certain conditions, for testing hypotheses (cf. Rueschemeyer, 2003, p. 307) and, of course,
desviada for illustrating deviant trajectories and specific properties of cases of special interest. Thus, there
is a fair chance that the combination of quantitative and qualitative methods increases the quality
of comparative (especially public policy, given our usual medium-N environments) research, and
many authors (e.g., Brady, Collier, & Seawright, 2006, p. 355; George & Bennett, 2004, p. 34;
Hall, 2003, p. 397;11 Kenworthy, 2007, p. 349; Kittel, 1999, p. 225; Lieberman, 2005, p. 450;
Luoma 2003, p. 12; Moses & Knutsen, 2007, p. 288;12 Pontusson, 2007, p. 328; Read & Marsh,
2002, p. 239; Siegel, 2007, p. 68; Wilson & Butler, 2007, 121f.; for a number of further refer-
ences see Mastenbroek & Doorenspleet, 2007, p. 3) have by now recommended to use this strat-
egy more often.
There are a number of caveats, however, and as they are concerned with the quality of
research of both types, that is, getting it right on both accounts, with the way the different anal-
yses are combined, that is, their nexus, and with the aspects of the social world researchers are
interested in, that is, their questions and answers, I will discuss them under these three headings.

Getting It Right on Both Accounts


First of all, the application of both kinds of methods needs to fulfill the respective quality stand-
ards (see also Brady et al., 2006, p. 359). This may sound simple, yet keeping track of the latest
refinements of quantitative modeling and MR estimation technique—requiring a nontrivial level
of econometric understanding13 as well as the maintenance of large databases—at the same time
as immersing oneself into the intricacies of several cases—usually requiring language skills as
well as securing and digesting vast amounts of primary sources and cultivating contacts with
expert interview partners—places a considerable burden on the shoulders of researchers.
Because there are controversies over many issues, this includes the formation of a standpoint
of one’s own.
Every application of quantitative as well as qualitative methods requires a solid theoretical
foundation (Criterion 1a in Table 1) and the availability of the necessary data, the quality of
which should also be gauged (Criterion 1c). Without them, many kinds of research simply do
not make sense. If theory testing is one of the aims, testable hypotheses should be formulated
(Criterion 1b; even though an alarmingly high percentage of papers with this goal fail to do
so, cf. Munck & Snyder, 2007, p. 21). Furthermore—this should actually go without saying—
scholars ought not to be preoccupied in favor of or against certain results, and the absence of
interrelations in question should be seen as just as relevant and desirable as their presence.
Regarding MR technique, it is not a good idea simply to follow routines advocated in ‘‘How to
do’’ publications. As Wilson and Butler (2007) have demonstrated, (pooled14) MR results are
regularly sensitive to the details of model specification. Therefore, it is crucial to establish the
reasons why a certain specification is deemed appropriate for a particular research question
and/or to demonstrate the degree of robustness15 of one’s findings over different model variants
(Criteria 1e and 1f in Table 1).16 Furthermore, auxiliary analyses of subpools17 down to single
periods and/or units and residual diagnostics ought to be employed. (The pattern of residuals, i.e.,
the difference between the values predicted by the model and the actual observations at each data
point, gives us a picture of the distribution of our relative explanatory success regarding different
148 Journal of Mixed Methods Research 4(2)

units and periods.) Regarding dynamics, the time-lag (or contemporaneity) between explanatory
and dependent variables, that is, the time it takes causal mechanisms to take effect, should also be
given some thought (Criterion 1g). Additionally, readers ought to be informed about the bivariate
relationships between explanatory and dependent variables (Criterion 1h) and—most basically—
be given an overview over the data used.18 Furthermore, if the language and technique of signif-
icance testing is invoked in studies of an N that is not a stochastic sample from a much larger N, estocastico
at least their meaning and function in this context, which can differ enormously according to no
determinist
researcher’s convictions and tastes, need to be pointed out (Criterion 1i).19 Even though these a
basic requirements are seldom fulfilled in published papers employing quantitative methods depende de
lo ocurrido
only, they will serve as criteria in the assessment of mixed methods comparative public policy y elementos
research in Section ‘‘A Review of Recent Mixed Methods Works in Comparative Public Pol- aleatorios
icy.’’ (Because the publications looked at there are monographs, at least space restrictions can
be ruled out as an excuse for not doing so.)
On the qualitative/small-N side, case selection (Criterion 1d in Table 1), whether by similar-
ity, dissimilarity, or by outcome (cf. Ebbinghaus, 2005, p. 141; see also Mahoney & Goertz,
2006, 239f.) or based on a most- or least-likely design (cf. Levy, 2007, 202f.) must be given spe-
cial attention, even though it is also often not discussed explicitly enough in quantitative large-N
comparisons (Levy, 2007, p. 207 and Ebbinghaus, 2005, pp. 134, 149; this is why I apply this
criterion to the MR parts as well). Case selection strategies are closely linked to the kind of con-
clusions aimed at20 (see Section ‘‘Questions and Answers’’), and they should be explained in
relation to them.
It is also especially important to point out whether the qualitative studies undertaken are
(exclusively or amendatorily) of an interpretative nature (cf. Mastenbroek & Doorenspleet,
2007, p. 8; more on this topic in the following section). If causal inference is one of the aims
of qualitative small-N studies, and thus the focus lies on causal-process information (CPOs),
the analytical value of these depends on ‘‘carefully constructed arguments as to why particular
CPOs [. . .] are especially diagnostic’’ (Brady et al., 2006, p. 367; Criterion 1k in Table 1).
‘‘When such arguments are lacking, CPOs may not be very useful for causal inference. When
they are present, CPOs can be a powerful tool’’ (Brady et al., 2006). Furthermore, the mediators
and markers associated with the causal mechanisms21 in question ought to be spelled out (Brady tener en
et al., 2006, p. 360; Criterion 1l in Table 1). It has to be borne in mind, however, that there are cuenta

‘‘radically different’’ positions on ‘‘the logic of causal analysis in small-N research’’ (Mahoney,
2000, p. 387; cf. Blatter & Blume, 2007 on this topic as well). The more cases are being looked
at, the higher is the risk of inadvertently slipping toward a logic usually associated with quanti-
tative large-N analyses:
More or less explicitly relying on Millian logic, many comparative case studies seem to
reject certain explanatory arguments and embrace others simply by matching causal var-
iables with outcomes across country cases. The conception of causality underlying such
work is often strikingly similar to the linear-additive conception that informs simple-
minded regression models, but with a highly deterministic twist. (Pontusson, 2007, p. 330)
Therefore, we can at least expect mixed methods researchers employing qualitative small-N
studies to spell out their personal stance in this matter. This includes being explicit about ‘‘the
specific type or types of within-case analysis they use’’ (Mahoney, 2000, p. 418; Criterion 1j in
Table 1). Hall (2008, p. 304) distinguishes three types of systematic process analysis, namely
‘‘historically specific, multivariate and theory-oriented,’’ characterized by whether the main
objective is ‘‘to explain the occurrence of a specific set of events’’ (p. 305), ‘‘to identify the
causal factors conducive to a broad class of events’’ (p. 305), or ‘‘elucidating and testing a theory
that identifies the main determinants of a broad class of outcomes’’ (p. 306).22 Obviously, the
Table 1. How the Reviewed Books Fare Regarding the Criteria Discussed in Section ‘‘The Promise of Combining Different Methods—and Three Crucial Caveats’’

Gourevitch Huber and


and Stephens Iversen Kenworthy Mares Rueda Swank
Shinn (2005) (2001) (2005) (2008) (2006) (2007) (2002)

Topic Political Welfare state Skills, social Feasibility Employment Insider–outsider The impact
determinants development protection, of growth, low performance phenomena in of capital
of corporate in the second and the inequality, and of advanced social democratic mobility on
governance half of the international high industrial labor welfare
law and 20th century division of employment economies market policy state
regulation labor spending
1. Getting it right
(a) Theoretical foundation + + + (+) + + +
(b) Testable hypotheses (+) + + (+) + + +
(c) Data quality + + + + + (+) +
(d) Case selection (MR/case studies) −/(+) (+)/+ − (+)/ − −/+ −/+ −/+
MRa (c-s only) (c-s only)
(e) Choice of model specification(s) − + + (+) − (+) (+)
(f) Discussion of robustness − (+) + + (+) (+) (+)
(g) Dynamics and lags − + + + (+) (+) (+)
(h) Bivariate relationships + (+) + + − − −
(i) Meaning of significance testing − − − + − − −
Case studiesb
(j) Type of qualitative analysis + + (+) (+) + + (+)
(k) Arguments for chosen CPOs (+) (+) (+) − + (+) +
(l) Mechanisms, mediators and markers (+) + − − (+) (+) (+)

2. The nexus
(a) Discussion of link(s) − (+) − − − − (+)
(b) Compatibility − (+) − − − − (+)
(continued)

149
150
Table 1. (continued)

Gourevitch Huber and


and Stephens Iversen Kenworthy Mares Rueda Swank
Shinn (2005) (2001) (2005) (2008) (2006) (2007) (2002)

(c) Order (+) (+) (+) (+) (+) − (+)


