Smyth, Russell --- "What Explains Variations in Dissent Rates?: Time Series Evidence from the High Court" [2004] SydLawRw 10; (2004) 26(2) Sydney Law Review 221

	Home \| Databases \| WorldLII \| Search \| Feedback Sydney Law Review

This article employs a multiple regression statistical model to examine the determinants of variations in the dissent rate on the High Court using annual time series data between 1904 and 2001. We hypothesize that variations in dissent rates over time depend on both institutional factors internal to the Court and socio-political complexity external to the Court. We find that the main determinants of variations in dissent have been prevailing norms on the Court as to acceptable levels of dissent, caseload, and the Australia Acts which removed the final vestiges of appeal to the Privy Council. We also find evidence of significant variation in dissent under the leadership of different chief justices.

* Professor, Department of Economics, Monash University. I thank Matthew Groves and two anonymous referees for helpful suggestions on an earlier version of this article and Brooke Watson and Olga Ryan for assistance with data collection.

1. Introduction

A large literature has emerged for courts in the United States which examines the reasons why appellate judges dissent. Some studies have employed what has been broadly termed the ‘attitudinal model’ to attempt to explain variations in dissent rates on the basis of the personal characteristics of the judges. The personal characteristics of judges that have been tested to explain differences in dissent rates between judges in the United States include political affiliation, social background (school, religion, class) and demographic and hereditary factors (age and birthplace).[1] Australian legal academics have been reluctant to embrace attitudinal models of decision-making, arguing that they are not apposite to the Australian context.[2] In the United States Supreme Court, which has traditionally been recognised as having a much more overt political role than the High Court, there is greater scope for personal characteristics to come to the fore. Existing attempts to explain dissent in the High Court using the attitudinal model support this conjecture. These studies suggest that the personal characteristics of judges including background have little power in explaining variations in dissent rates between judges.[3]

Another strand of the literature in the United States has focused on institutional and external factors, rather than the judges’ personal characteristics, as the main determinants of variation in dissent.[4] This literature has attempted to explain variations in dissent rates in terms of factors such as changes in caseload, whether the court has a discretion to select the cases it hears, leadership of the chief justice and the proportion of people living in urban areas, which is used to depict growing social complexity. The objective of this article is to draw on the hypotheses that have been tested in the literature for United States courts to examine the influence of changes in the High Court’s institutional arrangements, leadership and external environment on the Court’s annual dissent rate between its first full year of operation in 1904 and 2001. The article is premised on the belief that this exercise is likely to be more fruitful than looking at the personal characteristics of the judges, which has been the focus of past research.

A study such as this is useful because it contributes to our understanding of how the High Court functions and how institutions affect decision-making. To this point, while there has been much casual observation about variations in dissent rates on the High Court over time and the reasons for these fluctuations,[5] there have been few attempts to collect the hard data.[6] The absence of a consistent data set on the High Court’s dissent rate over time has meant that doing studies such as this, which seek to explain variations in dissent using statistical methods, has not been possible. This has been a limitation in our understanding of judicial behaviour because even for those who are otherwise reticent about the use of empirical studies in legal scholarship, quantitative analysis can be used as evidence to support, or disprove, speculation about trends in dissent and its causes.

The balance of the article is set out as follows. The next section proposes several hypotheses about judicial dissent under the broad rubric of external and internal institutional factors, building on the existing literature for the United States. Section 3 contains an overview of the data, including a discussion of how it was collected and a review of the statistical methods employed in the study. The results are presented in Section 4, while the final section concludes with some suggestions for future research.

2. Hypotheses

Following the extant literature for the United States we hypothesise that organisational factors, leadership on the Court, and social and political complexity have influenced variations in the High Court’s dissent rate over time. The organisational factors that we consider are the Court’s caseload, whether there is a norm of dissent operating on the Court, the enactment of the Australia Acts,[7] which made the High Court the final court of appeal for Australia, and the existence of case selection discretion, defined as whether the Court has a discretion to select the cases that it hears. In order to consider the effects of leadership on the Court we examine the tenure of each chief justice. To capture increasing social and political complexity over time we use the urbanisation rate, which is defined as the proportion of the population which is living in urban areas.

A. Caseload and Institutions of the Court

Numerous studies in the United States suggest that courts with discretionary power to control their own caseload are likely to experience higher rates of dissent.[8] One rationale is that with case selection discretion, run-of-the-mill cases are weeded out, leaving only complex cases where there is a higher likelihood that reasonable minds can reach different conclusions. Another explanation is that heavy caseload demands reduce the expression of dissent.[9] The underpinning argument is that in courts with high caseloads not only is there likely to be a higher proportion of relatively easy cases, but there is less time for the reflection and hair splitting which is likely to contribute to dissent.

The requirement of grant of special leave to appeal was introduced in the High Court in 1984 by Section 3(1) of the Judiciary Amendment Act (No.2).[10] This reduced the number of cases the Court heard and increased their complexity. Michael Kirby describes the effect of introducing special leave to appeal: ‘Virtually all of the cases which are chosen involve delicately balanced issues where there are powerful arguments for both sides. Quite frequently they are expressed in the majority and minority opinions of the court under appeal’.[11] This is likely to contribute to increased disagreement over outcome.

H2: The introduction of case selection discretion will result in higher rates of dissent.

Soon after the introduction of special leave to appeal to the High Court, the High Court became the final court of appeal for Australia. Whilst there had been abolition of appeals to the Privy Council from the High Court through enactments in 1968[12] and 1975,[13] until the enactment of the Australia Acts in March 1986 it was possible to appeal to the Privy Council directly from State courts exercising State jurisdiction.[14] The effect of abolishing appeals to the Privy Council on the dissent rate, however, is unclear.

