review attachment to complete the assignment

article3.pdf

Home >Psychology homework help >review attachment to complete the assignment

Criminal Justice Review ® 1990 College of Public and Urban Affairs Volume 15, Number 2, Autumn 1990 Georgia State University

THE DETERRENT EFFECT OF CAPITAL PUNISHMENT IN THE FIVE MOST ACTIVE

EXECUTION STATES: A TIME SERIES ANALYSIS

Scott H. Decker and Carol W Kohfeld

This study examines the effect of the death penalty on the murder rate. A 50-year time series is employed for the period 1930-1980 for the five states with the largest number of executions during this period: Georgia, New York, Texas, California, and North Carolina. Taken together, these five states accounted for 40 [wrcent of all the executions performed during this period. Incorporating a lag structure for the effect of executions, as well as several theoretically relevant explanatory variables for homicides, the study identifies no deterrent effect for executions. Several different policy-relevant analyses are performed, all with the same result. Neither the existence of the death penalty, its imposition, nor the level of imposition explains significant amounts of the variation in homicide rates in the 50-year period, 1930 to 1980.

Capital punishment is an issue that has prompted analysis from scholars in a variety of disciplines. Virtually every social science (and many physical sciences and humanities as well) has contributed to the debate over the existence, effect, and imposition of the death penalty. Such study now appears to have reached a historical high, prompted no doubt by the recent certification of the death penalty by the U.S. Supreme Court and the executions that have subsequently been carried out.

Three topics of debate have received the most attention. First, the legal merits of capital punishment have been debated in the courts, in state legislatures, and by the public. These debates have been most concerned with procedural issues and have concentrated in particulju- on Eighth Amendment concerns regarding "cruel and unusual punishment." While some have sought to ground these arguments in a more substantive framework, the bulk of the scrutiny in this regard has focused on the issue of the way in which the penalty has been imposed. Typically, the discriminatory or nonpatterned application of the death penalty has been the principal concern of these studies (Kleck, 1981; Paternoster, 1983). Thus, questions about the nature of the victim, characteristics of the offender, and contribution of aggravating circumstances to the sentencing decision have been the hallmarks of this approach.

As a second concern, many have debated the moral or ethical merits of the death penalty. This tradition is perhaps the oldest and most consistent theme in the analysis of the death penalty. Opponents have consistently

173

174 Scoit H. Decker and Carol W. Kohfeld

emphasized that executions represent little more than legalized killing performed in the name of the state. As such, they argue, these practices are without moral justification (Amsterdam, 1977; Conrad, 1983). Proponents, on the other hand, have argued that the failure to impose this severe penalty represents a serious moral error; indeed, such arguments have emphasized that it is a moral injustice not to impose the death penalty. These arguments stem from the traditional retributionist contention that those who have committed an offense have earned a penalty, and that, when such persons go unpunished, an injustice has been committed (Berns, 1979; Hook, 1961; Van den Haag, 1978).

Of all the debates, though, perhaps none has received more attention than the presumed deterrent effect of capital punishment on homicides. These deterrence studies have increased significantly in methodological and statistical rigor in the last decade. Earlier works employed the use of contiguous states as the basis for analysis. States with roughly similar social and demographic characteristics were compared in order to determine whether their homicide rates differed. Differences in this criterion variable were presumed to be the result of the primary differentiating feature—that one of the states had the death penalty and the other did not. Such analyses (Bailey, 1974; Sellin, 1958; Sutherland, 1925) consistently demonstrated that there was no difference in homicide rates between "similar" states that varied only with respect to the existence of the death penalty.

The next significant methodological advance came with the advent of correlational studies. The works of Bailey (1977) and Schuessler (1952) are the best-known examples of this trend. These studies sought to show a relationship between executions and the death penalty as well as other exogenous variables. They, too, failed to demonstrate the deterrent effect of executions identified by deterrence theory.

The most recent trend in deterrence studies has been the use of multivariate statistical tools. These analyses have incorporated the use of lag structures as well as a variety of techniques to minimize the effects of autocorrelation. In addition, such techniques lend themselves particularly well to the use of time series designs, a practice that represents an advance in data as well as method. By including a large number of points in time, deterrence studies can more accurately document the effects of executions. Such studies can also be categorized by the type of data used. Most deterrence studies in the 1970s used cross-sectional data (Bailey, 1974; Ehrlich, 1975); aggregates of jurisdictions, typically states, were grouped together for analysis. Recently, there has been a tendency to employ the use of a single jurisdiction in time series analysis. This preserves the advantages of the longitudinal design and eliminates some of the potential difficulties inherent in cross-sectional studies, particularly where policy inference is a likely application for the results.

The current study is consistent with these trends in deterrence studies. It proposes a time series design that incorporates a lag structure for the

Criminal Justice Review 175

analysis of the deterrence question in five states. A 50-year time series is used to assess the effect of executions on homicides in North Carolina, California, Texas, New York, and Georgia. These five states were chosen for analysis because of their historical use of the death penalty. They are the jurisdictions that have imposed the death penalty most frequently in this country. As such, they are likely candidates for analysis, from a policy standpoint as well as on methodological grounds.

LITERATURE REVIEW

The evidence on the existence of a deterrent effect of the death penalty is nearly unequivocal. Beginning with the work of Sellin (1958), most studies have failed to discover a deterrent effect. The Sellin study compared the homicide rates of similar contiguous states, one that had the death penalty and one that did not. The difference between the states' homicide rates in each case was negligible. This led Sellin to conclude that there was no deterrent effect for the death penalty.

