of causation suspect at best. In his analysis of causation, Mill presented tliree conditions liiat could be satislied in order to demonstrate a causal connection between 'variables':
(1) The Method of Agreement To paraphrase Mill, this method argues that, if the several observations of the phenomenon under investigation (dependent variable) have only one of several possible causal circumstances in common, then the circumstance in which all the instances agree is the cause of the phenomenon.
(2) The Method of Difference Again paraphrasing Mill, this is tlie argument that, if there is an occurrence and a non-occurrence of a phenomenon (llic dependent variable), and the circumstances in which these are observed (independent variables) are (lie same in all circumstances save one, then that one is the cause of tlie occurrence.
(3) The Method of Concomitant Variation If two variables tend to vary in the same pattern, then they are somehow linked, either causally or through some other pattern of connection (linked to some third variable).
These tliree methods for coping with the problem of causation were devised long before contemporary social science, and indeed Mill did not consider that his methods were appropriate for the social sciences. He (like many scholars now) believed tliac tlie complexity of the causal relationships encountered in social enquiry limited tlie possibility of discovering meaningful causal relations. Despite Mill's own cautions, his three approaches [o solving (lie problem of causation will be echoed rather strongly in some of the researcli methods discussed later in the book (see Chapter 9), as well as in the discussion of fundamental problems of comparative analysts in this chapter.
The basic problem for Mill, as well as for tlie contemporary social researcher, is to isolate, as far as possible, a factor or a limited number of factors that appear to produce (or at least arc strongly associated wnli) changes in the dependent variable. This problem is, in turn, a problem ofidciilifving possible ways to exclude the numerous possible confounding factors in llie relationship between variables a problem of llie greatest theoretical importance. The identification of tlie confounding influences is also in part a function of having adequate theoretical guidance about where to look for llic oilier variables (see Chapter 5). There are any number of possible causes for changes in

political conditions - ihe secret is to use theory to identify llic most likely sources of confounding variance.
Comparative Research Design
A more contemporary approach to the same problem addressed by Mill can be expressed in terms of the variance observed in the dependent variable in comparative research. Tlie fundamental litany for social research, which will keep appearing in the course of this volume, is:
Maximise experimental variance, minimise error variance, and control extraneous variance.
This is a relatively easy phrase to repeat, but a substantially more difficult task to implement in comparative research designs. In each dependent variable that we utilise for comparative analysis, there is always some of each of these three types of variance - llic question is how much. Empirically, it may be impossible to parcel out variance into these three categories. This impossibility is true, because of (he numerous possible sources of extraneous variance, and the numerous sources of error in any measurement, and the associated difficulty in knowing when any particular observation is aneclcd by error. Despite tlie impossibility of knowing how to attribute all the observed variance, the researcher must still examine his or lier methods with this basic question of research design in mind.
Expci imental variance is the observed diHi-reilces or changes in the dependent variable that arc a function of ilie independent variables identified as central to the analysis. For social-scientific analysis, even research of a qualitative nature, llierc must be a dependent variable -something \ve are attempting to explain. In some research situations, tlic dependent variable may be a simple dichotomy - revolutions occur or they do not fWickham-Crowlev, 1991), or people vole or they do not (Butler r/ ai., 1981). In oilier cases, the dependent variable ni;iy lie a range ol responses of governments to changing economic fortunes ((.oiirevilch, 1986). In otiier.s. it may be a con-[inuou'i variable, such as per capita expenditures for social welfare in a range of rouiuries (Hicks and -Swank, ty92; Castles and Mitchell,

1993) that appears much more like what we usually think of as a variable. For qualitative case studies, the dependent variable may be changes in observed configurations uf political power, or tlic nature of the decisions taken by political elites, with those ditlerenccs often being very subtle.
Whatever (lie nature of variable in llie research, tlic investigator is attempting to ensure lliat two characteristics exist for tliat dependent variable. One is to be certain that it docs vary. Some published political science research does not, in fact, utilise a dependent variable tliat varies - at least in tlic data collected for [lie purposes of the particular research. For example, a number of tlic recent studies ofdcniocratisation have examined only those cases tliat have democratised successfully, however that success is defined (Bova, 199]}. There is always (lie assumption tliat (lie oilier end of the continuum was diucrcnl, but tliat assumption was not really tested, ll may be that the variables used to 'explain' successful democralisation may actually have roughly tlie same values for those countries tliat did not democratise successfully.
This practice of selecting cases on llieir values on tlie dependent variable has been argued to be one of the mortal sins that can afflict comparative research (King el at., 1994), but tike most sins it is one which is practised frequently. It appears to make sense to examine cases that satisfy some criterion of relevance, but in fact this simply eliminates crucial variance in tlie phenomenon to be explained, Even were relationships to be discovered between successful dcmocratisa-tion and any presumed independent variables, we would not know if they were any different from the relationships tliat might be found for unsuccessful dcmoeratisations, or for authoritarian regimes llial did not attempt to democratise at all. These may simply be features of systems tliat have been authoritarian at one particular point in time. About the best result attainable using only cases with the same value on (lie dependent variable is to demonstrate tliat some independent variable appears to be a necessary condition for tlic occurrence of the dependent variable (Most and Slarr, 1989: 49-51).
As well as varying, il is good to have dependent variables that vary a lot. While this issue may itself be largely a part of tlie general question of measurement (sec Chapter 4), il is also a question for research design and case selection. If we waul to develop robust theoretical slalemenls that arc meaningful across a range of different types of circumstances (and we certainly do) then il is wise to begin the research willi lliat wide arrav of cases. Researchers would like to

