¹ Kelley, Stanley Jr., Ayres, Richard E., and Bowen, William G., “Registration and Voting: Putting First Things First,” American Political Science Review, 61 (June 1967), 359–79CrossRefGoogle Scholar; and Kim, Jae-On, Petrocik, John R., and Enokson, Stephen N., “Voter Turnout Among the American States: Systemic and Individual Components,” American Political Science Review, 69 (March 1975), 107–23CrossRefGoogle Scholar.

² Phillips, Kevin P. and Blackman, Paul H., Electoral Reform and Voter Participation (Washington: American Enterprise Institute for Policy Research, 1975), p. 14Google Scholar; and Tarrance, V. Lance, “The Vanishing Voter: A Look at Non-Voting as a Purposive Act,” in Voters, Primaries, and Parties, ed. Moore, Jonathan and Pierce, Albeit C. (Cambridge, Mass.: Harvard University Institute of Politics, 1976), pp. 11, 12Google Scholar. See also the minority views in Committee on Post Office and Civil Service, Voter Registration by Postcard, Senate Doc. 352, 93d Congress, 2d sess.

³ Senator McGee, Gale, Congressional Record, 13 March 1975, p. S6451Google Scholar. For other claims by politicians, see Voter Registration by Postcard; and the speeches by Senators McGee, Kennedy, Cranston, Williams, Montoya, and Humphrey on March 13, reported on pp. S6450–54 of the Congressional Record. Probably the most active and pervasive interest group working to loosen registration laws is the League of Women Voters. See, for example, Administrative Obstacles to Voting, Pub. No. 206 (Washington: League of Women Voters of the United States, 1972)Google Scholar. The Yankelovich polling organization suggested that registration laws kept 26 million people from the polls in 1972. This figure is an extrapolation from the survey respondents who told interviewers that they would have voted if they had not had to register. “A Study of the Registration Process in the United States (The Registered and the Non Registered)” by Daniel Yankelovich, Inc. for The Student Vote, Inc., 1973, p. 23Google Scholar.

⁴ Similar bills were introduced in 1971, 1973, and 1975. All of them foundered on opposition by Republicans and southern Democrats.

⁵ A description of sampling procedures and other technical matters is in Bureau of the Census, “Voting and Registration in the Election of November 1972,” Current Population Reports, Population Characteristics, October 1973, Series P–20, No. 253Google Scholar. We obtained the data tape from the Inter-University Consortium for Political Research. Neither the ICPR nor the Census Bureau is responsible for our tabulations and analyses.

⁶ Harper v. Virginia State Board of Elections, 383 U.S. 663 (1966).

⁷ Reitman, Alan and Davidson, Robert B., The Election Process: Voting Laws and Procedures (Dobbs Ferry, N.Y.: Oceana Publications, 1972), pp. 33–34Google Scholar.

⁸ Dunn v. Blumstein, 405 U.S. 330 (1972).

⁹ Marston v. Lewis, 410 U.S. 679 (1973); and Burn v. Fortson, 410 U.S. 686 (1973).

¹⁰ Campbell, Anguset al., The American Voter (New York: John Wiley, 1960), p. 277Google Scholar. The only legal variables studied were residency requirements and literacy tests in the North, and poll taxes and residency in the South (p. 269). Legal variables seemed to have more impact on blacks in the South, but the small number of cases precluded controls for education (pp. 279–80). For a similar conclusion, see Milbrath, Lester, “Political Participation in the States,” in Politics in the American States, ed. Jacob, Herbert and Vines, Kenneth N. (Boston: Little, Brown, 1965), p. 47Google Scholar.

¹¹ Report of the President's Commission on Registration and Voting Participation, November 1963, p. 13. Campbell and his associates said that residency requirements “prevented at least 3 percent of our respondents from voting” (The American Voter, p. 90n.).

¹² Kelley et al., “Registration and Voting.”

¹³ Some of the legal variables were assigned numerical values on the basis of the researchers' “guesses” about their impact on registration (Kelley et al., 366n.).

¹⁴ Ibid., pp. 373–74. Another study concluded that registration laws explained a good deal of the variations in turnout among the states in 1960. See Kim et al., “Voter Turnout Among the American States.”

¹⁵ Kelley et al. recognized that “inter-city analysis is less appropriate for an examination of the effects on registration of individual characteristics such as sex, age, education, and income, which can be better studied by treating the individual as the unit of observation” (p. 362).

¹⁶ Cf. Johnston, J., Econometric Methods (New York: Harper & Row, 1974), pp. 34–35, 129Google Scholar.