Sellin's conclusion has been called into question on a number of grounds by Ehrlich (1975). The comparative approach, however, has recently been resurrected by Lempert (1983), who addressed the Ehrhch criticisms by basing his comparison of states not on the mere existence of the death penalty but on the number of executions actually performed. Lempert used Ehrlich's well-known finding that each execution saves approximately eight lives (1975) as the basis for his hypothesis that states with more executions should enjoy lower homicide rates in the contiguous state comparisons. Lempert found results almost identical to those reported by Sellin 25 years earlier: There was no evidence that the death penalty or the use of executions served as a deterrent to homicides.

A few studies have purported to demonstrate the deterrent effect claimed by advocates, while others have failed to uncover such an effect. Works that have demonstrated a deterrent effect include the studies of Ehrlich (1975, 1977), Layson (1985), and Yunker (1976). Each author employed national data in his analysis. Ehrlich's well-publicized study demonstrated a strong deterrent effect, indicating that approximately eight lives had been saved for every execution that had occurred between 1933 and 1969. His study was called into question on a number of grounds by Barnett (1981), Bowers and Pierce (1975), Klein, Forst, and Filatov (1978), and McGahey (1980).

Among the primary criticisms of the Ehrlich work were the identification restrictions and the lack of homogeneity of structural relations over time (Klein et al., 1978). These are two serious problems in time series analyses that focus on selection of criteria measures. Additional criticisms included the use of data aggregated at the national level, the exogenous variables in the equation, and the years included in the time series.

176 Scott H. Decker and Carol W. Kohfeld

A significant criticism of the applications of Ehrlich's work to policy questions has been the focus on his use of the states aggregated as a single unit. This procedure has the unfortunate effect of commingling the effect of the independent variables, most of which show considerable variation from one jurisdiction to another. This produces the difficulty in locating the source of policy effect noted earlier.

Some have contended that Ehrlich's model was misspecified. In particular, the concern exists that the Ehrlich study failed to include many of the relevant control variables. Other criticisms of this work have argued either that Ehrlich failed to use enough independent variables or that he failed to include the correct ones. While one may quibble ad infinitum about variable selection, this criticism has some merit since the relationship between age and murder is a well-accepted tenet of criminology.

Ehrlich's time series ends in 1969, just after the suspension of executions by most states and shortly before the abolition of the death penalty by the Supreme Court in 1972. Thus he has included a period of high homicide rates with no executions at the tail of his time series. This is the segment of his time series that produces high negative correlations. Indeed, Kleck (1981), using a time series that extended into the 1970s, reported no evidence for the deterrent effect described by Ehrlich.

While Ehrlich *s work represents the most sophisticated of those that have shown a deterrent effect, it has been sufficiently criticized (see also Barnett, 1981; Forst, 1977; Friedman, 1979; Klein et al., 1978; Passell, 1975) as to render its conclusions highly suspect. Indeed, Bowers' 1984 replication of Ehrlich s analysis, using the same data and the same analytic techniques, failed to find the same results. Yunker (1976) demonstrated an even stronger deterrent effect than did Ehrlich. His method, data, and results were rigorously criticized by Fox (1977), who noted severe misspecification problems in the analysis. Thus it appears that those studies that have demonstrated a negative relationship between executions and homicides have been fraught with methodological problems.

Layson's work (1985) is of particular relevance in this context. Layson reanalyzed data similar to that used by Ehrlich in the 1975 study and concluded that the initial findings of that work had understated the deterrent effect of executions on homicides. His work is particularly noteworthy because it used vital statistical data as the measure of homicides, substituted OLS for the 2SLS procedure used by Ehrlich and others, tested a variety of models including a variety of explanatory variables, and used a moving average to establish the probability of punishment. Layson concluded by noting, "The regression results consistently support the deterrence hypothesis that increases in the probabilities of arrest, conviction, and execution reduce the homicide rate. Even murderers appear to obey the law of demand" (1985, p. 88).

Layson's study was endorsed by the National Center for Policy Analysis, and its findings were disseminated by that group. A response to the study.

Criminal Justice Review 177

formulated by Fox (1986), was presented to the Subcommittee on Criminal Justice of the U.S. House of Representatives. Fox's criticisms of Layson's work focused on the absence of a theoretical model for depicting how homicides may be deterred by executions. Without this theoretical underpinning, Layson, Ehrlich, and others who argue for a deterrent effect cannot accurately document how deterrence works, much less document that it exists at all. In short. Fox noted the inability of econometric models using nationally aggregated data to capture the way in which sanctions affect behavior. This general criticism was the primary basis for the testimony.

Additional criticisms of the kind that others have offered of Ehrlich's work were also made. These objections included concern over the quality of the data, especially the measures of sanction, the time period included in the study, the negative bias that results from using the same term in the numerator of the dependent variable and in the denominator of a sanction measure, and the aggregation bias in using macro data to identify micro behavior. These criticisms constitute substantial refutation of the method Layson used and of the findings determined by those methods. It is important to note that these are substantially similar to the criticisms of the Ehrlich work noted above. Fox concluded by agreeing with the assertion of Brier and Feinberg (1980) that econometric analyses of deterrence had added little to understanding the deterrent effect of sanctions on crime.