develop, for example, propositions that work not only for the wealthy industrialised world but also for at least tlie newly industrializing countries (NIC.S - Singapore, Soiilli Korea, Taiwan, and so on), and iT possible (hi- the less developed countries as well- Attempting to develop theories with a wide range of applicability is difficult (see pp. 1 1 7-26), but we should begin with this as a goal, and therefore be concerned in most research situations with selecting as wide a range of cases as possible. The 'as possible1 in tliat sentence then raises the issue of the judgment of the researcher, and [lie existing knowledge about variance within the research environment. In the end, there may be no realistic substitute for a researcher's good judgment,
Minimising error variance is less of a problem for case selection. Error variance is that portion of the variance observed in the dependent variable that is a function of random occurrences and errors in measurement. It is the slip of the pen of the data recorder, or by tile data entry technician, or sloppy fieldwork on the part of an interviewer, or a host of other occurrences that can produce data points that arc just inaccurate. In addition, no measurement of a social phenomenon is a perfect indicator of that underlying characteristic, so a certain amount of measurement error is almost inevitable. For qualitative research, this error variance could result from faulty observations by the researcher, or the misinterpretation of behaviours in different cultural settings, or jet lag, or again a whole range of things that can produce inaccurate observations.
Statistically, error variance is assumed to have a mean of zero, since it is random, and hence is assumed not to bias most statistical manipulations. Tlic comparative researcher cannot anbrd to be so sanguine about the effects of error. If the cases are selected inappropriately, or (lie observations made bv the researcher are faulty, there is litlle or no way to recover from the errors. Any results arising from research with such a poor design and execution are likely to be misleading, and therefore in some ways worse than no results at all. This potential for error simply argues for even greater attention to the selection of cases, and a very searching consideration of the comparisons ilial are planned at the beginning of [he research, Further, there needs to be extremely careful weighing of the modes of data collection to ensure thai they arc indeed compatible in the range of societies within which they ;irc being applied, and that the researcher is capable of doing the research in those countries.
Finally, there is tlic question of controlling extraneous variance. This is even more a concern (or [lie comparative researcher, because

this is systematic rallier than random error. With extraneous variance, there arc one or more variables that have a systematic relationship wilh the dependent variable, and perhaps also with the independent variables in an analysis. A researcher may find a relationship (statistically or less quantitatively) between A" and T. The problem is that lliere is a variable .<", that is systematically related to botli A' and V. For example, we may find thai democratic political systems spend more money on social welfare than do non-democratic regimes. The problem is that levels of economic development tend to be related to both democracy and welfare spending. Wealthier countries nnd it easier to maintain democracy and also to spend money for social programmes. Thus, assuming tliat democracy produces the welfare state may be only partially correct, if at all, but that relationship should rather be tested in the presence of a measure of the level of economic development.
In comparative research, there arc an almost infinite number of opportunities for extraneous variance to creep into the analysis. Researchers tend to be focusing upon whole countries or on large subnational governments, all of which come as 'data bundles' containing a huge number of variables and characteristics. John Stuart Mill referred to the nation as a 'permanent cause', which no amount ol additional variables would permit one to eliminate. When selecting a country for inclusion in a comparative investigation the researcher will receive all of those factors at once- Attempting to determine, therefore, whether the true 'cause' of the observed state of the dependent variable was the independent variables or not is extremely problematic. Statistically, at least, any variable lliat differentiates cases at the level ol the country (or other geopolitical entity) is as likely to cause the outcome as any other variable having the same effect. In llie conduct of comparative political research, minimising anrl/or detecting extraneous variance is a ditlicult problem, but not a totally insurmountable one.
One o( tlie nrsi defences against extraneous variance is theory- As Faure (1994: 313) points out, 'ocomparative method is therefore entirely dependent upon pre-existing criteria ol relevance such as concepts, propositions and llieories. These can be tested by the method but not discovered by tlic method.' If lliere is good theory in the research area then it should help the investigator determine whether the right control variables arc being considered for a statistical analysis, or whether the appropriate factors are being considered in .selecting llie cases in a more strictly comparative

design. Faure may have been too pessimistic given that, particularly willi statistical research, finding inconsistent or unexpected results may lead to the search for concepts that explain those findings, but, in general, researchers arc dependent u]x>n existing conceptions ol'llie political world,
Tins reliance on theory may he a catch-22 for the research - if the theory is already so good, why do we have to do the additional research? Further, can tlie theory ever be really tested if we assume that its guidance in the selection of cases and control variables is always reliable? Thus, if the research in question is to remain within the 'normal science' mode fKuhn, 1970), and assume tliat tlicrc is a reliable body of accepted and verified evidence, then existing theory is a good guide. If the research is more exploratory, or is intended to challenge the existing theories, llien (lie researcher is more alone in tlic liostilc world of extraneous variance. Again, the discovery or 'anomalies' (Kulin's word) may lead to the questioning and tlien improvement of the theory.
A second possible defence against the problems of extraneous variance is the use of time-series analysis, or using subnationa! units from within a single country. Neither of these approaches to analysis is a perfect prophylactic against extraneous variance, but each does provide some limited protection. Using lime-series analysis will do two useful things in the detection and potential elimination of extraneous variance- One is to allow for the detection of systematic relationships other than those being tested through the analysis. Suppose, for example, a researcher is using time-series regression and finds a significant correlation between tlie dependent and independent variables. That is very nice, but it is only the beginning of the research exercise. There are a variety of'statislical methods to lest whether tlic residuals (tlie difference between [lie actual and the predicted values of each case) of the regression analysis arc themselves systematically correlated across time. If there is significant autocorrelation then this systematic error in tlic relationship can be eliminated, cither through removing it by means of statistical manipulations, or by identifying the variables causing the relationship and including them directly in the analysis.
Examining residuals in this manner is most commonly done in time-series analysis, but a similar logic could be used to identify systematic error in other types of regression analysis- If [lie observations can be ordered by any variable in addition lo those included in tlic analysis, ;ind if tlic residual', are llien examined and are (bund to have a discernible pallern, llien tlierc is an obvious danger of

extraneous variance contaminating [lie findings. For example, assume there were ;> regression study of German lender that related per capita spending on education to income. The results of that analysis would yield residual errors, and these might be ordered by a potentially confounding factor, for example, tlic percentage of Roman Catholics in llic population (Wilcnsky, 1974). Of course, a more useful statistical result could be achieved through including the presumed confounding variable as a part of tlic regression equation, and determining how it performs relative to others explanatory variables. The virtue of examining tlic rc-siduals directly, however, is that they may present a pattern lliat can be discerned visually quite easily, but which would be more difficult to detect statistically. For example, a curvilinear pattern might not be as easy to detect statistically as witli a simple plot ol'residuals. Further, noting wlicre in the time-series cliange in tlie autocorrelation-occurs may suggest the source of tlic extraneous variance.
Using subnational units from within a single country also may lielp eliminate some problems of extraneous variance. As has been argued above, these units will tend to be more similar on a number of social and cultural dimensions than will the countries in cross-national research. Therefore, some of the potential sources of extraneous variance will have been eliminated bv selection of that locus of research. Tlic problems witli this strategy arc, among others, its relevance and tlic potential for a false sense of security. First, many theoretical relationships in which a researcher in comparative politics is interested will not be relevant in a subnational context. For example, legitimacy is largely a question for national governments, not for subnational governments, In many governments, however, subniitional units as a group may have more legitimacy than the national government. Tlie increasing disdain for (lie federal government in tlic United Slates may reflect one notable case.
Also, assuming that tlie problem ofextrancous variance is solved by using subnational units within a single country can lull the researcher into a false sense of security. There are often still significant dilfer-cnce.s among subnational units, even in countries that may appear to the casual observer to be reasonably homogeneous politically, fur example, tlic differences between Sicily and Lombai-dy in Italy. Therefore, extraneous relationships may well lie present in subna-tiona] data without llie researcher being as sensitive to them as lie or she might in a cross-national analysis, and therefore unnecessary error can creep into the analysis.