(d) Sequence (+) + (+) − (+) (+) +

3. Questions and answers


(a) Explanatory aims and interpretative goals (+) + (+) + + (+) (+)
(b) Kind of causal relationships − − − − − − +
(c) Level of aspired generalizations − (+) − + − − +
(d) Link between questions and methods (+) + (+) (+) (+) (+) +
(e) Relationship between results from different (+) (+) − − − − +
methods
(f) Uncertainties of conclusions (+) − (+) + − − −

Note: MR ¼ multivariate regression; + ¼ adequate documentation/statement/discussion/explanation (d/s/d/e) provided; − ¼ d/s/d/e not adequate; (+) ¼ d/s/d/e
partially satisfactory; 0 ¼ not applicable; c-s ¼ cross-section.
a. Iversen and Rueda also include one microlevel model (with a really large N) each, thereby exemplarily broadening the analytical foundation of their works. The
assessment here, however, is restricted to the macroquantitative models.
b. Iversen and Kenworthy do not present case studies as such, but a ‘‘comparative-historical analysis’’ (Iversen, 2005, p. 29) and an ‘‘examination of various types of
within-country evidence,’’ especially ‘‘on over time developments’’ (Kenworthy, 2008, p. 7), respectively.
Wolf 151

choice between these aims determines the particular proceedings. More than on the quantitative
side (even though it is not irrelevant there), drawing conclusions from qualitative research is ‘‘a
matter of judgement’’ (Hall, 2008, p. 311). Therefore, it is extremely important to give readers as
much information as possible about the way conclusions were arrived at.23 Of course, it is always
possible to get it wrong, when employing quantitative as well as when using qualitative meth-
ods.24 But mixed methods researchers (at least as much as anyone else) need to minimize that
risk and to inform their audience how they try to do so.
Given the usual time restrictions and pressure to publish, there certainly is a trade-off between
triangulation and refinement (the latter term is borrowed from Shalev, 2007a, 295ff.), that is,
between using several approaches or using one approach in a more sophisticated way, and it
is an open question which strategy offers higher ‘‘returns on scholarly investment’’ (Shalev,
2007a, p. 261). (Even if one subscribes to the usual SSCI (Social Science Citation Index) fetish-
ism, it is no longer clear which strategy is more promising.) Thus, sometimes it might be better to
stick to the approaches one is most familiar with than risking to produce substandard scholarship
by also engaging in other ones amateurishly. Yet although Shalev (2007a) is right in pointing out
that too many practitioners looking for quick fixes ‘‘have responded to this dilemma by looking
to ’best practice’ and following it faithfully—often with disastrous consequences’’ (p. 284), nei-
ther does his undifferentiated critique do justice to a lot of more enlightened studies employing
MR nor is his advice to substitute MR altogether with ’low-tech’ alternatives convincing (as Pon-
tusson, 2007; Scruggs, 2007, 310f.; and Swank, 2007, p. 362 also point out). At least some schol-
ars, at least sometimes, manage to be versatile ‘‘philosopher-technician[s]’’ (Rihoux, 2006, p.
333; see also Section ‘‘A Review of Recent Mixed Methods Works in Comparative Public
Policy’’).
One way of dealing with the triangulation versus refinement trade-off is the collaboration of
several researchers, each using the method(s) they are most familiar with (as suggested by George
& Bennett, 2004, p. 35). In this case, particular challenges lie in establishing a mutual understand-
ing, coordinating the efforts and, maybe most difficult, in aligning results into common conclu-
sions.25 Yet often this strategy will not be available, so I will give more attention to the efforts
of one or few authors who are engaged in qualitative and quantitative research at the same time.

The Nexus
The second caveat does not concern each of the methods employed themselves, but the mode of
their combination, or, in other words, the nexus between them. The goal here is ‘‘to search for use-
ful points of connection’’ (Morgan, 2007, p. 71). How can we ‘‘combine the two, so as to reap the
benefits from both approaches and correct their respective weaknesses’’ (Mastenbroek &
Doorenspleet, 2007, p. 4)? To start with, it is helpful to conceptualize different ways of doing
research as styles (cf. Brewer & Hunter, 2006, p. 12) rather than as paradigms. All research styles,
including several methods, risk being termed eclectic, but this eclecticism can be more or less
enlightened: Certainly, what the audience can ask for is an exposition of the links between the
applications of the respective methods (Criterion 2a in Table 1) and a discussion of their method-
ological compatibility (Criterion 2b). It is simply not enough just to do both MR and case studies
without pointing out how they relate to each other. Several variants are feasible and legitimate, but
all of them need to be set out and to be adapted to the respective research topic or problem.
One way is to start out with a large-N statistical analysis, which ‘‘can help identify outliers or
deviant cases, and case studies can then investigate why these cases are deviant, perhaps leading
to the identification of omitted variables’’ (George & Bennett, 2004, p. 34). Different degrees of
importance can be attached to the different methods employed in such a framework: For Esping-
Andersen (2007), the initial MR (only) serves as ‘‘as a diagnostic tool,’’ a ‘‘Popperian device’’:
152 Journal of Mixed Methods Research 4(2)

The strength of statistical association will not tell us much about the real causal mecha-
nisms at work, but the diagnostics that we can obtain from MR residual plots are a mine-
field of information, truly powerful instruments for fine-tuning and possibly correcting our
hypotheses, and subsequently for selecting alternative instruments. (p. 336)
A related possibility to link (pooled) MR results with case studies is the ‘‘focus on tipping
points: Qualitative analysis can explain turning points in quantitative time series and changes
over time in causal patterns established with quantitative data’’ (Tarrow, 2004, p. 174).
Criticizing the reliance on residual diagnostics as a nexus for quantitative and qualitative
methods, however, Shalev (2007b, p. 397) raises the question ‘‘of whether they are indeed
a mine of gold rather than a field of buried explosive devices,’’ and fears that ‘‘residual analysis
of [this] kind could encourage reading too much into prediction errors’’ (p. 397). To minimize
this danger, Pontusson (2007, p. 332) advises us to employ low-tech quantitative devices such
as tabular and graphical displays (which have been proposed by Shalev, 2007a as a substitute
for MR) as a ‘‘bridge between MR and theoretically informed, process-oriented case studies.’’
Lieberman (2005) has mapped out an elaborated scheme to combine large-N quantitative
analyses and qualitative small-N studies in a way he calls ‘‘nested analysis.’’ (Note that this is
not identical with the nested designs described in Creswell, Plano Clark, Gutmann, & Hanson,
2003/2008, 184f.) This mode of mixed methods research starts out with standard (frequentist)
forms of multiple regression (MR), QCA (qualitative comparative analysis, a technique to iden-
tify combinations of necessary and sufficient conditions employing Boolean algebra; also see
Note 2) or Bayesian modeling (an iterative estimation technique relying on assumed prior prob-
abilities). If the results are satisfactory, ‘‘model-testing’’ small-N analyses of ‘‘on the line’’
cases, that is, cases that can be explained by said model(s), are used to check the robustness
of their results, mainly but not exclusively by investigating the events and processes linking
the causes and effects assumed by the quantitative models.26 In the case of unsatisfactory results,
on the other hand, ‘‘model-building’’ small-N analyses of cases that lie on and off the regression
line are to be employed to improve the initial model. Although Lieberman (2005) claims not to
advocate ‘‘a single style of research’’ (p. 436)—indeed he is open to different ways of doing the
small-N and large-N studies and emphasizes their ‘‘distinct complementarities’’ (p. 436), several
authors like Mastenbroek and Doorenspleet (2007) are rather critical of nested analysis (and
advise to use it only in a comparison of most similar systems), mainly because they perceive
it to be dominated by the quantitative approach,27 oppose mixing deterministic and probabilistic
assumptions, and do not believe in the promise of nested analysis to enable improved concepts
and measurement (Mastenbroek & Doorenspleet, 2007, 4f., 16). Rohlfing (2008, p. 1497) intro-
duces the distinction between ‘‘regression-based’’ and ‘‘case study–based’’ nested analyses, and
he proposes to start anew with the latter (instead of doing model-building case studies as sug-
gested by Lieberman) when regression results in the former are unsatisfactory, because he is crit-
ical of selecting cases for case studies on the basis of misspecified models (or the respective
residuals), because in his eyes otherwise ‘‘mistakes travel through nested analysis’’ (p. 1501).
This danger, however, is not unique to regression-based nested analysis, but can also occur in
the opposite direction (p. 1509). In both cases, he concludes that ‘‘nothing is gained from the
combination of different methods’’ (p. 1511), which takes us back to the essence of the section
above—in mixed methods research, both parts must be done well, and diagnosing potential
model specification problems is particularly important.
In contrast to Lieberman, I would also argue that when MR results turn out generally satis-
factory, a closer look at off-the-line cases is often indicated, too. On the one hand, these cases
can be of particular interest in themselves, and on the other hand the reasons for their deviation
can often tell us something about the limits to the generalizability of the said MR results.
Wolf 153

