¹ Key, , Southern Politics in State and Nation (New York: Knopf, 1949Google Scholar).

² For a review of research that finds a closeness-turnout correlation, see Matsusaka, John and Palda, Filip, “The Downsian Voter Meets the Ecological Fallacy,” Public Choice 77 (December 1993CrossRefGoogle Scholar).

³ Rosenstone, Steven J. and Hansen, John Mark, Mobilization, Participation, andDemocracy in America (New York: Macmillan, 1993Google Scholar).

⁴ Ibid.

⁵ Amy, Douglas J., Real Choices/New Voices: The Casefor Proportional Representation Elections in United States (New York: Columbia University Press, 1993Google Scholar); Powell, G. Bingham Jr., “American Voter Turnout in Comparative Perspective,” American Political Science Review 80 (March 1986CrossRefGoogle Scholar).

⁶ Key (fn. 1); Morton, Rebecca, “A Group Voting Model of Public Good Provision,” Social Choice and Welfare 4 (May 1987CrossRefGoogle Scholar); idem, “Groups in Rational Turnout Models,” AmericanJournal of Political Science 35 (August 1991Google Scholar); Uhlaner, Carole, “Rational Turnout: The Neglected Role of Groups,” Amer ican Journal of Political Science 33 (May 1989Google Scholar); Cox, Gary W. and Munger, Michael, “Closeness, Ex penditures and Turnout: The 1982 U.S. House Elections,” American PoliticalScience Review 83 (March 1989Google Scholar); Matsusaka, John, “Election Closeness and Voter Turnout: Evidence from California Ballot Propositions,” Public Choice 76 (December 1993CrossRefGoogle Scholar); Rosenstone and Hansen (fn. 3); Nalebuff, Barry and Shachar, Ron, “Follow the Leader: Theory and Evidence on Political Participation” (Manuscript, Yale University, 1995Google Scholar).

⁷ Rosenstone and Hansen (fn. 3), 164, 169–70.

⁸ Riker, William and Ordeshook, Peter, “A Theory of the Calculus of Voting,” American Political Science Review 62 (March 1968CrossRefGoogle Scholar). Of course, doubts about the theoretical standing and empirical success of the pivotal voter have been expressed by, for example, Cox and Munger (fn. 6); Matsusaka (fn. 6); and Jackson, Robert A., “The Mobilization of Congressional Electorates,” Legislative Studies Quarterly 21 (August 1996CrossRefGoogle Scholar).

⁹ Elites might engage in both mobilization and demobilization (and persuasion). As long as the total turnout boost from mobilization exceeds the total turnout declinefromdemobilization (and perhaps from persuasion via negative advertising), turnout and closeness will covary positively, ceteris paribus.

¹⁰ Rosenstone and Hansen (fn. 3).

¹¹ Ibid., 164–65.

¹² Ibid., 171.

¹³ We use “social density” as a synonym for social capital at the aggregate level. Social capital has two components (at the individual level): social connectedness (being plugged into social networks, having many contacts with whom repeated interactions occur) and interpersonal trust (reflecting low transaction costs within the networks or a “cooperative disposition” fostering a sense of mutual obligation). See Rosenstone and Hansen (fn. 3); John Brehm and Wendy M. Rahn, “Individual-Level Evidence for the Causes and Consequences of Social Capital,” American Journal of Political Science (forthcoming). The text below provides one way to formalize these notions.

¹⁴ Suppose there are n voters and let c_j = 1 if voter j is contacted by the focal candidate, 0 otherwise. Let Δ_j be the increment to j's probability of turning out, if he is contacted; p_jk be the increment to the probability that j will urge k to vote, due to j's having been mobilized by the candidate; τ_kj. be the increment to k's probability of turning out, if he is contacted by j. For a given decision about whom to contact—(c₁, …, c_n)—the vote yield Y(c₁, …, c_n) is If p_jk and τ_kj correlate positively(j is more likely to contact those who are more likely to be persuaded to turn out by his urging), c_j correlates positively with Δ_j (the focal candidate is more likely to contact those more likely to be persuaded to turn out by his appeals), and c_j. correlates positively with p_jkτ_kj (the focal candidate is more likely to contact those who produce more secondary bang for the buck), then the lower bound stated in the text holds.