On the one hand, it is well-documented that abolition of appeals to the Privy Council acted as a catalyst for the Mason High Court to be more adventurous in its approach to decision-making. Sir Anthony Mason, himself, has suggested that many of the landmark decisions delivered during his tenure as chief justice would not have occurred if the High Court had still been bound by final review by the Privy Council.[15] When interpretation of the existing law is moving rapidly as it was in the Mason Court and is being driven by what some commentators regarded as overt judicial activism, there are likely to be strong differences in interpretation. In these circumstances, one might expect there to be higher levels of dissent as some judges hold on to traditional doctrines in the face of change.

On the other hand, when a court becomes a final court of appeal it takes on added responsibility for laying down the law of the jurisdiction. Some argue that with this added responsibility there is a need to ensure consistency in interpretation of the law. For instance, Lord Reid often expressed the view that the writing of dissenting judgments in final courts of appeal should be reserved for very important principles of law, although other judges, such as Michael Kirby, do not agree with this view.[16]

A related argument, sometimes used to explain the low rate of dissent on the United States Supreme Court before 1940, is that dissent weakens the authority of a final court of appeal. For example, famously Louis Brandeis is known to have withheld many dissenting opinions on the United States Supreme Court because he felt that random dissents would weaken the Court.[17] William Howard Taft, who was chief justice of the United States Supreme Court from 1921 to 1930, had a similar view. Like Lord Reid, he believed that ‘important questions of law should not be broken anymore than we can help by dissents’.[18] Taft is reported as stating: ‘I don’t approve of dissents generally, for I think that in many cases where I differ from the majority, it is more important to stand by the Court and give its judgment weight than merely to record my individual dissent’.[19] In contrast, Harlan Stone, who was chief justice of the United States Supreme Court from 1941 to 1946, when there was a sharp increase in the dissent rate, would not control dissent. Stone considered dissent critical because ‘legal principles .... never sprang full-fledged from the brains of any man or group of men. They are the ultimate resultant of the abrasive force of the clash of competing and sometimes conflicting ideas’.[20]

H3: The effect of the Australia Acts on the rate of dissent in the High Court is unclear. On the one hand there might be a proclivity for judges to dissent more in the face of rapid changes in legal interpretation. On the other hand, dissent might be viewed as undermining the authority of a final court of appeal, leading to less dissent.

Judicial decision-making is influenced by the prevailing norms of decision-making on the Court. The sharp increase in the dissent rate on the United States Supreme Court from the 1940s has been identified with a decline in consensual norms.[21] Previous research for the High Court suggests that given the long tradition of seriatim judgment writing and the absence of formal conferencing, unlike the United States Supreme Court, the High Court never developed a consensual norm.[22] Because the level of dissent will be influenced by norms of behaviour, established custom and precedent in a court’s decision-making practices will be important in explaining judicial behaviour. This suggests that, at any given point in time, accepted levels of consensus and dissent on the Court in the recent past will be an important precedent for current justices. Consistent with this conjecture, in a previous study Brian Goff found that on the United States Supreme Court the dissent rate in any given year had a strong positive effect on the dissent rate the following year. [23]

H4: Existing approaches and attitudes to a court’s appropriate decision-making norms will be important in explaining current judicial behaviour. The dissent rate in the Court’s recent past will be a good predictor of the Court’s current dissent rate.

B Role of the Chief Justice

The leadership of the court has often been identified in studies of United States courts as one of the major factors influencing consensus and dissent.[24] The importance of leadership is reflected in Figure 1, which plots fluctuations in the annual dissent rate on the High Court since its establishment. It shows that while the dissent rate has generally fluctuated between five and 15 judgments per 100 judgments delivered there have been substantial movements under different chief justices. For example, the second half of the 1920s and early 1930s, mid-to-late 1940s and the late 1990s were all periods of relatively high dissent. In contrast, in the early Griffith Court there was little dissent.

With the exception of the early Griffith Court (1903-06) and Gleeson Court, the High Court has not had a formal conference procedure.[25] Latham, Dixon and Barwick all experimented with informal conferences. Latham and Barwick are considered to have been largely unsuccessful in generating consensus through this means, although Dixon had more success.[26] The tenure of Knox, Gavan Duffy and Latham were all marked by strong personality differences on the Court which impeded attempts to build consensus and contributed to relatively high dissent rates in the 1920s–1940s.[27] When Evatt resigned in 1940 personal relations between the judges seemed to improve, but the Court did not function as a cohesive social unit until after Rich and Starke were replaced by Fullagar and Kitto following the defeat of the Chifley Labor government.[28]

Dixon is often regarded as the High Court’s most effective leader. His abilities to forge consensus were founded on the enormous respect he possessed as a jurist.[29] Compared with Dixon, Barwick is regarded as an ineffective chief justice. Barwick attempted to impose greater coordination in judgment writing and this was resisted by the puisne justices as an infringement on their individualism.[30]

The workings of the Court improved under Mason, Brennan and Gleeson. Meetings were held each month in the Mason Court to monitor progress with judgment writing and to exchange views and this practice was continued on the Brennan Court.[31] The anecdotal evidence which exists suggests that Mason’s approach to managing the Court resulted in a collegial atmosphere.[32]

H5: The ability of the chief justice to exercise leadership and foster consensus will influence the dissent rate. The dissent rate will be higher under chief justices who have failed to effectively exercise social leadership or foster consensual norms.