Research Design and Case Selection
In the bcsl of all worlds, we might be able to do experimental research to lest tlic propositions advanced in comparative politics. This is not however ilie best of all worlds, and there are a huge number of practical and ethical limitations on the capacity to expcrimciu on people and governments. Therefore, we are led with attempting to find the best possible substitute for the rigourous controls provided by [he experimental method. Tlie questions of case selection and rescarcli design here are presented as the substitute we have for not being able lo manipulate variables and randomise case assignments. Any non-experimental design is subject to a number of threats to validity (see below), and therefore political science researchers arc in the position of merely attempting to do the best they can, given die circumstances, 10 prevent contamination of their research findings,
The above solutions for die problem of controlling extraneous research are ways of skirting the principal issue, which is how to cope with this problem through the more direct means of research design and case selection. We will talk about how many cases and the problem of small As in the following chapter. Here, the principal question is not how many cases but which ones, although these issues cannot be separated entirely, Again, if we return to the fundamental differences between a comparative design and a statistical design, for the comparative design we must deal with the issues of controlling the sources of variance in the ex ante selection of the cases, rather than through ex post manipulations ofdaia. Comparative design tends to rely upon fewer cases, but ones tliat are selected purposefully rather than at random.
The mosi basic question is, \Vhal makes cases comparable? (See Lijpharl, 1975b). It is difllcull lo attend any academic meeting on comparative politics without hearing at least once (he phrase, 'But those cases really are not comparable.' What does that statement mean, and what criteria slioufd lie used for determining if political systems are indeed comparable? Some social scientists (Kalleberg, 19G6) liave been enamoured of the criteria assumed to apply in tlie natural sciences, and liave argued for very strict standards of comparabilily. In tills view, cases could not compared adequately unless they shared a substantial number of common properlics. This argument was an attack on the siructural-fiinclionalisl'i, such as Gabriel Almond, wlio sought lo include virtually all national cases

beneath their comparative umbrella (Almond and Coleman, 1960;
Almond and Powell, 196G). Structural-functionalism used concepts that were so broad tliat almost any political system could be compared, but sceptics wondered if (hat truly was comparison, or simply putting essentially incomparable cases into a common, and extremely nebulous, framework.
For the purposes of comparative politics, these criteria from the natural sciences (if indeed they arc actually operative there) are almost certainly too restrictive (DcFelice, 1980). First, we may well want to compare cases that display a certain properly with those that do not. Wlial factors appear to separate democratic from non-democratic political systems (I.ipsel, 1959; Prxcworski, i995a, b), or countries tliat experience revolutions from (hose tliat do not (Wickham-Crowley, 1991)? Comparative politics involves the development of theories explaining behaviour within groups of countries that are essentially similar (see pp. 18-19). It is also about contrasting cases that are different in any number of ways. Either focus of comparison - explaining similarities or differences - can let) tlie researcher a great deal about the way in which governments function.
Most Similar and Most Different Systems
One crucial question in the selection of cases has been advanced by Adam Przeworski and Henry Teune (1970). This is (lie difference between most similar and most different systems designs. The question here is how to select [lie cases for comparative analysis, given that most comparative work docs involve purposeful, rattier than random, selection of the cases. Does one select cases tliat are apparently the most similar, or should (lie researcher attempt to select cases tliat are the most different? Further, like much of the other logic of comparative analysis, this logic can be applied to both quantitative and qualitative work. Theda Skocpol (1979: 40-1), for example, argued in essence for a most different systems design in lier historical analysis of revolutions in France, Russia and Cliina. These systems all generated major revolutions, albeit arising within apparently very different political economic and social systems. The question for Skocpol then
to explain relationships among variables in samples of individuals. Tlic must different systems design is attempting to determine how robust any relationship among [liese variables may be - docs it hold up in a large number of varied places as if the observations were drawn frum (lie same population of individuals? If it docs, then we have some greater confidence thai there is a true relationship, not one produced by some unmeasured tliird or fourih or fifth variables that exists in all relatively similar systems.
Further, the basic logic of the most different systems is falsification, very much in tlie tradition of Popperian philosophy of science (Popper, 1959). The basic argument is lliat science progresses by eliminating possible causes for observed phenomena rather than by finding positive relationships. As noted above, there is no shortage of positive correlations in tlie social sciences; what there is sometimes a shortage of is research that dismisses one or another plausible cause for tliat phenomenon. By setting up tests in a wide range of settings, the most different systems design attempts to do just that, while the most similar systems design can identify many possible causes but can eliminate none- This problem can be seen in part in a study of the Mediterranean democracies (Lijpliart el n!., 1988). These systems were thought to be similar, yet once they were analysed differences in their transitions to democracy did emerge. Unfortunately, they were sunicicntly similar for it to be impossible to identify effectively the root cause of those differences.
The logic of this approacli is therefore fundamentally different from the most similar systems design. Whereas tlie most similar design dealt with control through careful selection of matclicd cases, tins design deals with [lie control issue by virtually ignoring it. In tlie most different case, there may still lie unmeasured extraneous sources of variance, but they will have to be very generic in order to survive in the range of social settings in which the research may be conducted. This .strategy is, however, also dangerous, given that it can create yet another false sense of security in the strength of the findings. Indeed, the findings may be gencralisable to a wide range of political and social systems, but [lie underlying causal process assumed to exist may not, even though it may appear from Berlin to Bombay to Bogota.
1 lie most similar ;>nd most different designs therefore do very dillcrem things. Tlie former deals more directly with countries ;is a unit of analysis. It allenipts to control for extraneous sources of variance by selecting cases in which tlii-s is not likely to lie a major problem, although tlie rese.m-her can rarely if ever know how big a

