What Can Two Robust Facts of Government Formation. Reveal about the Successful Operation of Democracy?

Transcription

1 What Can Two Robust Facts of Government Formation Reveal about the Successful Operation of Democracy? Michael D. McDonald* Department of Political Science Binghamton University, SUNY Bartle Library Bldg Binghamton, New York United States Voice: Fax: mdmcd@binghamton.edu * Some ideas and data presented here are part of a larger project on Organizing Democratic Choice being developed in conjunction with Ian Budge, Hans Keman, and Paul Pennings. They bear no responsibility for shortcomings in either the line of argument or the interpretations presented here. I thank Robin Best and Myunghee Kim for compiling data from the Comparative Study of Electoral Systems. Prepared for presentation at the ECPR Worshop on Party Government & Parliamentary Democracy in the New Europe, April 16-2, 29, Lisbon, Portugal.

2 1 Abstract Government selection can and often does play its own key instrumental role in producing responsive and congruent representation of voters with limited time resources. Why and how this is so arises from two robust facts about Western democracies. One is the tendency for incumbent government parties to lose votes in the election after they enter government (Nannestad and Paldam 22). The second is the plurality party s inclusion in over 8 percent of Western European governments in the first place (Muller and Strom 2). The thesis explored is grounded in an analytically derived theoretical model of democratic responsiveness and congruence when a government includes the plurality party. The model shows the plurality party selection and deselection has two countervailing, but in combination, propitious effects. The selection of the most popular party or a combination of newly popular previously opposition parties produce a short-run dynamic in response to changes in voter preferences. At the next election the tendency of the last election s more popular than normal party or parties to lose vote support produces enough alternation in government to create a medium-run policy centrism in proximity of median citizen preferences. Because the dynamics and stable centrism of the process rely on the ebb and flow of gaining more votes than usual to become the governing party or parties and then dropping back at the next election, voters do not have to be extraordinarily informed, diligent, and calculating in order to keep the rhythm and rhyme of democracy responsive and congruent. The democratic norm and institutional devices that go into including the currently most popular party or parties in government do much of the work.

3 2 I. Introduction For all the drama surrounding government formations and for all the fuss and fury associated with elections, two robust facts stand out. First, when governments form the plurality party usually is in government (see, e.g., Müller and Strøm 2). Second, parties in government usually lose votes from one election to the next (see, e.g., Paldam and Nannestad 22). This paper offers a preliminary exploration of the possibility that these widespread and persistent tendencies, these robust facts as I am calling them, have something important to tell us tell us about the successful operation of democracy. There is nothing especially remarkable about either fact standing alone. The usual presence of a plurality party in government could be said to reflect nothing more than the observation that in a democracy winning the most votes often leads to holding the most political power. The big surprise would be if that were not true: an essential truism for democracy is that votes power. Similarly, the usual loss of votes by incumbents might reflect nothing more than a second truism. Collective decisions require tradeoffs and cannot satisfy all the people all of the time. In societies with varied and conflicting interests the big surprise would be if a series of collective decisions were to make decision makers ever more popular. The potentially intriguing possibility comes from thinking about the robust facts in combination. In combination, possibly, the facts could produce an easy to overlook effect that keeps democratic decision making in line with the settled views of the public. How might this work? Under certain observable conditions, which are usually present in Western democracies, gaining power due to having received especially high vote support and losing power due to that higher than normal vote support slipping away indicates the existence of forces restoring party competition (Stokes and Iverson 1962). What are the restoring forces? Who is to know? Several generations of voting behavior specialists have been studying how voters arrive at their decisions (Berelson at al. 1954; Campbell et al. 196; Thomassen, ed., 25).

4 3 They have not come upon any easy answers, and probably no one outside voting behavior sub-field can pretend to know half as much as they do. But, whatever the restoring forces are, the two robust facts might be saying this. The intelligence of democracy under certain conditions is to rely on a decision rule that exploits the well known tendency summarized by this old saw, whatever goes up must come down. To social scientists this is better known by the term regression toward the mean. The purpose here is to explore whether the rhythm and rhyme of democratic politics among Western democracies is produced by the seemingly simple process of what an electorate first gives it soon thereafter takes away. Whether, how, and why these ebbs and flows of votes are important depends on certain conditions. Therefore section two is directed at establishing how common the certain conditions are in Western democratic societies. Section three presents a theoretical framework for how democracy could work in the presence of the conditions as shown in section two. Section four turns to an analysis of rhythm and rhyme of elections and government formation that result from a simple, almost automatic, regression to the mean process associated with some form of forces restoring party competition. Section five concludes by speculating about what all of this can tell us about the operation of democracy in general and about democratic development in Eastern and Central Europe more specifically. The essential overview can be described in this way. Citizen preferences along the predominant dimension of political perspectives and competition, left-right, are unimodally distributed with the mode near the center of a nation s political space. The major parties vying for the predominant position in government tend to stand to the left and right of the center. One consequence is for elections to produce governments incongruent with the median citizen preference. The incongruence can be mitigated to a very large extent if the policy making process is slow moving and if party control of governments alternates frequently. Here, a slow-moving process is assumed and the focus is on alternation. The supposition to be explored is this. The key to producing alternations is selection due to abnormally high

