Law and Human Behavior, Vol. 1, No. 3, 1977

Imprisonment v. the Death Penalty as a Deterrent to Murder William C. Bailey*

In this investigation we examine the simultaneous deterrent effect of imprisonment and executions on homicide. Examination of the census years 1920 to 1960 shows the certainty of execution and homicide rates to be generally unrelated. Also contrary to the deterrence hypothesis, the significant negative bivariate relationship between the severity of prison sentence and homicide rates found here and in earlier investigations is shown to be a statistical artifact resulting from a failure to control for the effects of alternative legal sanctions and sociodemographie factors. While neither imprisonment nor executions are found to have a significant deterrent effect on homicide, attention is called to some limitations of this investigation and the need for additional research.

Although most criminologists in this country have traditionally dismissed the question of the deterrent effect of the death penalty as a dead issue m the death penalty does not deter ~ recent years have shown a renewed interest in further exploring this issue? This renewed interest would appear to have resulted from at least three factors. First, while American criminologists have generally taken it as axiomatic that punishment does not deter, a series of investigations over the last six to eight years have forced serious students of crime to conclude that the punishment issue is far from settled? This body of research, although hardly convincing most criminologists of the merit of the deterrence doctrine, has contributed to an atmosphere where the deterrent *Department of Sociology, The Cleveland State University. 1As an illustration of this anti-punishment attitude, witness the following statement by Barnes and Teeters (1951,355). "Not a single assumption underlying the theory of capital punishment can be squared with the facts about human nature and social conduct that have been established through the progress of science and sociological thought in the last century and a half. In fact the whole concept of capital punishment is scientifically and historically on a par with astrological medicine, the belief in witchcraft or the rejection of biological evolution." Clearly, deterrence explanations of punishment are but "rationalizations of revenge." 2As Jeffery (1965, 299) has pointed out, traditionally, "one of the basic principles learned by every student of criminology is that punishment does not deter." 239 This journal is copyrighted by Plenum. Each article is available for $7.50 from Plenum Publishing Corporation, 227 West 17th Street, New York, N.Y. 10011.

240

BAILEY

effect of punishment (including capital punishment) is once again a legitimate area of inquiry. Along with this development, a second and parallel factor that would appear to have inspired renewed interest in capital punishment research has been the growing recognition, beginning in the mid- to late 1960s, of the serious theoretical and methodological weakness of earlier studies of the death penalty. Although Sutherland (1925), Schussler (1952), and Sellin (1954), in their now classic studies, did note some of the shortcomings of their research, due attention was not given to the seriousness of these limitations, nor were the limitations they noted exhaustive, as pointed out by Gibbs (1968), Bedau (1967, 1971), van den Haag (1969), Bailey (1974), and others, a As minor as these difficulties might appear to those of a certain persuasion about capital punishment, for those with even minimal methodological sophistication and an understanding of the deterrence thesis, no other conclusion could be reached than that the death penalty question has only been traditionally examined in the most narrow theoretical and methodological context (Jeffery, 1965; van den Haag, 1969). A third important factor prompting further exploration of the death penalty question was quite likely the United States Supreme Court's ruling of 1972 in the case of Furman v. Georgia. While the Furman decision did not focus directly upon the question of deterrence, the high court did express regret over the paucity of empirical evidence of a recent vintage on this important question. Unfortunately, the high court seemed unconcerned with the death penalty research that had been conducted since 1972 in arriving at its most recent ruling of July 2, 1976, on the constitutionality of capital punishment. 4 The earlier Furman decision did, however, prompt abolitionists, who were gearing-up for future litigation, and research criminologists to recognize the need for additional research in this area. 5 In sum, the above three factors would appear to be primarily responsible for a resurgence of research interest in the death penalty, and a desire to build upon the shortcomings of previous investigations so as to better understand this important 3Deterrence theory suggests that if punishment is to act as an effective deterrent, it must be: (1) severe enough to outweigh the potential pleasures that crime might bring, (2) administered with certainty, (3) administered promptly, and (4) be made public. Despite the presumed importance of each of these factors, traditionally death penalty investigations only focused upon the severity of punishment, with the certainty of the death penalty only receiving systematic attention in recent years (Bailey, 1974, 1975, 1976; Ehrlich, 1975; Passell & Taylor, 1975; Bowers & Pierce, 1975). To date, no investigations have systematically examined the deterrent effect of the celerity of the death penalty, with examinations of the deterrent effect of the publicity surrounding executions being highly impressionistic, i.e., in none of these studies has there been an attempt to directly examine the amount and intensity of publicity surrounding executions (Savitz, 1958; Dann, 1935; Graves, 1956). 'In the U.S. Supreme Court's most recent rulings on the death penalty (Gregg v, Georgia, Proffitt v. Florida, Jurek v. Texas, Woodson v. North Carolina, Roberts v. Louisiana), the focus of each decision was not upon the deterrent effect of the death penalty, but rather upon the constitutionality of sentencing procedures in death penalty cases. In addition, the high court ruled that use of the death penalty as a punishment for murder does not, per se, constitute cruel and unusual punishment in violation of the Eighth and Fourteenth Amendments to the Constitution. 5For a discussion of the impact of the Furman decision on a renewed research interest in the death penalty, the reader is referred to Bedau and Pierce's (eds.) Capital Punishment in the United States (1976), which provides the most comprehensive source available on recent research on the varied aspects of the death penalty question.