¹⁷ Another unsatisfactory approach is asking respondents why they were not registered. Answers to this question are in our data set. We tabulated the registration laws in each respondent's state of residence with the reasons offered for failure to register. The result is disillusioning for anyone who believes that human motivation can be discerned by asking people to explain their behavior. We found no relationship between the objective permissiveness of the laws and the proportion of respondents who said they were unregistered because they could not satisfy the residency requirement or had been unable to register. For example, in states where registration offices were required to remain open at night or on Saturday, 15 percent of those who were not registered attributed this condition to their inability to register, compared to 14 percent who gave this excuse where such extended hours were not required.

¹⁸ One recent study comparing turnout in the 1960 and 1972 elections is Ashenfelter, Orley and Kelley, Stanley Jr., “Determinants of Participation in Presidential Elections,” Law and Economics, 18 (March 1976), 695–733CrossRefGoogle Scholar. This article uses data on 925 cases drawn from the University of Michigan Center for Political Studies' national sample survey of the 1972 election. The estimates of the effect of registration laws are imprecise, perhaps because of the limitations of the Michigan sample for this purpose (see p. 26).

¹⁹ In 1962, five percent of the black voting-age population of Mississippi and 13 percent in Alabama were registered to vote. Ten years later the figures were 59 percent and 55 percent. Sources: for 1962, Peirce, Neal R., The Deep South States of America (New York: W. W. Norton, 1974)Google Scholar; for 1972, Lewis, John and Allen, Archie E., “Black Voter Registration Efforts in the South,” Notre Dame Lawyer, 48 (October 1972), 119Google Scholar.

²⁰ Data on turnout come from replies to this item: “This month we have some questions about whether people voted in the November 7th Presidential Election. Did (you) (this person) vote in the election held on November 7th?” For almost half the sample turnout data were provided by the respondent for another family member. Such second-hand interviewing raises the possibility of misreporting, as Tufte, Edward R. suggests (“Political Statistics for the United States: Observations on Some Major Data Sources,” American Political Science Review, 71 [March 1977], 306)CrossRefGoogle Scholar. This fear seems unjustified, however. Respondents who reported their own vote had a turnout rate of 66.7 percent. People in the sample for whom voting was reported by another family member had a turnout rate of 67.7 percent. We also estimated the effect of second-hand reporting by entering a dummy variable (1 = second-hand reporting) in the probit equation summarized in Table 2. The resulting estimate was .007 with a standard error of .010. This number is substantially zero, and no change occurred in the coefficients reported in Table 2. In short, we feel justified in assuming that any errors in the reporting of turnout (whether first-hand or second-hand reports) are uncorrelated with registration provisions.

²¹ The weighting procedure is described in Bureau of the Census, “Documentation of the Annual Demographic File,” 1975, Appendix A, pp. 4–12Google Scholar.

²² CPS 1972 American National Election Study (Ann Arbor, Michigan: Center for Political Studies, 1973), Variable 4Google Scholar.

²³ Bureau of the Census, “Voting and Registration in the Election of November 1972,” p. 13Google Scholar. For additional information on sampling procedures see Bureau of the Census, Current Population Survey: A Report on Methodology, Technical Paper No. 7, U.S. Department of Commerce, 1963Google Scholar; and “Documentation of the Annual Demographic File,” Appendix A; and Center for Political Studies, 1972 Study, p. xiiGoogle Scholar.

²⁴ Information on sampling error in the Census survey and the procedures for adjusting the standard error of estimates for groups of states below the regional level can be found in “Documentation of the Annual Demographic File,” Appendix A, pp. 13–45, 62–66; and “Voting and Registration in the Election of November 1972,” pp. 13–16.

²⁵ U.S. Senate, Committee on Post Office and Civil Service, as reported in the Congressional Record, 8 March 1972, pp. 7530–31Google Scholar; Thornton, Robert, “Election Legislation,” in The Book of the States, 1972–73, 19 (Lexington, Ky.: Council of State Governments, 1972), pp. 25–30Google Scholar; Reitman and Davidson, The Election Process; and League of Women Voters Education Fund, “Registration and Absentee Voting Procedures by State, 1972” (Washington, D.C., 1972)Google Scholar.

²⁶ One exception is the list compiled by the Congressional Research Service. Printed just after Dunn v. Blumstein, this publication claims that every state's residency requirement is governed by the 1970 VRAA for presidential elections and by Dunn for other elections. This blanket description proved to be wide of the mark. See Library of Congress, Congressional Research Service, Election Laws of the Fifty States and the District of Columbia, April 1972Google Scholar.