It is interesting to note that studies that have focused on a single state as the unit of analysis have failed to find a deterrent effect. Bailey's studies of North Carolina (1978a), Utah (1978b), Oregon (1979a), California (1979b), and Ohio (1979c) and Decker and Kohfeld's study of Illinois (1984) have all demonstrated a consistent finding—executions exert no deterrent effect on homicide rates. Bailey's work, however, is not without its own flaws. His time series spanned the years 1910 to 1962, but he does not have 53 time points, as would be expected if he used annual data. Instead, he used sociodemographic data from the census years as proxies for the real values for the two years preceding and following the diennial census years. Thus he included only 25 data points, less than half the total number of years in his time series. His last data point was 1962, more than 20 years ago, and his data are therefore somewhat dated given the change in homicide rates between 1962 and 1980. In addition. Bailey performed no test for autocorrelation and did not include a consideration of the possible effect of a time trend upon the analysis. Clearly, these methodological problems suggest that a replication of these analyses that deals with these problems at the state level would prove useful.

What is apparent from this literature review is that the overwhelming majority of results show no deterrent effect. This is even more convincing in terms of the policy debate regarding the death penalty. The burden of proof must rest with the proponents of the death penalty. That is, those

178 Scott H. Decker and Carol W. Kohfeld

who contend that there is a deterrent effect should be pressed to demonstrate the effect conclusively in the results of their work. This is consistent with both the basic assumption of the scientific method (that is, burden of proof on rejecting the null hypothesis) and the significance of the penalty. The studies reported here point strongly to a single conclusion: Executions have little impact on homicide rates.

This conclusion holds over a variety of studies and methodologies. Each of the studies that have found a deterrent effect has been severely criticized in the literature. The studies that have failed to support the deterrence notion have been varied in level of aggregation, data, technique, control variables, and site. Given the nature of social science analysis, it would be inappropriate to depend on a single study as the basis for guiding policy for such an important issue, as was the case with the Gregg decision. The variations in method and data among the studies that fail to Hnd a deterrent effect reflect the strength of those who claim that no deterrence exists.

DATA SOURCES

Some of the data for this study were provided by the U.S. Bureau of Prisons. Specifically, the authors have received from that agency an enumeration of the annual number of executions by state for the time period 1931 to 1980. These data facilitate both the time series nature of the analysis and the approach of using a single jurisdiction (states) as the unit of analysis. The dependent variable, annual rate of murder and nonnegligent manslaughter, was drawn from the Uniform Crime Reports. Other exogenous variables were drawn from the Statistical Abstracts. Selection of these variables was based on an analysis of the death penalty and homicide literature. There was a dual criterion applied for the selection decision: whether the variable was identified in previous research as having both empirical and theoretical relevance to variations in homicide rates. The controls included the proportion of the state population that was male and aged 15 to 29, the proportion of the state population living in urban areas, and the proportion of the state population in nonagricultural employment.

As noted earlier, these five states were chosen because of their high frequency of use of executions. Taken together, the five states account for 1,547 of the 3,863 executions performed in the United States between 1930 and 1981. Of the five states, Georgia has executed the most persons (366), followed by New York (329), Texas (297), California (292), and North Carolina (263). That these five states have been the most active can be easily demonstrated by looking at the next highest state in terms of frequency of executions, Florida, which executed 171 persons during the same time period. The method of execution varies for these states.

Criminal Justice Review 179

Currently, Georgia and New York use electrocution, California and North Carolina utilize lethal gas, and Texas makes use of lethal injection. All of the states except New York had persons on death row awaiting execution as of December 31, 1981. Indeed, during 1981 California had the greatest number of new death row inmates (40) of any state in the nation. Texas, Georgia, and California had the second, third, and fourth highest numbers of inmates on death row as of December 31, 1981. Thus the picture offered of these five states is that they not only have been active historically in imposing the death penalty but are continuing to actively pursue the imposition of this form of punishment.

The analysis performed here focuses on these states for a variety of reasons. First, they were selected for inclusion because, as was documented above, they represent the most active jurisdictions in the imposition of the death penalty. As such, they represent important sites for examining the impact of execution policy. Findings about the deterrent effect of executions in these states also have broader implications than would be the case for states with few executions. A second reason is more closely related to the type of analysis presented here. Because executions are typically not a frequent practice, states with a relatively low number of executions represent cases that are more problematic for study. That is, when the independent variable occurs infrequently, it is more difficult to draw inferences from results. Therefore, from a strictly methodological viewpoint, these five states are desirable units of analysis.

Clearly, such substantive differences underscore the need for single state analyses. In addition, several policy-related reasons exist for choosing this unit of analysis. It may be that when several states are aggregated the effect of the use of executions becomes commingled. That is, if executions deter in one or a few of the states in the aggregate, a strong negative effect in one state may create the appearance that the policy has an overall deterrent effect. The best example of this error occurs in studies that have aggregated several states (Ehrlich, 1975). When a deterrent effect is shown to result, one is in the difficult inferential position of having to claim that executions in death penalty states have a deterrent effect on homicides in abolition states. This is at best a tenuous position to maintain.

