An empirical analysis of life tenure: A response to professors Calabresi & Lindgren

Harvard Journal of Law and Public Policy

Volumn 30, Issue 3, 2007, Pages 792-830

  Stras, David R ^a   Scott, Ryan W ^b  

^a UNIVERSITY OF MINNESOTA LAW SCHOOL   (United States)

^b Office of the Solicitor General   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 36049019922     PISSN: 01934872     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Review

References (183)

1. Steven G. Calabresi & James Lindgren, Term Limits for the Supreme Court: Life Tenure Reconsidered, 29 HARV. J.L. & PUB. POL'Y 769 (2006). The article first appeared in a collection called Reforming the Court: Term Limits for Supreme Court Justices, edited by Roger C. Cramton and Paul D. Carrington and published in the spring of 2006. The revised version, published in the summer of 2006, contains updated data and additional material responding to critics.
2. Id. at 769 n.*.
3. Id. at 771.
4. Id. at 772. Elsewhere Calabresi and Lindgren have used even more provocative language, arguing that life tenure for Supreme Court Justices has produced "a gerontocracy-like the leadership cadre of the Chinese Communist Party."
5. Steven G. Calabresi & James Lindgren, Supreme Gerontocracy, WALL ST. J., Apr. 8, 2005, at A12.
6. See Calabresi & Lindgren, supra note 1, at 809-18, 824-54.
7. See Linda Greenhouse, How Long Is Too Long for the Court's Justices?, N.Y. TIMES, Jan. 16, 2005, at WK5;
8. Tony Mauro, Profs Pitch Plan for Limits on Supreme Court Service, LEGAL TIMES., Jan. 3, 2005, at 1;
9. Bruce Bartlett,. . . And Tenure Traps (July 6, 2006), http://www.ncpa.org/prs/cd/2005/20050706.htm.
10. Calabresi & Lindgren, supra note 1, at 809-13.
11. Id. at 813-15.
12. Id. at 815-18.
13. Id. at 777-78.
14. Id. at 778-79. To a lesser extent, the authors also rely on the average number of years between vacancies on the Supreme Court, a rough corollary of length of tenure. According to Calabresi and Lindgren, "from 1881 through 1970, the average number of years between commissions stayed consistent at about 1.6 to 1.8," but since 1970 "it has nearly doubled to 3.1 years."
15. Id. at 786. The authors also observe that the average age of Justices leaving the Court since 1971 has risen to 78.7 years, an increase of ten years compared with the average of 68.3 years for Justices leaving office before 1971.
16. Id. at 782. We have no quarrel with Calabresi and Lindgren's analysis of the data on age at retirement, as we share their concerns about mental and physical infirmity among Supreme Court Justices. This Article therefore focuses on their principal measure of the dramatic change in life tenure: average length of tenure for Supreme Court Justices.
17. David R. Stras & Ryan W. Scott, Retaining Life Tenure: The Case for a "Golden Parachute," 83 WASH. U. L.Q. 1397 (2005).
18. Id. at 1437.
19. Id. at 1418-20.
20. Id. at 1427.
21. See id. at 1428-29 charts 1 & 2.
22. See Calabresi & Lindgren, supra note 1, at 789-99.
23. See Stras & Scott, supra note 11, at 1430.
24. Id. at 1427.
25. Id. at 1430.
26. Id. at 1427.
27. Id. at 1427 chart 1.
28. Stras & Scott, supra note 11, at 1428 chart 2.
29. See id. at 791-94. Twice, Calabresi and Lindgren even imply that we either deliberately withheld details about our charts or affirmatively misrepresented the data.
30. Id. at 791 ("They do not report the cell counts for their chart dividing the Justices by decade, perhaps for a reason.");
31. id. ("Stras and Scott appear to be straining to find ways of presenting the data that lump Justices together in a way that will make the patterns in the data disappear.").
