gerring

AARHUS UNIVERSITET

FORSIDESKEMA Koplseet/kompendier til undervisningsbrug
kopieret efter ophavsretligt beskyttede udgivelser, jf. Rammeaftale af 1O.marts 2008 meJlem Danske Universiteter og Copydan om eksemplarfremstHling af ophavsretligt beskyttet materiale http:Uwww.au.dk/da/15-03.htm. De vigtigste uddrag af aftafen findes under skemaet (samt links til ydertigere information).

1. 2. 3.

Fakultet: Instltut: Titel a) Tidsskriftets navn eller bogens titel b) Artiklens tite! eller - ved redigerede boger - det pagceldende bid rags titet

Samfundsvidenskabeligt Institut for Statskundskab, Aarhus Universitet Case Study Research. Principles and Practices

4.
5. 6. 7. 8. 9. 10.

Forfatter(e) Forlag Ud 9 ivelsesa r

John Gerring Cambridge 2007 978-0-521-85928-8 University Press

ISBN/ISSN
Antal sider i seettet Antal eksemplarer Afgiftspligtige kopisider i alt

131

-

144

=

1411

-

=

a II

-

=

0

52
728

Undertegneder der har anmodet om - eller selv har foretaget - trykning/kopiering eller indskanning og digitallagring af dette undervisningsmaterialet, erklaerer herved, at fremstillingen efter mit bedste sken ikke indebaerer en kreenkelse af ophavsretten eller den indgaede aftale mellem Copydan og Danske Universiteter. • KOPIERET MATERIALE:
Koplseettet trykkes i reproafdelingen Hovedregler for kopieret materiale:
o o o o Af veerker, der er under 1 ar gamle, ma man hejst kopiere 10%, dog ikke over 25 sider Er veerket 1-5 ar gam melt, rna man heist kopiere 15%, dog ikke over 40 sider Er veerket eeldre end 5 ar, ma man hejst kopiere 20%, dog ikke over 75 sider Leesmere her ...

eller

pa instituttet efter rekvisition

,

.

(seer X)

INDSKANNET MATERfAlE:
Materiale indskannes af instituttet
o o o o

09 lag res pa password-beskyttet

Intranet (AULA)

,.

(seer x)

Hovedregler for indskannet materiale:
Der kan indskannes 20%, dog hejst 20 sider af en publikation proelev pr. semester Filerne skallagres pa et password-beskyttet intranet (fx AULA med adgang kun for deltagere) Udprint fra betatingsdatabaser samt aviser og dagblade ma ikke indskannes la:>smere her ...

24. januar 2011 dato

Lasse Lindekifde underskrift

Et eksemplar af dette forsideskema skal veerevedheeftet hvert enkelt solgt eller udleveret kopiseet. eller indskannet materiale Et eksemplar af forsideskemaet opbevares i institutsekretariatet. Et eksemplar af kopiszettet med forsideskema indsendes - efter aftale med fakultetet - til COPY"DAN.

Case Study Research Principles and Practices JOHN GERRING Boston University CAMBRIDGE UNIVERSITY PRESS .

rate. for The most-similar method. »--: Sometimes. may change as a researcher moves from an exploratory to a confirmatory mode of analysis.udies effect Techniques for Choosing Cases 131 e cases. The hope is that intensive study of these] cases will reveal one . or a tional given of a ~ autor (as uceds that treme riding lantly t hway r. 1975).ase IS tables cases r.'~~t. This sort of case illustrates the ca.. and yet demonstrate surprisingly different outcomes. in which case her research design is confirmatory (hypothesis-testing) from the get-go." after its inventor (Mill 1843/1872).ib~t~-d"~. . as illustrated in Table SA (A). The point is that the purpose of a most-similar research design. Lijpharr (1971. Often.~"~'~. Most-Similar Case vari(X1 is .:come. That is. it should be noted that when researchers refer to a particular case as~<!n"exampJe" of a broader phenomenon.'th. Gerring (2001: Chapter 9). If the study is ~ (i. r facr eth- vould rocra-neral Gould . Przeworski and Teune (1970). regardless of oned. However. as well as fodder for an exploration of causal mechanisms.sal relationship of interest in a particularly vivid manner. This is a common form of case selection at the initial stage of research.e. unlike the previous methods. the chosen pair of cases is similar in all respects except the variable(s) of iQterest. and therefore may be regarded as a common trope among case study researchers. she strives to identify cases that exhibit different scores on the factor of interest and similar scores on all other possible causal factors. _- section of the chapter: case-selection procedures often combine different logics. ~rated in tIie second (hypothesis-testing) diagram in Table SA (B).factors that differ across these cases. Despite the technical nature of this discussion. employs a minimum of two cases.or at most several . If she discovers such a case. For later treatments see Cohen and Nagel (1934). this pathuding 71 Sometimes the most-similar method is known as the "method of difference.. they are often referring to a ~y case. . the researcher looks for cases that differ on the outcome of theoretical interest but are similar on various factorstIi'at l~ighth'~~. Eggan (1954). and Skocpol and Somers (1980). a researcher begins with a strong hypothesis. and hence its basic set-up. fruitful analysis begins with an apparent anomaly: two cases are apparently quite similar. hypothesis-generating). These differing factors (Xl) are the putative causes. Meckstroth (1975).van[5. it is regarded as providing confirmatory evidence for the proposition." In its purest form.