IMPRISONMENT V. DEATH PENALTY

241

issue. In this light it is particularly ironic that post-Furman investigations have failed to get at the heart of the death penalty question; namely, whether there is a marginal deterrent effect achieved by executions over that accomplished by imprisoning convicted murderers? To illustrate this point we might briefly review the methodologies of a few recent deterrence investigations of homicide.

R E V I E W OF T H E L I T E R A T U R E The last few years have seen two rather distinct lines of research on homicide and deterrence. On the one hand, there have been a series of studies by sociologists and economists examining the deterrent effect of the certainty and severity of imprisonment on homicide rates. With slight variation, these investigations have consistently shown the certainty of imprisonment and state's homicide rates to be generally unrelated. Consistent with the deterrence argument, however, a substantial inverse relationship has been found between the severity of punishment (average length of prison sentence served by convicted murderers) and homicide rateso 7 Extreme caution must be exercised in interpreting these findings, however. While these findings are at least in part consistent with what deterrence theory would predict, in none of these investigations has there been an attempt to simultaneously examine (or control for) the deterrent effect of executions along with imprisonment in jurisdictions that provide for both types of sanctions. Rather, the statutory provision for the death penalty, and its actual level of use, have been completely ignored in these investigations. In other words, in this line of research the death penalty variable has been treated (although not being considered) as a constant, not as a variable in its statutory provision and level of use from state to state. This is clearly an ironic oversight in light of the deterrence argument for the death penalty? Due to this very unfortunate oversight, as it stands, we can only speculate to what extent (if any) the results of these deterrence studies of imprisonment were spuriously produced and are simply a statistical artifact resulting from alternative legal sanctions (executions) being artificially excluded from the analysis. In a second line of deterrence research, the focus has been upon just the opposite type of sanction. While examining the relationship between the certainty of execution and state's homicide rates, these studies are no less limited than those described 6As Bedau (1967, 1970) points out, the important question is not whether the death penalty deters, but rather, whether it provides a superior deterrent effect to that of alternative legal sanctions. Unfortunately, the questions of the deterrent effect of the death penalty per se, and the possible added deterrent effect achieved by the death penalty, have often been confused in the debate over capital punishment, 7For example, Bailey, Martin, and Gray (1974) report the following results from intercorrelating average length of prison sentences served by convicted murderers (severity of imprisonment) and state's homicide rates: 1950, r = -.396; 1960, r = -.332; 1964, r = -.165. Intercorrelating the certainty of imprisonment (the ratio of the total number of persons imprisoned for murder to the total number of reported murders) with homicide rates resulted in the following coefficients: 1950, r = . 120; 1960, r -- -.366; 1964, r = -.288, Similar results are also reported by Logan (1972) and Gray and Martin (1969). 8This "oversight" quite likely reflects the attitude of most criminologists that the death penalty does not deter (it's a dead issue), so why bother even considering it in examining the deterrent effect of imprisonmont.

242

BAILEY

above, for they too have rather consistently ignored the existence of alternative legal sanctions (imprisonment) for murder. In other words, by not also simultaneously examining variation in the use of alternative legal sanctions from state to state, jurisdictions are treated as if no such variation exists. Put in more concrete terms, by ignoring imprisonment as an alternative sanction to the death penalty, states are mistakenly treated as if they do not vary in the length of prison sentences handed down and actually served by convicted murderers. This oversight may well be of major importance. First, just as execution rates for homicide vary from state to state and year to year, so do imprisonment practices vary from jurisdiction to jurisdiction and year to year. Second, to ignore the imprisonment variable (variation in length of prison sentences, for example) is to overlook the fact that even when executions were at their peak during the 1930s, only a "small" proportion of homicides resulted in the death penalty. Third, to ignore imprisonment in this line of research is to overlook those investigations suggesting the severity of imprisonment to be a deterrent to murder. Recognizing this important limitation, Forst (1977) examined in a very recent investigation the relative deterrent effects of certainty and severity of imprisonment and execution rates on homicide. Each punishment variable, along with a number of sociodemographic variables, were examined cross-sectionally in a series of regression analyses, with the focus being upon changes in the levels (rates) of these variables from 1960 to 1970.9 (His analysis is in contrast to ours in that here we examine the same type of question cross-sectionally for states for individual years.) Utilizing a number of regression models and tests for the robustness of his parameter estimates, Forst consistently finds a nonsignificant relationship between changes (1960-1970) in the levels of execution and the severity of imprisonment and homicide rates. Consistent with the deterrence hypothesis, however, changes in certainty of imprisonment were found to be negatively, and significantly, associated with homicide rates. This finding leads him to conclude that not executions, but rather the reduced certainty of imprisonment for homicide from 1960 (41.3%) to 1970 (34.6%) is likely responsible for the increase in homicides (+ 53%) for this period. While Forst addresses an important, and largely neglected, question in the deterrence literature, his analysis is not without some important limitations. Briefly, first, due to problems of missing imprisonment data for 1970, his analysis is confined to only 32 states which may or may not be representative of jurisdictions not examined. Second, because Forst chose to examine the effect of executions by comparing levels for this variable for 1960 and 1970, the range of his execution index is ~Forst (1977) operationalizes his sanction variables as follows: (a) Execution rate = ratio of the number of executions to the number of homicide prisoners received from the court, (b) certainty of imprisonments = ratio of the number of homicide prisoners received from the court to the number of murders and nonnegligent manslaughters, (c) severity of imprisonment = mean time served (in months) of released homicide prisoners. The sociodemographic variables he considers include: (a) proportion of residents of the ages 21-24 years, (b) proportion nonwhite population, (c) proportion male population, (d) proportion urban residents; (e) employment rate, (f) school enrollment rate of persons 5-17 years, (g) population, (h) divorce rate, (i) median family income, (j) proportion of families in poverty, (k) a binary "Southern" variable with southern states weighted 1.0 and other states zero. The actual values utilized in the analysis consisted of the differences between 1960 and 1970 for the sanction and sociodemographic factors.