32. Id. at 790.
33. See Calabresi & Lindgren, supra note 1, at 781 chart 2, 783 chart 4, 788 chart 6.
34. Calabresi & Lindgren, supra note 1, at 798-99. Professor Lindgren is a Ph.D. candidate at the University of Chicago.
35. See James Lindgren, Curriculum Vitae, http://www.law.northwestern. edu/faculty/fulltime/lindgren/lindgrjacv.pdf. As a contributor to the popular law blog The Volokh Conspiracy, Professor Lindgren has said that he has "done tens of thousands of regression analyses." Posting of Jim Lindgren to The Volokh Conspiracy, http://volokh.com/posts/ 1162276977. shtml#155478 (Oct. 31, 2006, 10:40 EST).
36. Calabresi & Lindgren, supra note 1, at 777.
37. Id. at 786.
38. Id. at 779.
39. Id. at 775, 778.
40. Id. at 778, 779, 780, 782, 789, 798, 807, 832.
41. Calabresi & Lindgren, supra note 1, at 797.
42. Like Calabresi and Lindgren, we used an independent samples t-test for equality of means, without assuming equality of variance. See id. at 797 n.79. Here, we compared data from 1789-1820 with data from 1821-1850.
43. Here, we compared data from 1789-1820 with data from 1821-2006.
44. Calabresi & Lindgren, supra note 1, at 797.
45. Id.
46. Id. at 771.
47. Id. at 781, 783, 790, 793, 794.
48. Calabresi & Lindgren, supra note 3, at A12. 40. Id.
49. See id.
50. Calabresi & Lindgren, supra note 1, at 778 chart 1.
51. Id.
52. Ronald Brownstein, Time to Bring Down the Gavel on Lifetime Tenure for Justices?, L.A. TIMES, Oct. 17, 2005, at A10;
53. see also Jeff Jacoby, Don't Let Judges Serve for Life, BOSTON GLOBE, May 26, 2005, at A19;
54. Greenhouse, supra note 5, at WK5.
55. Calabresi & Lindgren, supra note 1, at 778.
56. SANFORD LEVTNSON, OUR UNDEMOCRATIC CONSTITUTION 128-29 (2006). Professor Levinson recognizes the period-selection problem, however, by noting that the 1941-1970 period included several Justices "who served unusually short terms because of death or moving to other positions of service." Id.
57. See Bartlett, supra note 5 (citing the increases between the last two periods in terms of average tenure, average age at retirement, and average interval between appointments).
58. See, e.g., Mark Alexander, Supreme Consequences, PATRIOT POST, No. 05-27, July 8, 2005, http://secure. patriotpost.us/Alexander/edition.asp?id=315;
59. Ken Bell, Lest Ye Be Judged, AUSTIN REV., July 6, 2005, available at http://www.austinreview.com/archives/2005/07/ lest_ye_be_judg.html;
60. Posting of Publius to Publius' Forum, http://pconservablogs.com/ publiusforum/2005/10/04/ scotus-stats-of-interest/ (Oct. 4, 2005).
61. Anxious about the role that period selection might have played in generating this particular figure, Calabresi and Lindgren distance themselves from it in the most recent version of their article. "[O]ther than in Chart 1 itself and in the one paragraph discussing Chart 1," they explain, "we made only one mention of the 12.2 year tenure of Justices who left office in the 1941-1970 period." Calabresi & Lindgren, supra note 1, at 795. We acknowledge that Calabresi and Lindgren have consistently emphasized the 14.9-year average for Justices retiring before 1971. As the sources cited in the last three footnotes demonstrate, however, scholars and commentators relying on their research have not been so careful.
62. Stras & Scott, supra note 11, at 1427.
63. Calabresi & Lindgren, supra note 1, at 793.
64. In calculating length of service for each Justice, we subtracted (1) the date of swearing-in from (2) the date of departure from regular active service on the Supreme Court, whether by death, resignation, or the election of senior status. We used the date of swearing-in, rather than the date of confirmation or commission, because swearing-in marks the start of a Justice's active service on the Court, even for recess appointees. This methodological choice represents a slight change from our method in our last article on this subject, where we used the date of a Justice's commission. See Stras & Scott, supra note 11, at 1428 n.164.
65. Meanwhile, we used the date of a Justice's death, resignation, or election of senior status to calculate date of departure because any one of those events triggers a new appointment opportunity for the President and marks the end of a Justice's active service on the Court. See 28 U.S.C. § 371(d) (2000). To convert the difference from days to years, we divided by 365.25.