C Social and Political Complexity

More complex economic, political and social environments are believed to produce more frequent expressions of disagreement in judicial institutions.[33] In the studies for United States courts the urbanisation rate is used as a measure of socio-political complexity.[34] Dean Jaros and Bradley Canon capture the reasoning behind using this variable:

This is true not only of the American historical experience, but also of immigration settlement in Australia. It can be hypothesised that urbanisation produces the conditions for increased demands and conflicts in the political system which will generate more legal conflict.[36] Increasing rates of urbanisation are also often associated with changing societal values, which have been tied to judicial dissent. In contrast to the common depiction of a dissenter as a judicial innovator, John Schmidhauser has argued that the typical dissenter on the United States Supreme Court is not an innovator, but ‘a tenacious advocate of traditional legal doctrines which were being abandoned during his tenure’.[37]

3 Empirical Specification and Methodology

To test the above hypotheses we apply multiple regression analysis to analyze time series data for 1904 to 2001. In multiple regression analysis, one variable, which is called the dependent variable, is expressed as a function of other variables, called the explanatory variables. In this study the dependent variable is the annual dissent rate on the High Court. The explanatory variables are variables designed to measure the effect of consensual norms on the Court, the Court’s caseload, the effect of the Australia Acts making the High Court the final court of appeal for Australia, whether the Court had discretion to select the cases it hears and the effect of leadership. The objective is to ascertain to what extent variation in each of these explanatory variables can explain variation in the dissent rate. With multiple regression analysis we can determine which of these explanatory variables has been most important in explaining variation in dissent rates and whether each of these variables has had a positive or negative effect on the dissent rate on the High Court holding the effect of each of the other variables constant.

The definition of each variable and expected sign on the explanatory variables are given in Table 1. The expected sign is either positive or negative or unclear. If the expected sign is positive this means that variation in the variable is expected to lead to an increase in the dissent rate. If the expected sign is negative, it is expected that variation in the variable will result in a lower dissent rate. If the expected sign is unclear the stated hypotheses suggest there could be either a positive or negative relationship.

We now discuss each of the variables in Table 1 in more detail, beginning with the dependent variable. The dependent variable, Dissent, is the annual proportion of dissenting judgments per 100 judgments in multi-judge panels in cases reported in the Commonwealth Law Reports decided between 1904 and 2001. The dissent rate was calculated through reading and recording the outcome of each case. A judgment was classified as dissenting if the Justice disagreed with the result proposed by the majority expressed in the orders of the Court. In adopting this approach we use the term ‘dissenting’ in a manner consistent with the recent literature on measuring (dis)agreement on the High Court, particularly the methodology suggested by Andrew Lynch.[38] In most cases determining whether a judgment is in dissent is relatively straightforward. However, as Lynch points out, in a few cases there are shifting majority opinions within one case and the researcher has to make choices to determine whether a judgment is in dissent.[39]

VARIABLE	DEFINITION	EXPECTED SIGN
DISSENT	The proportion of dissenting judgments per 100 judgments in multi-judge panels delivered in cases, which are reported in the Commonwealth Law Reports	Dependent variable
DISSENT(-1)	For any given year, the dissent rate, in the previous year.	Positive
URBAN	The percentage of the population living in urban areas	Positive
CASELOAD
TOTAL CASELOAD	Total number of appeals heard in the High Court	Negative
CASE PER JUDGE	Total number of appeals heard in the High Court as a proportion of the number of justices on the Court	Negative
DISCRETION	Dummy variable set equal to 1 when the High Court had discretion over which cases it heard; zero otherwise.	Positive
FINAL	Dummy variable set equal to 1 following the enactment of the Australia Acts eliminating all appeals to the Judicial Committee of the Privy Council; zero otherwise.	?
GRIFFITH	Dummy variable set equal to 1 for the period when Griffith was Chief Justice of the Court; zero otherwise.	?
KNOX	Dummy variable set equal to 1 for the period when Knox was Chief Justice of the Court; zero otherwise.	?
ISAACS/DUFFY	Dummy variable set equal to 1 for the period when Isaacs and Duffy were Chief Justices; zero otherwise.	?
LATHAM	Dummy variable set equal to 1 for the period when Latham was Chief Justice of the Court; zero otherwise.	?
DIXON	Dummy variable set equal to 1 for the period when Dixon was Chief Justice of the Court; zero otherwise.	?
BARWICK	Dummy variable set equal to 1 for the period when Barwick was Chief Justice of the Court; zero otherwise.	?
GIBBS	Dummy variable set equal to 1 for the period when Gibbs was Chief Justice of the Court; zero otherwise.	Benchmark category
MASON	Dummy variable set equal to 1 for the period when Mason was Chief Justice of the Court; zero otherwise.	?
BRENNAN	Dummy variable set equal to 1 for the period when Brennan was Chief Justice of the Court; zero otherwise.	?
GLEESON	Dummy variable set equal to 1 for the period when Gleeson was Chief Justice of the Court; zero otherwise.	?

Where there are multiple issues in the case, one approach would be to record a dissent if Justice X dissented on any issue, but this tends to exaggerate the level of dissent if Justice X agreed in the orders for the other issues in the case. Therefore, in such cases we decided on which was the most important issue or issues before the Court and recorded whether Justice X dissented on this issue or issues.

Turning to the first of the explanatory variables shown in Table 1, for any given year, Dissent(-1) represents the dissent rate in the previous year. For example, for 1905, Dissent(-1) denotes the dissent rate in 1904; for 1906, Dissent(-1) denotes the dissent rate in 1905 and so on. Dissent(-1) is used to test Hypothesis 4 that the dissent rate in the Court’s recent past will be a good predictor of the Court’s current dissent rate. If Hypothesis 4 holds, the sign will be positive meaning that an increase in the dissent rate in any given year is expected to have a positive effect on the dissent rate the following year.