Of course, there are many other ways to link the different methods within a mixed methods
design (for more variants, see Creswell et al., Hanson, 2003/2008 and Creswell & Plano Clark,
2007, chap. 4; for a list of different typologies28 of mixed methods designs cf. Teddlie & Tashak-
kori, 2003, pp. 27-30)—or to construct their nexus—that cannot be laid out in detail here.
Yet in any mixed methods study, the order, that is the relative importance or hierarchy (Cri-
terion 2c in Table 1),29 and the sequence, that is, the temporal succession (Criterion 2d),30 which
is not necessarily the same as the former, of the different methods to be employed need to be
established. This is best decided upon not on the basis of abstract reasoning alone, but depending
on the state of scholarship on the topic in question. Starting out with statistical analysis ‘‘is often
most useful when there is broad agreement on the basic causal processes behind an outcome and
dispute about the relative impact of particular factors within it’’ (Hall, 2008, p. 314).31 If, on the
other hand, ‘‘several theories alluding to rather different causal processes have been proposed to
explain the same phenomenon’’ (Hall, 2008, p. 314) starting out with process-tracing case stud-
ies is often more appropriate. Furthermore, qualitative small-N analyses can be used—in a con-
structivist sense—‘‘to prise open our imagination—to consider the possibilities and to encourage
new readings and understandings of the empirical literature’’ (Moses & Knutsen, 2007, p. 240).
Again, it is mainly a matter of having good reasons and being explicit about them rather than
following a standard protocol. And whatever the initial research design, actual research pro-
cesses will seldom be one-way streets, but be characterized by several feedback loops.32
Yet it is by no means self-evident that different methods are complements regarding the gen-
eration of answers to the research questions in question. Often, qualitative and quantitative work
will enrich each other, but there can also be contradictions. Mahoney (2000, p. 419) admonishes
us that we ‘‘must recognise that insights from quantitative research cannot always be directly
translated into helpful advice for small-N researchers.’’ On a more general level, the ‘‘unresolved
epistemological issues that arise from the attempt to do justice to historical particularity and at
the same time achieve theoretical generalization’’ (Mahoney & Rueschemeyer, 2003, p. 5) ought
to be addressed. This issue is dealt with in the following section on the third major caveat.

Questions and Answers


According to Morgan (2007, p. 64), ‘‘it was the increasing interest in combining qualitative and
quantitative methods that led to calls for greater clarity about the linkage between philosophical
commitments at the so-called paradigm level and practical procedures at the level of data collec-
tion and analysis.’’ In the following, I am going to discuss this link between ontology and epis-
temology on the one hand and methodology and methods on the other in light of the kind of ques-
tions social scientists are interested in and the type of answers they are hoping for, because these
are decisive when it comes to deciding on the appropriateness of triangulation strategies.
Mahoney and Goertz (2006, p. 228) ‘‘believe that qualitative and quantitative scholars share
the overarching goal of producing valid descriptive and causal inferences. Yet, we also believe
that these scholars pursue different specific research goals, which in turn produce different norms
about research practices.’’ First, it is striking that these authors associate scholars with either
group of methods. This leaves problem-oriented agnostics like myself out in the cold, and I guess
that such language use is not overly conducive to even thinking of triangulating methods. Sec-
ond, their statement about different research goals seems to be too simplistic. But we can derive
another criterion for good mixed methods research from it: Scholars need to be specific about
their research goals, and about the ways in which these are linked to their research practices.
The catalogue of ten dimensions in which there might be differing views offered by Mahoney
& Goertz in the same paper can be very helpful in this regard; for an overview see Mahoney
& Goertz, 2006, 229.) can be very helpful in this regard.) I would like to elaborate on four aspects
154 Journal of Mixed Methods Research 4(2)

at this point: When consuming mixed methods publications, readers at least ought to be informed
about (a) whether the aim is mainly to explain overall variation33 and/or to illuminate individual
cases of particular interest34 (Criterion 3a in Table 1); (b) which kind (if any) of causal relation-
ships, that is, deterministic35 or probabilistic ones, are purported (Criterion 3b); (c) which kind of
statements are striven for in the end, that is, which scope of generalizations is aspired to (Crite-
rion 3c); and (d) which method is meant to supply which part(s) of the results (Criterion 3d).36
All in all, this means that scholars engaged in mixed methods research need to be extra accurate
when formulating their research question(s) (cf. Kritzinger & Michalowitz, 2008, p. 199; for
a discussion of different ways to formulate mixed methods research questions see Tashakkori
& Creswell, 2007, p. 208). Doing this, and choosing a research design, poses a number of
trade-offs between incompatible goals (cf. Collier, Brady, & Seawright, 2004, p. 224). This
does not mean, however, that mixing methods is untenable. Rather,
the existence of such trade-offs means that no one set of methodological guidelines can
ensure the researchers will do good work. Diverse methodological tools will always be rel-
evant to any substantive problem. The best approach to trade-offs is to recognize them
explicitly, to acknowledge that there is usually no single ‘‘correct’’ resolution, and to iden-
tify the strengths and weaknesses of different combinations of goals and tools. (Collier
et al., 2004, p. 226)
Such an approach can also be termed order without orthodoxy (Brewer & Hunter, 2006, p. 11)
or not dogma, but disciplined thought (King, Keohane, & Verba, 1994, p. 7).37 Such an orderly
proceeding in mixed methods research needs to be, by its very nature, more problem-oriented
than method-driven (and maybe more attuned to middle-range theories38 than pure research
designs), but still methodologically sound and reflexive. As ‘‘research questions are not inher-
ently ’important’, and methods are not automatically ’appropriate’’’ (Morgan, 2007, p. 69),
again, there can be no standard protocol, yet this does not mean that anything goes.
Mastenbroek and Doorenspleet (2007) take the standpoint that (post)positivist (or naturalist,
to use the terminology adopted by Moses & Knutsen, 2007) and constructivist/interpretative
basic understandings are uncombineable (p. 9), so that ‘‘mixing is only possible within one
and the same paradigm’’ (p. 10; for a similar critique, see Blaikie, 1991, p. 125). According
to Marsh and Furlong (2002, p. 17), such ontological and epistemological positions are like
‘‘a skin, not a sweater.’’ If this metaphor were true, not only could researchers not jump back
and forth between approaches, but they could not even choose the camp they belong to. In
my eyes, however, actual research practices can be more nuanced and multifaceted than this
stance would like to make us believe. At least for some researchers in some contexts it ought
to be feasible to suppose, be interested in, and thus do combined research on, both probabilistic
regularities in social systems and how they relate to actors’ worldviews and constructions of real-
ity.39 Again, this is not to say that all ontological and epistemological creeds can always live in
mixed methods harmony with any other. But neither do we all necessarily have to dwell in par-
allel paradigmatic universes all the time. To the contrary, when ‘‘different paradigms give rise to
contradictory ideas and contested arguments’’ (Creswell & Plano Clark, 2007, p. 27) within
a mixed methods study, this tension can often be a source of scholarly progress. (It has to be
admitted, though, that manifestations of mixed methods research projects combining or bridging
several philosophical stances are much rarer than those within either perspective, and that there is
a need for further reflections on ways to conduct them; for the latter point, see Greene, 2008, p.
13). So whereas Moses and Knutsen (2007, p. 640), in response to Marsh and Furlong, go so far as
to state that ‘‘science is better served by researchers who master several methodologies, who can
self-consciously choose among concepts and theories, and who command many basic principles
of reasoning,’’ I do not subscribe to such a general superiority claim, but am of the opinion that
Wolf 155

there is a legitimate place for this breed of researchers (see also Johnson et al., 2007, 125f., who
state that ‘‘variation in particular philosophical commitments should be welcomed in mixed
methods research’’). So yes, there can be hazardous hotchpotches resulting from mixed methods
research, but this peril does not necessarily spring from being agnostic about ontological and
epistemological axioms in itself.
What I expect mixed methods researchers to do, however, is to make perfectly clear which of
their statements is based on which kind of reasoning,41 and how the conclusions they draw from
the different strands of their research relate to one another (Criterion 3e). As to ‘‘weighing the
results of each component’’ (Morse, 1991/2008, p. 155), there can be no general rule, because
the respective findings need to be ‘‘interpreted within the context of present knowledge’’
(p. 155) in order to achieve ‘‘inferential consistency’’ (Dellinger & Leech, 2007, p. 324). The
latter aspect is especially important when ‘‘the different methods produce different results’’
(Read & Marsh, 2002, p. 242). Then, either mistakes must have happened in the application
of at least one of the methods, or the ‘‘inadequacy of the applied theoretical concepts’’ (Erz-
berger & Kelle, 2003, p. 475) has to be admitted. Yet in the latter case (i.e., after ruling out
the former) more often than not the specific nature of the diverging results ought to serve ‘‘as
a pointer to new theoretical insights’’ (p. 475).
Finally, whatever the underlying research styles and philosophical paradigms, the uncertain-
ties our conclusions might be queried by should, as far as possible, be made explicit and their
magnitude estimated (Criterion 3f; cf. King et al., 1994, 8f.).