IMPRISONMENT V. DEATH PENALTY

243

seriously restricted. To illustrate, for 1970 there were n o executions for murder in the country, with a total of "only" 33 executions for 1960 for the states examined (Arizona = 1, California = 7, Colorado = 1, Connecticut = 1, Georgia = 6, Nevada = 1, New Mexico = 1, New York = 6, Ohio = 2, Oklahoma = 5, Utah = 1, Washington = 1). As a result, for nearly two thirds of the states Forst considers, there was no change in execution rates between 1960 and 1970 (0 - 0 = 0), and further, for the remaining states the changes are all in the same negative direction. In addition, even for executing jurisdictions in 1960, a comparison of the ratio of homicides to executions shows execution rates to be low and relatively uniform for most states. Third, because Forst only examined one time period (1960-1970), it remains unclear whether his findings hold equally well for time periods and years where executions (and the certainty and severity of imprisonment) were at higher levels. Due to these difficulties, and the sparse attention the relative deterrent effect of imprisonment versus execution has received, further research in this important area is clearly warranted.

THE PRESENT INVESTIGATION In an attempt to avoid some of the difficulties discussed above, we examine here the simultaneous deterrent effect of imprisonment and executions on homicide rates, controlling for a number of socioeconomic and demographic factors associated with horfiicide. This analysis will allow a clearer understanding of the deterrent effect of each type of sanction than revealed by past investigations and permit us to compare the relative deterrent effectiveness of imprisonment and executions. In line with the deterrence argument, we would hypothesize a substantial inverse relationship between both the severity of prison sentences and the certainty of execution and state's homicide rates. In addition, we would further hypothesize executions to be a more effective deterrent to murder than severe prison sentences. We can only guess, however, as to the added deterrent effect to be expected from the death penalty for this question is nowhere systematically addressed in the deterrence literature. At best, proponents of the death penalty are no more specific on this question than to argue that executions are a "much " more effective deterrent than imprisonment.

M E T H O D S AND P R O C E D U R E S To examine the above hypotheses, we have chosen to construct our measures of imprisonment, execution, and homicide utilizing available figures issued by the Federal Bureaus of Prisons and Investigation. Although the validity of official prisoner and crime statistics has come under some question in recent years, these data were utilized for two reasons. First, in the absence of alternative prisoner and crime figures of a national scope, we had no other choice but to make use of these data. Second, by drawing upon similar data as previous investigators, and constructing our punishment and offense rate variables in a similar manner, we will be better able to compare our findings with those of previous investigations.

244

BAILEY

First-Degree Murder: The Dependent Variable Like most previous investigators, in the absence of theoretically appropriate data for first-degree murder (the only form of homicide generally punishable by death), we have chosen to utilize F.B.I. figures for murder and nonnegligent manslaughter as an indicator of capital homicide rates.l~ Although this homicide offense category is much more inclusive than first-degree murder, we have been forced to utilize these data, for nowhere have there been compiled on a nationwide basis figures for capital homicides. As a result, it must be assumed that the ratio of capital to total criminal homicides is a constant from state to state so that the more general F.B.I. figures provide a reasonably good indicator of offenses punishable by death. 11 While some have objected to the practice of utilizing F.B.I. figures for homicide (and U.S. Public Health statistics for homicide, as well) in death penalty investigations, the error resulting from this practice remains unknown, for no one has succeeded in accurately counting the number of capital offenses hidden in these figures in order to test this question (Sellin, 1959; Bedau, 1967; van den Haag, 1969; Wilson, 1973). TM The bias introduced by using these F.B.I. figures for homicide may not be as serious as some claim, however; a recent study by Bailey (1974, 1975) has shown very similar results when both F.B.I. data and prison admission figures for first-degree murder were utilized as indicators of capital homicide rates, is

Execution Rate: An Independent Variable Also similar to previous investigations, certainty of the death penalty - execution rate - - is operationally defined here as the total number of executions for homicide divided by the total number of homicides-reported by the police. 14 This procedure resulted in an execution rate value for each state, for each year, that could theoretically range from zero (0.0) to unity (1.00). A value of 0.0 would indicate that there were no executions for homicide for the state and year in question, while a value of 1.00 would indicate an equal number of homicides and executions. It should be emphasized that this execution measure does not indicate the actual probability that any particular homicide will result in execution. Rather, this measure simply consists of the ratio of the total number of executions performed during a year 1~ Federal Bureau of Investigation includes in its offense category of murder and nonnegligent manslaughter all willful felonious homicides, as distinguished from deaths caused by negligence. "For a more complete discussion of the rationale for, and use of, Federal Bureau of Investigation figures for homicide as an indicator of first-degree murder (the only form of homicide punishable by death) see: Sutherland (1925), Sellin (1959, 1967), Bedau (1967, 1970), Schuessler (1952), Bailey (1974, 1975). 12Included in the U.S. Public Health Service homicide category are all deaths resulting from an injury purposely inflicted by another person. Intent to kill is not required to classify a death as a homicide. ~3Examining the five-year periods preceding 1967 and 1968, Bailey (1974, 1975) reports the following results when intercorrelating state's execution rates for murder (the ratio of total homicides to the total number of executions for each five-year period) and rates of murder and nonnegligent manslaughter: 1967, r = - . 166; 1968, r = -.039. Substitution of prison admissions for first-degree murder as an indicator of capital homicides produced the following results: 1967, r = -.137; 1968, r = -.194. ~4This measure of execution is similar to that used by Schuessler (1952), Bailey (1974, 1975, 1976), and Orsagh and Black (1976).