problem really docs exist. On tlie other hand, the most different design is not particularly interested in countries; this is more variable-based research, and is many ways closer to a statistical design than to llic true comparative design. Tlie principal task in this design is to find relationships among variables that can survive being transported across a range of very different countries. Given tlie statistical nature of tlie thinking here, controls for extraneous variance can be imposed
by the usual statistical techniques.
Tlie most similar and most dinerent research design strategies appear very reasonable strategies in theory, but perhaps more difficult to implement in practice. Giesele De Meur and Dirk Bcrg-Schlosser (1996) have demonstrated more clearly how tlie approach to analysis could be used. They were interested in the success or failure of democratic regimes during the post-war periods in Europe. They first divided the countries into groups of similarity based on the persistence or breakdown of democracy, and tlicn looked at the most different countries in each group, based on distance measures calculated from a large number of social and political variables- Thus, they could look at "most different, similar outcome' cases to see what variables were in common to explain these outcomes-
GtobaUsation and Gallon's Problem
A special case of tlie difficulties associated with controlling extraneous variance in research design arises from the increasing gtobalisalion of culture and politics. We can travel around tlie world and eat the same food, drink tlie same soft drinks, and watch MTV or CNN (depending upon taste) almost anywhere- We must also be concerned if lliere arc similar globalising trends in politics, so that what we observe in one country may not be llie result of any indigenous political process, but rather may be a product ofdifiusion. Of course, tlie diffusion does not make the occurrence of a phenomenon any less real, but it does influence any analysis interested in developmental patlcrn'1, or tlie relationship between economic and social conditions and political phenomena. For example, if [lie familiar argument alxJLil llie relationship of economic development and democraiisation (Lipset, 1959) is advanced, then any contemporary findings may be i-ont.iniinated by tlie diffusion of democratic ideals - forcibly through international organisations such as llie United Nations and informally through the media and oilier social processes. This problem may he furllier confounded by the differential receptivity of some

42 Comparative Politics
countries to diffused ideas, and tlic active scarcli by some for alternatives to (licir existing policies and structures. Tlic Japanese, for example, have been characterised as adept at borrowing constitutional ideas and [lien adapting them to their own circumstances.
Globalis.Hion is simply a contemporary means of stating a very familiar problem in llie social sciences, usually referred to as 'Gallon's problem'. Tilis is tlic methodological problem ol sorting out diliusion from other causes of variance in social systems. Gallon identified this problem in relationship to similarities among cultures he observed in tlie work of anthropologists, but it also arises in less exotic research situations. There arc some cases iu which observed patterns of social behaviour simply could not be the result of natural causes. 'Ibc prevalence of presidential forms of government in Latin American countries throughout rcceni history is a clear product of diffusion from the 'Colossus to the North' (Thibaut, 1993), rather than a natural process of institutional selection. Similarly, the prevalence of Westminster-style systems in former British colonies is a product of forced diffusion.
Oilier observed formats for government and public policy may be less obviously a product ofdifTusion. Take, for example, tlie formation of social insurance programmes in tlie countries of the world. \Vhilc tlic concept of social insurance could be invented only once, the pattern by which it spread appeared to follow a natural path of economic and social development rather than one of more forcible diffusion. Several of [lie wealtliier countries in Latin America, for example, adopted social security programmes long before tlic United States or many countries in Europe (Mcsa-Lago, 1978; 1989). Did this pattern result from (he simple diffusion of ideas, or were there socio-economic or political factors that made one group of countries more fertile ground for these programmes than were others? Further, even if there were diliusion involved, are societies and governments really just tlic 'passive receivers of traits' (Smith and Craon, 1977:
145), or is diffusion dcpendeni in part on socio-economic conditions, political traditions, or even cultural alTinitics thai facilitate diffusion? Those conditions may mean that tlic society is particularly receptive lo a programme that lias become 'an idea in good currency', or that there may be a greater than average need for adopting the policy in question.
If tlie former is (lie case llieti (here is little to say, except to attempt to understand diffusion processes belter (Rogers, 1995; Majonc, 1991;
Uolowity. and Marsh, 199G). If, on tlie other liand, there docs ;i|)pc;u

to be some systematic relationship of social and political factors to (lie adoption of social insurance programmes then we may have fodder for die tlicoretical mill on comparative public policy. The problem for research design is to be able to sort out tlie possible causes of the observed pattern as a way of addressing the theoretical question. For example, we might begin by looking ai diffusion in specific linguistic or cultural areas - do llicsc factors appear to make adoption of similar trails more likely? Or do factors from a more political theory appear

David Klingman (1980) investigated possible means of addressing
diffusion issues as they occur across botli time and space. The usual comparative strategy is loo look only at cases at a single point in lime and then infer a process, but Khngman wanted to demonstrate how to cope (statistically at least) with sorting out effects of both temporal and spatial diflusion. His argument is based upon the logic of autocorrelation in regression analysis. As with patterning of residuals produced by systematic relationships with a variable across time, patterning of residuals can be a function of systematic unmeasured relationships across space, and in this instance be a product of
diflusion.
Diffusion is a fact of political and social life, but it can confound our
comprehension of causation in comparative analysis. It is all to easy to assume that because attribute A' appears in country )' that it developed autonomously. This is especially true if operating from an implicit or explicit functionalist perspective in which any observed trait is presumed to fulfil some function for the society. That need not be the case, and some attributes may represent adoption of foreign traits and values virtually as quasi-expcriments or to conform to perceived stereotypes of values such as 'modernity' or 'democracy'. These political 'cargo culls' do show something of tlic power of ideas in changing behaviour, but also can confuse comparative analysis.
Levels of Analysis i
Another fundamental question for the design of comparative analysis is how to link tlic individual and the collectivity. Comparative politics t.s concerned about (lie behaviour of political systems, but it Es also concerned about llie behaviour of the individuals within those systems (Silverman, 1991). The problem is that it is very easy to slip Hiiu fallacious reasoning that attributes tlic properties discovered at

compiiralive research, studying a set of cultural attributes assumed to make democracy more or less viable in ;\ country. It found thai several of tin-countries included (West Oerm.my, Mexico and Italy)