A Review of Recent Mixed Methods Works in


Comparative Public Policy
In this section, I apply the criteria derived above to seven recent examples of triangulative
research from the comparative public policy literature (for an overview see Table 1) in order
to assess their particular mixed methods approaches.42 To begin with, two books will be pre-
sented in a little more detail in order to highlight the way in which their authors constructed
the nexus between the different methods, and to demonstrate a little more thoroughly how I
applied the criteria established above. After that, the order of appearance is alphabetical, and
the necessary brevity of assessments might sometimes make them sound a little harsher than
deserved. I would also like to stress that the following is not to be read as a general critique
or an overall judgment of the merits of these books, which are all very valuable contributions
to the literature. (Indeed, the very fact that they are felicitous works is why I chose them, and
of course, other reviewers might have come to slightly different judgments.) Rather, the aim
of this reviewing exercise is to identify ways in which future mixed methods research could learn
from the manifold strengths and fewer weaknesses of their design and implementation. Admit-
tedly, by only selecting monographs, and only better ones, I give mixed methods research the
best odds to convince. Such a ‘‘most-likely’’ approach cannot contribute to an assessment of
the state of mixed methods research in comparative public policy in general, but it is most suit-
able for appreciating its potential.

Duane Swank (2002): Global Capital, Political Institutions, and Policy Change in
Developed Welfare States
Swank analyses the impact of capital mobility on welfare state spending in three different groups
of welfare states, and he especially focuses on the way institutional settings dampen or catalyze
this impact. The author starts out his empirical analyses with panel regressions on 15 OECD
156 Journal of Mixed Methods Research 4(2)

(Organisation for Economic Co-operation and Development) countries, the selection of which is
hardly mentioned at all. Yet it is convincingly explained regarding the following seven (full-
scale) case studies43: Denmark, Finland, Norway, and Sweden are chosen to cover the social
democratic group of welfare states, and France, Germany, and Italy represent the continental
European, or corporatist, variety, while the liberal flavor serves ‘‘as a baseline for comparison
in both quantitative and qualitative analysis’’ (Swank, 2002, p. 11). The first two groups are
highlighted because on the one hand, the impact of higher capital mobility is said to be stronger
there, yet on the other, so is the political resilience of the welfare state (p. 121). Thus the partic-
ular nexus between the qualitative and the quantitative parts of this study is designed as it is for
theoretical rather than technical reasons (which is not intended to say that this is necessarily
superior, but a characteristic feature of Swank’s book).
In the quantitative part, Swank (2002) finds ‘‘little evidence for the conventional view that
rises in capital mobility are systematically related to retrenchments, rollbacks, and neoliberal
restructuring of the contemporary welfare states’’ (p. 117), with the only exception being that
globalization increased the (financial and political) costs of high budget deficits (p. 119). (In
turn, the explanatory power of national institutional factors like collective bargaining systems
as well as partisan politics are shown to be much higher than conventional accounts of global-
ization argue.) His case studies on the Northern welfare states unearth only ‘‘few temporally
ordered and direct linkages between rises in capital mobility and welfare reforms’’ (p. 153),
but that social democrats in all four countries came to emphasize balanced budgets more than
before (p. 154). Regarding the corporatist conservative welfare states of continental Europe,
Swank concludes his case studies with the remark that the impact of globalization has been
‘‘largely episodic and contingent’’ (p. 211) there. Moreover, he interprets the combined evidence
from MR and case studies as follows: ‘‘[T]he institutional structure of the corporatist conserva-
tive model [. . .] facilitated the ability of pro-welfare state interests to act to maintain present pro-
gram benefits or otherwise resist and shape proposed neoliberal reforms’’ (p. 212), which is in
sharp contrast to the developments in the Anglo-liberal welfare states (p. 282).
Not every reader will necessarily share Swank’s interpretations of his empirical findings, yet
the way these are presented allows for differently nuanced conclusions to be drawn from them by
other authors. Furthermore, his research design is especially convincing in terms of its theoretical
foundations, the formulation of hypotheses, the discussion of data quality, and also most of the
third group of criteria. The MR analyses and the case studies themselves are relatively homoge-
neous and of a fairly high standard, as is their nexus. That the author is not entirely specific about
the (supposed) absence of interpretative goals constitutes only a minor drawback, but three more
important ones are that he leaves his readers in the dark about bivariate relationships between the
variables, does not say what the significance of significance tests in a nonrandom sample is for
him, and finally that he does not provide more than incidental discussion of possible uncertain-
ties affecting his conclusions.

David Rueda (2007): Social Democracy Inside Out. Partisanship and Labor Market
Policy in Industrialized Democracies
Starting from the observation that the constituency of labor falls more and more into two distinct
groups with differing interests, namely the securely employed and those unwillingly not so, Rueda
examines whose interests social democratic labor market policies are more attuned to (and, in con-
cluding, offers some guidance as to the feasibility of egalitarian strategies) on the basis of a con-
vincingly constructed model. The quantitative part of this book consists of two steps: First, the
influence of individuals’ employment status on their preferences regarding employment protection
Wolf 157

is estimated using data on almost 19,000 persons from 13 EU (European Union) countries (selected
for reasons of data availability). Both active and passive measures are shown to be supported sig-
nificantly stronger by labor market outsiders, whereas insiders are in turn much more interested in
employment protection (Rueda, 2007, p. 68). Furthermore, the lower the latter are, the smaller are
the differences between the two groups (p. 68). In the second step, Rueda tests his hypothesis
according to which social democratic governments strengthen employment protection, while there
are no partisan differences regarding measures preferred by outsiders. This is done by using a series
of pooled MRs on 16 OECD countries (10 of the former plus Australia, Canada, Japan, Norway,
Switzerland, and the United States, apparently chosen for reasons of data availability). Results
offer support for Rueda’s hypothesis, even though maybe not as clear cut as he interprets them.
The qualitative part of his book consists of three case studies on Spain, the United Kingdom,
and the Netherlands (chosen to cover a broad range of values on the explanatory variables), which
underscore the main propositions in very different contexts.
Although this is the only book reviewed here in which the terminology of triangulation is
invoked (Rueda, 2007, p. 6), it is nevertheless much less concerned with the connection between
its qualitative and quantitative parts than with the MR and the three case studies themselves, which
are both of a fairly high standard. Additionally, it suffers from a certain lack of syntheses—the indi-
vidual chapters do not feature concluding sections, and the final chapter is rather short.

Peter A. Gourevitch and James J. Shinn (2005): Political Power and Corporate Control.
The New Global Politics of Corporate Governance
This book seeks to explain the variation in the laws and regulations regarding corporate gover-
nance, laying special emphasis on political determinants (which had been neglected by the liter-
ature on the topic so far). A total of 39 countries, the selection of which is hardly mentioned, are
analyzed quantitatively, and 18 of them are subjected to MR, whereas the choice of the nine
countries selected for cases studies (with varying thematic foci) is explained a little more exten-
sively. The authors leanly orient their readers as to the order and sequence of MR and case stud-
ies (with the former used to analyze cross-country variation and the latter to trace developments
over time), but the concrete links and the degree of compatibility between the two modes remain
underexposed.

Evelyn Huber and John D. Stephens (2001): Development and Crisis of the Welfare
State. Parties and Policies in Global Markets
In many ways, this award-winning monograph on welfare state development in the second half of
the 20th century served as an early benchmark for mixed methods research in comparative public
policy. Both the quantitative and the qualitative parts of the approach followed here taken for
themselves, namely, testing theoretical arguments about long-term change in social policy pat-
terns and substantiating them by establishing agency and tracing processes, are done in a largely
convincing manner. Yet (results from) the MR analyses on 18 states and the in-depth studies of 9
countries could have been linked with each other a little more systematically and explicitly in
order to structure the dialogue between methods intended by the authors.

Torben Iversen (2005): Capitalism, Democracy, and Welfare


Linking the development of a new theoretical framework with a comparative-historical analysis
followed by four related, but distinct quantitative empirical studies, Iversen examines the
158 Journal of Mixed Methods Research 4(2)

interplay between investment in human capital, social protection regimes, and the international
division of labor. The particular strengths of this book lie in its innovative theoretical reasoning
and sophisticated formal model construction and testing, whereas the incipient comparative his-
torical analysis could have extracted the underlying causal processes more thoroughly and could
have dwelled on particular cases more than cursorily. In turn, the findings from the qualitative
part could have been taken up more systematically in the following quantitative studies.44
Among the further aspects not treated entirely satisfactorily, the absence of an explanation of
case selection (in combination with the varying number of countries looked at) is maybe the
most irritating.

Lane Kenworthy (2008): Jobs With Equality


Kenworthy analyses how and to what extent growth and low inequality are feasible at the same
time. After a theoretical discussion of the relevance of equality, the route via high employment is
identified as the most promising strategy to achieve it by different quantitative micro analyses.
Thereafter, various determinants of its success are analyzed using MR (and additional lower tech
quantitative tools) on macro data from 12 OECD countries. The level of reflection on the use of
quantitative methods is outstanding here (although the presentation of MR results is unorthodox),
and likewise is the discussion of explanatory aims and the (limited) generalizations intended.
Qualitative within-country evidence, however, is interspersed on a somewhat arbitrary basis,
and the insights gained by analyzing it as well as their relation to the quantitative analyses remain
less clear.

Isabela Mares (2006): Taxation, Wage Bargaining, and Unemployment


This book on the employment performance of advanced industrial economies starts out with the
very convincing construction of a formal model. This is followed up by MR analysis of a panel of
14 OECD countries testing the model, yet regarding estimation technique it simply refers to the
Beck and Katz’s (1995, 1996) standard. Then, three profound case studies on the developments
in Sweden, Germany, and the United Kingdom, chosen to reflect the different types of wage bar-
gaining systems, are presented. Most important, these studies illustrate how in all cases wage
moderation lost effectiveness as welfare states matured. Unfortunately though, the linkage
between the quantitative and the qualitative analyses and between their results is underdeveloped
in this book, both in absolute terms and in comparison to the quality of these analyses them-
selves: Mares offers only very scarce bridging passages between them, and instead of a synthesis
of her results she wrote an outlook on new social pacts in the concluding chapter.