THE CURRENT ANALYSIS

In light of these findings, several different analyses were performed. The time series nature of these data made it possible to address the deterrence issue from several vantage points. Mean homicide rates were compared for three different eras corresponding to the death penalty experience in each of the five states. These periods are referred to as (a) "use years," in which the death penalty was in force and executions were carried out, (b) "threat years," in which the death penalty was in force but there were no

180 Scott H. Decker and Carol W. Kohfeld

executions, and (c) "abolition years," in which there was no death penalty in force. This natural division allowed the assessment of three distinctively different eras, each of which can be viewed as representing a particular policy. This permitted a distinction to be drawn between the effect of the actual use of the sanction and the effect of the mere threat of its use, in addition to the more typical questions about the effect of its abolition. These comparisons were elaborated in a number of ways, beginning with zero-order correlations and then proceeding to testing for autocorrelation, which lead to the judgment that generalized least-squares estimation is required only in California.

In the search for consistency in deterrent effects for executions, three modal variants were investigated. First, a polynomial in time was used as a surrogate for all other controls. Second, regressions were run in each state using executions timed contemporaneously and also timed with lag. Several controls that had been demonstrated to have both theoretical and empirical relevance to homicide rates were included in the models. Third, on the hypothesis that contiguous states' crime rates and executions might proxy unmeasurable demographic factors in the five states, regressions were run using such unique independent measures. These searches for consistency in the effect of executions were uniformly unsuccessful.

FINDINGS

In Table 1 the mean homicide rate for each state during the three eras identified above is presented. This allows a comparison of the behavior of the dependent variable during years in which there were executions (use years), years in which the death penalty was in force but there were no executions (threat years), and years in which the death penalty was not a legally available sanction (abolition years). Because each state has executed different numbers of prisoners, the threat and use estimates are different for each state.

The most interesting observation to be derived from this table is concerned with the lack of consistency of effect across the states. This inconsistency may be linked to the fact that this type of analysis fails to include a consideration of the control variables that may affect this relationship. Deterrence theory would predict that the homicide rate should descend from highest to lowest respectively in abolition, threat, and use years. Thus the threat of the sanction should show a deterrent effect when compared with years in which no executions were allowed. This result occurs in only two of the jurisdictions. New York and California. In each of these states there is a considerable increase in the dependent variable as one moves from use to threat to abolition eras.

The results for North Carolina and Georgia, however, are somewhat different. In each of these states the homicide rate is lowest for threat

Criminal Justice Review 181

Table 1

A Comparison of Mean Homicide Rates per 100,000 Residents for New York, California, North Carolina, Georgia, and Texas, 1933-1980, for USE, THREAT, and ABOLITION Periods of Capital Punishment

State New York

Mean Standard Deviation N

California Mean Standard Deviation N

North Carolina Mean Standard Deviation N

Georgia Mean Standard Deviation N

Texas Mean Standard Deviation N

USE

3.07 .74

4.24 .96

15.96 5.91

19.30 6.83

12.5 3.9

THREAT

7.76 3.44

8.00 3.75

9.63 1.32

13.44 2.11

12.09 2.98

ABOLITION

10.92 .19

9.60 .73

12.20 .79

16.40 2.10 5

12.86 .66

years, followed by abolition years, and is highest for use years. This suggests the rather anomalous conclusion that having the death penalty but not using it is the most successful deterrent and that abolishing the death penalty proves to be a greater deterrent than using it.

The results in Texas do not fit either of the patterns identified above. As one compares the means for the three eras within Texas, the obvious conclusion is that there are no differences in any of the eras. This set of findings presents us with a pattern that will be replicated throughout this paper: These five states show no consistent pattern for the effect of the death penalty and of executions on homicide rates. In fact, the results are often internally contradictory and frequently run counter to deterrence theory.

The deterrence hypothesis suggests that sanctions—that is, executions- should deter homicides. When the zero-order correlation coefficients between murder and nonnegligent manslaughter rates and executions are examined, we expect to find a negative relationship if the deterrence hypothesis is correct. If executions have an immediate effect, then a stronger relationship should be observed for the year in which the executions occur; if the deterrent impact is delayed, then we might expect the negative relationship to be stronger when the sanction variable is lagged one or even two years. In Table 2, simple correlations are presented for the five states for the years 1933 to 1980. In this set of tables simple correlations between the murder and nonnegligent manslaughter rate and

182 Scott H. Decker and Carol W. Kohfeld

sanction measures (executions, executions lagged one year, and a dummy variable—policy—that measures the presence or absence of executions) as well as several demographic variables (discussed more completely below) are examined.

The pattern within states is consistent for the sanction measure and its lags. That is, the lagged and nonlagged executions and policy have the same effect (sign) within each state. Across states, however, the results are inconsistent. For the industrialized states. New York and California, the relationship is both negative and significant for all three sanction measures. It appears, at least for this simple relationship in these two states, that executions deter murders and indeed continue to deter them for at least two years after the executions occur. In the more rural Southern states of North Carolina and Georgia, the correlations are just as strong but are positive, while in Texas the relationships are positive but weaker. Thus, in these states, executions and lagged measures of executions could be interpreted as encouraging an increase in murder rates.

How does one explain these results? Rather anomalous conclusions emerge if we take these correlations at face value. Executions deter murder in some states and incite murder in others. An alternative explanation might be that cultural differences between states account for the differential impact of executions. This explanation is reminiscent of the South and non-South differences that have been observed in a wide variety of studies. Cultural factors are difficult to measure, however, partly because they are not readily identifiable. Furthermore, an appeal to cultural factors is usually a confession of ignorance or inability to measure. We pursue a path of explicit measurement here.