Both the United States and Canada inherited English political cultures. persistent over many years. II. Hamilton Lipset (1968). Consequently. Canada has highly disciplined parties whose members vote together on the floor of the House of Commons. Y = the outcome of interest. let us consider Leon Epstein's classic study of party cohesion. the notion of a "most-similar" analysis is usually understood as a tool for understanding a specific X1/Y relationship.72 Several caveats ___________. Epstein first discusses possible causal factors that are held more or less constant across the two cases. when published. to P( el se ti: 11l t. see Brenner (1976). Doing Case Studies Most-similar analysis with two case types (Y-centered): Te Oil str (A) Hypothesis-generating fe( In re: r" (B) Hypothesis-testing (X/Y-centered): th (0 Xl controls. while the United States has weak. the United States and Canada. both are federal. n For further examples of the most-similar method. Moulder (1977). both have large territories and heterogeneous populations.132 TABLE SA. As an example. while the United States is presidential. look like a hypothesistesting research design. Where they differ is in one constitutional feature: Canada is parliamentary.. a b 51 ti o ( a e a b a ( . apply to any most-similar analysis (in addition to the usual set of assumptions applying to all case study analysis). undisciplined parties whose members often defect on floor votes in Congress. And it is this institutional difference that Epstein identifies as the differentiating cause (XIl. "l ru ~ P( where one begins. These are the "control" variables (X2). Miguel (2004). which focuses on two similar countries. Question marks have been removed: (A) becomes (S) in Table SA. the results. In explaining these divergent outcomes.. First. and Posner (2004). (1977). and both have a fairly loose party structures with strong regional bases and a weak center.

: Hamilton (1977).aving strong regional bases of power. This can be a ) huge advantage over large.~Cl_sure_contr.tl'l.g. However. Specifically~ neces.. and Y) or virtually identical (on X2). At the same time. The problem of coding continuous variables in a dichotomous manner is threatening to any most-similar analysis. She assumes.oJ~f!. they are more or less constant across the two cases. even to those who are members of the opposing party). Arguably. and in these instances the logic of most-similar comparison becomes questionable. whatever they are. it sets up a most-difficult research scenario.. In one respect. the empirical universe does not always oblige the requirements of Millean-sryle analysis. it is often the case that variables of concern in the model are continuous (e. this element of nonidentity does not challenge Epstein's conclusions. present/absent). where each case must be r() assigned a specific score on all relevant control variables . dominated in the latter twentieth century by direct primaries (open to all who declare themselves a member of a party and. Unfortunately. For example. y t(J. in some states.N cross-case methods. a factor that is probably more significant in recent Canadian history than in recent American history. Epstein's description of Canadian and American parties as "loose" might be questioned. because regional bases of power should lead to weaker parties. that these unknown factors have been neutralized across the treatment and control groups by randomization. Epstein describes both the United States and Canada as '. American parties. This is straightforward if the underlying variables are also dichotomous (e. as discussed earlier. rather than to stronger parties. Indeed. rather. First. party cohesion). These are : constitutional is presidential.tent over many 'e held more or es and Canada tories and het~ a fairly loose :nter. are considerably more diffuse than Canadian parties.g. This is relatively unproblematic if the actual scores on this dimension are quite different (on X. the researcher must "dichotomize" the scoring of cases so as to simplify the two-case analysis. one must code cases dichotomously (high/low. One can simply assert that.~jsar abies (at least not with a high degree'q~clsion) in order tQ_conJ. federal/unitary). This is similar to the technique employed in a randomized experiment.often a highly questionable procedure...:£Q_Uor Ii) ~f two countries can be assumed to have similar cultural heritage one needn't worry about constructing variables to measure that heritage. However. es as the differin addition to analysis). and one that must impose strong assumptions . where the researcher typically does not attempt to measure all the factors that might affect the causal relationship of interest. the requirements for case control are not so stringent..ezgCase Studies Techniques for Choosing Cases 133 ~ a hypothesisd: (A) becomes liar" analysis is X11Y relationstudy of party United States ~members vote e United States :t on floor votes . however. In this setting. Some flexibility is admissible on the vector of controls (Xl) that are "hcld constanf" across the cases:1\JOi1Td~i~~ tolerable if the deviat~ runs counter to the predicted hyPothesis.