Lessons to Be Learned and Some Suggestions


The overall picture that emerges from the review is a positive one. In all columns of Table 1, the
plus symbols outnumber the minuses, and in each case the application of different methods has
enriched the findings as well as brought the discipline forward. Yet the cross-sectional dimension
is of primary interest here: Looking at the rows in Table 1, we can identify the eight areas where
a majority of the books reviewed scored unsatisfactorily—this is where mixed methods research
in comparative public policy offers the greatest room for improvement. These areas are case
selection for quantitative analyses, the interpretation of significance tests in nonrandom samples,
the kind of causal relationships assumed, the level of generalizations aspired to, handling the
uncertainty of conclusions, and three aspects of the nexus between quantitative and qualitative
methods, namely: linkage, compatibility, and the relationship between the results from the
Wolf 159

different methods. Especially the latter three lie at the heart of triangulation strategies, and future
reasoning on research design would do well to focus on them. Regarding all three, I would like to
refer back to Sections ‘‘The Nexus’’ and ‘‘Questions and Answers,’’ whereas regarding linkage, I
would also like to offer some additional suggestions based on my own research. None of the
monographs reviewed used the relatively simple, but often highly instructive tools of residual
diagnostics. This surprised me, because I have found them to be extremely helpful devices to
link quantitative and qualitative parts with. In an analysis of the determinants of education spend-
ing in the German Länder (cf. Wolf 2006; Wolf 2007; Länder are the states in the German federal
system, with Land being the singular), for example, plots of residuals versus fitted values as well
as boxplots of residuals by countries and years45 demonstrated that explanatory success of MR
analyses was much lower (and residuals larger) regarding two Länder, Brandenburg and Thür-
ingen, and in particular years at the beginning of the period covered. Therefore I selected these
Länder (besides four others) for case studies and, among many other things, asked several pol-
iticians and experts in interviews how they would interpret this pattern. Thus I found out that
Thüringen had been the Land with the most ambitious school law in the transitory period
between the fall of the socialist regime and unification, that it had been the only Land that
had sustained the GDR (German Democratic Republic, the former socialist state on Eastern Ger-
man soil) tradition of day nurseries attached to schools (and paid from the school budget),46 and
that the minister of education had had a particular strong standing in his party, the Christian
Democratic Union (CDU). In Brandenburg, on the other hand, the Social Democratic Party
(SPD) minister of education had been one of the weaker cabinet members, and higher education
of this Land surrounding Berlin had been particularly restrained by the assumption that students
from Brandenburg would study in the capital as commuters. Although these insights did not gen-
erate generalizable conclusions (and were, strictly speaking, neither model building nor model
testing), they enabled me to paint a more encompassing picture of the topic under scrutiny,
and they also generated feedback for further MR analyses, for example, including the quota
of teachers with lifelong contracts as an additional independent variable.47 Of course, skew resid-
ual distributions should first of all trigger a search for omitted variables of theoretical substance.
But once we run out of these, and if we are interested in explaining as much of the variance in our
dependent variable as possible, it makes sense to trace such unit- or period-specific factors in
data points with large residuals. (In order not to overemphasize exceptional cases, however, I
also selected ‘‘on the line’’ cases and tried to trace the causal processes there as well.)
Another relatively simple measure that I found helpful—and did not come across in the mono-
graphs reviewed—is jackknifing regression models, that is, removing the Länder in turn from the
estimated model, comparing the results, and in this way identifying which unit is responsible for
which conclusion from the full model to what degree. Thereby I could demonstrate that the neg-
ative effect of public indebtedness depended on the ruinous financial situation of the Land Berlin
alone.48 Maybe all the authors whose books I looked into have silently employed this tool, yet if
so it would be a little surprising if none of them had found any noteworthy phenomenon of this
kind. Of course, these two examples by no means exhaust the list of possible ways to link dif-
ferent parts of mixed methods designs, yet I hope they spur readers’ imaginations to think of
other ones fitted to their research needs.

Conclusion: Toward Tailor-Made Triangulation?


This article set out to answer the question whether the trend toward triangulation and mixed
methods in comparative public policy is a development worth welcoming wholeheartedly, and
thus whether we should want to see more of the mixed methods research this discipline has
become known for. First of all, it can be stated that triangulation49 is promising in many research
160 Journal of Mixed Methods Research 4(2)

contexts, and in the felicitous (if not perfect) manifestations reviewed above it has certainly
broadened the scope of insights, and brought the discipline forward. Yet it is also a particularly
demanding research strategy. It is no panacea, and using it superficially just because it seems to
be trendy will not lead to satisfactory results. Simply presenting the evidence from macroquan-
titative analyses and case studies alongside each other without linking them adequately fails to
exploit the possible synergies to their full extent, and it also brushes aside critical methodological
issues. This rather critical assertion is not intended to discourage comparative public policy
researchers from applying different methods in general, however, but to advise them to do so
only with care and after considering the perils as well as the potentials involved. As one strategy
among others, mixed methods research definitely has a role to play. Yet before choosing it,
scholars must pose themselves the basic question whether they have the time that is necessary
to seriously engage in different methods, which will often include further training. Otherwise,
time restrictions will necessarily act to the detriment of analytical depth. (More often than
not, this will already be the decisive obstacle.) Then, in case they have decided to use triangu-
lative strategies, researchers should be aware of the steps that separate hazardous hotchpotches
from enlightened eclecticism: In addition to the criteria for good research in single-method con-
texts, these include clearly stating the (potentially various) aims pursued and the level(s) of gen-
eralizations aspired to, pointing out which kind(s) of causal relationship(s) are assumed, explic-
itly linking (sub)questions with methods, and discussing the relationship between the results
obtained from the different methods. And furthermore, or perhaps foremost, the nexus between
the qualitative and quantitative analyses needs to be carefully established. Here, my recommen-
dation is not to subscribe to ‘‘one-size-fits-all’’ solutions, but to use tailor-made triangulation
strategies fitted to the respective research topics, questions, and interests.50 By doing so, unlike
the valiant little tailor in the Grimm brothers’ fairy tale, scholars might not always succeed to
strike seven flies with one blow, but hopefully more analytical goals than otherwise.

Acknowledgments
The author would like to thank the directors and participants of the workshop on‘‘Methodological Plural-
ism? Consolidating Political Science Methodology’’ at the ECPR Joint Session of Workshops in Rennes,
April 11-16, 2008, as well as the editors and three anonymous reviewers for various comments and sugges-
tions. They helped to improve the paper a great deal.

Declaration of Conflicting Interests


The author declared no conflicts of interest with respect to the authorship and/or publication of this article.

Funding
The author received no financial support for the research and/or authorship of this article.

*Notes
1. In the course of the several revisions and discussions this article went through, I have learnt that many
readers feel uneasy if the author does not spell out his creed at the very beginning. So, somewhat hes-
itantly: I believe that at least some relevant social phenomena can be explained, that actors’ worldviews
and constructions of reality can be important explanatory factors of political processes (and their
results), and that trying to understand the meaning human beings attach to actions (and institutions,
and other things) can be a worthwhile endeavor. So probably this makes me some kind of realist, yet
I do not want to be part of a new paradigm that once again treats others as inferior. Rather, I would
Wolf 161