¹⁵ We assume that higher values of y are dominated by the secondary mobilization terms rather than by the primary mobilization term. Rosenstone and Hansen (fn. 3) and Robert Huckfeldt and John Sprague argue that secondary mobilization effects dominate even in the U.S., a largely untraditional “mass” society. See Huckfeldt, and Sprague, , “Political Parties and Electoral Mobilization: Political Structure, Social Structure, and the Party Canvass,” American Political Science Review 86 (March 1992CrossRefGoogle Scholar).

¹⁶ Note that we focus on the simplest case in which the vote yield of k contacts is known with cer tainty: it is just ky. Allowing for risky returns to contacting does not change anything essential.

¹⁷ Because k_s is taken as given in Tanaka's problem, one can renormalize her utility function by adding the constant Pr(V_T > V_s + k_sy)B, and dividing by the constant B, yielding Pr(V_T + k_Ty > V_s + k_sy) - ck_T/B. A little algebra then yields the formulation in the text.

¹⁸ Tanaka will wish to contact some positive number of supporters if the marginal utility of contact ing another supporter is positive, when k_T = 0. Since dU_T/dk_T = f(k_Ty - k_sy)y - c/B, Tanaka's mar ginal utility is positive when f(-k_sy)y > c/B, which reduces to the expression in the text when k_s = 0.

¹⁹ Rosenstone and Hansen (fn. 3).

²⁰ Note that the pivotal voter model, which posits direct effects of closeness on voter behavior (unmediated by elite mobilizing efforts), does not predict these results.

²¹ Denver, David and Hands, Gordon, “Marginality and Turnout in British General Elections,” British Journal of Political Science 4 (January 1974CrossRefGoogle Scholar); idem, “Marginality and Turnout in General Elections in the 1970s,” British Journal of Political Science 15 (July 1985Google Scholar); Berch, Neil, “Another Look at Closeness and Turnout: The Case of the 1979 and 1980 Canadian Elections,” Political Research Quarterly 46 (June 1993CrossRefGoogle Scholar); Matsusaka and Palda (fn. 2).

²² Watanuki, Joji, “Patterns of Politics in Present-day Japan,” in Lipset, Seymour Martin and Rokkan, Stein, eds., Party Systems and VoterAlignments (New York: Free Press, 1967Google Scholar); Curtis, Gerald L., Election CampaigningJapanese Style (New York: Columbia University Press, 1971Google Scholar); Flanagan, Scott C. and Richardson, Bradley M., Japanese Electoral Behavior: Social Cleavages, Social Networks, and Partisanship (London: Sage, 1977Google Scholar); Shupe, Anson, “Social Participation and Voting Turnout: The Case of Japan,” Comparative Political Studies 12 (July 1979CrossRefGoogle Scholar).

²³ Sasago, Hiroto, Abe, Kazuyoshi, and Muraoka, Katsuya, Seijiientin noiozu (The structure of political finances) (Tokyo: JICC shuppan kai, 1990Google Scholar), 59–60; Ishikawa, Masumi and Hirose, Michisada, Jiminto: choki shihai no kozo (The LDP: The structure of long-term dominance) (Tokyo: Iwanami shoten, 1989Google Scholar); Ramseyer, J. Mark and Rosenbluth, Frances, Japans Political Marketplace (Cambridge: Harvard University Press, 1993Google Scholar).