Hence. fact as tl are i 1 olo~ .1 Statistical estimates of causal effects based on matching techntques have been a major topic in quantitative methodology over the last twenty-five years. a simple 1 difference-of-means test is often sufficient to analyze the effects of a treatment variable (Xl) across groups.q. in observational studies it is unusual to find cases that differ on Xl but not on various background characteristics (Xl) that might affect the outcome of interest. 76 Ho et at. countries that are stro~~ democratic (or strongly authoritarian) are Ii e y to be simi ar in more t~_anone res pest. 75 Hahn (1998). Thus are observational studies translated into the lexicon of experimental analysis. 81 ." Matching techniques are based on an extension of experimental logic.- scor simi on t see Ho et al. For example. abler the c - marc science. experiment. the treatment group and the control group have a high probability of being similar in their background characteristics (X2). In observational studies where the hypothesized causal factor (Xj ) is dichotomous. and to cases with "low" scores as members of the control group. Morgan and Harding (200S). for a large enough selection of cases. elaborate statistical models are unnecessary for causal inference because. 74 Rosenbaum and Rubin (1985). For purposes of discussion. Rosenbaum (2004). first in statistics?" and subsequently in econornetrics " and political !h0 ~ . the situation is superficially the same. (2004).== a 77 H e) e) a{ . But this standard-issue technique requires a strong set of assumptions about the behavior of the various factors introduced into the model. see Abadie et aL (2001). 73 For good introductions. Imai (2005). we shall refer to cases with a "high" score on Xl as members of the treatment group. This great y complic~ t e ana ysis of Xl'S indepen ent effect on the outcome. However. For a discussion of matching procedures in Stata. Rosenbaum (2004).£ va vana Cross-Case Technique The most useful statistical tool for identifying cases for in-depth analysis in a most-similar setting is some variety o~matchin~" strategy.134 about the shape of the underlying to be linear). The traditional approach to this problem is to introduce a variable for ea~J2_otential~~r in a regression model of causal relationships. (2004). II. and Rosenbaum and Silber (2001). Matching techniques have been developed as an explicit alternative to broa the t vide set 0 l~ In a randomized case! like]: l +u knofails may exar GDI the J II inste grm are. Doing Case Studies causal relationship (usually presumed Tech the c 9.

This procedure typically fails for continuous variables such as wealth... This approach suggests a somewhat different iefinition of 'similarity than the previous two. age. By contrast. Rather than focusing on sharing scores on the matching variables. but not all. f:!!o~r_"">.~he cases are to be matched.U. . The benefits of matching extend only so far as equivalence on the variables explicitly included and any unmeasured variables that fortuitously happen to be similar across the cases. matching is clearly inferior to a well-designed and well-executed randomized experiment. olirical .-'-"'''-'''-!!:!::1 the treatment roup. itive to senbaum zdures in ~::. methodologists have developed an alternative procedure known as. nnecesf cases. propensity-score matching focuses or. researchers may instead employ ap roximate matching. as well as the fact that. In situations where exact matching is infeasible. Major weaknesses of this approach include the fact that the definition of "close enough" is inevitably arbitrary. the treatment-group and the matching control-group cases should provide a good estimate of the causal effect.sharin~ a similar estimate£! probjlbility of havW been~atment group. and distance. es have tty-five the control-variable approach. known as exact matching. the researcher identifies as many cases as~ble from t e contr grou WIt the exact same scores on the matching variables (t e covariates). Then. To deal with situations in which exact matching is impossible. since there may be no two cases with the same score on a continuous variable. Note that the larger the number of covariates. Even in a situation in which the set of matching variables includes some.::o 77 However. proper randomization handles all unmeasured variables. few treatment cases are likely to have even approximate matches.2ropensitj!: score matching. the average difference between . the lower the likelihood of finding exact matches. For example. the researcher looks at the difference a the dependent variable between the cases in the treatment group and the matching cases i~ the control group.Studies esumed Techniques for Choosing Cases 135 h analltegyJ. If the set of mat~hing variables iSl' broad enough to mclude all confounders.. Finally.~ing may . ility of simple a treatis of disbers of : of the lexicon (Xl) t differ 1 ariable -lationssurnpmodel. the relatively simple matching procedure just described. there is no undemocratic country with the exact same per capita GDP as the United States...I logic.e_roduce better causal inferences than regression models because ~ases that match on a set of explicitly selec~ariables are also more likely to H sImIlar on unmeasured confoundersr II Unfortunately. This alternative begins by identifying a set of variables (other than the dependent variable or the main independent . confounders.. for large sets of matching variables. is often impossible. where cases from the control group t at are close enough to matching cases from the treatment group are accepted as matches.