5 4 voter support which recedes and brings a 5:5 chance of alternation at the next election. The ultimate effect is for a nation s governance to emanate from a left-right location in the vicinity of the median citizen. 2. Citizens and Parties in a Nation s Left-right Space Figure 1 shows the distributions of citizen self-placement along the left-right dimension for sixteen Western democracies. The details of the data underlying these distributions are provided in Appendix A. With the exception of Germany, the data come from the Comparative Study of Electoral Systems (CSES), module 2. The German data are from the CSES module 1 because the German surveys in module 2 were conducted in two different forms, one as a mail survey and the other as a telephone survey. The data underlying the graphs are presented in Appendix Table A1. [Figure 1 about here] While each nation s precise details differ from the others, the common feature among all sixteen is a unimodal form with a mode in the center. Part of the similarity among nations is likely to be an illusion due to respondents norming their position to the centre of their own country s political spectrum. That is, a location at position 5 in Australia likely has a meaning different from position 5 in Norway. But, regardless of doubts about the direct crossnational comparability of the numbers, the principal message is all but selfevident. Each nation s median voter position is somewhere in the middle of a nation s political space; a clear majority is within two units of the median on an eleven-point scale; and relatively small percentages report being extremists. This view of the electorates in Western is in no sense idiosyncratic or ephemeral. Comparisons to the left-right distributions reported for the 198s by G. Bingham Powell, relying mostly on the 1981 World Values Study, show similar unimodal forms (see Powell 2, 168). Figure 2 traces the stability of mean locations in the left-right space for the six nations on which James Adams and his colleagues report annually from 1976 through 1998, relying in their case on Eurobarometer surveys (Adams et al., 24, 596). The

6 5 prominent feature is stability. On this evidence, median/mean citizens are in the center of each nation s political space and stay there from year to year, with only modest dynamics. 1 [Figure 2 about here] What we know about the policy positions parties strongly, univocally, indicates party systems tend to present left-right options with the major parties standing astride the position of mean or median voters (Castles and Mair 1984; Laver and Hunt 1992; Huber and Inglehart 1994; Budge et al. 21; Klingemann 26; Benoit and Laver 26). The citizens themselves agree. The CSES asked the same respondents who self-placed on the leftright dimension to place their national parties. Figure 3 shows the citizen placements of the two largest parties in the nation s election at the time of the CSES surveys, with two exceptions. For Belgium the Flemish and Walloon party systems are shown separately, and for the Netherlands the four leading parties are shown, as two of the three typically largest parties ran third and fourth behind List Pim Fortuyn in 22. The party position data underlying the graph are presented in Appendix Table A2. [Figure 3 about here] In most nations it is apparent that the two major parties present electors with a choice between a position to the left and to the right of where the median citizen is located. The upshot is the election of parties to parliament and, in turn, the selection of parties in government that are too far left or right for the taste of a nation s median voter. This is the crux of the theme about collective representation that has emerged over the past 15 years (Huber and Powell 1994; Powell and Vanberg 2; Powell 2; McDonald, Mendes, and Budge 24, McDonald and Budge 25). To be sure, that research shows the degree of incongruence varies by electoral system type, and more recent research shows it varies by time within single- 1 Slight trends are visible in Italy, the Netherlands, and the United Kingdom. Autoregressive equations for all six indicate that those three plus France show statistically significant predictability from year to year. That, some systematic change is occurring four of the six beyond variability attributable to sampling error.

7 6 member district systems (Powell, forthcoming; Kim, Powell, and Fording forthcoming). And, it should be noted, in all likelihood the graphic depiction in Figure 3 understates the extent to which incongruence is likely to arise. 2 This is because citizen placements vis-à-vis expert placements have a tendency to draw parties toward the center (see Alvarez and Nagler 24; Golder and Stramski 27). Reviewed briefly, citizen political preferences along a left-right dimension form something similar to a normal distribution. From the details of six nations, the central tendencies of distributions appear to have only mild if any systematic dynamics through time. And, the major parties competing for electors votes tend to offer left-right policy positions a half to a full standard deviation to the left and to the right of the median/mean citizen. The consequence is for governments usually to form at some incongruent location to the left or to the right of the position of the median citizen. 3. Policy Representation How could democracies deal with this persistent incongruity between citizen preferences and policy positions pursued by parties in government? The existing evidence shows coalition formation reduces its degree, as multiparty proportional representation systems show less congruence than predominantly two-party SMD systems. But that only makes for a relative solution and leaves one to wonder what the Anglo-American democracies are doing about incongruence. Another possibility is for major parties to take it upon themselves to converge on the center. This has been shown to be the case, on average, for some SMD systems in the 195s and early 196s and then again in period between 1995 and 25. But that only makes for a fleeting solution. The possibility pursued here is to derive a set of theoretical expectations that holds for democracies generally, regardless of electoral system format and regardless of any peculiarities that come and go from time 2 Especially noteworthy is the dubious placement of the Forza Italia to the left of center.