One likely explanation is that simple correlations are really model artifacts that are evidence of model misspecification. It is well understood that time series can contain trends that are not independent of the processes being studied and can influence the correlations by their omission from the model. Reporting practices vary for some offenses, both across states and over time within states. These problems have been documented in a variety of sources. But reporting differences are less problematic with homicides, both within and across states, because murder tends to be a more uniformly defined and fully reported crime. There are well-documented and obvious time trends for violent crimes, and murder rates tend to track similarly to other violent crimes.

It is not the case, however, that crime rates have simply risen everywhere across this 50-year period. In Figure 1, the murder and nonnegligent manslaughter rates (murders per 100,000 population) are plotted for North Carolina and New York across the 50-year period. During the first 30 years. New York's homicide rate stays essentially constant while North Carolina's can be characterized as steadily decreasing. Both tend to increase during the years of rapidly increasing violent crime (1960 to 1973)

Criminal Justice Review 183

Table 2

Correlation Coefficients for Murder and Nonnegligent Manslaughter Rates (MNMAN) With Executions (EXEC), Executions Lagged One Year (LEXEC), Proportion of the Population Urban (URBP), Level of Industrialization (INDUS), the Occurrence of Executions, a Dummy Variable (POLICY), and the Proportion of the Population Male, Aged 15 Through 29 (M1529P) for the Five States—Texas, North Carolina, New York, Georgia, and California: 1933-1980

TEXAS

MNMAN EXEC LEXEC URBP INDUS POLICY

EXEC

.26

NORTH CAROLINA

MNMAN EXEC LEXEC URBP INDUS POLICY

NEW YORK

MNMAN EXEC LEXEC URBP INDUS POLICY

GEORGL4

MNMAN EXEC LEXEC UREP INDUS POLICY

CALIFORNIA

MNMAN EXEC LEXEC URBP INDUS POLICY

EXEC

.77

EXEC

-.55

EXEC

.55

EXEC

-.62

LEXEC

.21 .70

LEXEC

.67

.72

LEXEC

-.60 .70

LEXEC

.67

.73

LEXEC

-.64 .66

URBP

-.52 -.76 -.69

URBP

-.71 -.79 -.75

URBP

.08 -.71 -.66

URBP

-.77 -.84 -.76

URBP

.59 -.75 -.69

INDUS

.21 -.74 -.77 .80

INDUS

-.39 -.63 -.67 .94

INDUS

.34 -.25 -.09 .14

INDUS

-.48 -.76 -.79 .84

INDUS

.84 -.58 -.63 .56

POLICY

.09

.76

.73 -.71 -.81

POLICY

.54

.69

.64 -.86 -.79

POLICY

-.79 .71 .58

-.31 -.36

POLICY

.38

.74

.73 -.71 -.87

POLICY

-.71 .82 .72

-.77 -.64

M1529P

.65

.04

.02 -.41 .23

-.24

M1529P

.62

.63

.61 -.42 .04 .32

M1529P

.15

.53

.55 -.87 .24

-.01

M1529P

.54

.23

.17 -.45 .30

-.22

M1529P

.49 -.31 -.32 .01 .54

-.46

184 Scott H. Decker and Carol W. Kohfeld

and settle down during the late 1970s and converge to approximately the same rate in 1980. Thus, across the observed time period. North Carolina's rate starts higher and decreases, while New York's rate starts lower and increases. To better understand the anomalous results from the simple correlations and the time trends, we turn now to regression models that include some demographic and socioeconomic control variables.

THE REGRESSION MODELS

Preliminary results using ordinary least-squares estimation showed that the autocorrelation that is often found in time series was a problem in these data with models that did not include demographic and socioeconomic control variables. For all the states except California, with the better- specified models including controls, the autocorrelation coefficients were below the .3 cutoff (Hanushek & Jackson, 1977), where correcting for autocorrelation would make little difference in the results. The autocorrela- tion problem in California was corrected by using the generalized least- squares procedure provided in the statistical package SAS. This procedure reduced the autocorrelation coefficient for California to within acceptable limits but did not change the results obtained using ordinary least-squares in any important substantive respect. The models reported in the tables were estimated by ordinary least-squares for all states except California. The models for California were estimated using the SAS generalized least- squares procedure.

Simple models for each state that included only time as a series of increasing integers and a sanction measure as the independent variables yielded inconclusive results. Time was included in these limited models as a surrogate for all the other controls that were not included. The models were run once with the number of executions in the current year as the sanction measure and once with the number of executions from the previous year as the sanction measure (that is, sanctions lagged one year). There is some argument that executions have their impact over a longer period of time, and thus it is argued that executions of last year have more impact this year. The effects of sanctions and of sanctions lagged, when controlled in the simple models with a time variable, continued to have an inconsistent impact across these five states, and, as mentioned earlier, autocorrelation was a serious problem. The uneasy result still remained that executions deterred murder in Georgia, incited murder in North Carolina, and had essentially zero impact in California, New York, and Texas.

An alternative to including time as a simple linear function is to fit a polynomial of chronological time to the homicide rates for each state, for the period that executions were used but not thereafter. This included the years through the early 1960s for all five states. Models were examined that

Criminal Justice Review 185

186 Scott H. Decker and Carol W. Kohfeld

included time, time squared, and time cubed, with the sanction variable measured both in the current year and in the preceding year. The results continued to be inconsistent across states. The only significant relationship between the sanction measure and homicide rates was in North Carolina, but it was positive. For the rest of the states, the results were an inconsistent mixture of nonsignificant positive and negative relationships with no clear pattern emerging.