The final step in the process is to choose matches for each case in the treatment group.. so they can be examined and their stories can be told [ individually.r dee~ppropriate. This is accomplished by a two-stage analysis. Modern matching methods involve statistical modeling and combinatorial algorithms. conditional on its scores on the matching variables. ~ dependent variable and the matching variables as independent variables. the variable of theoretical rest. at least on average. Xj. Instead. the researcher wishes to select a case is as similar as possible to Costa Rica in background variables. are the "most-sImilar" ca'SeS. 78 berwe the ca plex. cities. thick and than cedu in tl diffe outc tatic of tl Indeed.would have been as likely to be in the treatment group as actually chosen cases. Suppose that. Thes.4: crati most racy. doz to ( assi . .'~o the qualitative termi'nology. Ibid. the first stage of which approaches the key independent variable. pro. ostensibly comparable patterns are compared directly. in matching. democracy. people retain their integrity as people. or indeed any paired cases in the social sciences. researchers look for cases in the control group that . This is accomplished by selecting cases from the control group with similar propensity scores.before the score on the independent variable is known .136 II. and the outcome of interest. may Matching.79 the authors. where a two-stage approach to causal inference is adopted. Rosenbaum and Silber summarize the results of recent medical studies: Unlike model-based adjustments. In matching. political parties. which tell us the probability of that case being assigned to the treatment group. where patients vanish and are replaced by the coefficients of a model. the same matching techniques that have been used successfully in observational studies of medical treatments might also be adapted to the study of nation-states. carter try's 1 syster legal navia distal Th ~ (the i able. but the end result is a collection of pairs or sets of people who look comparable. Doing Case Studies Techn In I conditional on the matching variables. (This is similar in spirit to selection models. in order to study the relationship between wealth and democracy. the coefficient estimates are disregarded. while ng as different as possible on per capita GDP. rather than inhibits. The end result of this procedure is a set of matched cases that can be compared in whatever way the ~rcl. the second stage of the analysis employs the fitted values for each case. "facilitates. when looking for ~atch for a specific case in the treatment group. and ( prop.) Once this model has been estimated. In other words. 7B 79 hisn me 11 dep: Rosenbaum and Silber (2001: 223). prop' quite of $1 $5. These fitted values-are referred to as propensity scores. conclude description. one by one.

lie.tist~. At the same time. German legal heritage. An in-depth analysis of these two cases may shed light on the causal pathways between economic development and democracy. one's assumptions are rendered transparent. if not thousands of cases vie for inclusion . Costa Rica. one must ~rri~e at ~ st.-.63.5. and cases with similar propensity scores are interpreted as matching.-rof the causes of a country's wealth. as a t variables. is 7. A superficial model like the one used here may produce rather superficial matches. al studies: placed by the are compared nodchng and irs or sets of people retain . A country's wealth will be assumed to be a function of the origin of its legaD system (measured by dummy variables for English legal heritage. Examining the Propensity-score data. the of the analobability of m its scores as propenor each case es from the :es that can riate. Since this is an illustrative example. Costa Rica and Benin may be viewed as most-similar cases for testing the relationship between wealth and democracy. Indeed.a formal approach to case selection offers significant advantages. and Scandinavian legal heritage) and its geographic endowments (measured by the distance of each country's capital city from the equator).58 quite similar to Costa Rica's. socialist legal heritage. thick successfully dapted to the paired cases relationship select a case . one sees that Benin has a propensity score of 7. These ermi'nology..where dozens.s can be told Techniques for Choosing Cases 137 In order to select most-similar cases for the study of the relationship between wealth and democracy. At the very least. and historical experience). Yet.r~~-d. Hence.486. It is important to keep in mind that the quality of the "match" depends entirely on the uaiit of th -1 sed to ~Wrte t le propensity scores..163 is substantially different from Costa Rica's per capita GDP of $5. fJf _J . as are their democracy scores in 1995 (Benin is much less democratic than Costa Rica). while )f theoretical . French legal heritage. region. such a proposition is complex. we shall be satisfied with a cartoon model that includes only a few independent variables. This means that the differences on the variable of theoretical interest (GDP per capita) and the outcome (democracy) can be given a causal interpretation . in a large-N context .an interpretation that would probably not be suggested by a qualitative assessment of these two countries (which are quite different in culture. The propensity score for our focus case.iables. Xl.ase Studies looking for rs look for ndenr varinent group ie analysis. e approach imated. Obviously. Benin's per capita GDP of $1. these two cases are probably more informative than other two-case comparisons precisely because the case-selection procedure has identified countries whose other attributes are roughly equal in their propensity to democracy/authoritarianism. The first step in selecting most-similar cases is to run a nonparametric (> regression with thes 'nde en ent variables and logged per capita GDP ( (the independent variable of theoretical interest) as-the dependent vari: able. as illustrated in Table 5. The fitted values from this regression serve as propensity scores.- I hibits.