8 7 to time. We arrive at these theoretical expectations by specifying a set of initial conditions and using a simulation approach to see what unfolds Simulations Ignore for the moment the valid and persuasive reasons for recognizing the position of a plurality party does not always record the policy desires of an electoral majority. If it did, providing congruent representation requires policy to move immediately to the plurality party s position after an election and to stay there until the next one. Granted, objections might be lodged against policy adoptions that track the passions of transient pluralities. But these objections are beside the point here. They stem from beliefs about the need for good government to balance policy stability and policy dynamism. When the concern is congruent representation, which it is for the purpose of this discussion, rapid change is called for. Switching focus to congruent representation of a median citizen, where, as the facts in require us to acknowledge, parties stand divergent and voters show but a limited tendency to make policy deterministic selections, the opposite is true. The most congruent outcome for the median citizen comes from sticking with something like the long-run mean, as the mean is the number that minimizes the equally weighted deviations. 3 Sticking to the mean requires never changing. Thus, when faced with divergent parties and multi-motivated voters often making unpredictable collective choices based on policy position taking alone, congruence with median citizen position is best served by never changing (except to keep pace with a trend, see fn. 3). Set-up. To illustrate how the tradeoff works and what it means for congruent representation in theory, we take control of the relationships between concepts and measured reality by introducing a series of simulations. 3 We are assuming here that median citizen positions are moving about as-if randomly. There might be something like a trend embedded in the movements. It is best to interpret this discussion of the median citizen mean positions as referring to either a stationary series or a de-trended non-stationary series.

9 8 We simulate two- and three-party systems. Party and voter positions are on the CMP metric. Zero marks the center; left is negative; right is positive. Left Party positions have a mean of -13; Right Party positions have a mean of +13; and Center Party positions have a mean of. Each party moves around its long-run mean so that its positions are normally distributed with a standard deviation of The median voter also moves around: normally distributed with mean =, and standard deviation = 5. 5 All movements of the three parties and median citzens are uncorrelated with one another. Also, the plurality party is randomly selected, to reflect outcomes that appear to have no single dimensional policy deterministic voting behind them. Such uncorrelated movements, even with de-trended series, and such completely random selection slightly overstate the case compared to what could be estimated empirically. About half the within-nation party positions show statistically significant, though usually modest, inter-party correlations (evidence not shown), and electorates show but a modest tendency to support the more centrist party in two-party systems (Cremona and McDonald 26) or to add support to parties that make centrist moves in multi-party systems (Ezrow 25). Nevertheless, there is only a small cost in terms of verisimilitude and a very large payoff to assuming randomness, at least for the time being. One payoff is to allow us to dispense with strict concern for a left-right dimensionality. While we have referred to a left-right dimension at several points in the discussion thus far, the randomness assumptions make 4 These numerical features rely on observations of party positions recorded by the Comparative Manifesto Project (CMP) (see, Budge et al. 21; Klingemann et al. 26). 5 Setting the electoral volatility at five is based on work by Robert Erikson and his colleagues. Erikson et al. (22) relate liberal-conservative positions in American party platforms to the liberal-conservative mood of the public. On the CMP metric, the two major American parties stand about 25 units apart on average, and movement of the public s liberal-conservative mood stays inside those bounds (Erikson, MacKuen, and Stimson 22, 265). The standard deviation of mood is 4.45 (Erikson, MacKuen, and Stimson 22, 219). That is, public opinion moves left and right to an extent slightly less than one-fifth of the distance that separates the parties. One-fifth of 25 is five.