one level ol'analysis l() the otiier level (see p. lit). On the one hand, researchers (ire confronted with the 'ecological fallacy', hi which the properties or a collectivity arc assumed to characterise individuals within them. For example, in his classic explication of the ecological fallacy, Robinson (1950) pointed out that if we identify political units with higher rates of literacy and then find tliat lliose same units have high rates of foreign-born residents then it may not be that foreign born citizens are more literate on average.
One more subtle version of tills fallacy, which is an all-to-common trap for tlic comparative researcher in political science, is to assume that something we call a country is a homogeneous unit, We know that the countries in Africa that were carved out by colonial powers for their own purposes are not homogeneous units ~ the outbreaks of ethnic warfare in Burundi and Rwanda are unfortunate reminders of that. The lack of homogeneity lends to be forgotten for more developed political systems in which variables such as religion may cut across national boundaries. For example, Germany may appear to lie a homogeneous country from outside, but the religious cleavage remaining from the seventeenth centurv still influences the manner in which politics and policy-making function in the two parts of the country today. Other countries such as ttie Netherlands, Belgium and Switzerland are divided by religion, language, or both (see Lane and Ersson, 1994a).
Furthermore, an entire country may be characterised as an outlier in some distribution, because one part of it (geographically or socially) lias very extreme values, while the remainder is quite 'normal', lying within the distribution- For example, the United Stales lias a substantially higher level of infant mortality than might be expected given its level of economic development and expenditure for health care. If, however, thai figure is disaggregated by race then the white population has an infant mortality rate equivalent to that found in many European countries, while the non-white population lias a mortality rate more similar to transitional or Third World countries. Similarly, the slate of New Hampshire has an infant mortality rate of 5.6 deaths per thousand live births, wliilc Mississippi lias a rate of 1 1,3 - more than double that in the lowest stale.
On the other hand, researchers also can fall into the individualistic fallacy in which tlie collectivity is assumed lo have the properties of llie individuals that comprise il. Perhaps llic most important example of tills fallacy would lie The (,'ivic Culture (Almond and Verba, 1967;
see Key '1'exl 2.1). This was one of tlic major pieces of behavioural

Key Text 2.1 Gabriel Almond and Sidney Verba, and the Civic
Culture
The Civic Culture has been widely praised and widely damned, but what is certain is that this study has had a profound influence on comparative politics. The study set out to identify the cultural characteristics that made democracy successful in some countries and less successful in olhers. In the end the argument was that a country such as the United Kingdom with a mixture of participative and deferential components in its political culture would be more successful than would countries that were more participative (the United States) or more 'subject' oriented (Italy or
Mexico).
The Civic Culture was one of the first large-scale comparative surveys
conducted as comparative politics moved from its earlier institutional focus to a more behavioral focus in conducting research. It was therefore in some ways a model for the (then) newer approaches to the discipline, and demonstrated the potential of mass political surveys as a research tool for comparative politics. H also demonstrated the potential of linking empirical research methods with normative questions (democratic performance) in comparative analysis. Finally, the project developed several interesting concepts about political attitudes and behavior that have
informed subsequent research in the field-
Although extremely influential in comparative politics, and the source of numerous dissertations and secondary analyses, this project was also roundly criticised. In the first place, although it purported to be objective and 'scientific' there was a clear prescriptive element in the work, pointing out what a good democracy was, and also what political development entailed. Further, there were methodological questions about how to link individual attitudes with the political culture of a country. Finally, The Civic Culture apparently was not as careful about questions of the comparability and meaning of political terms within different cultures as it might have been, so that some of the findings are
suspect.
The above having been said, it is hard to diminish the importance of
The Civic Culture. It was a major source of encouragement lor comparative political scientists who wanted to use the methods usually confined to national elections or national public opinion polling. It also demonstrated thai empirical comparison was really possible using those methods, even given the problems that inevitably arose. did not have the attitudinal support for democracy that others did, the inference being th;>t democracy would not survive tlicre. Thus, ttie characteristics of the system as a whole were being derived from the attributes if individuals - inaccurately as it turns out.
Depending upon the purposes of tlic researcher, any level of analysis will do well. The problem is simply not to mix the levels, or, if one dues mix them, to pay careful attention to the possible misinterpretations that may arise. Cross-level inference should be regarded as a crucial question in comparative politics. As Heinz Eulau (1963: 47) argued, perhaps the most important task in the discipline is:
to build, by patiently linking one unit to another, the total chain of interrelationships which link individuals to other individuals, individual to primary group, primary group to secondary group, secondary group to secondary group, secondary group to organization, organization to organization, and so on.
Ifsuch a sequence can be established then we will have a much better idea of how politics works, whether at the individual, the group or the state level-Further, this statement by Eulau points out that whenever we begin an analysis we are cutting into a complex chain of causation. Simply focusing on tlie individual voter, for example, may appear to be the right level of analysis for understanding electoral behaviour, but we also know that llici-c are important enecl-s of community and neighbourhood on voting. Indeed, much of the analysis of voting behaviour lias depended upon class as the explanatory variable (Nieuwbccrta, 1995). On the other hand, we can obtain reasonably good ideas about how civil service systems behave by looking at patterns of individual recruitment and the training and ideas of the individual members of the service (Abcrbach el al., 1981; Suleiman and Mcndras, 1995).
Threats lo Validity in J^on-Experimental Research
flie litany we have hccn discussing for coping with variance is another way of expressing the problem of validity in social research. \ alidity is the simple question ofwhellicr we are measuring wlial we think we arc measuring, or whether (lie observations we make (especially on our dependent variables) are a function of oilier factors not included in tlic analysis. The 4'xpcrimental design charai irri.slic