66. Like Calabresi and Lindgren, we excluded current members of the Court from our calculations even though we know a certain minimum length of service for each of them. Justice Stevens, for example, has already served for more than thirty years. On the other hand, Chief Justice Roberts and Justice Alito have served for approximately four years combined. Because the current average on the Court falls well below the historical average, adding current members to the data setas if all of them retired immediately-would distort the data. Likewise, projecting an exit date for current members, perhaps based on life expectancy or historical averages, would both distort the data and defeat the point of the analysis by assuming an answer to the question that we are analyzing. We therefore follow Calabresi and Lindgren in limiting our analysis to Justices who have already departed from the Court.
67. Calabresi & Lindgren, supra note 1, at 792.
68. We are reminded of Darrell Huff's description of the " flattened" precursor to "the Gee-Whiz Graph," in How to Lie with Statistics: "That is very well if all you want to do is convey information. But suppose you wish to win an argument, shock a reader, move him into action, sell him something. For that, this chart lacks schmaltz." DARRELL HUFF, HOW TO LIE WITH STATISTICS 62 (2d ed. 1993).
69. That was, in part, our reasoning for the period selection in our earlier article, which included a close approximation of Chart 5. See Stras & Scott, supra note 11, at 1428.
70. This time, heeding Calabresi and Lindgren's criticism, see Calabresi & Lindgren, supra note 1, at 791-92, we have disclosed the sample size for each decade.
71. Calabresi and Lindgren argue that even a chart using ten-year periods "tends to support, rather than reject, [their] hypothesis" because "three of the four decades with highest mean tenure are three of the last four decades." Calabresi & Lindgren, supra note 1, at 792. That claim, however, reflects period selection too. The decade with the fifth-highest tenure is the 1860s (24.5 years), which is noticeably higher than the 1980s (18.5 years). In any event, Calabresi and Lindgren keep moving the target. How can they seriously maintain that life tenure "worked well" before 1970, when it produced decades like the 1830s and 1860s, but "has broken down" now that it has produced decades like the 1970s and 1990s?
72. Calabresi & Lindgren, supra note 3, at A12. Only period selection allows the authors to explain away the long tenures of so many mid-nineteenth-century Justices.
73. Calabresi & Lindgren, supra note 1, at 778, 781. Unsurprisingly, period lengths of very close to thirty years, including thirty-five years and twenty-five years, produce charts quite similar in shape to Calabresi and Lindgren's chart. No period length, however, results in a final-period rise as stark as that created by thirty-year periods.
74. Id. at 775.
75. The revised version of Calabresi and Lindgren's article concedes that there would be a smaller increase in average tenure if they used a slightly later cutoff date. See id. at 779. They calculate that moving the cutoff to 1975 to include Justices Black, Harlan, and Douglas would increase the penultimate-period average to 15.0 years while reducing the final-period average to 25.1 years, resulting in an increase 27% smaller than the one reported in their chart.
76. See id. The same is true for an earlier cutoff date. Moving the cutoff to 1965, for example, changes the penultimate-period average to 12.8 years while reducing the final-period average to 22.1 years, resulting in a 33% smaller increase than the one reported by Calabresi and Lindgren. Moving the cutoff date by even a single Justice in either direction also affects average tenure in the final two periods. Moving Justice Black makes the increase in the final period smaller by 13%, while shifting Chief Justice Warren renders the increase smaller by 4%.
77. The revised version of Calabresi and Lindgren's article includes a series of charts showing the "lagging average" for consecutive groups of nine Justices for each of their three metrics, which are meant to "smooth" the data so that it is less susceptible to significant fluctuations. See id. at 780-81, 783, 788 charts 2, 4, & 6. Because the new charts appear in a separate section of the article, it is not clear whether Calabresi and Lindgren intended the lagging-average analysis as a direct response to our period-selection criticism.
78. Nonetheless, they argue that the lagging-average charts show the data "without any period selected by the researcher." Id. at 780. That claim is misleading, however, because the lagging-average charts suffer from a different form of the same problem. Instead of depending on the number of years per period, the lagging-average charts depend on the number of Justices per group.