For a more completely specified model, some of the variables that have been shown in studies to be significantly related to murder rates were included. The subgroup of the population that is most likely to be involved with murders is that of young men between the ages of 15 and 29. Higher murder rates in some states might be due to disproportionate numbers of young men in the population. Thus, controlling for the proportion of the population in this age group should help in making a determination as to whether executions have significant deterrent effect on murder rates. A measure of males in these states between the ages of 15 and 29 is included here as a proportion of the population as a whole (Messner, 1983).

Another change that occurred during this time period is increased urbanization (Archer & Gartner, 1984). In Northern states, urbanization is usually associated with increased crime rates. In Southern states, however, it is the more rural-dominated culture that is associated with increased violent crime rates. Urbanization is related to the murder rates in the state models; it seems important to include it as a control for these changes in living environments in these states over this time period so that its impact on violent crime is not confused with that of sanctions.

Similarly, some measure of economic conditions should be included to control for their change throughout this time period. The first choice for an economic measure was unemployment. But unemployment measures are not available for states annually before 1960, and there is considerable doubt about how well these measures are estimated across the years and across the states as well as within the states. Since the time series for murder rates exists from 1933, it was decided that another measure of economic enterprise, which was available as early as 1939, would be used. A crude measure of labor force composition was chosen, which was calculated by norming the number of nonagricultural employees to the total population in each state annually. This still truncated the time series, but less than would have occurred using an unemployment measure. Although this is a rough measure, it should control for major changes in the labor force structure that occurred especially in the South during this time period.

Thus, in the more complete model used to estimate the effect of sanctions upon murder rates, the following are included as control measures: the percentage of males between the ages of 15 and 29, the percentage of urbanization, and a measure of the labor force composition.

Criminal Justice Review 187

Two versions of the model are estimated for each state: one that includes the three controls and executions in the current year and one that includes the same three controls and executions lagged one year. These models are estimated using ordinary least-squares for all states except California, where a generalized least-squares procedure was used, and the results are reported in Tables 3 and 4.

First, note that these models explain a significant amount of the variance in homicide rates over time in each state as evidenced by the R^ This is not surprising, since in previous studies and in the zero-order correlations presented in Table 2 most of these variables have been shown to be related to murder rates at least cross-sectionally, and in some cases over time. What is important for our purposes is to note the impact, or perhaps we should say the lack of impact, of sanctions on murder rates when controls are instituted. In Table 3, where executions in the current year are included, the signs for the sanction coefficients are the same as in the simple correlations (Table 2), except for Georgia where the positive simple correlation turns negative in the presence of controls. The positive

Table 3

Results of Generalized Least-Squares Regressions for California and for Ordinary Least-Squares for New York, North Carolina, Georgia, and Texas

[The dependent variable is Murder and Nonnegligent Manslaughter Rate (per 100,000 population)— MNMAN. The independent variables are Proportion of Population in Nonagricultural Employment (INDUS), Percent Population Which Is Male Between 15 and 29 (M1529P), Proportion Urban (URBP), and Number of Executions (EXEC). All are annual measures.]

Independent Variables

Intercept

M1529P

EXEC

URBP

INDUS

.58

-22.0 (7.18) -3.07»*

.77 (.43) 1.78

-.070 (.069)

-1.013

8.65 (6.81) 1.27

36.8 (11.0)

3.33**

.57

80.0 (102.9)

.78

.81 (.74) 1.10

-.49 (.083)

-5.91***

-103.5 (118.5)

-.87

16.6 (24.0)

.69

'Unstandardized Regression Coefficients "Standard Error Regression Coefficient 't scores

* p < .OS ** p < .01

• • • p < .001

.57

15.6 (16.7)

.93

.592 (.908) .65

.080 (.136) .58

-58.2 (44.0)

-1.32

37.2 (40.1)

.93

.66

35.0 (19.2)

1.83

1.09 (1.04) 1.05

-.28 (.18)

-1.57

-62.5 (26.8)

-2.33*

4.36 (30.1)

.14

.59

13.2' (10.4)"

1.27<

.230 (.601) .38

.046 (.127) .36

-28.8 (9.0) -3 .20"

53.9 (13.9)

3 .88*"

188 Scott H. Decker and Carol W. Kohfeld

Table 4

Results for Generalized Least-Squares Regressions for California and for Ordinary Least-Squares for New York, North Carolina, Georgia, and Texas.

[The model reported here is essentially the same as that presented in Table 3 with the exception that the sanction variable (Executions) is lagged one year—LEXEC]

Independent Variables

R̂

Intercept

M1529P

LEXEC

URBP

INDUS

'Unstandardized "Standard Error t scores

* p < .05 ** p < .01

*** p < .001

.58

-23.0 (7.18) -3.21**

.82 (.42) 1.92

-.050 (.067)

-7.40

9.07 (6.81) 1.33

36.9 (11 .1)

3 .32"

.68

38.3 (84.9)

.45

1.01 (.63) 1.60

-.514 (.066)

-7.74**»

-63.2 (98.3)

-.64

32.7 (20.7)

1.58

Regression Coefficients Regression Coefficient

.61

20.1 (16.2)

1.24

.0077 (.91) .01

.248 (.125) 1.98

-61.3 (41.5)

-1.48

50.4 (38.9)

1.29

.66

-2.31 (18.6)

-.12

2.37 (1.03) 2.30»

.26 (.17) 1.47

-15.2 (25.4)

-.60

-16.4 (29.4)

-.56

.60

22.1' (9.69)'' 2.28'

-.10 (.56) .18

-.12 (.11)

-1.05

-34.7 (8.49) -4.09***

54.4 (13.6)

3.99»«*

correlations between homicide and measures of sanction is consistent with the finding of "brutalization" of Bowers and Pierce (1980). While it would be premature to label these findings as evidence confirming the existence of such an effect, this is an issue that should be addressed by data able to measure such processes more effectively.