like to subscribe to a pluralist view that welcomes different basic assumptions as at least potentially
enriching in certain research contexts (cf. Moses & Knutsen, 2007, p. 289)—notwithstanding the
need to discuss possible incompatibilities within every concrete research design.
2. Qualitative or configurative comparative analysis and event history analysis, in contrast, have been
increasingly recommended, but not become as widespread so far.
3. One factor conducive to this development is the presence of several rival, yet principally combinable
theoretical accounts of public policies, highlighting, among others, the role of institutions, parties, inter-
est groups, and globalization.
4. Eclecticism by no means needs to be a derogatory label, yet it can be a more or less enlightened practice.
5. This definition is based on an analysis of 19 definitions from contributors to the literature (for a list of
these definitions, see Johnson et al., 2007, pp. 119-121). For examples of authors who distinguish
between mixed methods and mixed models, see Mastenbroek and Doorenspleet (2007, 6f.).
6. On the origins of the triangulation metaphor in techniques for locating one’s position, and for a critical
discussion of its usefulness in the social sciences, see Blaikie (1991, 117f., 122f.). In a similar vein,
Hammersley (2008) reminds us that
[i]n the case of navigation, the second measurement does not provide verification or validation of
the first, but rather is a necessary complement in order to identify relative location. [. . .] So, trian-
gulation in navigation is not a device for detecting and discounting error; indeed, any error in iden-
tifying the landmarks or calculating the bearings will vitiate the triangulation process. (p. 24)
7. It is not to be seen, though, as a technique that provides ‘‘guaranteed truth or completeness’’ (Hammers-
ley, 2008, p. 32). Furthermore, progress is not only made when the search for convergence is successful.
And in any way, it will indeed often be ‘‘a delicate exercise to decide whether or not results have con-
verged’’ (Jick, 1979/2008, p. 113), and researchers should beware of assuming so too readily.
8. Thus it can also be read as a contribution to the formation of a more self-conscious sub-‘‘community of
practice’’ (cf. Denscombe, 2008, p. 276).
9. This is illustrated by the number of methodological papers reanalyzing the works of other scholars (e.g.,
Beck & Katz, 1995, 1996, 2001; Green, Kim, & Yoon, 2001; Kittel & Winner, 2002; Plümper, Troeger,
& Manow, 2005; Wilson & Butler, 2007).
10. ‘‘Cause implies narrative and relational pathways. It is narrative because explanation must be embedded
in time and move through time. Indeed the success of any explanation resides in accounting for tempo-
rality and sequence’’ (Somers, 1998, p. 771).
11. A pretty typical rationale is the following: ‘‘We perceive the world only dimly, and all techniques for
testing causal theories are imperfect. In this context, we need more, not fewer, weapons in our meth-
odological arsenal’’ (Hall, 2003, p. 399).
12. While advocating ‘‘individual methodological pluralism’’ (Moses & Knutsen, 2007, p. 291), these
authors caution that not everyone should engage in it: ‘‘Specialists should avoid playing the role of
the general practitioner, and vice versa’’ (p. 290). Thus, they seem to suggest mixed methods research
as a unique form of specialization in its own right.
13. Shalev (2007a, 262) uses the apt term statistical literacy.
14. Pooling, that is, combining time-series and cross-sections into a data set with T × N observations, itself
needs to be justified in every research context it is applied to, because it is neither unconditionally indi-
cated nor always, but only ‘‘sometimes worth its price’’ (Stimson, 1985, p. 945)—especially in mixed
methods research.
15. Robustness ‘‘usually refers to the reliability of the derived estimates in the presence of a mistaken
assumption, such as the underlying normality of the distribution, dependence of the error on unobserved
effects, or misspecifying the structure of the error’’ (Sayrs, 1989, p. 62), and in the context of pooled
data this also includes ‘‘the degree to which an estimate does not change by the introduction of new
cases and time points, or by the omission of particular cases and time points’’ (p. 32).
16. Somewhat naively pleading for the presentation of single, simple models with no more than three
explanatory variables and seeing physics as a role model, as Taagepera (2007, p. 122) does, deprives
the reader of the possibility to assess the robustness of the results for herself. Furthermore, the striving
162 Journal of Mixed Methods Research 4(2)

for parsimonious models, as also advocated by King et al. (1994, p. 29), comes with an in-built spec-
ification error.
17. These should include jackknifing MR models, that is, removing units in turn from the analysis to iden-
tify which unit is responsible for which result to what degree, and what Pontusson (2007, p. 328) calls
‘‘moving windows analysis,’’ that is, reestimating the same model for different overlapping groups of
time periods.
18. For suggestions on how to do this, see, for example, Kenworthy (2007, 344f.).
19. Kenworthy (2007, p. 349) is one of very few authors (rightly) insisting on this.
20. ‘‘If your goal is to estimate average causal effects for large populations of cases, it makes sense to
include all types of negative cases and treat them as equally important for drawing conclusions about
causal effects. But if your goal is to explain outcomes in particular cases, it does not make sense to
select cases without regard for their value on the outcome.’’ (Mahoney & Goertz, 2006, p. 240)
21. It must be noted that although the concept of causal mechanism has found wide acclaim, there is no
single consensual definition of it. George and Bennett (2004) provide the following:

[W]e define causal mechanisms as ultimately unobservable physical, social, or psychological pro-
cesses through which agents with causal capacities operate, but only in specific contexts or condi-
tions, to transfer energy, information, or matter to other entities. In doing so, the causal agent
changes the affected entity’s characteristics, capacities, or propensities in ways that persist until
subsequent causal mechanisms act upon it. (p. 137)
On the one hand, moving beyond correlations of causes and effects in this manner broadens the
range of aspects of causality, thereby tapping ‘‘additional sources of causal inference’’ in the
Humean sense, namely ‘‘spatial contiguity and temporal succession’’ (George & Bennett, 2004,
p. 140). On the other, ‘‘it is precisely because connective mechanisms are unobservable—unlike
correlations of empirical indicators—that positivism has militantly rebuked their inclusion in the
realm of scientific theory’’ (Somers, 1998, p. 726). So invoking the concept of causal mechanisms
turns a researcher into a kind of realist, albeit without necessarily having to subscribe to its whole
paradigm. Inadvertently, though, one has to bear the ‘‘stigma of what classical positivist scepticism
[. . .] has considered the mere ’psychological appeal’ of ’ideas’ about causation (beyond conjunc-
tion)’’ (p. 746). What I spare you and me at this point is the potentially infinite regress caused by
asking how the ‘‘causal powers’’ referred to by George and Bennett were to be defined. Rather, let
us simply state that ‘‘it is very often useful to think in terms of mechanisms, and very often not’’
(Stinchcombe, 1993, p. 40)—it is up to individual scholars to decide and explain their decision
whether to do so or not in any particular research context.
22. For a discussion of inferences via congruence analysis, see Blatter and Blume (2007, pp. 8-13). This
approach focuses on ‘‘drawing inferences from the (non)congruence of concrete observations with pre-
dictions deduced from theories to the relevance of theories’’ (p. 8).
23. For advice on case studies with special emphasis on theory building, see also George and Bennett
(2004).
24. Shapiro (2005), for instance, has a lot to say on this topic.
25. Kenworthy and Hicks (2008), for example, present a volume that combines different methodological
approaches applied by different researchers to ‘‘a single substantive topic’’ (2008, p. 1), namely, the
employment performances of affluent countries. In their introduction, the editors explicitly state their
aim to ‘‘illustrate in practical fashion the advantages and drawbacks of these analytical strategies’’
(p. 1). Unfortunately, however, the volume does not include the chapter on ‘‘Methodological Conclu-
sions’’ that was originally announced (and neither the chapter on ‘‘Substantive Conclusions’’).
26. Lieberman (2005, p. 436) assumes an interest in ‘‘both the exploration of general relationships and
explanations and the specific explanations of individual cases and groups of cases.’’
27. Lieberman (2005, p. 440) invites this kind of criticism by explicitly stating that in his approach ‘‘[t]he
SNA [small-N analysis] should be used to answer those questions left open by the LNA [large-N anal-
ysis].’’ While Tarrow (1995, 471f.) charges King et al. (1994), and even more the following hype
around this book, for not doing justice to the genuine contributions qualitative analyses have to offer,
Wolf 163

Mastenbroek and Doorenspleet (2007, p. 6) also criticize Tarrow for not including in his list of ways to
combine methods the variant in which mixed methods research starts on the qualitative side.
28. Teddlie and Tashakkori (2003, p. 26) point out, however, that these typologies cannot be expected to be
exhaustive, given the diverse nature of mixed methods research.
29. Alternatively, Creswell and Plano Clark (2007, p. 81) use the term weighting. For a distinction of qual-
itative dominant, equal status, and quantitative dominant mixed methods research, see Johnson et al.
(2007, p. 124). Even within an equal status approach, however, the exact relation between the different
elements needs to be established.
30. For an early discussion that first distinguished these two dimensions, see Morse (1991/2008).
31. Shalev (2007b, p. 400) has a clear and general preference to start with qualitative studies.
32. Brady et al. (2006, p. 356) also argue that causal-process observations can be useful in various ways at
any stage of the research process, and that their ‘‘balance and sequencing’’ with data set observations
(their term for the currency of quantitative analyses) ‘‘may vary greatly.’’ The same can certainly be
said about case study evidence in general.
33. Here, researchers using pooled MR should also spell out in which ways they are interested in variation
between units and over time (cf. Kenworthy, 2007, p. 346), although both kinds of variation are often
interrelated, as Pontusson (2007) points out regarding the example of partisan differences (but also rel-
evant to many other variables):
[I]f we believe that government partisanship is a cause of cross-national differences in levels of
social spending, then we must also believe that government partisanship affects changes in social
spending in ways that do add up. In other words, time-series variation is just as relevant to the prop-
osition that government partisanship matters as cross-sectional variation. (p. 328)
34. Within the context of congruence analysis, for example, ‘‘crucial cases’’ are selected with regard to their
likelihood to reflect on the predictions flowing from a dominant theory (Blatter & Blume, 2007, p. 28).
35. Small-N comparisons aimed at causal inferences usually imply deterministic relationships, yet demon-
strating high levels of probability can sometimes justify the use of a quasi-deterministic model (cf. Lie-
berson, 1991, 318f.). According to Mahoney (2000, p. 389), ordinal strategies in small-N comparisons
are ‘‘more compatible with a probabilistic understanding of causation and, thus, more consistent with
the assumptions that guide large-N research.’’ To adopt both a deterministic and a probabilistic under-
standing within the same (mixed methods or methodologically pure) research project, however, would
be ‘‘not logically consistent’’ (p. 408) in his eyes. For a thorough discussion of this aspect, see also Kühn
and Rohlfing (2009).
36. The accusation of selection bias, more frequently raised against qualitative studies, is risked only when
the latter are formulated too ambitiously. Or, put differently, ‘‘the ’thickness’ of case studies creates
only a trade-off [between extensity and intensity] when we want to generalise towards a wider popula-
tion of comparable cases’’ (Blatter & Blume, 2007, p. 1).
37. Whether ‘‘political inquiry that is simultaneously scientific and critical, rigorous and heterodox, struc-
tured and patchwork,’’ as championed by Topper (2005, p. 181), is able to fulfill this requirement would
have to be judged by its concrete realizations.
38. According to Rueschemeyer (2003, p. 330), a proponent of those, ‘‘[t]he most important advances in
social and political theory [. . .] are not found in empirically testable and tested specific propositions
of universal applicability but in theoretical frameworks of tested usefulness for the study of particular
areas of analysis.’’
39. As concerns the compatibility of concepts, this should also neither be taken for granted nor denied out of
hand, as Mastenbroek and Doorenspleet (2007, p. 13) do as well.
40. For further discussion, see Moses and Knutsen (2007, pp. 288-291).
41. Simply ignoring this issue is indeed ‘‘untenable’’ (Read & Marsh, 2002, p. 240)—demanding ‘‘episte-
mological accountability’’ (Somers, 1998, p. 755) is not asking too much, but an important criterion for
good research.
42. Please note that the authors of these works would not necessarily claim that their books are intended
to meet the criteria I assess them by. The books all share (or are at least compatible with) a realist
worldview—without necessarily declaring so.
164 Journal of Mixed Methods Research 4(2)