Key Text 2.2 Phillipe Schmitter and the Politics of Interest Intermediation
The study of interest groups during the 1950s and 1960s in comparative politics was heavily nifluenced by American political science and its pluralist assumptions. Although it was clear that in many countries interest groups were organised very differently and played a different role in politics, those differences were not adequately conceptualised. Phillipe Schminer's 1974 article describing corporatism and positing different types (stale and societal) of corporatism was a landmark in the understanding of both interest-group politics and the politics of continental European countries (see also Heilser. 1974).
Schmitter argued that rather than being the open universe of pressure groups that was the heart of pluralist theory the consteallation of interest groups in many countries was actually more tightly organised and restricted. In particular, government played a central role in selecting the interest groups with which they wished to interact in makign policy, tn this mode!, a limited number of peak organisations would be granted direct access to decision-making in exchange for some acquiescence to the decisions once they are made.
This basic model of corporatism focussed on macro-economic policy through tri-partile bargaining, although the corporatist notion appeared in principle applicable to a broader range of issues.
The original Schmitter article spawned a number of extensions and elaborations of the basic corporatist argument. For example, there were a number of discussions of the boundary conditions for a corporalist system to exist and why countries such as the United States and Japan had little or no formalised corporatism. There was also a discussion of 'meso-corporatism' for issues other than macro-economic policy . Similarly, the revival of Rokkan's ideas about 'croporate pluralism' (see Key Text 3.1) in which a large number of interest groups are granted direct access to policy-making extended the concept of corporatism.
Despite the impact of the concept and the associated research there have been a number of critiques. One of these has centered on the generalisability of the concept; just as pluralism was used excessively. so loo has corporatism. A similar critique is that the concept has been stretched to ocover situations in which it is probably inappropriate. That stretching has been argued to be true in bolh geographical and policy terms.
of tlie nalural sciences is supposed ti* solve ihosc problems lor those disciplines, and some disciplines thai bridge tlie gap between social and naluriil science for example, psychology. As already explained,

is

most social scientists rarely have the luxury of conducting true experiments, and therefore must find ways to make noil-experimental designs as valid as possible
When we speak of validity, \ve must be careful to separate two dlllerent forms of (jig concept. One type is internal validity. This is llie capacity to be sure [hat, if the independent and dependent variables
co-vary, then indeed the changes in A' did cause [lie observed changes in /. T|^g question is a manifestation of the problem of extraneous variance, which we have been talking about throughout much ol tins chapter. Internal validity is generally not a problem for experimental research, since every effort is made to liold as many laclois as possible constant through random assignment, and the use of control groups permits checking whether that cfTort has been successful. In "on-experimenlal research, however, other mechanisms must be develop^ ^ cope with llie threats to internal validity and the possible influence of extraneous factors on observed findings.
External validity is a very different issue in research. It refers to the question o( wilder any relationship observed among a set of variables in the research setting is generalisable to the 'real world'. ror survey research and statistical analysis, this amounts to whether or not the investigator is justified in generalising from the sample to Ihe population, \v\^ for strictly comparative analysis tlic issue is whether the cases selected do reflect adequately the dimensions of dillcrcnt variables assumed by the researcher. Experimental design actually performs very poorly for this type of validity problem. The artthciality of the laboratory setting for social research that provides it witli ihc ability to control extraneous influences also divorces (hose findings from the social context within which they have meaning. If
/- i i 'o "
\ve nnci mat people behave in a certain way in a psychological or
economic experiment can we have any confidence that they will
i i ' i
ueiuvc in the same way in their everyday lives? The answer is
probably not.
Internal versus p xternal validity is therefore yet another trade-on in comparative research. Experimental designs are generally tlie gold standard for internal validity, but are of little use for proving (lie external validity of ^ finding. Indeed, many people would consider the hnclings of laboratory experiments about important political issues to be mor<- misleading than valuable. On tlic other hand, the type ol lesearcli more common in political science tends to have a good deal of extern;i] validity - it is based on observing the real world - hut can have a number of serious deficiencies in internal validity.

The use of experiments for many of the important questions of
political science may not he practical, so we are then left with attempting to cope with the internal validity problems that arise from non-experimental designs.
The obvious answer is to attempt to do both, and to test the findings oftlie laboratory in a more natural setting, or to attempt to replicate observed findings in a more controlled environment. One interesting example of this strategy is the attempt to match experimental findings in international relations theory witli real cases of foreign policy decision-making. Kaufmann (1994) derived a series of propositions from rational and psychological theories of foreign policy, then tested them by examining evidence in the Imperial German archives. While such testing involves introducing another potential source of error - the coding of the researcher - it still constitutes an interesting way to address the internal and external validity dilemma.
Sources of Invalidity
The principal approach to coping with the numerous threats to internal validity in social research is to be cognizant of their existence, and to use tliat knowledge to implement strategies to avoid as many pitfalls as possible. When the pitfalls cannot be avoided entirely then conservative interpretations of the findings can save the researcher and colleagues most future difficulties. Donald Campbell and Julian Stanley (1967) havf enumerated the.se potential problems in an extremely useful manner. Some of (lie threats they mention have little relevance for most comparative research, exerting their major influence on individual level research, but some are endemic in comparative research. These should be discussed at some greater length as means of arming die prospective researclier.
History
One very common tlircal to llie validity of comparative research i what Campbell and Stanley refer to as history. This is the simple fact that, while researchers arc observing a case and llie presumed interaction between an independent and a dependent variable, there may be a number of other changes going on around them tliat are impacting the observations being made. l-'or example, any attempt to understand the p<ililii-;i! behaviour oftr.ide unions in llie 1990s must take into .iccouiii the possibly confounding enects of economic

Key Text 2.3 George Tsebelis and Game Theory
In addition to institutional theory (see Key Texts 5.3 and 6.2), there has been some growth in rational choice accounts ol comparative politics. The assumption of this Iheoretical approach is thai the best way to understand the behaviour of individuals in politics is to think of them as rational utility maximisers, much as an economist would assume about their economic activilies. In particular, Tsebelis applies concepts of game theory to comparative politics, demonstrating that many common political problems can be conceptualised as 'games' in which individual players choose strategies, with the outcomes being determined by the structure of the payoffs available and the strategies adopted by the competitors in the game.
Tsebelis was particularly interested in 'nested games', that is games that occur are piayed within multiple arenas simultaneously. These games often generate suboptimal outcomes even when the individual players are acting rationally. As well as describing these games and their dynamics, Tsebelis also points to their implications for institutional design. The utility of the approach is illustrated with several examples drawn from comparative politics, including consociationalism and coalition formation.
Rational choice analysis has a number of critics in political science. The critics argue that political actors are not (necessarily) the rational utility maximisers presented in these models but ralher operate from a number of motives, personal utility being only one of many. Further, the critics contend that these models oversimplify complex realities found in politics, especially the behaviour of political institutions. The result of that oversimplification is argued to be results that are either trivial or misleading. Further, it is argued that rational choice approaches place too much emphasis on individual actors and not enough on the role of collective actors in politics, largely because of the problems of aggregating preferences and rationalities in institutions.
Tsebelis acknowledges the criticisms and provides a vigorous defence tor the approach. The defence may not be entirely convincing to the more committed opponents of rational choice, but it will give even them reason to reflect on their assumptions about politics. This book is a crucial slarting point for anyone attempting to understand rational choice approaches in political science and their applicability to comparative politics.
insecurity in a globalising economy. So, while trade unions may appear much less activist lli.in in the past, thai may have nothing to do with Factors suc-li as changes in leadership or changes in manage-