79. See id. at 780. Calabresi and Lindgren have elected to display the data using a lagging average of nine Justices but they might just as easily have chosen a larger or smaller number. Thus, although they have avoided selecting a "period," it is not entirely accurate to suggest that these charts involve no selection "by the researcher."
80. Id.
81. In a footnote, Calabresi and Lindgren anticipate this critique, and state that "[i]f [they] had included more Justices in [their] lagging averages, such as a twelve-Justice lagging average, Chart 2 would have resembled Chart 1 even more closely, and the recent rise in judicial tenure would have appeared even more dramatic." Id. at 781 n.39. To test their claim that "includ[ing] more Justices" only makes their argument stronger, we created lagging-average charts for groups of three, five, seven, nine, eleven, thirteen, fifteen, seventeen, and nineteen Justices. We then generated the same charts treating each Justice's tenure as an observation at the date of appointment rather than the date of departure. We found that lagging-average charts for groups of nine, eleven, or thirteen Justices show a trend in length of tenure that is consistent with Calabresi and Lindgren's empirical claim. Because that conclusion holds true for several values and does not depend on a particular cutoff date, the lagging-average charts lend some support to Calabresi and Lindgren's decision to use thirty-year periods in Chart 1. Like the charts displaying average tenure over various time periods, however, the lagging-average charts support Calabresi and Lindgren's empirical claim only when particular group sizes within a narrow range are selected. Lagging averages for groups of fewer Justices tend to refute the claim of an "unprecedented" recent increase in length of tenure. With three Justices, lagging averages in the mid-nineteenth century reached 29.7 years, longer than any of the lagging averages since 1971, and five of the eight values longer than twenty-seven years occurred between 1830 and 1880. With five Justices, the peak in the nineteenth century reached 28.1 years, higher than every value in history except 2006, and the cluster of lagging averages higher than twenty years is just as impressive in the nine-teenth century as in the decades since 1971. Even with seven Justices, the lagging average in 1873 came within 0.2 years of the highest value in 1994 and was higher than the lagging averages in 2005 and 2006. Likewise, lagging averages for groups of more Justices also tend to refute the claim of an unprecedented recent increase in length of tenure. With seventeen and nineteen Justices, lagging averages in the mid-nineteenth century were higher than any value in history. Even with fifteen Justices, the nineteenth-century lagging averages were higher than all but the 2005 and 2006 values, falling short of the 2005 and 2006 values by less than two years.
82. Moreover, like the period charts, the lagging-average charts depend on Calabresi and Lindgren's assumption that length of tenure should be treated as an observation on the date of death, retirement, or resignation from the Court rather than the date of appointment. See infra Part II.B. Changing that assumption eliminates any astonishing and unprecedented trend in the lagging-average charts, regardless of the number of Justices in each group.
83. Stras & Scott, supra note 11, at 1428.
84. Calabresi & Lindgren, supra note 1, at 790.
85. Id.
86. Stras & Scott, supra note 11, at 1428 (emphasis added).
87. Id.
88. Calabresi and Lindgren also made an arithmetic error in the most recent version of their article. They claim that the largest change resulting from a one-year shift in any of their periods is a "trivial" change for the period from 1941 to 1970. Calabresi & Lindgren, supra note 1, at 790. "[I]f the 1941-1970 period had started in 1942 instead of 1941," they argued, then "the mean tenure for Justices leaving the Court in that... period would have been 13.5 years, rather than 12.2
89. years . . . ." Id. Ironically, they have overestimated the effect of the shift. In fact, moving the start of the 1941-1970 period to 1942 reduces the average tenure to 11.4 years by excluding Justices McReynolds and Hughes, and moving both the start and end dates to create a 1942-1971 period increases average tenure to only 12.9 years by including Justices Black and Harlan.
90. See David H. Kaye & David A. Freedman, Reference Guide on Statistics, in REFERENCE MANUAL ON SCIENTIFIC EVIDENCE 113 (Fed. Judicial Ctr. ed., 2d ed. 2000) (noting that when "the shape [of the graph] can be altered somewhat by changing the size of the bins," critical readers might find it "worth inquiring how the analyst chose the bin widths").