43. Qualitative evidence from other countries is additionally offered along the course of the argument.
44. Admittedly, though, the rational choice institutionalism the book is based on probably implies less inter-
est in the not-so-generalizable.
45. These are two graphical ways to link MR with case studies (and select cases for in-depth studies) not
mentioned by Lieberman (2005).
46. This factor alone accounted for more than 10% of the unexplained variation for this Land.
47. Because the Länder do not have to pay contributions to the major social insurance systems for these
teachers, current (as opposed to future) spending is lower where there are more of them.
48. Yet another way of assessing the leverage of particular data points is, of course, computing dfbetas.
49. As was discussed in the introduction, the terms mixed methods research and triangulation are used
interchangeably throughout the article. (For a detailed review of possible meanings of triangulation
as well as a critique of the conflation of triangulation and mixed methods research, again, see Hammers-
ley, 2008.)
50. For similar, albeit much more solidly grounded advice on the choice of methods in general, see Moses
and Knutsen (2007, p. 15).

References
Amenta, E. (2003). What we know about the development of social policy: Comparative and historical
research in comparative and historical perspective. In J. Mahoney & D. Rueschemeyer (Eds.), Compar-
ative historical analysis in the social sciences (pp. 91-130). Cambridge, UK: Cambridge University
Press.
Author a: see file ‘completetext’
Author b: see file ‘completetext’
Beck, N., & Katz, J. N. (1995). What to do (and not to do) with time-series-cross-section data in compar-
ative politics. American Political Science Review, 89, 634-647.
Beck, N., & Katz, J. N. (1996). Nuisance vs. substance: Specifying and estimating time-series-cross-section
models. Political Analysis, 6, 1-36.
Beck, N., & Katz, J. N. (2001). Throwing out the baby with the bath water: A comment on Green, Kim, and
Yoon. International Organization, 55, 487-495.
Berg-Schlosser, D. (2006). Political science in Europe: Diversity, excellence, relevance. European Political
Science, 5, 163-170.
Blaikie, N. W. H. (1991). A critique of the use of triangulation in social research. Quality & Quantity, 25,
115-136.
Blatter, J., & Blume, T. (2007, September). Beyond the co-variational template: Alternative directions in
case study methodology. Paper presented at the 4th ECPR General Conference, Pisa, Italy.
Brady, H. E., Collier, D., & Seawright, J. (2006). Toward a pluralistic vision of methodology. Political
Analysis, 14, 353-368.
Brewer, J., & Hunter, A. (2006). Foundations of multimethod research. Synthesizing styles. London: SAGE.
Collier, D., Brady, H. E., & Seawright, J. (2004). Critiques, responses and trade-offs: Drawing together the
debate. In H. E. Brady & D. Collier (Eds.), Rethinking social inquiry. Diverse tools, shared standards
(pp. 195-227). Lanham, MD: Rowman & Littlefield.
Creswell, J. W., & Plano Clark, V. L. (2007). Designing and conducting mixed methods research. Thousand
Oaks, CA: SAGE.
Creswell, J. W., Plano Clark, V. L., Gutmann, M. L., & Hanson, W. E. (2008). Advanced mixed methods
research designs. In V. L. Plano Clark & J. W. Creswell (Eds.), The mixed methods reader (pp. 161-196).
Thousand Oaks, CA: SAGE. (Original work published 2003)
Dellinger, A. B., & Leech, N. L. (2007). Toward a unified validation framework in mixed methods research.
Journal of Mixed Methods Research, 1(4), 309-332.
Denscombe, M. (2008). Communities of practice. A research paradigm for the mixed methods approach.
Journal of Mixed Methods Research, 2(3), 270-283.
Dye, T. R. (1976). Policy analysis: What governments do, why they do it, and what difference it makes.
Tuscaloosa: The University of Alabama Press.
Wolf 165

Ebbinghaus, B. (2005). When less is more: Selection problems in large-N and small-N cross-national com-
parisons. International Sociology, 20, 133-152.
Erzberger, C., & Kelle, U. (2003). Making inferences in mixed methods: The rules of integration. In
A. Tashakkori & C. Teddlie (Eds.), Handbook of mixed methods in social & behavioral research
(pp. 457-488). Thousand Oaks: SAGE.
Esping-Andersen, G. (2007). Multiple regression in small-N comparisons. In L. Mjøset & T. H. Clausen
(Eds.), Comparative social research: Vol. 24.Capitalisms compared (pp. 335-342). Amsterdam: JAI
Press.
George, A. L., & Bennett, A. (2004). Case studies and theory development in the social sciences. Cam-
bridge: MIT Press.
Gourevitch, P. A., & Shinn, J. J. (2005). Political power and corporate control. The new global politics of
corporate governance. Princeton, NJ: Princeton University Press.
Green, D. P., Kim, S. Y., & Yoon, D. H. (2001). Dirty pool. International Organization, 55, 441-468.
Greene, J. C. (2008). Is mixed methods social inquiry a distinctive methodology? Journal of Mixed Methods
Research, 2(1), 7-22.
Hall, P. A. (2003). Aligning ontology and methodology in comparative politics. In J. Mahoney &
D. Rueschemeyer (Eds.), Comparative historical analysis in the social sciences (pp. 373-404). Cam-
bridge, UK: Cambridge University Press.
Hall, P. A. (2008). Systematic process analysis: When and how to use it. European Political Science, 7,
304-317.
Hammersley, M. (2008). Troubles with triangulation. In M. Bergman (Ed.), Advances in mixed methods
research: Theories and applications (pp. 22-36). Thousand Oaks, CA: SAGE.
Huber, E., & Stephens, J. D. (2001). Development and crisis of the welfare state. Parties and policies in
global markets. Chicago: University of Chicago Press.
Iversen, T. (2005). Capitalism, democracy and welfare. Cambridge, UK: Cambridge University Press.
Jick, T. D. (2008). Mixing qualitative and quantitative methods. In V. L. Plano Clark & J. W. Creswell
(Eds.), The mixed methods reader (pp. 107-120). Thousand Oaks, CA: SAGE. (Original work published
1979)
Johnson, R. B., Onwuegbuzie, A. J., & Turner, L. A. (2007). Toward a definition of mixed methods
research. Journal of Mixed Methods Research, 1(2), 112-133.
Kenworthy, L. (2007). Toward improved use of regression in macro-comparative analysis. In L. Mjøset &
T. H. Clausen (Eds.), Comparative social research:Vol. 24. Capitalisms compared (pp. 343-350).
Amsterdam: JAI Press.
Kenworthy, L. (2008). Jobs with equality. New York: Oxford University Press.
Kenworthy, L., & Hicks, A. (2008). Introduction. In L. Kenworthy & A. Hicks (Eds.), Method and sub-
stance in macrocomparative analysis (pp. 1-28). New York: Palgrave Macmillan.
King, G., Keohane, R. O., & Verba, S. (1994). Designing social enquiry. Scientific inference in qualitative
research. Princeton, NJ: Princeton University Press.
Kittel, B. (1999). Sense and sensitivity in pooled analysis of political data. European Journal of Political
Research, 35, 225-253.
Kittel, B., & Winner, H. (2002). How reliable is pooled analysis in political economy? The globalization-
welfare state nexus revisited (MPIfG Discussion Paper 02/3). (This paper was also published 2005 in
European Journal of Political Research, 44(1))
Kritzinger, S., & Michalowitz, I. (2008). Methodologische Triangulation in der europäischen Policy-
Forschung (Methodological Triangulation in European Policy-Research). In F. Janning & K. Toens
(Eds.), Die Zukunft der Policy-Forschung. Theorien, Methoden, Anwendungen (The Future of
Policy-Research: Theories, Methods, Applications) (pp. 191-210). Wiesbaden, Germany: VS Verlag.
Kühn, D., & Rohlfing, I. (2009, September). Causal explanation and multi-method research in the social
sciences. Paper presented at the 105th APSA annual meeting, Toronto, Ontario, Canada.
Levy, J. S. (2007). Qualitative methods and cross-method dialogue in political science. Comparative Polit-
ical Studies, 40, 196-214.
Lieberman, E. S. (2005). Nested analysis as a mixed-method strategy for comparative research. American
Political Science Review, 99, 435-452.
166 Journal of Mixed Methods Research 4(2)