ment practices but a great deal more lo do with the changes in the economy.
Observing a single case may make research particularly prone to the clFccts of history, but using multiple cases may be no guaranteed way to avoid the confounding effects ol'llial problem. For example, dillci-cnt countries may react very din'crently to changing economic circumstances as a result of historical experiences. There is some (Schmoldcrs, 1960) evidence, for example, that the British public is extremely concerned about any threat of imcmpioymciit, while the German public is mure concerned about inflation. These concerns can be seen as a function of the economic histories of (lie two countries. Britain was plagued by mass unemployment between the two wars, while Germany suffered two massive inflations that wrecked the economy. Younger generations may be less allccled by these fears, but llicre is some sense that they do persist. Thus, a change in economic variables would have a difTcrenlia) impact in these two systems [hat might confound the findings of research attempting to assess the effects of economics on political behaviour (Lewis-Beck, 1988; Andersen, 1995).
Selection Bias
We have been discussing the selection of cases in much of this chapter, so that it may appear that there is little else to say on llie subject. One point still to be mentioned, however, is that there is a natural bias in comparative research arising from the tendency to select tlic cases that the researcher knows best, and to attempt to make the theory fit the cases rather than vice versa. Tins bias is very natural, and in some ways logical, given that it prevents other types of errors of interpretation and the barriers of foreign language to meaning research. The same logic also creeps into the recruitment of multi-person research teams on dinercnl countries, in which the tendency is always to 'round up the usual suspects', lor example, researchers willi whom the organiser is familiar and who have shown themselves to be reliable contributors in the past. A case may be selected because of the availability of scholars rather than its intrinsic merit. Again, that can be an antidote to otiier types of practical problcm-s arising later in (lie research.
The above said, however, the preference for I'ainili.n- cases can present severe problems in testing theory wilti the d.ua collected. The cases with which we are most familiar arc nut necessarily llie best for testing theories that are intended to have genera! applicability.

^
.6
Further, almost by dennition, ifwe want to think about theory testing ,o' through din'icull cases, ihen the less familiar countries and (lie places s with perhaps die least welt-known scholars on the international \i 'circuit' will probably be the most interesting. Comparative politics r, is sometimes described by [is critics as the search for the most exotic locales and for academic adventure, but in practice re-searchers sometimes are wedded to (lie least exotic and the excessively familiar, and that may be detrimental to theory development-In looking at instances of selection bias in comparative analysis it is easy to assume that tills is a problem for small-n, qualitative studies. Unfortunately, elegant quantitative analysis can be used to mask some fundamental problems in case -selection. All the analysis in the world, however, cannot make up for a 'sample' that contains some systematic bias. Take, for example, Hahm elal.'a (1996) study of fiscal policy. This study used a high selective sample of nine countries to lest the influence of the nature of fiscal institutions on deficits. That sample excluded Sweden and Norway, however, which have quire similar administrative systems, but have had wildly different fiscal policy fortunes in recent years- It also excluded Ireland, which has an administrative system not dissimilar from thai of (lie United Kingdom, but which again had liad very different policy outcomes. One can only wonder at tlie utility of these findings. Similarly, much of the research arguing for the negative impact of presidential ism on democratic stability appears to be beset by some selection bias (see Power and Gasiorowski, 1997).
Instrument Kins
Instrument bias in comparative politics may be closely related to selection bias, in that our selections are sometimes made on the basis of available 'instruments'. Tlie crucial instrument for comparative political analysis is tlie individual scholar- That scholar may employ a variety of other instruments, sucli ;is questionnaires or interview protocols, in Ins or her work, but scholarly judgment and knowledge are the crucial factors. The problem is that scholars all come with some built-in biases. These biases may be theoretical, national or temperamental, among others, but they will lie present. Further, given that comparative analysis is often interpretative, those biases arc likely to influence the conclusions of the work. Even if the research is quantitative, tlie selection of the research question and tlie particular quantitative measures may lie influenced by theoretic-ill and methodological biases dial exclude oilier possible explanations.

One common strategy in comparative research projects is to have one national expert write a chapter on his or her country about what ever the issue at hand may be. This practice often produces a published collection in which, as Christopher Hood once commented, me major independent variable is llie author. Such a strategy mav be especially problematic wlien tlie experls arc recruited from tlie nations themselves, and therefore tend to replicate the prevailing national analysis of llieir own system. In sncli a case, any implicit comparison with other systems may be lost, and it is only the compiler of the volume who can attempt to make any comparative sense out of the national studies. Many editors have done tills successfully, but just as many appear to have failed ratiier badly.
Maturation
As well as tlie external circumstances changing around a research site, the internal circumstances can change as well. For example, take the case of a scholar who is interested in decision-making in cabinets, and spends time following tlie deliberations of a particular cabinet, even if from afar, and through interviews with participants after tlie fad. Lei us now assume that there is a reshuffle (a very frequent occurrence in parliamentary governments ~ Italy lias had over 40 since 1945). In this case, the research site has changed in a fundamental way, particularly if a key figure is shuffled out, or a particularly eneclive new minister is shulllcd in. Depending upon tlie particular focus of the research, tlie project could well be ruined, and liave to start again wiili the new cabinet.
The change in a cabinet is a very visible change in the research site, and that is probably fortunate. In that instance, the researcher can easily notice that something lias happened in tlie research setting, and then can make a decision as to whether- that change really impacts his or her investigation. On tlie other hand, subtle and gradual changes in individuals or organisations present a more dillicull collection of threats to the external validity of findings, simply because they are. likely to occur unnoticed- For example, after lenglliy bargaining over an amendment to a piece of legislation, the legislators involved finally reach agreement. Is this because the demands of all (lie participants wen; satisfied in the linal proposals, or did they simply become tired? In sliort, individuals and groups do change over lime, and those changes can anecl llic validity of findings. The only real defence against thesf threats appears to be awareness of the problems.