the case study performs a somewhat different ro e. It may e per ect y reasona e to ppropriate this large-N method of analysis for case study purposes. Epstein'S study of American ~d Canadian political parties is notable ~for its principal finding. Here. the usual purpose of a case study analysis in this setting is to corroborate an initial cross-case finding. Yet. as suggested in the foregoing discussion. that is. Sl c: o o t( d .st Conclusion The most-similar method is one of the oldest recognized techniques of qualitative analysis. harking back to J. the current popularity of matching among statisticians . as identified by previous research or theoretical hunch. the general causal relationship is usually clear. matching statistics are a relatively new technique in the arsenal of the social sciences. Perhaps the causal pathways from economic development to regime type are difficult to identify.relative.138 TABLE 5. S. If Rosen ba urn and Sil ber are correct. Mill's classic study. By contrast. we will be more interested in die CoVariational patterns that disc ed between Xl and Y. and have rarely been employed for the purpose of selecting cases for in-depth analysis. To be sure.based" approach to causal analysis. We know from our cross-case study that GDP per capita is strongly associated with democracy. Indeed. the purpose of a case study is somewhat different in situations where a large-N cross-case analysis has already been conducted. the case study analysis may give us reasons to doubt. II.at least none of a formal nature . Doing Case Studies Paired cases resulting from matching procedure th tit hi B) ca re hi pI ra . to garden-variety regression models . Here.5. there is a strong presumption of causality..] S( are . System of Logic (first published in 1834). Even so. are simply not in evidence. there may be a fruitful interchange between the two approaches. that the underlying cause of party cohesion is to be found in the structure of it! re pi III q I ul Sf a( JU r. there is no rior cross-case investigation . By contrast . Thus.rests upon what qualitative researchers would recognize as a "case. Of course. Perhaps the presumed causal pathways.

As with other methods of case selection. his principal focus is on "scoring" the relevant variables. idy. if Epstein had already conducted a large-N crosscase analysis prior to his case study.N method I erent in situaen conducted. By the same token." following its inventor. Evidently. 1975). Moore (1966). 'e know from . the most-similar method is ~ prone to problems of non-representativeness. Meckstroth (1975). itself. the case e causal pathult to identify. the problem of potential bias can be addressed by assuring a choice of cases that are not extreme outliers. the function of the most-similar case study shifts subtly but importantly when the case-selection procedure is.N sample. S ecifically. as well as the dependent variable (Y). covary. vious research the usual purIrate an initial .sociated with curse. t. a mode of analysis. fruitful interpopularity of variety regresd recognize as and Silber are rge. J. as discussed.hough some scope f~r deviance around the regression line may be acceptable for purposes of finding a good fit among cases. Mill (1843/1872). see Collier and Collier (199112002).. Karl (1997). 80 The most-different method is sometimes referred to as the "method of agreement. :terns that are American and ding: that the e structure of Most-Different Cases A final case-selection method is the reverse image of the previous method. Skocpol (1979). However. For examples of this method. Case Studies Techniques for Choosing Cases 139 techniques of the executive (parliamentary/presidential). as Judged by their residuals in the full model. variation on independent variables is prized. Przeworski and Teune (1970). Most-similar cases should also be"typical" cases. Converse and Dupeux (1962). and Yashar (2005: 23). Rather than looking for cases that are most-similar. If this technique is employed in a qualitative fashion (without a systematic cross-case selection strategy). If the researcher emplo s a matcmng technique of case selection within a large.at least none ~e. Gerring (2001: 212-14). potentia! biases in the chosen cases must be addressed in a spec. . his two-case analysis of the United States and Canada (cases that we presume would have very similar propensity scores) would now serve a rather different purpose. System of tics are a reIa- rd have rarely epth analysis . ~~ ulatlve way. Epstein spends relatively little time in this article discussing possible causal mechanisms. and if this cross-case analysis had revealed a strong pattern between executive type and party cohesion. most of these studies are described as combining most-similar and most-different methods. S. Indeed. while variation on the outcome is eschewed. offering strong prima facie evidence of a causal relationship. one looks for cases that are most-different. while all other lausible factors X show different values. See also Def-elice (1986). Lijphart (1971. and Skocpol and Somers (1980). Here. the researcher tries to identify cases where just one independent variable (Xl).