Only the sanction coefficient in New York is both negative and significant and thus supportive of deterrence theory. In addition, the uneasy conclusion from Table 2, that executions incite murder in North Carolina, is tempered because the coefficient is no longer significant. Executions continue to show differential but insignificant impact across the states, except for New York, even when young male population, urbanization, and labor force composition measures are included in the model. One would be hard pressed, using this evidence from the states that have made the most use of executions, to recommend using the death penalty as a deterrent to murder, except in New York.

From Table 4, where the sanction measure was lagged one year, we find no significant differences in the amount of variance explained in the murder rates across states. Again, sanctions—executions lagged one year— do not have a significant impact upon murder rates in any state except New

Criminal Justice Review 189

York. For New York, lagged sanctions have a significant {p < .001) and negative impact on murder rates. The lagged sanction measure is now negatively related in Texas and positively related in Georgia, which is the reverse of the nonlagged measure results. In neither model, however, is the effect of sanctions significant, and therefore the sign changes can be attributed to chance or cited as evidence of an unstable relationship between sanctions and homicide rates. Again we must conclude, at least from the evidence reported here, that executions do not exert any long-term deterrent effect on murder rates.

One other analysis was attempted to try to find consistency in the deterrent effect of sanctions across these states. Exogenous variation within each of the five states was proxied by including execution and homicide rates of a similar contiguous state. Models were examined using both current measures and lagged (one year) measures for the contiguous states: for New York, Pennsylvania; for North Carolina, South Carolina and Virginia; for Georgia, Florida; for Texas, Louisiana; and, for California, Arizona. A good case can be made for similarity for all pairs except perhaps California and Arizona. Once more, the results were inconsistent across states. The sanction measures were negative and significant for New York and California, which supports deterrence, but for Georgia and North Carolina the relationships were positive and insignificant, and for Texas the relationship was both positive and significant.

DISCUSSION AND CONCLUSIONS

What is one to make of the several different sets of results presented here? The first and most obvious comment to be made must point to the anomalies among the findings. A consistent pattern of findings within each state was not observed; nor can such statements be made about the states as a group. It would be expected that a certain commonality of outcomes should emerge for these five states, given their high level of activity in the area of executions. The conclusion to be drawn from the early section of the analysis—where a deterrent effect was observed in some jurisdictions, the opposite effect was observed in others, and in still others no effect was discernible—is that the penalty has a varying effect depending on where it is measured. A deterrent effect, however, was observed for executions only in the preliminary examinations of means and correlations and in the regression analysis for New York. This renders the disparities identified in the early section of the paper somewhat easier to explain.

When a more appropriately specified model is presented, the effect of executions on homicides is significant only in New York. That is, sociodemographic variables account for the variation in homicide rates in four of the five states. The sanction measure, executions, accounts for virtually none of the explanatory power of the models. This conclusion

190 Scott H. Decker and Carol W. Kohfeld

holds true, whether lagged or unlagged effects are examined, and is robust for varying specifications designed to deal with autocorrelation problems. When surrogate demographic controls are introduced using contiguous state homicide and execution measures, the results sustain themselves unimpaired. In short, we could not remove the inconsistent pattern of execution effects.

In this respect these results are consistent with the majority that have preceded them. Of particular note in this context are the studies by Bailey for California (1979b) and North Carolina (1978a), two states included in this analysis. Bailey found no deterrent effect, and the bulk of the variation was explained by sociodemographic variables. In many ways, such findings should not be surprising. Even in these five active states there have been relatively few executions over a 50-year period, compared to the number of homicides. In addition, there is a great gap in the certainty of application of the penalty. It may also be that the level of analysis of the current approach faiils to capture the real essence of deterrence, if one argues that only by measuring individual perceptions of the certainty of apprehension and the severity of punishment will deterrence questions ever be adequately addressed. One conclusion seems inescapable from the results presented here: Executions have failed to exert a consistent deterrent impact on homicides in the five states most likely to execute.

REFERENCES

Amsterdam, A. G. (1977). In opposition to the death penalty. In H. Bedau (Ed.), The death penalty in America (pp. 346-358). New York, NY: Oxford University Press.

Archer, D., & Gartner, R. (1984). Violence and crime in cross-national perspective. New Haven, CT: Yale University Press.

Bailey, W. C. (1974). Murder and the death penalty. Journal of Criminal Law and Criminology, 65, 416-423.

Bailey, W. C. (1977). Imprisonment v. the death penalty as a deterrent to murder. Law and Human Behavior, 1, 239-260.

Bailey, W. C. (1978a). An analysis of the deterrent effect of the death penalty in North Carolina. North Carolina Central Law Journal, 29-52.

Bailey, W. C. (1978b). Deterrence and the death penalty for murder in Utah: A time series analysis. Journal of Contemporary Law, 5, 1-20.