10 9 dimensionality an mostly unimportant consideration. Given the randomness, all that follows holds for any policy on which the major parties take divergent positions. 6 Another payoff is a matter of conceptual and terminological convenience. Selecting a government follows these rules. In two-party systems the randomly selected plurality party s position is the position of the parliament and government. In three-party systems, the governing position is formed in each of the three different ways single party plurality alone, the median parliamentary party (MPP) alone, and a coalition as a weighted combination of either the plurality party and MPP or of the plurality-mpp single-party with a randomly selected party on its left or right. Finally, from among 2 simulated elections, we selected one election at random to serve as the starting point and let the process of governing develop for a series of a hundred years through twenty-five elections on a four-year cycle. We assume before the first election that the governing position is at zero, matching the long-run mean position of the median citizen. 7 With these various conditions in place, we ask how congruent the governing positions are to the positions of the median citizen and modal voters. Pace of Policy Change. As rapid versus no change is an important feature when it comes to representing median citizen versus modal voter, time needs a place in the analysis. It enters in the following way. A role of elections is to change the policy target. A changed target is installed shortly after an 6 What we are about to show could be undone by convergent non-centrist positions (e.g., U.S. Republicans tacit and Democrats explicit approval to continue a ban on alcohol in 1928) and by long periods without party alternation in government (e.g., notably, the American South from 188 to 197, as well as, though less consequentially, social democratic control in Sweden from the 194s into the 197s, conservative control in France from the late-195s to early 198s, and conservative control in Britain from the late 197s to late-199s). 7 This is not of much consequence. We alternatively did just as we say but entered the process at a randomly selected election, so its preceding position is not zero, and performed the same analysis reported below. The results are essentially the same.

11 election, but a governing position emerges from a series of policy changes each one of which typically involves an adjustment to the status quo policy inherited by the new government. This is consistent with the analysis of American policy making described and analyzed by Erikson and colleagues. They find a firm status quo base a unit root contribution in statistical parlance and policy adoptions/innovations are added to the base (Erikson, MacKuen, and Stimson 22, 325-8, esp ). The process of policy change is specified to follow this general formulation. Governing positiont = (1 β) Targett + β Governing positiont-1. It says the Left-Right governing position today, time t, is set according to a target value coming from the most recent election (i.e., Targett) i.e., from the Left-Right (or any policy dimension) position of the party or coalition controlling parliament and government and from the carryover policy status inherited by the newly installed government (i.e., Governing positiont-1). Whether policies consistent with the target take hold soon after an election depends on the pace of policy change, set by the value (1 β). 8 Importantly, in analyses and commentaries that have taken congruence to be focal point, the implicit concern is for representation as voice: asking, do the policy viewpoints match up? This, quite rightly, implies no carry forward from the previous governing position i.e., β =, and thus (β * Governing positiont-1) =. Where, as here, the concern is with representation as policy substance, the carry forward from the previous governing position has some weight, probably a good deal of weight. If, for example, the carry forward has a hefty weight, which enters in terms of a large β, the pace of change (1 β) is slow. A slow pace of change, say from β =.8, means movement toward the target will be slow i.e., =.2. 8 Aligning these terms with those Erikson and his colleagues use to describe the American policy process, our Targett is their Policy Activity, our Governing Positiont is their Policy, and the difference between our Governing Positiont and Governing Positiont-1 is their Policy (see Erikson, MacKuen, and Stimson 22, Chapters 8 and 9).

12 11 Movement toward a target occurs at a rate of.2 units of the difference between the last year s governing position and the target. As a simple example, take a current governing position of., a degree of policy stickiness so that a pace of change of (1 β) =.2, and a new target of + as set by the newly elected party. In the year after the election the governing position will not move immediately to + in such a slow-paced system. Rather, it will head toward + at a pace of.2 units of the difference between where it has been (Governing positiont-1, zero in this hypothetical) and where it is ultimately going (Targett, + in the hypothetical). In year one it will go from to +2 (last year s zero position plus.2 of the difference between + and ); in year two it will go from 2 to 3.6 (last year s +2 position plus.2 of the difference between + and +2); in year three it will go to (last year s +3.6 position plus the.2 of the difference between + and +3.6). That is, after three years the governing position will have moved not quite halfway to the target. After another three years, the position will again move halfway, to about 7.5. But, possibly, before this happens the governing party with a target of + is liable to be replaced by the other party with a governing position of, say, -. When that occurs, if it does, the governing position moves back toward and through the zero point. For median citizen positions consistently around zero, the governing positions will not ever have drifted too far from it as long as government control alternates. What will be the state of affairs for representation of the modal voter position? A slow pace of change, such as.2, keeps the process from reflecting it for a long time. Slow-paced change represents the median position reasonably well and the modal position poorly; fast-paced change represents the modal position well and the median position poorly. In short, there is a trade-off Compromise, in the Course of Time Figure 4 reports traces of incongruence values for median citizen and modal voter positions across eleven different paces of change, ranging across the entire spectrum from no change at all to full-scale immediate change.