Yet they both score near the bottom of various cross-national indices intended to measure the prevalence of civic engagement in the current era.6.140 TABLE 5. Doing Case Studies Most-different analysis with two cases Techm organi ment x numb: genng isfacti deteri: develc Re~ sumec Bulga: solves ~ Howe the in nist 0 societ civil s varia] Xl = the variable of theoretical interest. .civil society.~\iI!Q~enti_gLr~~l_~~~~ As an example. during the Soviet era. Thus. X2a~d = a vector The simplest form of this two-case comparison is illustrated in Table 5. Howard wonders why this relationship is so strong and why it persists. which explores the enduring impact of communism 01). help to explain the laggard status of civil society in post-communist polities: "the mistrust of communist Howard (2003). In order to answer this question. Howard claims. Howard's case selection procedure meets the requirements of the most-different research design: variance is found on all (or most) dimensions aside from the key factor of interest (communism) and the outcome (civic engagement). In the following discussion I treat the terms "social capital. stratified samples of Russians and East Germans. _s intere -." though they are s!l1]:Lli!! j~. as East Germany was absorbed into West Germany. 82 What leverage is brought to the analysis by this approach? Howard's case studies combine evidence drawn from mass surveys and from indepth interviews of small. in countries that are no longer socialist or authoritarian.F' Cross-national surveys show a strong correlation between former communist regimes and low social capital. of how quantitative and qualitative evidence can be fruitfully combined in the intensive study of several cases. of controls. These two countries were quite different .--IS gen 1actor ~ that i: ~ w"""1thl are m N{ on th intern ety rn exhai M dilem insigl ---ciViC I ""tf:i'eC Indee well meth 83 84 81 82 Ibn rbi. Cases A and B are deemed "most-different." and "civic engagement" interchangeably. incidentally. Russia and East Germany. I follow Marc Howard's recent work. It is a strong result. Howard (2003: 6-9).) The product of this analysis is the identification of three causal pathways that. and in the post-Soviet era.6." "civil society. Y = the outcome of interest. controlling for a variety of possible confounders. (This is a good illustration. II. he focuses on two most-different cases. and perhaps even strengthens.in all ways other than their communist experience prior to the Soviet era.

.. it is generally difficult to reach conclusions about the causaIStatUS~ ~s on the basis of the most-different analysis alone. the author employs two other small-N cross-case methods. "84 Results obtained from the analysis of East Germany and Russia are presumed to apply in other post-communist polities (e. years a ter their disappearance. not just communist ones. and other factors does not exhaust this subject. Indeed." "civil organizations. they "'i're not properly regarded as causal..agement). East Germany may be regarded 83 84 1 ~l Ibid. a general satisfaction with their own personal networks (accompanied by a sense of deteriorating relations within society overall).communism and its putative causal pathways..82 . Howard solves potential problems of re~ .83 Simply put. Bulgaria. the persistence of friendship networks. .Jl_J2le~ g . Indeed. based on the foregoing methodological dilemmas.ation of three laggard status of communist al capital. He wishes merely to show that whatever influence on civil society might be attributed to economic. and the disappointment with post-communism.Ehe outcome.at ave never experienced communist rule.hiLr. Ibid. Nor does it seem possible to conclusively eliminate rival hypotheses on the basis of this most-different analysis. ~ that is that the three causal athways identified by Howard also operate wit in po mes t . 122.:erest .&iu . isJ!!at the most-differen_t~n.tri~$_~ . ~ntativene. Poland. Equally problematic is the lack of variation on key causal factors of in. vhicI. this is not Howard's intention. by choosing a heterogenous sample. 145.nexplor~ s-naflbnal surst regimes and ounders. Howard concludes. Fortunately. this is not the only research deslgn employed by Howard in his admirable study..es. he focuses on two countries st experience est-Soviet era. My considered judgment. ley both score to measure the Howard's case ferent research E: from the key .g Case Studies Techniques for Choosing Cases 141 :rated in Table 'ley are ~il!!ilit! i. The chosen sample is truncated [censored] on the dependent variable). If so. ! However.. which is intended to cover all countries. Lithuania. these methods do most of the analytic work. For this reason. Indeed. . this sample is not representative across the entirepopulation of the inference. and disappointment in the developments of post-communism.. "a great number of citizens in Russia and Eastern Germany feel a strong and lingering sense of distrust of any kind of public organization. as wen as a large-N cross-country statistical analysis. It is a trong and why are no longer . cultural.£!ovides only minimal insight into the problem of why communist s stems a ear to Sil ress CIVIC engagement.ch? Howard's and from inand East Geriantitative and :nsive study of . In my opinion. (To argue that communism impedes the development of civil society is to imply that noncommunism stimulates the development of civil society.. Albania).g.