IMPRISONMENT V. DEATH PENALTY

245

to the total number of homicides reported during that year. In addition, it should also be noted that while some death penalty investigators have chosen to build in a time lag of a year or more between homicides and executions in constructing a certainty index (due to the typical delay of at least one year between the offense and the possibility of execution, if it does result), we view it as more consistent with the general deterrence argument to examine the relationship between these two variables within a more immediate time interval. 15 That is, in line with the general deterrence thesis (in contrast with the primary deterrence argument) it could well be argued that the general public, including potential killers, is more affected (deterred) by its impression of the current level of homicides and the current level of executions, than by (I) the actual proportion of homicides committed during a year (t) that may result in execution during the next year (t + 1), or in subsequent years (t + 2, t + 3 . . . . . t + x), and/or (2) the actual proportion of homicides committed the previous year (t - 1), or during early years (t - 2, t - 3. . . . . t - x), that resulted in execution this year (0.18 In addition, it is also our impression that for purposes of examining general deterrence, it is preferable to utilize an execution rate measure like ours over one that solely considers the conditional probability of convicted and imprisoned homicide offenders being executed. While measuring execution rates in terms of the proportion of convicted and imprisoned murderers who are subsequently executed may have some merit in examining questions of primary deterrence and recidivism, this matter is not at the heart of the larger general deterrence debate over the death penalty. In all fairness it must also be pointed out that our measure of execution is only one of many possible operationalizations, and only through experimenting with different measures will it be possible to determine which measure is preferable. Unfortunately, the theoretical literature on deterrence provides us with very little help in dealing with this question.

Severity of Prison Sentence: An Independent Variable The severity of prison sentence is operationally defined here as the median length of prison sentence (in months) served by convicted homicide offenders. As with the execution figures, these data were secured from publications issued by the Federal Bureau of Prisons. Although most previous investigators of homicide and imprisonment have used these, or similar data, this measure of severity is obviously not without its faults. First,

~For example, Ehrlich (1975), Bowers and Pierce (I 975), Passell and Taylor (1975), and Forst, Filatov, and Klein (1976) built in a one-year time lag between homicides and executions in constructing their measures of the certainty of the death penalty in order to take into consideration the typical delay of at least one year between the offense and execution, if it does result. With the exception of Ehrlich's (1975) study, utilizing this measure of execution has produced almost exactly the same results as those found in studies where executions and homicides have been examined within the same year. ~6In the deterrence literature a distinction is commonly drawn between general deterrence and primary deterrence. Whereas the focus of primary deterrence is upon the future criminal behavior of those who have been subject to legal sanctions, the concern of general deterrence is with the effect that the threat and application of sanctions has upon would-be offenders. For a detailed discussion of primary and general deterrence see: Andenaes (1966, 1974, 1975) and Zimring and Hawkins (1973).

246

BAILEY

in the absence of more detailed information about imprisonment, it remains unclear how well these median figures reflect state's imprisonment practices for homicide. While the median is certainly less subject than the arithmetic mean to distortion due to the effect of extreme cases, it is simply not possible to presently gauge how adequate these median figures are as an indicator of the severity of state's incarceration practices. Second, and possibly of even greater importance, in considering the deterrence hypothesis we have been forced, due to a lack of alternative data, to examine the relationship between (1) state's homicide rates for year (t), and (2) the median length of prison sentences served by convicted homicide offenders who were released during the year (t). Although it would clearly seem more in line with the general deterrence argument to examine the relationship between current levels of sanctions and current levels of crime, it was not possible to conduct such an analysis for there are no nationwide court or prisoner statistics available on the length of prison sentences received by convicted murderers. As a result, in utilizing the prisoner statistics that are available, it must be assumed that variation from state to state in the length of prison sentences served by homicide offenders released during any given year provides a reasonably good indicator of the variation from state to state in the severity of sentences received by homicide offenders convicted during that year. Although adequate data are simply not available to systematically test this assumption, the high degree of consistency in variation in the length of prison sentences served by released homicide offenders for the years where figures are available (1951, 1960, and 1964) would seem to at least lend some support to this assumption. That is, since variation in the length of prison sentence served by convicted murderers is at least in part (a large part) a function of the length of prison sentences received upon conviction, using the available data on the actual length of prison sentence served may not introduce serious bias into the analysis. With figures only available for the above three years, however, a more conclusive answer to this question is not possible at this time.