Regression Towards {he Mean
Another source of potentiiil invalidity in research is referred to ^ 'regression towards llie mean*. This defect will result when (lie cascg for research are selected on the basis of extreme values on l^
dependent variable. Given tlic errors inherent in any social measure. incut, all subsequent tcsis arc likely to be less extreme. Further, givp^ the extreme values in tlie initial measurement, llicrc is nowhere else for the observations to go, except back towards (lie mean, when there lias been any en-or and extraneous variance in (lie initial measure. ment. For example, suppose students arc selected Ibr an investigation on the basis of poor scores on a test of educational skills, and are tlit-n given a special class and then tested again. The second measurement is likely to show that they have improved, and that tlie class was 3 success. This observation could, however, be totally an artefact of regression towards the mean, rather than revealing any real impact of the educational programme.
Sucli a source of invalidity represents a more individual-level version of (tie problem in comparative politics of selecting cases on the dependent variable. If we select cases of successful democratism. tion, for example, then examine them across time, there is likely to bg a significant failure rate of democracy. The 'sample' will, in essence have regressed towards tlie mean of the distribution of democratic political behaviour in tlie world. Thus, any 'treatment' we ai^ inferring from (lie environment of these political systems - for example, changing economic fortunes or increased ethnic- tensions-is suspect as a cause of tlie declining rate of democracy. Similarly, the 'breakdown' of institutionalised party systems in democr;uic countries (Dallon el nl., 19^-1; Lawson and Mcrld, 1988) may merely represent a regression, of an aberrant form of parly system observed for only a short period of time, towards the mean of parly systems in the generally more fractious political world. We have cautioned against this practice in a number of places in this volume, but the abov^ discussion provides a somewhat more formal justification for those admonitions. We should, however, also point ouE that selecting cas^s on (lie dependent variable mav be acceptable and desirable as a means of generating hypotheses, just not for testing them.
Regression towards tfie mean can occur wlien llic 'sample' used in tlic research has been selected on tlic basis of its extreme values on the dependent variable. While King et al. (1994), among others, liavc argued against selecting on tlic basis of values on tlic dependent variable, there arc some instances in which this praclice may lie

Key Text 2.4 Vincent Wright and the Reform of the State
Most of Ihe attention on transformation ol political systems has been directed toward the spread of democratic politics and markel economics in the former Communist countries of Central and Eastern Europe (see Key Text 7.1). To the West of those countries there also has been a major process of transformation and reform, less dramatic but nonetheless significant. This article, along with the others in the same special issue of West European Politics (1994) describe and analyze those changes very welt. A thorough reading of these articles will provide an enhanced understanding of the intellectual debate over reform in West Europe and the other developed democracies.
Vincent Wright and the other authors in this special edition demonstrated that the state in Western Europe is becoming fundamentally different from what it had been in the past- The numerous political and economic pressures for reform have required some rethinking of jusl what government can and should do in these countries. Further, the availability of ideas for changing the State, especially the important of the market as an alternative to the hierarchy of State bureaucracies, has facilitated the process of change. Finally, the different policy challenges to government, many arising from globalisation and Europeanisation. appear to require a different set of structures and practices in order lor government to be effective.
What is perhaps most important in this discussion is that although these states are changing in some important ways they remain in many ways unchanged. The basic Ftechtstaat character of most European states is still in place and, despite the numerous 'marketising' and other reforms bureaucracies remain a central part of the apparatus for making and delivering services.
Further, despite privatisation of many public firms the State remains a major socio-economic actor and continues to influence the economy in many ways. These articles help the reader to see that despite (he change government remains the principle source of guidance and control in most Western countries.
acceptable or even desirable. Tlic regression argument makes it manifest lhat one condition is clearly that iherc should be no inference based on tlic retcsl of tlie same 'sample1, especially if tlial sample is extreme in its values. This is rarely a problem for comparative politics, olher than for some ci-oss-nalionat studies of public policies, in large part because of the inability lo h;ivc a clearly defined mttTvenlion that would makr a retest meaningful.

Conducting valid political research is never an easy task. There are a large number of threats to th;it research lurking in practices [hat appear normal, and even desirable (for example, looking only at cases that illustrate a particular trait). Other threats 10 the validity of comparative researcli, for example, maturation, occur as very normal parts of the political process, hut still constitute significant threats to good researcli. Again, these threats are almost inevitable in non-experimental research, so that the major benefit to be gained from ibis description of [hose threats is simply to be aware of [lie threats, in order to be able to discount their possible impact on results, if not always avoid them. Thus, if a researcher knows that a particular llircal to validity is present, and can estimate the direction in which that threat is likely to have operated, then he or she can make more conservative interpretations of the results.
Conclusion
Comparative political analysis is heavily dependent upon tlie cases selected for analysis. This selection is often done in a haphazard maniier, or on the basis of familiarity; as in Casablanca, we tend 10 round up [lie usual suspects when starting on a project. Tliese may well be llie right cases, but often they are not! Any project on European politics appears to have to have [lie big [hree or four counlrics, although the smaller ones are much more relevant theoretically. A researcher should lie able to justify the choice of cases on theoretical grounds rattier than on convenience grounds, but there are few consumer warning labels on comparative research projects.
There are several views on how to select cases for analysis, the dominant being (lie 'most similar' and the 'most different' systems design'1. These two designs enable [lie researcher to examine different form of relationship between independent and dependent variables, and therefore are suitable for different types of enquiry. Again, however, one design or [he other oflcn is selected on the basis of ease or familiarity with cases, rather than through thinking about the nature of the type of comparative results to be sought. Without attention to (lie projected outcomes of analysis, the clioice of designs and of cases is unlikely to be (lie best one.

There is no single answer to all tliese questions. Although Prze-worski and Tcunc (1970) liavc rather strong opinions about the best way to design analysis, oilier scholars appear to have diametrically opposed views. What appears to be most important is that the researcher consider his or her options carefully before selecting a researcli design and (lie cases that will he fitted into that design. It may be that llie purpose of the research is not so much to find the cause of a particular phenomenon, but rather is more exploratory, looking at an interesting phenomenon and attempting to understand better something of its nature- In that case, selecting 'most similar systems', and selecting on llie dependent variable, might be a very reasonable thing to do. \Ve tend to focus attention on llie uses of comparative research for theory testing and falsification, and often forget that there are more exploratory uses that are equally valid.

• [ На главную ]
• [ Библиотека ]
• [ Reader ]
• [ Программы ]
• [ Абитуриенту ]
• [ Фотоальбом ]
• [ О ФП ]
• [ Гостевая книга ]
• [ Ссылки ]
• [ Обратная связь ]