91. Stras & Scott, supra note 11, at 1428.
92. Justin Crowe and Christopher Karpowitz suggest yet another reason to be skeptical of Calabresi and Lindgren's empirical claim. See Justin Crowe & Christopher Karpowitz, Where Have You Gone, Sherman Minton? The Decline of the ShortTerm Supreme Court Justice (Princeton Law & Public Affairs Working Paper Series, Working Paper No. 06-014, 2006), available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=948813. According to the authors, any increase in long-term tenure is driven in part "by the fact that fewer justices are serving relatively short terms."
93. Id. at 1. During the current period, identified by Calabresi and Lindgren as having the most dramatic increase in average tenure in history for Supreme Court Justices, Crowe and Karpowitz find that "we have witnessed the longest period without a short-term justice (37 years) in the Court's history."
94. Id. at 8. Even using Calabresi and Lindgren's period lengths, removal of short-term Justices from all earlier periods shows that, "though still a historical high, [the period of 1971-2005] is not radically out of line with any earlier period in the Court's history and is only one and a half years higher than the previous peak."
95. Id. at 9. In other words, any increase in average tenure may be attributed more to the recent absence of short-term Justices, who arguably are not representative of ordinary life-tenured justices
96. see LEVINSON, supra note 46, at 128-29, than to any alarming recent change in the operation of life tenure.
97. Calabresi & Lindgren, supra note 1, at 792.
98. Id.
99. See David R. Stras, The Incentives Approach to Judicial Retirement, 90 MINN. L. REV. 1417, 1439, 1444-45 (2006).
100. Calabresi & Lindgren, supra note 1, at 792.
101. Type II error occurs when the null hypothesis is false, but a statistical test fails to reject it, "i.e., there is a false negative." Kaye & Freedman, supra note 66, at 176. Type I error, by contrast, occurs when the null hypothesis is true, but a statistical test rejects it, "i.e., there is a false positive."
102. Id.
103. Calabresi and Lindgren may have intended their critique about Type II error as a commentary solely on the second chart that we reported in our first article, which displayed average tenure in non-overlapping groups of five Justices. See Stras & Scott, supra note 11, at 1429. In that chart, we selected groups of five Justices in an effort to "flatten[] th[e] periods" used in both Calabresi and Lindgren's charts and our own.
104. Id. at 1428. In a spectacular mischaracterization, Calabresi and Lindgren chastise us for deliberately "'flatten[ing]' [the] effects" and "'flatten[ing]'[] the unprecedented length of tenure" in the data.
105. Calabresi & Lindgren, supra note 1, at 790, 791 n.57, 793 (emphasis added). As our text made clear, however, we sought in our second chart only to "flatten[] th[e] periods," i.e., to use groups that contain an equal number of Justices, rather than periods of years that contain a variable number of Justices.
106. See Stras & Scott, supra note 11, at 1428 (emphasis added). Our goal was to eliminate a potential source of distortion caused by period selection. We did not hold, and never expressed, a desire to flatten effects in the data.
107. Calabresi & Lindgren, supra note 1, at 779. Political scientists also recognize that the decision of whether to retire is one that is made continuously and over time. Thus, most studies that have analyzed the retirement behavior of Supreme Court Justices have used each year of a Justice's service as a separate observation.
108. See Peverill Squire, Politics and Personal Factors in Retirement from the United States Supreme Court, 10 POL. BEHAV. 180, 184 (1988);
109. Albert Yoon, Pensions, Politics, and Judicial Tenure: An Empirical Study of Federal Judges, 1869-2002, 8 AM. L. & ECON. REV. 143, 150 (2006) (using judgeship-year as the relevant unit of analysis to "account[] for secular and individual-level changes from year to year, with an eye towards examining which factors, if any, explain when judges created a judicial vacancy");
110. Christopher J.W. Zorn & Steven R. Van Winkle, A Competing Risks Model of Supreme Court Vacancies, 1789-1992, 22 POL. BEHAV. 145, 150 (2000).
111. But see Timothy M. Hagle, Strategic Retirements: A Political Model of Turnover on the United States Supreme Court, 15 POL. BEHAV. 25, 27 (1993) (criticizing Professor Squire's approach because of the large disparity between voluntary retirements and total observations).