Socioeconomic and Demographic Control Variables As noted above, in order to control for factors previously associated with variation in homicide rates, and to determine the independent deterrent effect of the severity of imprisonment and executions on homicide, eight socioeconomic and demographic variables were introduced into the analysis as control variables. These variables include: (1) population, (2) population density, (3) percent nonwhite population, (4) percent male population, (5) percent population 20-40 years of age, (6) percent urban population, (7) median family income, and (8) median education. These data were gathered from figures compiled by the Bureau of the Census, U.S. Department of Commerce. 17

17Socioeconomic and demographic data were secured from the following sources: (1) Census of the Population: 1950. Characteristics of the Population... United States Summary (Tables 6, 9, 59, 60, 62, 72, 84) and (2) Census of the Population: 1960. Characteristics of the Population... United States Summary (Tables 12, 59, 107, 115, 119, 138).

IMPRISONMENT V. DEATH PENALTY

247

Certainty of Imprisonment: A Control Variable While previous deterrence investigations of homicide and imprisonment have treated the certainty of imprisonment (the ratio of the total number of reported homicides to the number of persons imprisoned for homicide) as an independent variable, we shall consider this factor as both an independent variable and a control variable in examining the deterrent effect of the severity of imprisonment and execution rates on homicide. By introducing the certainty factor as a control variable, it will be possible to determine the relative deterrent effectiveness of imprisonment and executions holding constant (controlling for) the certainty of imprisonment. As with the severity of imprisonment data, certainty figures were also secured from Federal Bureau of Prisons statistics.

Missing and Incomplete Data In considering a question as serious as the deterrent effect of imprisonment and the death penalty, one would, of course, want to conduct as thorough an analysis as possible and be perfectly cognizant of the investigation's limitations. The theoretical limitations of our execution, imprisonment, and homicide indexes have been briefly discussed above. In addition, the data used to construct these measures were also found to be incomplete. As in all secondary analyses, the investigator must work within the confines of the data that are available. While it was initially our hope to examine the deterrence question for as many retentionist states and years as possible, we quickly found that reasonably complete data for each of our measures were only available for 1950 and 1960.18 Although generally complete homicide and execution figures are available for most states and years from about the 1920s, the required prisoner data are only available for 1951, 1960, and 1964. l~ In addition, the required figures for the socioeconomic and demographic control variables are only available for the census years. As a result, we have been forced to confine our analysis to 1950 and !960, for these are the only years where figures are available for the independent, dependent, and control variables (severity of imprisonment figures for 1951 are used as an estimate for 1950).20 It will be possible, however, to at least speculate about the

tsWe, of course, confine our analysis to retentionist states for it is not possible to compare the relative deterrent effects of imprisonment versus executions in abolitionist jurisdictions. 19Homicide figures were secured from the Federal Bureau of Investigation Uniform Crime Reports for 1950 and 1960. Execution data were obtained from the Teeters-Zibulka Inventory of executions under state authority, which was only recently published in its entirety by Bowers (1974b). Prison admission and release figures were secured from the Federal Bureau of Prisons' National Prisoners Statistics series for 1950, 1951, and 1960. ~~ and certainty of imprisonment figures wel"e,found to be complete for each retentionist state for 1950 and 1960. Severity of imprisonment data were not available for some states for each year, however. For 1950, Vermont and Idaho had to be dropped from the analysis because there were no convicted murderers released from prison during this year. Similarly, for 19613length of imprisonment figures were not available for New Hampshire, Vermont, Idaho, Wyoming, and Utah due to the small (N < 10) number of murderers released during the year. As a result, these states had to be dropped from the 1960 analysis. Unfortunate/y, we were unable to secure release data for neighboring years in order to estimate severity figures for the missing eases.

248

BAILEY

deterrence hypothesis for other years, by comparing, for example, (1) the bivariate relationship between executions and homicides for 1950 and 1960 with this relationship for other years where data are available for these two variables, and (2) the multivariate results for 1950 and 1960 with the results for earlier census years where only the imprisonment data are unavailable. Obviously, such comparisons will not allow us to directly address the question of the deterrent effectiveness of imprisonment v. execution for years other than 1950 and 1960. If the general pattern of the relationships is found to be similar for 1950, 1960, and other years, however, we may be able to at least speculate about this matter.

Method of Analysis To examine the above hypotheses, the socioeconomic, demographic, imprisonment, execution, and homicide variables were fitted into a stepwise multipleregression analysis. While in the typical stepwise analysis the independent variable most strongly associated with the dependent variable enters first into the regression, then followed by the next best predictor of the dependent variable (and so on until all of the independent variables have been considered), we have departed from this procedure in one respect. To determine the independent effect of executions and severity of prison sentences on homicide, and to compare their relative deterrent effectiveness, the socioeconomic and demographic variables were allowed to enter into the regression first. Next, the certainty of imprisonment was entered into the analysis, and finally, the execution and severity of imprisonment variables were allowed to enter the analysis in their order of importance. By structuring the analysis in this manner we will first be able to examine the effect of the socioeconomic and demographic factors, and the certainty of imprisonment, on homicide rates, and the variation in homicide yet to be accounted for by other factors. The introduction of these variables at this point will also have the effect of holding constant (controlling for) the effects of these factors in considering subsequent variables. Second, by allowing the severity of imprisonment and execution rate variables to enter next, it will be possible to determine (1) which of these two variables is a better deterrent to homicide, (2) the independent effect of each factor on homicide rates, and (3) the combined deterent effect of both types of sanctions on homicide.