The next explanatory variable given in Table 1 is Urban, which is defined as the percentage of the population that is living in urban areas in Australia. The expected sign on the Urban variable is positive, meaning that an increase in the urbanisation rate is expected to result in a higher rate of dissent on the High Court (Hypothesis 6). The urbanisation rate from 1904 to 1959 is sourced from Mukherjee et al[40] and the urbanisation rate from 1960 to 2001 is taken from the World Bank World Tables.[41]

To measure the effect of caseload on the dissent rate we use two alternative variables. Total Caseload is the total number of appeals heard in the High Court for each year between 1904 and 2001. Case Per Judge is the total number of appeals heard as a proportion of the number of judges on the Court for each year between 1904 and 2001. Case Per Judge is designed to correct for the fact that the number of justices has not been constant throughout the Court’s history and that an increase in caseload might reflect more sitting justices. The expected sign on both variables is negative, which implies that an increase in the Court’s caseload is expected to have a negative effect on the dissent rate (Hypothesis 1). The sources for the caseload figures are the Commonwealth of Australia Yearbooks and the High Court of Australia Annual Reports.

The remaining explanatory variables are depicted as what are known as dummy variables. Data may either be quantitative or qualitative. Quantitative data is that which takes a number for each year. For example, in any given year we can assign a number to the Court’s caseload or the urbanization rate. Qualitative data refers to a characteristic which either exists or does not exist at a given point in time. Examples are: Did the Court have case selection discretion? Was the Court the final court of appeal for Australia? Was Griffith chief justice? These are questions to which the answer is either ‘yes’ or ‘no’ for any given year. Dummy variables are used to operationalise these characteristics. In other words, dummy variables are used to quantify the effect of specific institutional characteristics of the Court at particular periods of time on variations in the dissent rate.

Each dummy variable takes the value of one or zero for each year from 1904 to 2001 corresponding to whether a specific institutional characteristic of the Court was in place. For each year, if the answer is ‘yes’ the variable is assigned the value one. If not, the variable is assigned the value zero. The first dummy variable described in Table 1 is Discretion which depicts whether the Court has a discretion to select its own cases. It takes the value zero for each year from 1904 to 1983 when the Court did not have a discretion to select its own cases and one for each year from 1984 to 2001 when the Court did have discretion to select its own cases. The expected sign on Discretion is positive, meaning that the introduction of case selection discretion is expected to have a positive effect on the dissent rate. To put it differently, holding the effect of all the other explanatory factors constant, the dissent rate is expected to be higher after 1984 when the High Court was given the discretion to select the cases that it can hear (Hypothesis 2).

The next dummy variable in Table 1 is Final, which is designed to measure the effects of the Australia Acts on the dissent rate. It takes the value zero for each year from 1904 to 1985, prior to the enactment of the Australia Acts and the value one for each of the years 1986 to 2001. The expected sign on Final is unclear, meaning that the Australia Acts could have either a positive or negative effect on the dissent rate (Hypothesis 3).

The other dummy variables denote the tenure of the chief justices and are designed to capture the effect of leadership on the Court. For instance, the dummy variable for the Griffith Court takes the value one for each year from 1903 to 1919 and the value zero for the years 1920 to 2001. The dummy variable for the Knox Court takes the value one for each year from 1920 to 1929 and the value zero for every other year. The dummy variables for the other chief justices were constructed in a similar fashion.[42] When examining the effects of leadership on the Court we are interested in whether the dissent rate under each chief justice is higher or lower than under the other chief justices. Multiple regression analysis allows us to ascertain whether the dissent rate under each chief justice is higher or lower relative to the term of a specific chief justice, which is treated as a benchmark. In the analysis below we treat Gibbs’ period as chief justice as the benchmark.[43] Thus, the findings for the signs for the dummy variables for the other chief justices have to be interpreted relative to Gibbs’ period as chief justice.

To illustrate, if the sign on the dummy variable for Barwick’s term as chief justice is positive this means that the dissent rate was higher under Barwick than Gibbs, holding the other explanatory variables constant. To take another example, if the sign on the dummy variable for Dixon is negative, this implies that the dissent rate when Dixon was chief justice was lower relative to when Gibbs was chief justice holding the other explanatory variables constant. If this is the case, in terms of Hypothesis 6 we could conclude that Dixon was more effective than Gibbs at forging consensus. In Table 1 the expected sign for each Chief Justice’s term relative to Gibbs is given as unclear because while anecdotal evidence coupled with visual inspection of Figure 1 might suggest that the expected sign should be positive or negative for specific chief justices there is no theoretical reason to expect positive or negative signs on each of the variables.

Table 2 presents the mean (or average), standard deviation and minimum and maximum values for each variable. Some interesting points are worth noting. The average dissent rate over the period of the study was 9.8 per 100 judgments with a minimum of 1.05 per 100 judgments, which was in 1905, and a maximum of 23.46 per 100 judgments, which was in 1944. The average total caseload of the Court has been 86.07 cases per annum with a low of 29 cases and a high of 134 cases per year. Given that the dummy variables can either take the value zero or one, the minimum value for each dummy variable is zero and the maximum is one. The mean value for most of the dummy variables is close to zero which reflects the fact that there are more ‘zeros’ than ‘ones’ for each of the variables. To illustrate, consider the dummy variable Final denoting the Australia Acts. It takes the value one for 16 years (1986–2001) and zero for 82 years (1904–1985).

Table 3 shows the correlation coefficients between the dissent rate and each of the explanatory variables. As a first approximation, this is a measure of the degree of association between the two variables. The correlation coefficient can either have a positive or negative sign and can take values between -1 and +1. Values close to -1 suggest a strong negative relationship. Values close to +1 suggest a strong positive relationship. Most of the correlation coefficients in Table 3 are in the range -0.3 to +0.3 suggesting a weak association between each of the variables and the dissent rate.

Variable	Mean	Standard Deviation	Minimum	Maximum
Dissent	09.80	03.73	01.05	023.46
Urban	74.21	11.23	57.20	091.10
Caseload
Total Caseload	86.07	23.74	29.00	134.00
Case Per Judge	13.26	04.08	04.10	029.70
Discretion	0.18	0.39	0	1.
Final	0.16	0.37	0	1.
Griffith	0.16	0.37	0	1.
Knox	0.10	0.30	0	1.
Isaacs/Duffy	0.06	0.24	0	1.
Latham	0.16	0.37	0	1.
Dixon	0.12	0.33	0	1.
Barwick	0.17	0.38	0	1.
Gibbs	0.08	0.24	0	1.
Mason	0.08	0.28	0	1.
Brennan	0.03	0.17	0	1.
Gleeson	0.04	0.20	0	1.