²⁷ In Arizona, for example, a recent county registrar of voters required prospective deputy registrars to take a rigorously graded examination. His successor brought about a substantial increase in the number of registrars by relaxing these standards. (We are indebted to Bruce B. Mason of Arizona State University for this information.)

²⁸ The frequent erroneous responses in our mail survey of state election officials suggest that it would be virtually impossible to gather accurate information on how state laws are administered in the nation's thousands of counties and cities.

²⁹ Downs, Anthony, An Economic Theory of Democracy (New York: Harper & Row, 1957), pp. 255–56, 273Google Scholar.

³⁰ Raymond E. Wolfinger and Steven J. Rosenstone, Who Votes? Forthcoming.

³¹ Using the entire sample would have required about 50 minutes of computer time for each probit equation estimated. Although we did not use all respondents for the crucial stage of our analysis, it should be noted that a national sample of such magnitude is necessary to provide adequate state-by-state subsamples.

³² Other things being equal, increasing the variance decreases the standard error of the estimate. See Theil, Henri, Principles of Econometrics (New York: John Wiley, 1971), pp. 90–91Google Scholar; and Finney, D. J., Probit Analysis, 3d ed. (Cambridge: Cambridge University Press, 1971), pp. 33–37Google Scholar.

³³ We excluded respondents in the District of Columbia from all data analyses reported hereafter. Since their franchise is limited and an inordinate number of District residents maintain legal voting residences elsewhere, we thought we would avoid unnecessary complications.

³⁴ See Theil, Principles of Econometrics, Ch. 8. All probit equations were estimated on the subsample of 7,936 actual respondents. While the case weight provided by the Census Bureau should be applied for descriptive purposes (e.g. cross tabulations), it is not appropriate to apply it for purposes of estimation. See Koch, Gary G., Freeman, Daniel H. Jr., and Freeman, Jean L., “Strategies in the Multivariate Analysis of Data from Complex Surveys,” International Statistical Review, 43 (April 1975), 59–78CrossRefGoogle Scholar; and Porter, Richard D., “On the Use of Survey Sample Weights in the Linear Model,” Annals of Economic and Social Measurement, 2 (June 1973), 141–158Google Scholar.

³⁵ Our initial analysis also included these variables: (a) the number of days before the election an absentee ballot had to be obtained; (b) whether the absentee ballot had to be notarized; and (c) whether voting machines were used. These variables had less than a one percent impact on the probability of voting and their coefficients were not statistically significant. We deleted them after this early stage of the analysis.

In addition, estimates were made for nonlinear functions of the residency requirement and closing date. This did not appreciably improve the fit. We also estimated interaction terms between each provision and education, between deputy registrars and county registration, and between deputy registrars and irregular office hours. All were insignificant.

Finally, we examined the independent effect on turnout of income and party competition. Poorer people, of course, are less likely to vote, and party competition is often said to result in higher turnout. Thus any estimates of the relationship between registration laws and turnout that omitted consideration of these two variables might lead to a spurious conclusion.

We used the following procedure to determine whether these two variables in the error term should be included in the analysis. For one of these variables to bias the coefficient of a particular registration provision (which we will call X₁), two conditions had to be met: After controlling for the other independent demographic variables, the variable in question had to be correlated both with X₁ and with turnout. If either of these conditions were not met (i.e., if one of the partial correlations was zero), then the variable could safely be left out of the analysis without biasing the estimated coefficient for variable X₁.

Using this procedure, we found that income was not correlated with any registration provision, once education, region, and age had been controlled. The average magnitude of the partial correlations of the registration provisions was .029. The partial correlation between competition and turnout was −.009. Therefore, excluding these variables from Equation II does not bias the estimated coefficients for the registration laws. The measure of party competition we used was adopted from Ranney, Austin, “Parties in State Politics,” in Politics in the American States, 2d ed., ed. Jacob, Herbert and Vines, Kenneth N. (Boston: Little, Brown, 1972), p. 87Google Scholar.

³⁶ As explained in Appendix C, the estimated effect of an independent variable in probit analysis depends on the probability that the individual would otherwise vote.

³⁷ For the evidence, see Wolfinger and Rosenstone, Who Votes?

³⁸ This reweighting was accomplished by multiplying, for each case, the original weight by the inverse of the sub-sampling proportion for the state from which the case was selected. This yields a reweighted N of 126,591. This number is slighty less than the weighted N mainly because respondents from the District of Columbia were deleted from the probit analysis and secondarily because of the random selection of cases in the sub-sampling procedure.