²⁴ Tsujimura, Akira, “Shimbun ga moteasobu 'senkyo yosoku,' I” (Newspapers trifle with “election forecasts,” part I), Shokun 18 (September 1986Google Scholar). One often hears the laments of the candidates slotted in first place. A politician complained to a newspaper reporter after an election, “My vote total went down by ten thousand votes because you said I was a shoo-in.” Kabashima, Ilcuo, Seiji Sanka (Political participation) (Tokyo: Tokyo University Press, 1988Google Scholar), 172. As one campaign organizer explained, “The campaign effort falters as soon as news about your front-runner status gets out People in the lower reaches of our campaign organization do not think they have to work as hard, and people we are counting on to vote feel less obligation to show up.” Tsujimura, Akira, “Shimbun ga moteasobu 'senkyo yosoku,' II” (Newspapers trifle with “election forecasts,” part II), Shokun 18 (October 1986Google Scholar), 188.

²⁵ Flanagan, Scott C., “Mechanisms of Social Network Influence in Japanese Voting Behavior,” in Flanagan, Scott C. et al., eds., The Japanese Voter (New Haven: Yale University Press, 1991Google Scholar), 453–57.

²⁶ Bradley Richardson, “The Japanese Voter: Comparing the Explanatory Variables in Electoral Decisions,” in Flanagan et al. (fn. 25), 383–84.

²⁷ Kabashima (fn. 24) mentions in passing that competitiveness and turnout move together, the point is also made by Sato, Seizaburo and Matsuzaki, Tetsuhisa, Jiminto seiken (The LDP administraboa) (Tokyo: Chuo koron sha, 1986), 116Google Scholar–17.

²⁸ Research into turnout has focused (1) on changes in national-level turnout over time-Kabashima (fn. 24), 171–80; Miyake, Ichiro, Tohyo Kodo (Voting behavior) (Tokyo: Tokyo University ftess, 1989Google Scholar), 227; Arai, Kunio, Senkyo, Joho, Yoron (Elections, information, public opinion) (Tokyo: NHK Books, 1988Google Scholar), 32–35; (2) on broad demographic effects-Arai, 84–92; Blaker, Michael, Japan at the Palls (Washington, D.C.: American Enterprise Institute, 1976Google Scholar), 149; Flanagan (fn. 25), 174–78; Flanagan and Richardson (fn. 22), 37; Hrebenar, Ronald J., TheJapanese Party System: From One-Party Rule to Coalition Government (Boulder, Colo: Westview Press, 1986Google Scholar), 15–19; and (3) on anecdotal evidence that “nothing apparently draws the voters like a good conservative blood-letting.” Thayer, Nathaniel, How the Conservatives Rule Japan (Princeton: Princeton University Press, 1969Google Scholar), 137.

²⁹ Pharr, Susan J., Political Women in Japan: The Search for a Place in Political Life (Berkeley: University of California Press, 1981), 25–26Google Scholar.

³⁰ Ibid, 32.

³¹ Sumikolwao, , TheJapanese Woman: TraditionalImage and Changing Reality (Cambridge: Harvard University Press, 1993Google Scholar), 234.

³² Flanagan (fn. 25), 158.

³³ In 1975,66.2 percent of women aged twenty to twenty-four were in the workplace, compared with only 43.9 percent of women aged thirty to thirty-four. In 1990 the corresponding figures were 75.1 percent and 51.7 percent, respectively. So while the number of young women in the workforce has increased over time, the number who quit their jobs when they marry or have children has increased as well. These women, along with those who never enter the workforce, are the “young mothers* to whom we refer, who are active in local social networks. See Iwao (fn. 31), 63. See also Imamura, Anne E., UrbanJapanese Housewives:At Home and in the Community (Honolulu: University of Hawaii Press, 1987Google Scholar). Imamura (pp. 116–20) points out that membership in the PTA and other groups related to children is typically automatic and that most women feel they have to take a turn as an officer. While many women return to work once their children are older, they tend to stay closer to home or work shorter hours. See also Brinton, Mary C., Women and the Economic Miracle: Gender and Work PostwarJapan (Berkeley: University of California Press, 1992Google Scholar). Brinton reports (p. 176) that only 31 percent ofworking women aged forty to forty-four work full time, while 34 percent work part time or as temps, 16 percent are self-employed, 15 percent are family enterprise workers, and 4 percent do piecework. And see also Lam, Alke, “Equal Employment Opportunities for Japanese Women: Changing Company Practice,” in Hunter, Janet, ed., Japanese Women Working (London: Routledge, 1993Google Scholar).