FINDINGS Table 1 reports zero-order (r) and partial (B and beta) coefficients between each of the independent and control variables and homicide rates, as well as the combined multiple effect of these factors on homicide. Examination of the simple r correlations reveals a picture very similar to that reported in a number of earlier investigations. Consistent with the deterrence hypothesis, we find a statistically significant inverse zero-order relationship between the severity of imprisonment and homicide rates for both 1950 (-.486b) and 1960 (-.457b), but a less substantial, nonsignificant negative association between the certainty of imprisonment and homicides for both years (1950 = -.051, 1960 = -.240). Also similar to previous findings, neither for 1950 (.129) nor for 1960 (-.041) are the r coefficients for execution rates negative and as

I M P R I S O N M E N T V. D E A T H P E N A L T Y

249

substantial as the deterrence thesis would lead us to expect. In fact, for the former year, the results show a slight tendency for states that have made the greatest use of the death penalty to have higher homicide rates. Inspection of the partial coefficients (B and beta) reveals a slightly different picture. With the introduction of the control variables into the analysis, the relationship between the severity of imprisonment and homicide remains in the predicted negative direction, but the coefficients are very slight, and not statistically significant for either 1950 (-.010) or 1960 (-.008). These findings indicate that the significant zero-order correlation between these two variables is a statistical artifact resulting from not controlling for the socioeconomic, demographic, and certainty of imprisonment factors. After controlling for these variables, the partial coefficients (B values) show that a one-unit (month) increase in the length of imprisonment is only associated with about a one-hundredth of a person reduction in the number of homicides per 100,000 population for either year. Put differently, the partial regression coefficients for severity suggest that a 100-month increase in the median length of imprisonment of convicted murderers would have the effect of reducing the homicide rate by about one (1.0) person per 100,000 population. Certainty and Severity of Prison Sentence, Execution Rates, Selected Sociodemographic Factors and State Homicide Rates, 1950 and 1960

T a b l e 1. R e l a t i o n s h i p b e t w e e n

Independent variable

Simple r

Multiple R

Multiple R2

Change in R ~

B coeff,

Beta coeff.

1950 Pct. nonwhite Mdn. income Pct. 20-40 yrs. Pop. density Pct. urban Pct. male Mdn. education Population Cert. of prison Sev. of Prison Execution rt.

.795c" -.66% -. 139 -.228 -.367a -. 183 -.57 lc .000 -.051 -.486b .129

.795c .814c .829c .822c .828c .833c .838c .839c .854c .858c .861 c

.631 .662 .671 .676 .685 .694 .702 .703 .729 ,736 .741

.631 ,030 ,009 ,004 ,010 .009 .007 .002 .026 .007 .005

23,565c -.004a - 1.209 -.0003 9.970 68.008 -.525 .0000 .851a -.010 .250

.493 -.505 -.011 -.094 .354 .164 -. 133 .001 .210 -,149 .087

1960 Pct. nonwhite Pop. density Pct. 20-40 yrs. Pct. urban Mdn. education Mdn. income Population Pct. male Cert. of prison Sev. of prison Execution rt.

.850c -.288a .214 -.279a -.633c -.614c -.059 -. 167 -.240 -.475b -.04i

.850c .871c .881 c .890c .901c .905c .905c .906c .906c .914c .914c

.723 .758 .776 .793 .812 .819 .819 .819 ~821 .835 .836

.723 .034 .018 ,017 .019 .007 .000 .000 .003 ,014 .001

20.124c -.0006b t 2.295 8.824a -.538 -.0009 -.0000 1.066 -.024 -.008 .603

.593 -.226 9132 .353 -.174 -9 -.034 .024 -. 107 -. 158 ,026

"a = P < .05; b = P < .01;c = P < .001,

250

BAILEY

Examination of the partial coefficients for certainty of imprisonment reveals a mixed pattern. For 1960, the partial B coefficient (-.024) is in the same negative direction as the respective zero-order r correlation and suggests that a 1% increase in the certainty of imprisonment for homicide would have the effect of reducing the homicide rate by roughly 2.4 hundredths of a person per 100,000 population. For 1950, the partial coefficients is positive (.851a), and statistically significant, and indicates a 1% increase in the probability of imprisonment to be associated with roughly an 8.5 tenths of a person increase in the number of homicides per 100,000 population. This finding for 1960 and the very slight inverse relationship found for 1950 are clearly contrary to what deterrence theory would lead us to expect. As with the zero-order analysis, the partial regression coefficients for the execution rate variable are also contrary to our hypotheses. For both 1950 (.250) and 1960 (.603) the coefficients are of a positive sign, thus indicating increases in the execution rate to be associated with increases in the homicide rate. In sum, while generally consistent with the results of earlier investigations, the above findings do not lend support to claims of the deterrent effectiveness of severe prison sentences and executions on homicide. Nor do we find any support for claims of the greater deterrent effectiveness of the death penalty over imprisonment. In fact, to the contrary, while severity of imprisonment is very slightly, negatively, associated with homicide rates, execution rates are positively associated with homicides.