13 12 The upper-left figure is for a two-party system. At the conclusion of a four year term in office the median and modal voter incongruence values intersect when the pace of change is slow, somewhere between.1 and.2. Thus for a two-party system to operate by representing the median citizen and modal voter equally well at the conclusion of a government, without trading more congruent representation of one for less congruent representation of the other, there would have to be a hefty carry forward effect of policies inherited by a new government. To be sure, there also needs to be some dynamism in terms of policy adoptions, but wholesale revamping of policies trades away a large amount of congruent representation of the median citizen position for near perfectly congruent representation of the modal voter position. [Figure 4 about here] The three-party case with a plurality party having control over government and policy making (lower-left), shows an intersection in the same vicinity, a pace of change between.1 and.2. The level of incongruence is slightly lower than in the two-party case (around 11 versus 12.5), because a viable centrist third option for voters offers a prospect of modestly tighter congruence In the three-party case where the MPP has complete control, representation is decidedly different. There is no intersection; congruent representation of the modal voter position is poor regardless of the pace of change; and the pace of change matters some but less for congruent representation of the median citizen position than in the three other cases. A system operating in this way continuously foreordains, at least implicitly, reasonably congruent representation of a median citizen position while operating as if the modal voter position holds no information about electoral preferences. Where the foreordination applies plausibly well, perhaps Belgium or Denmark (though, there, in much larger party-system contexts), congruent representation describes the process reasonably well. On the other hand, where the plurality party choice carries information about the leanings of preferences, perhaps Germany, it might be reasonable to think of forming a coalition, as in the fourth and final case.

14 13 Turning to the fourth case, a coalition government of either a plurality party with an MPP or with a singular plurality-plus-mpp operating in coalition with a party to its left or right, we see a trade-off similar to the twoand three-party cases in the two left figures. That is because, unlike the allpower-to-the-mpp case, the coalition takes some account of the plurality position. It compromises that information by taking account also of the MPP and thus has more congruence (lower incongruence) than the two figures on the left, concentrating as those cases do on the plurality party. The intersection of median-modal voter incongruence occurs at a level around, compared to 12.5 for the two-party plurality case and 11 for the three-party plurality case. Similar to both plurality-selection cases, the pace of change at the intersection is slow. It is, however, slightly faster;.8 as compared to approximately.85 in the two left-side graphs. The reason, as we explain immediately, comes from a built-in compromise of modal voter and median citizen positions through the coalition arrangement Centrist Tendencies A slow pace of change in the three cases where median-modal incongruence levels intersect is more than coincidence. The intersection occurs when the median and modal positions contribute equal weight to Governing positiont. Equal contributions come from splitting the difference between immediate change, which serves the modal voter position well, and no change, which serves the median citizen position well. Thus, to compromise the two positions at the conclusion of a time interval of k years between elections, the pace of change has to be set to.5. Writing the dynamic equation as a generalization across k years, we have Governing positiont = (1 β k ) Targett + β k Governing positiont-1. The inherited governing position and the newly installed target carry equal weight to produce the intersection i.e., β k = (1 β k ) =.5. The equal weight is considered by itself when the modal voter position is the exclusive target.

15 14 When the modal voter position is in coalition, having a pre-specified weight of.75 in the simulations, then [(1 β k ) x.75] =.375. If there is just one year between elections and a government is formed by the plurality party alone, the incongruence of the median and modal positions intersect simply when β =.5. When, as in the analysis which produced Figure 4, the time interval between elections is four years, then β 4 =.5. The fourth root of.5 is.841, and the pace of change is or.159, which is the approximate pace of change where the intersection occurs in two left-side graphs of Figure 4. Where elections are more frequent than every four years, the pace of change required to produce the intersection is faster. In three-year intervals, the cube root of.5 is.794 so the pace of change is.26; in two year intervals, the square root of.5 is.77 so the pace of change is.293. A longer interval between elections requires a reduced pace of change e.g., the fifth root of.5 is.871 so the pace of change is.129. In general, the pace of change is the k th root.5, where k is the number of years between elections. The longer the interval between elections the slower is the pace of change needed to reach an equalizing compromise between how congruent a governing position is with positions of the median citizen and modal voter. As elections tend to occur every two to five years in most democracies, the pace of change that brings the governing position approximately equidistant from the two indicators of majority preferences is slow, between.1 and.3. In coalition governments all that we have said in the preceding paragraph holds, except the k th root is of the quantity.5 multiplied by the weight the plurality party holds in government. In the coalition result in Figure 4 (lower-right), this means taking the fourth root of.375. And.375 ¼ =.783, which is close to the pace of change at the intersection in the lower-right graph of Figure 4. What is more, as targets swing left, then right, then left,, the switching ensures, on average, that the governing position passes through the long-term median citizen location. Focusing on the case of the two-party system for a simple example, one can calculate that a movement halfway