4. Results

Table 4 presents the results from the multiple regression of the dissent rate on the explanatory variables in Table 1 and a constant term.[44] We begin by discussing how to interpret the results in Table 4 and then proceed to discussing the hypotheses. The first column lists each of the explanatory variables. The columns headed I and II present the results of two alternative sets of regressions. In the column headed I Case Per Judge is used to measure caseload and in the column headed II Total Caseload is used to measure caseload. Note that as these variables are both being used to measure caseload they are not included in the same regression. Total Caseload is excluded from the first set of regression results and Case Per Judge is excluded from the second set of regression results. The other variables are included in both sets of regressions reported in I and II. The constant term, which is also reported in Table 4, can be interpreted as the average effect on dissent rates of all variables omitted from the regression model.

For each explanatory variable, for each regression (reported under I and II) we provide the value and sign (positive or negative) on the coefficient and the t-value, which is reported in parenthesis next to the coefficient. To illustrate how to read Table 4, consider Dissent(-1). In the first regression under column I, the coefficient on this variable is 0.33 and the t-value is 3.08. In the second regression under column II, the coefficient on this variable is 0.38 and the t-value is 2.81. The value of the coefficient is an indicator of the responsiveness of the dissent rate to a change in that variable, holding all the other explanatory variables constant. The coefficient on Dissent(-1) can be interpreted as follows. Because the coefficient on Dissent(-1) has a positive sign in both sets of results an increase in the dissent rate last year will have a positive effect on the dissent rate this year. We can say that for any given year, a 1 per cent increase in the dissent rate last year will result in a 0.33 per cent increase (according to the results in I) or a 0.38 per cent increase (according to the results in II) in the dissent rate this year, holding all the other variables constant.[45] Note that this is slightly smaller than the corresponding rate for the United States Supreme Court between 1800 and 1994 where Goff found that, for any given year, a 1 per cent increase in the dissent rate last year results in about a 0.50 per cent increase in the dissent rate this year.[46] The coefficients on the other explanatory variables in Table 4 for both sets of regressions can be interpreted in a similar fashion.

The t-value, which is given in parenthesis, measures the statistical significance of the coefficient or, to put it differently, the reliability of the estimate of the coefficient.[47] Standard levels of statistical significance are 1 per cent, 5 per cent and 10 per cent. These are denoted in Table 4 by one, two and three asterisks respectively, next to the t-values.[48] If a coefficient is statistically significant at 1 per cent, it can be regarded as highly significant, at 5 per cent moderately significant and at 10 per cent weakly significant.[49] If a coefficient is not statistically significant we can conclude that the variable does not have any effect, which can be reliably measured, on the dissent rate. Thus, the variables which can be regarded as having an important effect on the dissent rate, either positive or negative, in Table 4 are those variables which are statistically significant.

Overall, the fit of the model is reasonably good. The adjusted coefficient of determination (which is depicted in Table 4 as Adjusted R2) is a measure of how much of the variation in the dissent rate can be explained by variation in the explanatory variables. The adjusted coefficient of determination is 0.41 in column I and 0.39 in column II, which suggests that about 40 per cent of the variation in the dissent rate can be explained by the specified model. Table 4 also presents the F-statistic, which is a measure of the overall significance of the regression. It is statistically significant at the 1 per cent level in the regression results reported in both columns I and II, which suggests that the explanatory variables are jointly important in explaining variations in the dissent rate.[50]

Variable	I		II
Constant	2.26	(4.67)*	2.19	(3.23)*
Dissent(-1)	0.33	(3.08)*	0.38	(2.81)*
Urban	-9.47	(1.30)	-8.37	(1.21)
Total Caseload	_	_	-0.19	(1.36)
Case Per Judge	-0.33	(1.75)***	_	_
Discretion	-0.03	(0.36)	-0.03	(0.33)
Final	-0.08	(3.08)*	-0.09	(3.53)*
Griffith	-0.27	(2.04)**	-0.28	(1.80)***
Knox	0.20	(1.67)***	0.20	(1.69)***
Isaacs/Duffy	0.15	(1.35)	0.09	(0.78)
Latham	0.20	(1.50)	0.16	(1.15)
Dixon	0.15	(0.94)	0.12	(0.79)
Barwick	0.13	(1.52)	0.14	(1.60)***
Mason	0.26	(2.75)*	0.30	(3.45)*
Brennan	0.20	(1.01)	0.28	(1.62)***
Gleeson	0.22	(1.32)	0.26	(1.61)***
R2	0.50		0.47
Adjusted R2	0.41		0.39
F-statistic	5.76	[0.01]	5.30	[0.01]

Notes: Gibbs was treated as the reference point for the leadership dummy variables. Figures in round parenthesis are Newey-West Heteroskedastic-consistent t-values. Figures in square parenthesis are probability values. *(**)(***) indicates coefficients are statistically significant at the 1%(5%)(10%) statistical significance level.

We now discuss each of the hypotheses set out above in light of the empirical results. The first two hypotheses are that the dissent rate will be inversely related to the caseload of the Court (Hypothesis 1) and that the introduction of case selection discretion will result in higher rates of dissent (Hypothesis 2). Beginning with Hypothesis 1, both measures of caseload have the expected negative sign. Case Per Judge is statistically significant at the 10 per cent level in the first specification, while Total Caseload is statistically insignificant in the second specification. This suggests that an increase in caseload does have a reliably measurable negative effect on the dissent rate when we control for changes in the number of justices on the Court. A 1 per cent increase in annual caseload, controlling for the number of justices on the Court, results in a 0.33 per cent reduction in the dissent rate. There is, however, no support for Hypothesis 2. While the sign on Discretion is negative, it is statistically insignificant in both specifications. The introduction of case selection discretion has not had any reliably measurable effect on the High Court’s dissent rate, holding each of the other explanatory variables constant.