A Time-Lag Model In considering the relationship between execution rates and homicide rates in the above analysis, we chose to examine the association between these two variables within the same year( rather than building a time-lag between the independent and dependent variables. Accordingly, it might be questioned whether our negative findings for the death penalty are a result of not permitting a sufficient period of time to elapse between executions and rate in order to allow the deterrent effect of the death penalty to be fully realized. To test this question, and to assess both the immediate as well as possible delayed deterrent effects of executions, we examined the relationship between this variable for year t (1950 and 1960) and mean homicide rates of year t and t + 1 (1950 and 1951, and 1960 and 1961, respectively), again controlling for the socioeconomic, demographic, and certainty of imprisonment variables. (This analysis is simply a replication of that reported in Table 1, but with the substitution of a different measure of homicide.) Results of this analysis are reported in Table 2. The findings presented in Table 2 almost exactly parallel the results reported above. Again, we find statistically significant and negative zero-order correlations between severity of imprisonment and homicide for 1950 (-.489b) and 1960 (-.465b), and less substantial, nonsignificant zero-order correlations for certainty of imprisonment for these two years (1950 = -.050, 1960 = -.224). Similarly, we also find comparable zero-order correlations for executions (1950 = .097, 1960 = -.043). Examination of the partial regression (B) analysis also reveals the same general pattern when the control variables are introduced into the analysis. Again, we find a significant positive association between the certainty of imprisonment and homicide for 1950 (1.186a), and a very slight negative association for 1960 (-.032). Also, like the results of the earlier analysis, the partial coefficients for severity of imprisonment are

IMPRISONMENT V. DEATH PENALTY

251

again negative, but very slight (1950 = -.014, 1960 = -.012) each year. Likewise, the partial coefficients for the execution rate variable are again positive for 1950 (.246) and 1960 (.964). In sum, this analysis also leads us to the same general conclusion. Neither the severity of imprisonment, nor the certainty of the death penalty, would appear to be an effective deterrent to homicide.

A Second Time-Lag Model As Bowers (1974a) has pointed out in a recent discussion of deterrence research, examinations of the relationship between the certainty of sanctions and offense rates have been flawed by not considering how levels of crime during a year (t - 1), may affect legal responses to crime the next year (t), and accordingly, how this factor may distort (bias) examinations of the relationship between the certainty of punishment and offense rates for year (t). To illustrate Bowers' argument, community A, with a low level of homicides (t - 1), may respond to this situation with a low level of executions (t). If the level of Table 2. Relationship between Certainty and Severity of Prison Sentence, Execution Rates, Selected Sociodemographic Factors and Mean State Homicide Rates, 1950 and 1960 ~ Simple Independent variable

r

Multiple R

Multiple R~

Change

Beta

in R ~

B coeff~

coeff.

1950 Pct. nonwhite Mdn. income Pct. 20-40 yrs. Population Pct. male Pct. urban Mdn. education Pop. density Cert. of prison Sev. of Prison Execution rt.

.819cb -.690c -. 153 .012 -.219 -.374a -.607c -.216 -.050 -.489b .097

.819c .839c .846c .849c .853c .858c .859c .861c .876c .880c ,881c

.671 .704 .715 .720 .727 .736 .738 .742 .767 .774 .776

.67i .033 .011 .005 .007 .009 .002 .004 .025 .007 .002

35.443c -.006a -3.200 .0000" 90.411 13.613 -.814 -.0004 1.186a -.014 .246

.518 -.490 -.020 .011 .152 ,338 -.144 -.089 .205 -.139 .060

1960 Pct. nonwhite Pop. density Pet. 20-40 yrs. Pct. urban Mdn. education Mdn. income Pct. male Population Cert. of prison Sev. of prison Execution rt.

.856c -.296a .191 -.293a -.642c -.630c -. 173 -.053 -.224 -.465h -.043

.856c .877c .884c .891c .902c .907c .907c .907c .908c .916c .916c

.732 .769 .781 .795 .814 .822 .823 .823 .824 .838 .839

.732 .037 .012 .013 .019 ,009 .001 .000 .001 .014 .001

28.520c -.0008b 15.941 12.133a -.724 -.001 .675 -.0000 -.032 -.012 .964

.585 -.223 .110 .337 -.163 -.241 .011 .02t -.099 -,158 .029

aMean homicide rates are examined for the years 1950 and 1951, and 1960 and 1961, respectively. ba = P < .05;b = P < .01;c = P < .001.

252

BAILEY

homicides does not change in community A from year t - 1 to year t, the relationship between execution rate and homicide will be positive. Conversely, community B, with a high level of homicides ( t - 1), may respond by increasing the level of executions (t). If the level of homicides in community B is not immediately reduced within the year t, then, the association between executions (t) and homicides (t) will be positive. While in both of these types of situations executions may be accomplishing a deterrent effect to some degree, responses to homicides like those noted here would confound the certainty-rate relationship by introducing "positive pressure" into a correlation analysis between executions and rates. To examine Bowers' argument, we introduced the previous ( t - 1) year's homicide rate into the analysis as a control variable by first regressing homicide rates (t - 1) against execution rates (t) and then regressing the residual in execution rates (t) against homicide rates (t), controlling for the socioeconomic, demographic, and imprisonment variables. Results of this analysis are reported in Table 3. As with the two previous regression analyses, figures reported in Table 3 provide no support for either of our two hypotheses. In fact, the results are almost identical with those shown in Tables I and 2. Again, the zero-order coefficients for the severity Table 3. Relationship between Certainty and Severity of Prison Sentence, Residual Variation in Execution Rates, Selected Sociodemographic Factors and State Homicide Rates, 1950 and 1960

Independent variable

Simple r

Multiple R

Multiple R~

Change in R 2

B coeff,

Beta coeff.