112. Admittedly, observing length of tenure at the date of appointment suffers from the opposite problem by frontloading the data. Using date of appointment, however, is an equally defensible methodological choice because both approaches create similar distorting effects. As a compromise, researchers could observe length of tenure at the exact midpoint of a Justice's career. That method also yields results that undercut Calabresi and Lindgren's hypothesis of a dramatic and unprecedented recent increase in average tenure. See infra note 136.
113. Using the date of exit backloads the data for other periods as well. For example, the first Justice Harlan joined the Court in 1877, but Calabresi and Lindgren group him with the pre-World War II retirees because he remained on the Court until 1911.
114. See Calabresi & Lindgren, supra note 1, at 809-13.
115. Id. at 793.
116. Id.
117. Id. at 794.
118. Id. at 792.
119. The best reason we can imagine is that length of tenure is not known at the date of appointment, but it is known (indeed, it finally becomes known) at the date of death, retirement, or resignation. As a result, if a Justice's term in office counts at the date of appointment, the death of a long-serving Justice will retroactively affect an earlier period. These factors, however, have never deterred statisticians from using start date (most commonly birth date) rather than end date (death date) when reporting life expectancy data, which are closely analogous to data on length of tenure. In both cases, researchers cannot be certain of the date of death (or departure) until it has occurred, and must retroactively attribute new cases to past dates (birth or appointment). Yet life expectancy is uniformly reported as a value reflecting the average lifespan (tenure) of a person born in a particular period rather than the average lifespan (tenure) of a person who died during that period. See, e.g., Elizabeth Arias, United States Life Tables, 2003, in NAT'L VITAL STATS. REPS. 2006, at 34-35 tbl.12, available at http://www.cdc.gov/nchs/data/nvsr/nvsr54/nvsr54_14. pdf.
120. A robust trend in length of tenure should be apparent regardless of the method of observation. For example, if average length of tenure were to remain constant for the next fifty years when measured from date of appointment, then any perceived trend during the same period, when measured from date of departure, would simply be an accident of timing in deaths, retirements, and resignations.
121. See Calabresi & Lindgren, supra note 1, at 794.
122. Id. at 797.
123. Id. at 797-98.
124. Id. at 798.
125. See id. at 799 chart 1. We replicated Calabresi and Lindgren's model and, as far as we can tell, our results are identical. Differences in rounding explain any slight disparities in our reported values.
126. Calabresi & Lindgren, supra note 1, at 798.
127. Id.
128. See Daniel L. Rubinfeld, Reference Guide on Multiple Regression, in REFERENCE MANUAL ON SCIENTIFIC EVIDENCE, supra note 66, a 181.
129. See Kaye & Freedman, supro note 66, at 171.
130. The best-fitting line or curve is usually calculated by using the "ordinary least squares" method of estimation, which minimizes error by minimizing the sum of the squared residuals - the difference between the predicted value and actual value - for each observation. See MICHAEL O. FINKELSTEIN & BRUCE LEVIN, STATISTICS FOR LAWYERS 333 (1990).
131. Kaye & Freedman, supra note 66, at 122.
132. Rubinfeld, supra note 92, at 226.
133. The t-statistic serves a similar function for each parameter. If the absolute value of t for a particular parameter estimate is greater than 1.96, then the estimate is statistically significant. Id. at 214.
134. Calabresi & Lindgren, supra note 1, at 798.
135. FINKELSTEIN & LEVIN, supra note 94, at 345.
136. Id.
137. Id. at 346.
138. Id.
139. Rubinfeld, supra note 92, at 215-16.
140. Kaye & Freedman, supra note 66, at 136.
141. Rubinfeld, supra note 92, at 216. A few statisticians have proposed rules of thumb, but such rules should be received with great caution.
142. See, e.g., R. SENTER, ANALYSIS OF DATA: INTRODUCTORY STATISTICS FOR THE BEHAVIORAL SCIENCES 433 (1969) (describing a system wherein values of r between 0.40 and 0.70 demonstrate a "substantial" relationship, while values of r less than 0.20 demonstrate a "slight, almost negligible" relationship);
143. 2 exceeds 0.09, thereby "explaining 9% or more of the observed variation").