16 15 toward a target value (halfway to +13) followed by a switch in the other direction halfway between the endpoint of +6.5 and 13, must pass through on its way to Therefore, on average, with a transition from Right to Left and vice versa, the governing position passes close to the long-run position of the median citizen, responding at each turn to the position of the modal voter. The same holds in the three-party plurality and coalition cases, but the calculations are a little more complex. Finally, as we trace the path farther through time, again using the two-party case as a simple example, it is apparent that the endpoint of the transitions i.e., the concluding year for one set of governors before governing moves in the opposite direction the next year tend to collapse toward the political center. Set the first transition, as above, to 6.5, and allow successive transitions in the two-party case to run between +13. The sequence of end-state governing positions equilibrates at on the Right and 4.33 on the Left. The sequence is +6.5, 3.25, +4.88, 4.6, +4.47, 4.27, +4.37, 4.32, +4.34, 4.33, +4.33,, That is, the equilbria are twofold: one on the left, located one-third of the distance between the median citizen position and the average governing position on the left, and one on the right, located one-third of the distance between the median citizen and the average governing position on the right. The world does not work in this simple way of left-right-left-right- transitions with each successive election, but the two general principles just reviewed are important to recognize. Governing positions tend to track around and through the center, with a tendency much more toward the center than the idea of governing positions lurching immediately to the position of newly installed governing parties, as in analyses that concentrate their attention on representation as voice Formalizing the Centrist Tendencies Clearly there is a structure to the representational process we have been describing. An important element in it is the positions of the parties, especially the viable plurality parties. Party positioning is not all that

17 16 matters, however. We count five important conditions: (1) symmetrical bracketing of the median citizen by the plurality parties, (2) coalition formation, (3) distance between the modal parties on the left and right, (4) frequency of alternation in government by the plurality parties, and (5) pace of policy change. We begin our discussion of the five with a focus on elements (5), (4), and (3): the pace of change, the frequency of alternation, and a set of positive and negative target values. We designate the governing position as G, the target as T, and the stickiness of policy as β where stickiness is the flip side of the pace of change (1 - β). For simplicity at the start we set the time interval between alternations to one year, so that k = 1. The result shows that the equilibrium interval is a proportion of the magnitude of the target values. We have, under the same change process as before, Gt = (1 β k = 1 ) Tt + β k = 1 Gt-1. At the time of alternation across the position of the median citizen (where underscoring across reminds us of the bracketing condition to which we need to return below), a positive Gt follows a negative Gt-1 or vice versa. Thus, for example, Gt = (1 β) Tt + β (-Gt-1). Rearranging, Gt β (-Gt-1)= (1 β) Tt. or Gt + β (Gt-1) = (1 β) Tt. And, because +Tt and Tt have the same magnitude, (1 + β) +Gt = (1 β) +Tt. Therefore, Gt = [(1 β) / (1 + β)] +Tt. or +Gt = [(1 β) / (1 + β)] Tt.

18 17 In words, the size of the equilibrium interval is a proportion of the positive and negative values set as targets by the plurality parties, and that proportion is determined by the degree of stickiness of policy change. If there is no carry forward of the governing position from the previous governing position, β =. Then, continuing to assume k = 1 (the target value is reached in one year), the equilibrium interval reaches all the way to the positions of the modal parties on the left and right since [1 ) / (1 + )] = 1. If, however, the stickiness is.5, then the equilibrium interval is one-third of the distance to each of the target values since [(1 -.5) / (1+.5)] =.333. Therefore, the more polarized the major parties on the left and right or the faster the pace of change, the wider the policy equilibrium. The role of party polarization is intuitive, but the less obvious pace of change can play the same role. A fast paced system, e.g., β =.2, with parties standing at +12 has a policy equilibrium interval, +8 i.e., (.8/1.2) (+ 12) = +8. A slower paced system, e.g., β =.6, with parties in the much more divergent positions of +32 has the same equilibrium interval i.e., (.4/1.6) (+ 32) = +8. Most governments of Western democracies last more than a year, and when short-lived governments change they do not often alternate from left to right. To account for longer time intervals between alternations, a per-year pace of change needs to be added to our considerations as do durations of more than one year between alternations i.e., for k > 1. We choose values for the slowness of annual policy change between.5 and.95 i.e.,.5 < β <.95. We set the frequency of alternation between plurality parties on the left and right at 4, 8, 12, and 16 years. And we again set the major parties at The four graphs in Figure 5 show how wide the policy equilibrium interval would be under these various conditions. [Figure 5 about here] A slow pace of change, in the neighborhood of.5 per year, creates a narrow and centrist equilibrium interval, between + 5. How large the interval is with faster-paced change depends on the frequency of alternation in government. With frequent alternations, say every four to eight years, an annual pace of change up to.15 creates an interval between center-right and