The third hypothesis concerns the effect of the Australia Acts on the dissent rate on the High Court. We hypothesised that there are competing considerations suggesting that the Australia Acts could have either a positive or negative effect on the dissent rate. Final is statistically significant at 1 per cent with a negative coefficient in both specifications. This means that the introduction of the Australia Acts has had a strong negative effect on the dissent rate, holding each of the other explanatory variables constant.

The fourth hypothesis is that judicial decision-making is influenced by the prevailing norms of decision-making on the Court. Therefore, for any given year, an increase in the dissent rate last year will have a positive effect on the dissent rate this year. There is strong support for this hypothesis. As noted earlier a 1 per cent increase in the dissent rate last year will result in a 0.3–0.4 per cent increase in the dissent rate this year. This result is statistically significant at the 1 per cent level in columns I and II making it reliable.

The fifth hypothesis examines the effect of leadership on the Court. The coefficient on Griffith has a negative sign, while the coefficient for the other chief justice variables have a positive sign. If the tenure of a particular chief justice has an important effect on dissent rates, relative to the benchmark, which is Gibbs’ term, this will be reflected in a statistically significant coefficient for that chief justice’s tenure. The dummy variables for Griffith, Knox and Mason are statistically significant in the results in both columns I and II, while the dummy variables for Barwick, Brennan and Gleeson are statistically significant in the results in column II. This suggests that, holding the effect of the other explanatory variables on the dissent rate constant, we can reliably conclude in a statistical sense that the dissent rate was lower when Griffith was chief justice relative to when Gibbs was chief justice, while the dissent rate was higher under the leadership of Knox, Barwick, Mason, Brennan and Gleeson, relative to Gibbs’ term as chief justice.

The findings for the Griffith, Knox and Gleeson dummy variables are expected. The dissent rate on the Griffith Court, particularly the early Griffith Court, was very low relative to later years, while Knox presided over the first sharp increase in the dissent rate in the second half of the 1920s. The dissent rate also spikes in the late 1990s. This seems to reflect a ‘Kirby effect’ with just under one-third of Kirby J’s judgments being in dissent, which makes him the highest dissenter in the history of the Court.[51] Interestingly, the spike in the dissent rate in the first half of the 1930s under Gavan Duffy and in the mid-to-late 1940s under Latham does not show up in the form of statistically significant effects. Similarly, the leadership of Dixon does not have a statistically significant negative effect on the dissent rate relative to the leadership of Gibbs.

The final hypothesis is that an increase in the urbanisation rate will have a positive effect on the dissent rate. There is no support for this hypothesis with the urbanisation rate statistically insignificant in both specifications. Thus, in a statistical sense, the urbanisation rate has no measurable effect on the dissent rate. This finding is consistent with the most recent study of the determinants of dissent on State Supreme Courts in the United States by Paul Brace and Melinda Hall,[52] which also found that the urbanisation rate had no reliably measurable effect. Earlier studies which have found that urbanisation has a positive effect on dissent have used cross-sectional data. Brace and Hall suggest: ‘It is likely that whatever relationships, which were found between [urbanisation] and dissent rates in the past were time bound’,[53] meaning that they do not take account of the time series dimension. Another possibility is that while the urbanisation rate has been widely used, it might be too crude to capture socio-political complexity.

5 Conclusion

A clear picture that emerges from this study is that the Court’s institutions are better predictors of the dissent rate than the external environment. While the urbanisation rate does not have a statistically significant effect on the dissent rate, several factors internal to the Court are important predictors. These are norms on the Court as to acceptable dissent, with an increase in the dissent rate last year resulting in a higher dissent rate this year for any given year, caseload after controlling for changes in the number of justices throughout the Court’s history, and the enactment of the Australia Acts removing all appeals from Australian courts to the Privy Council. There are also statistically significant differences in the dissent rate under the leadership of different chief justices.

This study should be seen as a first step in unravelling the role that environmental and institutional factors have had on dissent in Australian courts. There are at least two avenues open to future research. First, future analyses could incorporate more complex and varied environmental factors. A potential problem is that some of the external factors which have been found to have the most predictive power in the United States, such as partisan competition, are clearly not relevant to Australia where judges are not elected. However, other factors which might be used to examine increased socio-political complexity and changing social attitudes in lieu of urbanisation could include indicators such as immigration levels reflecting ethnic diversity or possibly even divorce rates. Second, more sophisticated modelling could be employed to combine institutional variables with judges’ personal characteristics. While existing evidence on the application of the attitudinal model to the High Court suggests it might not have much power, the modelling was fairly crude.[54] Future studies could exploit the cross-sectional and time series dimensions of larger data sets to examine how personal characteristics, institutions and external factors interact in influencing variations in the dissent rate.

[1] See Sidney Ulmer, ‘Dissent Behaviour and the Social Background of Supreme Court Judges’ (1970) 32 Journal of Politics 580; John Schmidhauser, ‘Stare Decisis, Dissent and the Background of the Justices of the Supreme Court of the United States’ (1962) 14 University of Toronto Law Journal 194; Saul Brenner & Harold Spaeth, ‘Ideological Position as a Variable in the Authoring of Dissenting Opinions in the Warren and Burger Courts’ (1988) 16 American Politics Quarterly 317; Orley Ashenfelter, Theodore Eisenberg & Stewart Schwab, ‘Politics and the Judiciary: The Influence of Judicial Background on Case Outcomes’ (1995) 24 Journal of Legal Studies 257; C Neal Tate, ‘Personal Attribute Models of the Voting Behaviour of US Supreme Court Justices: Liberalism in Civil Liberties and Economic Decisions’ (1981) 75 American Political Science Review 355; Dean Jaros & Bradley Canon, ‘Dissent on State Supreme Courts: The Differential Significance of Characteristics of Judges’ (1971) 15 Midwest Journal of Political Science 322.