³⁴ Bestor, Theodore C., Neighborhood Tokyo (Stanford, Calif.: Stanford University Press, 1989Google Scholar).

³⁵ Iwao (fn. 31), 244.

³⁶ Richardson, Bradley, “Urbanization and Political Participation: Th e Case of Japan,” American Political Science Review 67 (June 1973CrossRefGoogle Scholar), 436; Matsubara, Nozumu and Kabashima, Ikuo, “Tanaka ha assho Jiminto taihai no kozu” (The logic of an overwhelming Tanaka Faction victory amidst an LDP defeat), Cbuokoron 99 (February 1984Google Scholar).

³⁷ Richardson (fn. 36), 437.

³⁸ Flanagan (fn. 25), 154; Richardson (fn. 36), 438.

³⁹ Richardson (fn. 36), 437–38.

⁴⁰ Cox and Munger (fn. 6); Denver and Hands (fn. 21,1974,1985); Berch (fn. 21); Matsusaka and Palda (fn. 2).

⁴¹ The campaign period lasted twenty days for the 1958 through the 1980 elections, fifteen days for the 1983 through the 1990 elections, and fourteen days in 1993.

⁴² Iwai, Tomoaki, Seiji shikin no kenkyu (Research on political financing) (Tokyo: Nihon keizai shimbun-sha, 1990), 66–67Google Scholar.

⁴³ Personal communications with Home Affairs Ministry bureaucrats and Japanese journalists. For a discussion of the plausibility of campaign period spending reports, see also Cox, Gary W. and Thies, Michael F.. “The Cost ofIntraparty Competition: SNTV and Money Politics in Japan,” Compar ative Political Studies 31 (June 1998Google Scholar).

⁴⁴ Hrebenar (fn. 28), 48.

⁴⁵ In various years there were also a couple of two-seat and six-seat districts and one one-seater. We drop them from the anlaysis for the sake of simplicity. In early 1994 Japan abandoned this electoral system (in use from 1925 through 1993, with a brief interruption in the 1940s) in favor of one combining single-member constituencies with large proportional representation districts.

⁴⁶ Another well-known issue is whether ex post measures of closeness correlate highly with ex ante measures, the latter being the theoretically relevant ones. If the election has equilibrated-in the sense that (1) everyone shares the same expectations about who will finish where in the poll (and hence how close the race will be), (2) everyone is behaving optimally in light of those expectations; and (3) optimal behavior is consistent with the postulated expectations-then ex post measures of closeness will simply be realizations of a particular draw from an ex ante distribution. We follow a host of others in assuming that the ex post measure is acceptable if not ideal; see, e.g., most of the twenty-five studies cited in Matsusaka and Palda (fn. 2).

⁴⁷ For our data, see http://dodgson.ucsd.edu//lij. For the Home Ministry data, see Jichisbo stnkyo kyoku. Shuugtin Giin Sosenkyo Saiko Saibansho Saibankan Kokumin shinsa: Kekkacho (The results of elections to the House of Representatives and of the national referendum on Supreme Court justices) (Tokyo: Home Affairs Ministry Public Elections Bureau, various years).

⁴⁸ One reason for worrying about redistricting in the present analysis is that we include the lagged dependent variable in our estimations, and, obviously, this procedure would not make much sense if district number 31 in 1976 were substantially different from district number 31 in 1972. The reason that it does not make much practical difference whether we include redistricting years or not is that the Japanese redistrictings affected only a small minority of districts in any substantial way.