³⁹ For each respondent, Equation II was used to compute a probit estimate which was then converted to a probability by evaluating that number on the cumulative standard normal distribution. (The International Mathematical and Statistical Library includes a fortran subroutine which will compute this value.) The same procedure was followed, altering the values for the variables in Equation II to simulate changes in the registration provisions. For each respondent these two probabilities were subtracted. The arithmetic mean of these individual probabilities is the estimated aggregate percentage.

⁴⁰ The standard error of this estimate is 2.1. Thus there is only a 5 percent chance that the true coefficient is less than 4.9 percent or more than 13.3 percent.

⁴¹ North Dakota did not require registration at all in 1972. There were not enough respondents from this single, sparsely populated state to permit precise estimates of the effects of its unique voting laws. By 1976 four other states permitted election day registration.

⁴² Table 4 also shows the effect of liberalized registration laws by income and age. Poorer people would be most affected, as we would expect because of the relationship between education and income. The variation among income categories is much smaller than among people with different educational attainment, however. This reflects the much stronger relationship between education and turnout. Variations by age are small. People in the age groups with the highest turnout would be affected only slightly less than those with the lowest voting rates–citizens under 32 and over 78.

⁴³ Robinson, John P. and Converse, Philip E., “Social Change in the Use of Time,” in The Human Meaning of Social Change, ed. Campbell, Angus and Converse, Philip E. (New York: Russell Sage, 1972), pp. 74–75Google Scholar. See our Who Votes? for more discussion of the relationship between free time and turnout.

⁴⁴ For representative comments, see Polsby, Nelson W. and Wildavsky, Aaron, Presidential Elections, 4th ed. (New York: Scribner, 1976), p. 129Google Scholar. For a dissenting view, emphasizing “the general insensitivity of partisanship to large changes in turnout,” see Converse, Philip E., “The Concept of a Normal Vote,” in Campbell, Anguset al., Elections and the Political Order (New York: John Wiley, 1966), p. 29Google Scholar.

⁴⁵ San Francisco Chronicle, 16 May 1977, p. 38Google Scholar. The Republican National Committee formally opposed Carter's plan, and a Gallup Poll in the spring of 1977 revealed that rank-and-file Republicans were more hostile to it than Democrats, who themselves were unfavorable to the proposal. (San Francisco Chronicle, 9 May 1977, p. 8.Google Scholar) The pronounced partisan cleavage on this issue among Congressmen is described in Congressional Quarterly Weekly Report, 14 May 1977, pp. 909–11Google Scholar.

⁴⁶ The composition of the actual electorate was estimated by using Equation II to compute a probit estimate for each respondent. The probit estimate was then converted to a probability. The percentage of the electorate comprised of people with a given characteristic (e.g. a college education) was computed in the following manner:

where j = the number of respondents with a given characteristic (i.e. in that category on the variable); and n = the total number of respondents. The same procedure was used to calculate the projected composition of the electorate, with the values for the registration provisions altered to simulate the liberalized conditions.

We used Equation II rather than the reported turnout to estimate the composition of the actual electorate. If we had used the reported turnout, then the difference between the projected electorate and the actual electorate would be due both to the simulated changes in registration laws and the residual for each individual in Equation II. These residuals cancel out if Equation II is used for calculating the composition of both the actual and the projected electorates.

⁴⁷ The data were made available by the Inter-University Consortium for Political Research.

⁴⁸ See Polsby, and Wildavsky, , Presidential Elections, pp. 239–41Google Scholar; and Key, V. O., Politics, Parties, and Pressure Groups, 5th ed. (New York: Thomas Y. Crowell, 1964), p. 590Google Scholar.

⁴⁹ These projections, as well as those summarized in Tables 6 and 7, are based on the 1972 National Election Study of the University of Michigan Center for Political Studies. For this sample we used Equation II to estimate both the actual and projected composition of the electorate, as described in note 46. We deleted 13 cases from this sample: the 10 respondents from the District of Columbia and 3 respondents whose education was not reported, and for whom turnout therefore could not be predicted. Equation II correctly predicted turnout for 74.3 percent of the cases in this sample, compared to 71.4 percent of the cases in the Current Population Survey subsample. The differences between the actual and projected electorates summarized in Table 5 can be replicated within 0.4 percentage points by using Equation II with the Michigan data.

⁵⁰ This classification is based on Keith, Bruce E., et al., “The Myth of the Independent Voter” (paper delivered at the 1977 Annual Meeting of the American Political Science Association)Google Scholar.

⁵¹ Candidate popularity was measured by the proportion of those rating McGovern higher than Nixon on the candidate “feeling thermometers.”