Lieberson, S. (1991). Small N’s and big conclusions: An examination of the reasoning in comparative stud-
ies based on a small number of cases. Social Forces, 70, 307-320.
Luoma, P. (2003, September). New options in crossing the methodological borders: From quantitative to
qualitative analysis and vice versa. Paper presented at the 2nd ECPR General Conference, Marburg,
Germany.
Mahoney, J. (2000). Strategies of causal inference in small-N analysis. Sociological Methods & Research,
28, 387-424.
Mahoney, J., & Goertz, G. (2006). A tale of two cultures: Contrasting quantitative and qualitative research.
Political Analysis, 14, 227-249.
Mahoney, J., & Rueschemeyer, D. (2003). Comparative historical analysis: Achievements and agendas. In
J. Mahoney & D. Rueschemeyer (Eds.), Comparative historical analysis in the social sciences (pp. 3-
38). Cambridge, UK: Cambridge University Press.
Mares, I. (2006). Taxation, wage bargaining, and unemployment. Cambridge, UK: Cambridge University
Press.
Marsh, D., & Furlong, P. (2002). A skin, not a sweater: Ontology and epistemology in political science. In
D. Marsh & G. Stoker (Eds.), Theory and methods in political science (pp. 17-41). Basingstoke, UK:
Palgrave Macmillan.
Mastenbroek, E., & Doorenspleet, R. (2007, September). Mind the gap! On the possibilities and pitfalls of
mixed methods research. Paper presented at the 4th ECPR General Conference, Pisa, Italy.
Morgan, D. L. (2007). Paradigms lost and pragmatism regained: Methodological implications of combining
qualitative and quantitative methods. Journal of Mixed Methods Research, 1(1), 48-76.
Morse, J. M. (2008). Approaches to qualitative-quantitative methodological triangulation. In V. L. Plano
Clark & J. W. Creswell (Eds.), The mixed methods reader (pp. 151-158). Thousand Oaks, CA:
SAGE. (Original work published 1991)
Moses, J., & Knutsen, T. L. (2007). Ways of knowing. Competing methodologies in social and political
research. Basingstoke, UK: Palgrave Macmillan.
Moses, J., Rihoux, B., & Kittel, B. (2005). Mapping political methodology: Reflections on a European per-
spective. European Political Science, 4, 55-68.
Munck, G. L., & Snyder, R. (2007). Debating the direction of comparative politics: An analysis of leading
journals. Comparative Political Studies, 40, 5-31.
Onwuegbuzie, A. J., & Johnson, R. B. (2008). The validity issue in mixed research. In V. L. Plano Clark &
J. W. Creswell (Eds.), The mixed methods reader (pp. 273-298). Thousand Oaks, CA: SAGE. (Original
work published 2006)
Pickel, S. (2009). Die triangulation als methode in der politikwissenschaft. In S. Pickel, G. Pickel, H.-J.
Lauth, & D. Jahn (Eds.), Methoden der vergleichenden Politik- und Sozialwissenschaft. Neue
Entwicklungen und Anwendungen (pp. 517-542), Wiesbaden, Germany: VS Verlag.
Plümper, T., Troeger, V., & Manow, P. (2005). Panel data analysis in comparative politics: Linking method
to theory. European Journal of Political Research, 44, 327-354.
Pontusson, J. (2007). Methods in comparative political economy. In L. Mjøset & T. H. Clausen (Eds.),
Comparative social research: Vol. 24. Capitalisms compared (pp. 325-333), Amsterdam: JAI Press.
Read, M., & Marsh, D. (2002). Combining quantitative and qualitative methods. In D. Marsh & G. Stoker
(Eds.), Theory and methods in political science (pp. 231-248). Basingstoke, UK: Palgrave Macmillan.
Rihoux, B. (2006). Two methodological worlds apart? Praises and critiques from a European comparatist.
Political Analysis, 14, 332-335.
Rohlfing, I. (2008). What you see and what you get. Pitfalls and principles of nested analysis in comparative
research. Comparative Political Studies, 41, 1492-1514.
Rueda, D. (2007). Social democracy inside out. Partisanship and labor market policy in industrialized
democracies. Oxford, UK: Oxford University Press.
Rueschemeyer, D. (2003). Can one or a few cases yield theoretical gains? In J. Mahoney & D. Ruesche-
meyer (Eds.), Comparative historical analysis in the social sciences (pp. 305-336). Cambridge, UK:
Cambridge University Press.
Sartori, G. (2004). Where is political science going? Political Science & Politics, 27, 785-787.
Sayrs, L. W. (1989). Pooled time series analysis. Newbury Park, CA: SAGE.
Wolf 167

Scruggs, L. (2007). What’s multiple regression got to do with it? In L. Mjøset & T. H. Clausen (Eds.), Com-
parative social research: Vol. 24.Capitalisms compared (pp. 309-323). Amsterdam: JAI Press.
Shalev, M. (2007a). Limits and alternatives to multiple regression in comparative research. In L. Mjøset &
T. H. Clausen (Eds.), Comparative social research: Vol. 24. Capitalisms compared (pp. 261-308).
Amsterdam: JAI Press.
Shalev, M. (2007b). Rejoinder: Affirming limits and defending alternatives to multiple regression. In
L. Mjøset & T. H. Clausen (Eds.), Comparative social research: Vol. 24. Capitalisms compared
(pp. 391-409). Amsterdam: JAI Press.
Shapiro, I. (2005). The flight from reality in the human sciences. Princeton, NJ: Princeton University Press.
Siegel, N. A. (2007). When (only) money matters: The pros and cons of expenditure analysis. In J. Clasen &
N. A. Siegel (Eds.), Investigating welfare state change. The ‘‘dependent variable problem’’ in compar-
ative analysis (pp. 43-71). Cheltenham, UK: Elgar.
Somers, M. R. (1998). ‘‘We’re no angels’’: Realism, rational choice, and relationality in social science.
American Journal of Sociology, 104, 722-784.
Stimson, J. A. (1985). Regression in space and time: A statistical essay. American Journal of Political Sci-
ence, 29, 914-947.
Stinchcombe, A. L. (1993). The conditions of fruitfulness of theorizing about mechanisms in social science.
In A. B. Sørensen & S. Spilerman (Eds.), Social theory and social policy. Essays in honor of James S.
Coleman (pp. 23-41). Westport, CT: Praeger.
Swank, D. (2002). Global capital, political institutions, and policy change in developed welfare states.
Cambridge, UK: Cambridge University Press.
Swank, D. (2007). What comparativists really do. In L. Mjøset & T. H. Clausen (Eds.), Comparative social
research: Vol. 24. Capitalisms compared (pp. 361-372). Amsterdam: JAI Press.
Taagepera, R. (2007). Predictive versus postdictive models. European Political Science, 6, 114-123.
Tarrow, S. (1995). Bridging the quantitative-qualitative divide in political science. American Political Sci-
ence Review, 89, 471-474.
Tarrow, S. (2004). Bridging the quantitative-qualitative divide. In H. E. Brady & D. Collier (Eds.), Rethink-
ing social inquiry. Diverse tools, shared standards (pp. 171-179). Lanham, MD: Rowman & Littlefield.
Tashakkori, A., & Creswell, J. W. (2007). Editorial: Exploring the nature of research questions in mixed
methods research. Journal of Mixed Methods Research, 1(3), 207-211.
Tashakkori, A., & Teddlie, C. (2008). Introduction to mixed method and mixed model studies in the social
and behavioral sciences. In V. L. Plano Clark & J. W. Creswell (Eds.), The mixed methods reader (pp. 7-
26). Thousand Oaks, CA: SAGE. (Original work published 1998)
Teddlie, C., & Tashakkori, A. (2003). Major issues and controversies in the use of mixed methods in the
social and behavioral sciences. In A. Tashakkori & C. Teddlie (Eds.), Handbook of mixed methods in
social & behavioral research (pp. 3-50). Thousand Oaks, CA: SAGE.
Topper, K. (2005). The disorder of political inquiry. Cambridge, MA: Harvard University Press.
Wilson, S. E., & Butler, D. M. (2007). A lot more to do: The sensitivity of time-series cross-section analyses
to simple alternative specifications. Political Analysis, 15, 101-123.
Wolf, F. (2006). Die Bildungsausgaben der Bundesländer im Vergleich. Welche Faktoren erklären ihre
beträchtliche Variation? Münster: LIT Verlag. (The German Länders? Educational Expenditure: Which
Factors Explain their Substantial Variation?)
Wolf, F. (2007). Die Bildungsausgaben der Bundesländer:Bestimmungsfaktoren und sozialpolitische Rel-
evanz. Zeitschrift für Sozialreform, 53(1), 31-56. (The German Länders? Educational Expenditure:
Determinants and Relevance in Terms of Social Policy Translation of Journal Name: Journal of Social
Policy Reserach)

You might also like