⁴⁹ We prefer using the lagged dependent variable to using the Prais-Winsten (or other) GLS estimator for the reasons articulated in Beck, Nathaniel and Katz, Jonathan N., “What to Do (and Not to Do) with Time-Series Cross-Section Data,” American Political Science Review 89 (September 1995), 634CrossRefGoogle Scholar–47. Reestimating all our models using Prais-Winsten does not affect any of our conclusions. We report Huber cluster robust standard errors in Tables 1, 2, and 3 (implemented through Stata's cluster and robust options in its “regress” procedure). These are similar to a restricted version of Beck and Katz's panel-corrected standard errors. Use of the latter is not indicated in our case because we have a relatively short time series (T = 8).

⁵⁰ Arguably, a 1 percent gap by this measure means something different in districts of varying magnitude (M): 1 percent of the registered electorate is 2 percent of a Droop quota in a one-seat district, but 5 percent of a Droop quota in a four-seat district, for example. We correct for this by dividing by the Droop quota in each district or, equivalently, by multiplying by M + 1. This allows us to pool the observations from districts of varying magnitude and estimate a common slope.

⁵¹ Compare Cox, Gary W. “Closeness and Turnout: A Methodological Note,” Journal of Politics 50 (August 1988CrossRefGoogle Scholar). We tried using the margin and margin squared, as suggested by Cox, but found no significant nonlinear effects. Note that Cox's recommendation that a squared term be included matters from a practical standpoint only if the data actually have observations for which the raw vote margin as a percentage of the electorate (Mc) is large. But the largest observed value of Me is 18 percent in Japan (much lower than in the U.S.). So, theoretically speaking, the relationship between turnout and Mc could be negative over the entire range of observations. And, as it turns out, it is: the squared term is never significant and not even consistently positive in our analyses.

⁵² In order to bring the magnitude of the coefficients on NCPER and NLDPPER into line with the others in the model, we use the number of candidates per hundred thousand electors, rather than just per elector. This of course affects only the placement of the decimal point in the coefficient estimate.

⁵³ Hrebenar(fh.28),48.

⁵⁴ This is calculated by dividing the coefficient on MARGIN by the Droop quota (equivalent to multiplying by M + 1 = 5 for a four-seat district), which gives us .191 x 5 = 0.956. Recall that we standardize the closeness measure so that the same measure can be used in districts of three, four, and five seats.

⁵⁵ To investigate whether or not the increase in expenditures due to closeness is driven by the candidates actually on die margin, we reestimated equation 1 using the per elector expenditures by the two candidates on the cusp between winning and losing (the last winner andfirstloser) as die dependent variable. (This specification required some modification of the right-hand-side variables as well.) Our results (available on request) show that the two marginal candidates do indeed account for a disproportionate share of the overall effect.

⁵⁶ The government's official measure of urbanness is the percentage of the population living in “densely inhabited districts,” analogous to the “standard metropolitan statistical area” used in the United States.

⁵⁷ We have also used variables tapping the occupational structure of districts as controls, but these turn out to be highly correlated with the urbanness of the district and so are excluded here.

⁵⁸ Cox and Munger (fn. 6). The variability in total expenditures is larger in the U.S., because the variability in competition is larger. If one compares the impact of equal expenditures, a different comparison arises. Using an exchange rate of about 250 yen per dollar for the early 1980s, one more dollar (250 more yen) per voter would lead to an increase in turnout of .038 x 250 = 9.5 percent. This compares to afigureof 4.0 percent from Cox and Munger for the U.S. case.

⁵⁹ Given the substantial changes over time in the percentage of women working, one might wonder whether the impact of the social indicators that we use might change over time. We have investigated this matter and found no systematic change in effect over time. For example, the effect of %pop < 15 does not vary significantly from year to year. This may be due in part to the fact, noted above, that employment change has been much greater among women without young children than among women with young children.

⁶⁰ Riker and Ordeshook (fn. 8); Green, Donald and Shapiro, Ian, Pathologues of Rational Choice Theory (New Haven: Yale Uuiversity Press, 1994Google Scholar).