144. See JOHN E. FREUND, MATHEMATICAL STATISTICS 334-35 (2d ed. 1971).
145. Rubinfeld, supra note 92, at 215.
146. See SANFORD WEISBERG, APPLIED LINEAR REGRESSION 22 (2005). Prediction intervals are a range of values that have a specified probability (usually 95%) of containing the value of the dependent variable based on the observed value of an independent variable. See id. at 21. Prediction intervals are similar to confidence intervals, except that the latter estimates an unobservable population parameter, while the former predicts the distribution of individual points.
147. See id. at 20.
148. See Calabresi & Lindgren, supra note 1, at 798.
149. Id.
150. Compare id. at 778 chart 1 (rendering the tenure data in thirty-year periods)
151. with id. at 799 chart 7 (displaying Calabresi and Lindgren's cubic regression model).
152. Cf. Posting of James Lindgren to The Volokh Conspiracy, http://www.volokh.com/ posts/1162276977.shtml (Oct. 31, 2006, 12:42) (faulting a study using regression analyses for reporting that internet access at home reduces the incidence of rape, when in fact "the observed rape increasing effect of computer access is even more highly significant than the observed rape decreasing effect of internet access") (emphasis added).
153. 2 of 206.
154. See Calabresi & Lindgren, supra note 1, at 798.
155. Id.
156. See, e.g., DAVID RIESMAN ET AL., THE LONELY CROWD: A STUDY OF THE CHANGING AMERICAN CHARACTER 7 (abr. and rev. ed. 2001);
157. HENK A. DE GANS, POPULATION FORECASTS 1895-1945: THE TRANSITION TO MODERNITY 53, 99 & nn. 50-51 (1999).
158. Calabresi & Lindgren, supra note 1, at 798.
159. Id. at 798.
160. Id. at 791.
161. Stras & Scott, supra note 11, at 1427.
162. Id.
163. See WEBSTER'S THIRD NEW INTERNATIONAL DICTIONARY 790 (1961) (defining "exaggerate" as "to misrepresent on the side of largeness (as of size, extent, or value): overstate the truth").
164. Stras & Scott, supra note 11, at 1427 (emphasis added).
165. Calabresi & Lindgren, supra note 1, at 797.
166. BURTON G. MALKIEL, A RANDOM WALK DOWN WALL STREET (8th ed. 2003).
167. Stras & Scott, supra note 11, at 1430 (footnotes omitted) (emphasis added).
168. See Calabresi & Lindgren, supra note 1, at 797-98.
169. Id. at 798.
170. Id.
171. See COMPREHENSIVE STATISTICAL SOFTWARE 7, available at http://www.spss.com/ pdfs/S15SPClr.pdf (marketing brochure for SPSS, boasting that eleven trend-regression models are available: "[l]inear, logarithmic, inverse, quadratic, cubic, compound, power, S, growth, exponential, and logistic"). SPSS is a leading statistics software package for social scientists.
172. See id. at 1.
173. See supra notes 78-83 and accompanying text (noting that because Calabresi and Lindgren's critique of life tenure focuses largely on the role of the political branches, which view tenure length ex ante rather than ex post, it is equally defensible to count each observation as occurring at the date of appointment rather than the date of departure).
174. See supra notes 84-85 and accompanying text.
175. Rubinfeld, supra note 92, at 195.
176. The power model has a slope and shape similar to the S-curve model discussed in Part II.C.2.
177. See SENTER, supra note 105, at 433;
178. cf. Jeffrey S. Kinsler, The LSAT Myth, 20 ST. LOUIS U. PUB. L. REV. 393, 398 (2001) (describing a model in which LSAT performance "accounted for less than 4% of the variance witnessed in law school performance" as documenting "a very weak correlation by any standard").
179. 2=.147), with a similar shape to the curve in Chart 8. Again, the S-curve model is statistically significant at a higher level of confidence, explains more of the variance in length of tenure, and has greater predictive reliability. It also fits the data better, with an average prediction interval of 13.72 years at the 95% confidence level, roughly 33% lower than the average prediction interval of 18.67 years for the cubic model.
180. See Calabresi & Lindgren, supra note 1, at 798.
181. See Stras & Scott, supra note 11, at 1398.
182. Id. at 1419-20.
183. David R. Stras, Why Supreme Court Justices Should Ride Circuit Again, 91 MINN. L. REV. (forthcoming 2007).