19 18 center-left i.e., + 7.5, something in the range of party position taking where one would expect to find Christian democratic and most liberal parties. The combination of less frequent party alternation in government and any moderately fast-paced change, however, pushes the equilibrium interval to policy alternations that swing from the major party positions on the right and left. 4. Alternation in Government With median citizens standing in the middle of a nation s political space and with the major party alternatives usually standing to the left and right, alternation in government is one of the key elements in keeping the policy substance in line with the preferences of median citizens. Of particular interest is whether the government selection rules and norms of Western democracies play a role in producing alternation Theoretical Payoff The reasons to care whether selection rules and norms have this role touch upon both normative and empirical concerns in democratic theory. In the normative sense, having the ebb and flow of votes producing reasonably accurate representation removes the obligation on citizens to be the fully informed, policy deterministic voters. To ask that of voters is to require them to devote resources to the political process that are almost surely beyond their capabilities. Few voters have either the time or high quality information sources to become fully informed policy deciders. And, were some particular voter tempted to think he or she could (should?) make time to do so, a moment s thought would tell him or her that the probability of making a difference is too low to commit to such an enterprise. But, then, this realization only amounts to a critique of electoral democracies if such fully informed, rational deterministic voters are necessary for producing reasonably accurate representation. If reasonably accurate representation can be produced by much simpler and less time consuming means, the prospects for democracy on the scale of a nation-state look much brighter.

20 19 From an empirical theoretical perspective, entertaining the possibility that a decision rule or norm is pivotal significantly reorients one s thinking toward the vital role that parties play in the process of representative democracy. This is what E. E. Schattschneider had in mind, I suppose, when he touted political parties as the creators of modern democracies and unthinkable without them (Schattschneider 1942, 1). He soon thereafter expanded on his claim in this way. They [pre-democratic theorists] made the mistake of underestimating the difficulties arising from the numbers, preoccupation, immobility, and indifference of the people. Everyone took it for granted that the people themselves would assume responsibility for the expression of their own will as a matter of course without so much as dreaming of the intervention of syndicates of self-appointed political managers and manipulators who for reasons of their own might organize the electorate and channelize the expression of popular will (Schattschneider 1942, 14). Party systems capable of channelizing the popular will could take a Downsian form, where parties converge to the position of the median voter (Downs 1957). This would allow voters safety and freedom, effectively, to choose at random. Whatever the voters choose, the empowered party or parties are committed to pursuing the policy preferences of the median citizen/voter. But, of course, we know that parties do not often converge. We also know that voters make choices on many grounds other than policies on offer candidate appeal, general dispositions toward parties, particular but transient issues such as corruption, and the like and thus parties do not actually have good reason converge based on the premise that their policy offerings are all that matter. The back story to the proposition here holds that parties standing at divergent positions on each side of the median citizen allow voters to choose as if by random selection and still empower left and right alternatives that produce something similar to the Downsian version of party democracy.

21 2 Parties hold the responsibility for nearly all that happens, because the role of the decision rule enters in such a way that the as if random selections occur. Accurate representation follows, while citizens give most of their attention to the joys and trials of their daily lives Rhythm and Rhyme of Elections The first column of Table 1 reports the change in vote percentages for incumbent parties from the preceding election to a current one for sixteen nations. 9 [The details of the data used see McDonald and Budge 25, 1-13.] In fifteen of the sixteen countries, incumbent parties vote percentages declined. On average, across the 215 elections, the decline is 2.33 percentage points. Incumbent parties generally are not capable of maintaining the level of support they received before entering government. This could possibly signal that governments have a difficult time keeping together the coalition of voters they had been able to stitch together before entering governments. Voters may have become disaffected by the performance of the economy (see, among many others, Taylor and Dorussen 2) or because other hopes prompted by pre-election party rhetoric have been dashed (Mueller 197; Stimson 1976). Or, perhaps, many macro polities react as Erikson, MacKuen, and Stimson say about the American electorate. Electoral choices are like a thermostat that turns on and off, right to left or vice versa, as policies move too far from the middle for comfort (Erikson, MacKuen, and Stimson 22, ). [Table 1 about here] Alternatively, perhaps all of the lost momentum of incumbent parties was to be expected in the first place, because the reason for getting into government was a consequence of a confluence of good fortunes that are unlikely to repeat themselves from one election to the next. There are, in other words, forces restoring party competition. What the particular forces are vary from election to election, because it was often the idiosyncratic that 9 France has been excluded from the first column of this table because it is not always clear (to this author) which parties are the incumbent parties.