[2] See the discussion in Roger Douglas, ‘Courts in the Political System’ (1969) 2 Melbourne Journal of Politics 47.

[3] For the most recent and explicit test of the attitudinal model see Russell Smyth, ‘Explaining Historical Dissent Rates in the High Court of Australia’ (2003) 41 Commonwealth and Comparative Politics 83.

[4] See Edward Beiser, ‘The Rhode Island Supreme Court: A Well-Integrated Political System’ (1974) 8 Law and Society Review 167; Paul Brace & Melinda Hall, ‘Neo-Institutionalism and Dissent in State Supreme Courts’ (1990) 52 Journal of Politics 54; Bradley Canon & Dean Jaros, ‘External Variables, Institutional Structure and Dissent on State Supreme Courts’ (1970) 4 Polity 185; Lawrence Friedman, Robert Kagan, Bliss Cartwright & Stanton Wheeler, ‘State Supreme Courts: A Century of Style and Citation’ (1981) 33 Stanford L R 773; Henry Glick & George Pruet, ‘Dissent in State Supreme Courts: Patterns and Correlates of Conflict’ in Sheldon Goldman & Charles M Lamb (eds), Judicial Conflict and Consensus: Behavioral Studies of American Appellate Courts (1986); Melinda Hall, ‘Constituent Influence in State Supreme Courts: Conceptual Notes and a Case Study’ (1987) 49 Journal of Politics 117; Melinda Hall & Paul Brace, ‘Order in the Courts: A Neo-Institutional Approach to Judicial Consensus’ (1989) 42 Western Political Quarterly 391; Steven Peterson, ‘Dissent in American Courts’ (1981) 43 Journal of Politics 412; Kevin Stack, ‘The Practice of Dissent in the Supreme Court’ (1996) 105 Yale Law Journal 2235; Paul Wahlbeck, James Spriggs & Forrest Maltzman, ‘The Politics of Dissents and Concurrences on the US Supreme Court’ (1999) 27 American Politics Quarterly 488; Roger Handberg, ‘Leadership in State Courts of Last Resort: The Interaction of Environment and Procedure’ (1978) 19 Jurimetrics Journal 178.

[5] See generally, Andrew Lynch, ‘Dissenting Judgments’ in Tony Blackshield, Michael Coper & George Williams (eds), Oxford Companion to the High Court of Australia (2001).

[6] Andrew Lynch has made a start in this direction with his methodological piece on how dissent should be measured and an attempt to measure dissent in the Gleeson Court. See Andrew Lynch, ‘Dissent: Towards a Methodology for Measuring Judicial Disagreement in the High Court of Australia’ (2002) 24 Syd LR 470; Andrew Lynch, ‘The Gleeson Court on Constitutional Law: An Empirical Analysis’ [2003] UNSWLawJl 2; (2003) 26 UNSWLJ 32. An example of an earlier attempt at collecting data on dissent rates is Anthony Blackshield, ‘Quantitative Analysis: The High Court of Australia, 1964–1969’ (1972) 3 Lawasia 1. However, there are no studies which have collected data on dissent rates through the history of the High Court.

[7] Australia Act 1986 (Cth); Australia (Request and Consent) Act 1986 (Cth); the Australia Acts Request Act 1985 of each state; Australia Act 1986 (UK).

[9] See John T Wold & Gregory Caldeira, ‘Perceptions of ‘Routine’ Decision-Making in Five California Courts of Appeal’ (1980) 13 Polity 334; Burton Atkins & Justin Green, ‘Consensus on the United States Courts of Appeals: Illusion or Reality?’ (1976) 20 American Journal of Political Science 735.

[10] Judiciary Amendment Act (No.2) 1984 (Cth). For a discussion of its operation see D Jackson, ‘The Role of the Chief Justice: A View From the Bar’ in Cheryl Saunders, Courts of Final Jurisdiction: The Mason Court in Australia (1996).

[14] See David Jackson, ‘Leave to Appeal’ in Tony Blackshield, Michael Coper & George Williams (eds), Oxford Companion to the High Court of Australia (2001).

[22] Russell Smyth, ‘Historical Consensual Norms in the High Court’ (2002) 37 Australian Journal of Political Science 255.

[23] Brian Goff, ‘Supreme Court Consensus and Dissent: Estimating the Role of the Selection Screen’, Department of Economics, Western Kentucky University, Mimeo, December 2002.

[24] See, for example, David Danelski, ‘Causes and Consequences of Conflict and its Resolution in the Supreme Court’ in Sheldon Goldman & Charles Lamb (eds), Judicial Conflict and Consensus: Behavioral Studies of American Appellate Courts (1986); Robert Steamer, Chief Justice: Leadership and the Supreme Court (1986); Stacia Haynie, ‘Leadership and Consensus on the US Supreme Court’ (1992) 54 Journal of Politics, 1158–1169; Walker et al above n21, Caldeira & Zorn above n21. For a recent Australian study of the effects of leadership on consensus in the High Court, see Paresh Narayan & Russell Smyth, ‘Hail to the Chief! Leadership and Structural Change in the Level of Consensus in the High Court of Australia’, Journal of Empirical Legal Studies (forthcoming).

[25] For a more extended discussion of differences in leadership on the High Court and their relevance for building consensual norms see Smyth, above n3 at 88–91.

[26] See Troy Simpson, ‘Conferences’ in Tony Blackshield, Michael Coper & George Williams (eds), Oxford Companion to the High Court of Australia (2001).