close association with Western European culture and politics).. Gerring 2001. Howard's conclusions are justifiable.) Ibid. It has all the attributes normally assumed to foster civic engagement (e. Karl usually relies on contrasts between petro-states and non-petro-states (e.g.g. there is little left upon which to base an analysis of causal relations (aside from the large-N cross-national analysis). In my opinion. Chapter 10). e. Yet if one strips away the pathway case (East Germany) and the most-similar analysis (EastlWest Germany). a growing economy. showing that some countries (e. Indonesia) manage to avoid the pathologies brought on elsewhere by oil resources.. most scholars who employ the most-different method do so in conjunction with other merhods. Karl (1997). but nonetheless shows little or no improvement on this dimension during the post-transition era.. as Howard does. Moore (1966).. a stand-alone merhod. This was recognized. 23). Surely. The first concerns the (usually) obstructive role of oil in political and economic development. by Mill (1843/1872: 258-9).142 II. The contrast between East and West Germany provides a mostsimilar analysis. Doing Case Studies Technic Let! lean st] the ana cases ill mous f close e as a causal-pathway case (as discussed earlier). issues that formed the basis of Chapter Three. Skocpo! (1979).. This variation is also deftly exploited by Howard.. and Yashar (200S. 8. Karl (1997). a free press. but not on the basis of mostdifferent analysis. I do not wish to dismiss the most-different research method entirely. which affects to be a most-different system analysis (20). The second SOrt of inference concerns variation within the population of petro-stares. very little use is made of the mostdifferent research design. Indeed. makes two sorts of inferences. 2 88 Anol .~~If ~ do not dently that th sary c. explicit defenses of the most-different method are rare (but see DeFelice 1986). (1 leave aside issues faced by all case study analyses. Collier and Collier (1991/2002).g. at least implicitly. if ever. considr lar can higher IS cons enough a judgi analysi In a . In short. When attempting to explain the constraining role of oil on petro-states. Cohen and Nagel 1934: 251-6. 85 It is plausible to attribute this lack of change to its communist past.V present interpr. See. Howard's findings are stronger with the intensive analysis of Russia than they would be without.. I offer the following summary remarks on the most-different method of case analysis.g. Her study. not Xl is not: 85 86 Usu cation this se tions ( grven r -the pn ~ 87 COl ease. Skocpol and Somers 1980). Conclusion Generalizing from this discussion of Marc Howard's work. Norway. since the two polities share virtually everything except a communist past. multiparty competition. focused ostensibly on petrostates (states with large oil reserves). Indeed. elemen elimin: uig ne the sat Table. Only when attempting to explain differences among perro-states does she restrict her sample to petro-srares.g. Skepticism has been echoed by methodologists in the intervening years (e. civil liberties. is a particularly clear example of this.I" It is rarely.