22 21 produced the good fortune. Whatever they are, however, the effect is a statistically significant vote loss of 2.33 percentage points. Now consider column 2 in Table 1. The numbers there are the difference between incumbent party votes at the previous election, the one that preceded the current incumbents entry into government, and the average vote percentages the incumbent party or parties receive in the period covered i.e., between the late-194s and the mid-199s. In all sixteen countries the vote percentages won before entering government exceeded expectations. Parties apparently are incumbent parties, in part, because their vote support exceeds expectations. Indeed, on average the vote percentages in excess of expectations nearly match the average vote percentage loss at a current election. The small difference, -.26%, between incumbent parties current vote and what that party or those parties could expect is statistically insignificant. The vote support in excess of normal expectations appear to have something important to do with propelling parties into government, and the return to normality may have something equally important to say: incumbent retention in office becomes nothing better than a 5:5 bet. Is that so? Among seventeen nations, with France now included, there appears to be something to the notion of a 5:5 prospect. Table 2 reports the proportion of times a party or parties in government stand to the right of a nation s center of course, meaning also that one minus the proportion is the relative frequency of a government standing to the left of center. It is not that the overall proportion of right versus left-standing governments is.53 that is important, for the proportions of interest are those within each nation. On the within-nation score, twelve of the seventeen nations have right governments following an election between 35% and 65% of the time. France, Italy, and New Zealand have governments standing to the right of center after 7% or more of their elections. Sweden and Belgium The mean party votes are standing in here for simplicity. I have performed the same calculation after adjusting expected party votes for trends (see McDonald and Best 26). The results are essentially the same.

23 22 tilt the other way; between 73% and 8% of the governments empowered after an election stood to the left of center. [Table 2 about here] Well known are the facts that many of these nations have experienced long runs of parties on one side or the other of the center Sweden s Social Democratic party dominance from the 195s (in these data, but from the 193s in full) until the mid-197s, the Australian Liberals-plus-County parties from the 195s in the 197s, Gaullists in France from the late 195s into the 198s, and British Conservatives from the late 197s until In fact, most countries have parties with runs in government steadily on the left or steadily on the right for twelve or more years. But a close inspection of steady runs on the left or the right indicate it is inferentially risky to make too much of those seemingly long, steady runs. Column 2 of Table 2 reports the number of changes in direction, from right to left or vice versa, and the number of total opportunities to switch directions. 11 Indicated also is whether the number of switches is statistically significantly greater or less than could be expected in a series of 5:5 chance selections. 12 A statistically significant large number of switches indicates too many direction changes to be a chance-like outcome, as if nearly every time a left government forms it would all but necessarily be followed by a government of the right. A statistically significant small number of switches indicates too few direction changes for the outcomes to be chance-like, because no, one, or only a few switches shows that being a government on the left or right has too much staying power to be something that chance forces would produce. Among the seventeen nations only the small number of switched directions in Australia, three of the seventeen opportunities, is statistically 11 Changed directions are assessed based on government positions to the left or right of the average value of a nation s median voter, where both the government and the median voter positions are scored using the Comparative Manifesto Project data (see, McDonald and Budge 25, ). 12 The significance test is the nonparametric runs test, perhaps best known today in the form of Geary s test for residual auto-correlation (Geary 197).

24 23 significant. Australia is the one nation where the opposition has been unable to gain government power as frequently as could be normally expected. Noteworthy in this particular case is the fact that three times between 195 and 197 Australia s Liberal-Country electoral alliance ran second to Australian Labor but did not have to relinquish power because the electoral system translated the Liberal-Country second place finish into a parliamentary majority. Thus, but for the peculiarities of the electoral translations in this one nation, all of the series of elections in the seventeen countries show electoral choices as if they have forces restoring party competition to extents that produce right-to-left and left-to-right alternations in government that might be able to keep policy choices somewhere in the vicinity of the median citizen. 6. Conclusion The exploration here has sought to check on how plausible it is to think a particular descriptive theory of democracy in Western society might be working. In summary form the thesis is this: democracy can be infused with the energy to change and sustained with the stability of centrist policymaking by relying on the statistical tendency of today s most popular party or parties to regress at the next election. There is something to the thesis, or at the very least there is no certain evidence against it. All the bits of evidence presented here tend to point in a direction consistent with this descriptive theory s possibility, although it must be said that they are not yet so tightly knit together as to indicate this is how Western democracies have been operating. The theoretical circle cannot be closed until and unless the apparent regression to the mean component is directly and unequivocally linked to alternation in governments. On the evidence here, it looks as if it can be. But, that linkage remains somewhat elusive. It hinges in part on putting a firm hand on the possible idiosyncrasies that give a push to higher than normal popularity at one time only to recede. That the sources are idiosyncratic makes the task daunting, which one can only hope is not to say non-falsifiable.