Bailey, W. C. (1979a). Deterrence and the death penalty for murder in Oregon. Willamette Law Review, 16, 67-85.

Bailey, W. C. (1979b). The deterrent effect of the death penalty for murder in California. Southern California Law Review, 52, 743-764.

Bailey, W. C. (1979c). The deterrent effect of the death penalty for murder in Ohio: A time series analysis. Cleveland State Law Review, 28, 51-81.

Barnett, A. (1981). The deterrent effect of capital punishment: A test of some recent studies. Operations Research, 29, 346-370.

Bowers, W. (1984). Legal homicide: Death as punishment in America, 1864-1982. Boston, MA: Northeastern University Press.

Bowers, W., & Pierce, G. (197S). The illusion of deterrence in Isaac Ehrlich's research on capital punishment. The Yale Law Journal, 85, 187-208.

Bowers, W., &. Pierce, G. (1980, October). E>eterrence or brutalization: What is the effect of executions? Crime and Delinquency, 26, 453-484.

Criminal Justice Review 191

Brier, S. S., & Feinberg, S. E, (1980). Recent econometric modeling of crime and punishment: Support for the deterrence hypothesis? Evaluation Review. 4, 147-191.

Conrad. J. P. (1983). The state as killer. American Bar Foundation Research Journai, 2, 451-464. Decker, S., & Kohfeld, C. (1984). A deterrence study of the death penalty in Illinois: 1933-1980.

Journal of Criminai Justice, 12, 367-379. Ehrlich, I. (1975). The deterrent effect of capital punishment: A question of life and death. American

Economic Review, 65, 397-416. Ehrlich, I. (1977). Capital punishment and deterrence: Some further thoughts and additional evidence.

Journal of Political Economy, SJ(4), 741-788. Forst, B. E. (1977). The deterrent effect of capital punishment: A cross state analysis of the 1960s.

Minnesota Law Review, 61, 743-767. Fox, J. A. (1977). The identification and estimation of deterrence: An evaluation of Yunker's model.

Journal of Behavioral Economics, 2, 225-242, Fox, J. A. (1986). Persistent flaws in econometric studies of the death penalty: A discussion of

Layton's findings, testimony to Subcommittee on Criminal Justice, Committee on the Judiciary, U.S. House of Representatives, May 7, 1986. Loyola Law Review (forthcoming).

Friedman, L. S. (1979). The use of multiple regression analysis to test for a deterrent effect of capital punishment: Prospects and problems. In E. Bittner & S. Messinger (Eds.), Criminology review yearbook. Beverly Hills, CA: Sage Publications.

Hanushek, E. A., & Jackson, J. E. (1977). Statistical methods for social scientists. New York, NY: Academic Press.

Hook, S. (1961). The death sentence. The New Leader, 44, 18-20. Kleck, G. (1979). Capital punishment, gun ownership and homicide. American Journal of Sociology,

84, 882-910. Kleck, G. (1981). Racial discrimination in criminal sentencing: A critical evaluation of the evidence with

additional evidence on the death penalty. American Sociological Review, 46(6), 783-805. Klein, L., Forst, B., & Filatov, V. (1978). The deterrent effect of capital punishment: An assessment of

the estimates. In A. Blumstein, J. Cohen, & D. Nagin, Deterrence and incapacitation: Estimating the effects of criminal sanctions on crime rates. Washington, DC: National Academy of Sciences.

Layson, S. (1985). Homicide and deterrence: A reexamination of the United States time series evidence. Southern Economic Journal, 52, 68-89.

Lempert, R. (1983). The effect of executions on homicides: A new look in an old light. Crime and Delinquency, 29, 88-115.

McGahey, R. (1980). Dr. Ehrlich's magic bullet: Econometric theory, econometrics, and the death penalty. Crime and Delinquency, 26, 485-502.

Messner, S. F. (1983). Regional and racial effects on the urban homicide rate: The subculture of violence revisited. American Sociological Review, 88, 997-l(X)7.

Passell, P. (1975). The deterrent effect of the death penalty: A statistical test. Stanford Law Review, 28, 61-80.

Paternoster, R. (1983). Race of victim and location of crime: The decision to seek the death penalty in South Carolina. Journal of Criminal Law and Criminology, 3, 278-363.

Schuessler, A. (1952). The deterrent effect of the death penalty. Annals, 54, 284-297. Schwarzschild, H. (1982). In opposition to death penalty legislation. In H. Bedau (Ed.), The death

penalty in America. New York, NY: Oxford University Press. Sellin, T. (1958). The death penalty. In Model Penal Code: Reprint, tentative drafts nos. 8, 9, and 10

(tentative draft no. 8). Philadelphia, PA: The American Law Institute. Sellin, T. (1959). The death penalty. Philadelphia, PA: The American Law Institute. Sutherland, E. (1925). Murder and the death penalty. Journal of Criminal Law and Criminology, 15,

522-536. Van den Haag, E. (1978). In defense of the death penalty: A legal-practical moral analysis. Criminal

Law Bulletin, 14(1), 51-68. Van den Haag, E. (1982). In defense of the death penalty: A practical and moral analysis. In H. Bedau

(Ed.)i The death penalty in America. New York, NY: Oxford University Press. Yunker, J. A. (1976). Is the death penalty a deterrent to homicide: Some time series evidence. Journal

of Behavioral Economics, 5, 45-81.