they are both low.but also no less . Howard assumes that this divergence is minimal enough to be understood as a difference of degree rather than of kind.6. Recall that.. as Howard ivides a mostryrhing except }y Howard. there is Ins (aside from 'S who employ r methods. ~h most-similar analysis. Usually. but only if assumptions of "causal uniqueness" hold.than most-similar analysis. in Table 5. sociation with )WS little or no ra.. The problem of "degrees" is deadly if the variables under 1 consideration are by nature continuous (e. Skocpol (1979). In Table 5.g. this would be a set of cases that satisfy conditions X2a-d. Thus. but not Xl. the most-diff~~~-~-. However. high/low. even the presumed strength of the most-different analysis is not 50 strong. present/absent). a judgment that might be questioned.g. This' is a particular concern in Howard's analysis. high/high. Xl is necessary. X2a-d in Table 5. In this setting. not on the elimination of possible causes. most-different analysis is no more secure .cases must be sizeable enough to be interpretable in an essentially dichotomous fashion (e. differen~es acr~-~~ . one would presumably seek out cases that register the same outcomes and have maximum diversity on other attributes. The first concerns . -ent system analysis stensibly on petro.e.( Case Studies the 'attributes ling economy. In one respect.6.g.t" It Techniques for Choosing Cases 143 . Note that the defining feature of this method is the shared element across cases . the most-different technique is useful. In basis of mostethod entirely. GDP).are evidently unnecessary for the production of Y. )f the most-different Let us begin with a methodological obstacle that is faced by both Millean styles of analysls~'::: the necessity orai-tROfomizing every variable in the analysis..6 . case study analysis is focused on the identification (or clarification) of causal relations. and similarities must be close enough to be understood as essentially identical (e. This feature does not help one to eliminate necessary causes. By this I mean a situation in which a given outcome is the product of only one cause: Y cannot occur except in t e presence of 1.pment. if one were focused solely on eliminating necessary causes. where East Germany scores somewhat higher than Russia in civic engagement. res that formed 6). In these respects. though Russia is considerably lower. Otherwise the results of a Millean-sryle analysis are not interpretable.g. most-different analysis is superior to most-similar analysi~~ If the coding assumptions are sound. it does not follow that the most-different method is the best method for eliminating necessary causes. present/present). alysis of Russia iway case (East nany). The second itates.85 It is plaust.X. showing that iologics brought on lining role of oil on nd non-petro-states ng petro-states does s made of the mostSkepticism has been Nagel 1934: 251-6. and in some situations (given certal f 88 Another way of stating this is to say that X is a "nontrivial necessary condition" of Y. Indeed..rch _siesign maybe useful foi 'iEminat~n& !1ecessary cau$~ausal factors that do not appear across the chosen cases . . I offer the folIf case analysis.

has many causes. as Marc owar does. It supposes t at the isease arises from the sam ' In any setnnz: 1S may e a reasonable operating assumption when one is dealing WIth certain natural phenomena like diseases. one of which has developed the disease and the other of which has not. For this reason. This is the most-different research design. the most-different meth. However. for example. one might wish to encompass a wide varier of revol . and SOCiologitstshav1e multiple f causes. it would not occur to us to look for most-different cases of high mortality around the world. And it is for this reason that most-different analysis is rarely applied in social science work and._______. Death. From this perspective. This assumption rarely holds in social scientific settings. the most-different desi n demands stronger assumptions about e rl trig . There are many ways to win an election. to get into a war. A second approach focuses on (highly dissimilar} communities where the disease has appeared across the two countries and searches for any similarities that might account for these similar outcomes. In this setting. to overthrow a government.J" (so-called) might be better labeled a diversecase method. several new cases of the disease surface in a single community. to bui d a we are state. and we can imagine epidemiologists employing them simultaneously.actors at work. Even so. If the unusual o~ is revolutlOn. is rarely convincing. as explored earlier. Arguably. If this seems a tad severe. The first examines two similar communities within Country A. This is the most-similar style of case comparison. I the unusual outcome is post-communist civil society. I. In Country B.to build a strong civil society. for most outcomes of interest to anrhro- o In be it 01 (c id tc m S( c SJ b S< f. located at the other end of the world. we can imagine two sorts of Millean analyses. economists. this is not a pure "method" at all but ( merely a supplement.the factor held constant across the diverse cases is the only possible cause of Y (see Table 5.. . . where applied. I .144 II. in one's ana ysis. a way of incorporating diversity in the subsample Qf _ cases that provide the u~sual outcome of interest. there are many exceptions. and focuses accordingly on the identification of a difference between the two cases that might account for variation across the sample. there is a more charitable way of approaching the most-different method. Both are plausible approaches to this particular problem.6). f f t s \"1 . political scientists.returning to Marc Howard's work . pologists. the researcher must assume that Xl . In order for the most-different research design to effectively identify a causal factor at work in a given outcome. Doing Case Studies Tt hundreds of infected persons across dozens of affected communities in that country. or .it seems appropriate to include a diverse set of post-communist olities in one's samp eo case studies.

Sign up to vote on this title
UsefulNot useful