[27] Amelia Simpson & Troy Simpson, ‘Personal Relations’ in Blackshield, Coper & Williams (eds), Oxford Companion, above n5. On personal relations in the Latham Court see C Lloyd, ‘Not Peace, but a Sword! The High Court under JG Latham’ [1987] AdelLawRw 9; (1987) 11 Adel LR 175.

[28] Roger Douglas, ‘Latham Court’ in Blackshield, Coper & Williams (eds), Oxford Companion to the High Court of Australia (2001).

[29] Leslie Zines, ‘Dixon Court’ in T Blackshield, M Coper & G Williams (eds), Oxford Companion to the High Court of Australia (2001).

[32] See, for example, Anthony Mason, ‘Personal Relations: A Personal Reflection’ in Blackshield, Coper & Williams (eds), Oxford Companion to the High Court of Australia (2001); Gerard Brennan, ‘A Tribute to Sir Anthony Mason’ in Cheryl Saunders (ed), Courts of Final Jurisdiction: The Mason Court in Australia (1996).

[34] Jaros & Canon, above n1; Henry Glick & Kenneth Vines, State Court Systems (1973); Kenneth Vines & Henry Glick, ‘State Courts and Public Policy’ in Herbert Jacob & Kenneth Vines (eds), Politics in the American States (3rd edn, 1976).

[38] See Andrew Lynch, ‘Dissent: Towards a Methodology for Measuring Judicial Disagreement’, above n6. See also Lynch, above n5; Michael Coper, ‘Concurring Judgments’ in Blackshield, Coper & Williams (eds), Oxford Companion to the High Court of Australia (2001); Michael Coper, ‘Joint Judgments and Separate Judgments’ in Blackshield, Coper & Williams (eds), Oxford Companion to the High Court of Australia (2001).

[39] Lynch, ‘Dissent: Towards a Methodology for Measuring Judicial Disagreement’, above n6.

[40] Satyanshu Mukherjee, Anita Scandia, Dianne Dagger & Wendy Matthews, Sourcebook of Criminal Justice Statistics (1988).

[42] When a new chief justice assumed office mid-year, the chief justice who presided for the majority of the year was coded 1. Because Isaacs was chief justice for a period shorter than one year (April 1930 to January 1931) his term was coupled with that of Gavan Duffy.

[43] There is no statistical rule for choosing the benchmark and equally one of the other chief justices could have been chosen. If one of the other chief justices had been used as the benchmark the results would be interpreted relative to the chosen chief justice. The Gibbs Court was chosen as the benchmark in this study because visual inspection of Figure 1 suggests that dissent under Gibbs was ‘average’, meaning that it was neither very high or very low. This should make it easier to document variations from the average under other specific chief justices such as Griffiths & Dixon (below average dissent) or Latham & Barwick (above average dissent) if they are measured relative to Gibbs’ term.

[44] Because we are using time series data it is necessary to pre-test each of the variables other than the dummy variables for stationarity. Time series data are often non-stationary. This means that if there is a sudden shock to the data, such as, in the case of the dissent rate, a change in leadership or change in the institutional arrangements on the court, it will tend to drift aimlessly and not return to its equilibrium level. Non-stationary data is sometimes likened to a drunkard’s walk where a drunk, having left a bar, will walk in a certain direction for a distance, stop and then walk in a different direction. If one or more variables are non-stationary the results will not be reliable unless the variable is made stationary. If one or more variables are non-stationary, they can be made stationary by taking the first difference. To investigate whether the variables are stationary we used the Augmented Dickey-Fuller and Phillips-Perron unit root tests with and without a trend, which are the most common tests for this purpose. The two unit root tests gave the same results; namely, Dissent, Dissent(-1), Case Per Judge and TOtal Caseload are each stationary, while Urban is non-stationary. Therefore, in the analysis below Urban is estimated in first difference form. The results of the unit root tests are available on request.

[45] The variables, other than the dummy variables, were expressed in natural logs which has the advantage that the coefficients on the explanatory variables can be interpreted as elasticities.

[47] The reported t-values are adjusted for heteroskedasticity using Newey-West robust estimator.

[48] The asymptotic critical values are 1.64 (10 per cent), 1.96 (5 per cent) and 2.32 (1 per cent).

[49] Technically, each coefficient represents a point estimate of an unknown population. Statistical significance is an indicator of the reliability of the estimate. Instead of relying on the point estimate alone we can construct an interval around the point estimator, such that this interval has α per cent probability of including the true parameter value. If the coefficient is significant at 1 per cent, there is a 99 per cent probability that the interval contains the true value; if the coefficient is significant at 5 per cent, there is a 95 per cent probability that the interval contains the true value, and if the coefficient is significant at 10 per cent, there is a 90 per cent probability that the interval contains the true value.

[50] Formally, the results from the F test mean we can reject the null hypothesis that the true slope coefficients are zero. In addition standard tests for serial correlation and stationarity of the residuals, not reported in Table 4, indicate that there are no statistical problems with the model.

[51] See Lynch, ‘The Gleeson Court on Constitutional Law’, above n6; Matthew Groves & Russell Smyth, ‘A Century of Judicial Style: Changing Patterns in Judgment Writing on the High Court 1903–2001’, Monash University, Unpublished Paper, 2003.

[54] See Smyth, ‘Explaining Historical Dissent Rates in the High Court of Australia’, above n3.

Variable	Pairwise Correlation Coefficient With Dissent
Urban	0.29
Caseload
Total Caseload	-0.15
Case Per Judge	-0.37
Discretion	0.20
Final	0.22
Griffith	-0.49
Knox	0.19
Isaacs/Duffy	0.01
Latham	0.12
Dixon	-0.10
Barwick	0.11
Gibbs	-0.09
Mason	0.13
Brennan	0.13
Gleeson	0.16