1950 Pct, nonwhite Mdn. income Pct. 20-40 yrs. Pop. density Pct. urban Pct. male Mdn. education Population Cert. of prison Sev. of Prison Execution rt.

.795a a -.669c -. 140 -.228 -.376a -. 183 -.57 lc -.0004 -.051 -.486b .077

.795c .814c .819c .822c .828c .833c .838c .839c .854c .858c ,860c

.631 .662 .671 .676 .685 .694 .702 .703 .729 .736 .739

,795 ,030 .009 .005 .010 .009 .007 .001 .026 .007 .003

24.137c -.004a - 1.262 -.0003 10.124 75.244 -.508 .0000 .850a -.010 .303

.505 -.513 -.011 -.090 .360 .181 -. 129 .007 .210 -.143 .069

1960 Pct. nonwhite Pop. density Pct. 20-40 yrs. Pct. urban Mdn. education Mdn. income Population Pct. male Cert. of prison Sev. of prison Execution rt.

.850c -.288a .214 -.279a -.633c -.614c -.059 -. 167 -,240 -.457b -,034

.850c .871c .881c .890c .901c .905c .905c .906c .906c .914c .914c

.723 .758 .776 .793 .812 .819 .819 .821 .821 .835 .836

.723 .035 .018 .017 .019 .007 .001 .000 .000 .014 .001

20.117c -.0006b 13,303 8.815a -.538 -.0009 -.0000 1.068 -.024 -.008 .098

.593 -.226 .132 .352 -.174 -,214 -,034 .025 -. 107 -. 158 .027

aa = P < .05;b = P < .01;c = P < .001.

IMPRISONMENT V. DEATH PENALTY

253

of imprisonment are statistically significant, and in the predicted direction, but this variable is reduced to nonsignificance when the control variables are introduced. Also contrary to the deterrence thesis, the partial coefficients for the execution rate variable are positive for both years, and not negative, as hypothesized. In sum, this analysis does not provide support for Bowers' argument, or for claims of the deterrent effectiveness of imprisonment and executions for homicide. Nor, we might add, does Table 4 show this situation to be altered when we substitute mean homicide rates for years (t) and (t + 1) as a measure of the dependent variable, as we did in the previous (Table 2) regression analysis.

S U M M A R Y AND CONCLUSION In this investigation we have attempted to avoid some of the methodological limitations of previous deterrence investigations of homicide by (1) examining the deterrent effects of both imprisonment and executions on homicide simultaneously, Table 4. Relationship between Certainty and Severity of Prison Sentence, Residual Variation in Execution Rates, Selected Sociodemographic Factors and Mean State Homicide Rates, 1959 and 1960 a

Independent variable

Simple r

Multiple R

Multiple R2

Change in R ~

B coeff,

Beta coeff.

1950 Pct. white Mdn. income Pct. 20-40 yrs. Population Pct. male Pct. urban Mdn. education Pop. density Cert. of prison Sev~ of Prison Execution rt.

.819cb -.690c -,153 ,012 -,219 -,374a -.607c -.216 -.050 -.489b .044

.819c .839c .846c .849c .823c .858c .859c .861c .876c .879c .880c

,671 .704 .715 .720 .727 .736 .738 .742 .767 .774 .775

.671 .033 .015 .005 .007 .009 .002 .004 .025 .007 .001

36,172c -.006a -3,264 .0000 100,084 13.798 -.790 -.0003 1.186a -.013 .268

.528 -.498 -.021 -.018 .168 .342 -. 140 -.085 .205 -.133 .043

1960 Pct. nonwhite Pop. density Pct. 20-40 yrs. Pct. urban Mdn. education Mdn. income Pct. male Population Cert. of prison Sev. of prison Execution rt.

.856c -.296a .191 -.293a -.642c -.630c -.173 -.053 -.224 -.465b -.038

.856c .877c .884c .891c ,902c .907c .907c ,907c ,908c .916c .916c

.732 .769 .781 .794 .814 .822 .823 .823 .824 .838 .839

.732 ,037 ,012 ,013 ,019 .009 .001 .000 .001 .014 .001

28.510c -~0008b 15.953 I2,120a -.724 -.001 .678 -.0000 -.032 -.012 .156

.584 -.223 .110 .337 -.163 -.242 .011 -.021 -.099 -.158 ,030

aMean homicide rates are examined for the years 1950 and 1951, and 1960 and 1961, respectively. ba = P < .05;b = P < .O1;c = P < .00l.

254

BAILEY

and not considering each type of sanction separately as have other researchers, and (2) introducing into the analysis a number of sociodemographic control factors in order to determine the independent (and combined) deterrent effect of imprisonment and executions on homicide rates. In addition, four separate regression analyses were performed utilizing two different measures of the certainty of the death penalty and homicide rates. Briefly, this analysis resulted in the following findings. First, the significant negative zero-order correlation between the severity of prison sentence and homicide rates found here and in earlier investigations would appear to be a statistical artifact, for when control variables were introduced into the analysis this relationship was reduced to nonsignificance. Second, and also contrary to the deterrence thesis, our analysis failed to provide any support for claims of the deterrent effect of executions. To the contrary, execution rates and homicide rates were found to be positively associated when control factors were introduced into the analysis. Third, no evidence whatsoever was found to support claims of the superior deterrent effect of executions over imprisonment. Rather, the analysis shows neither type of sanction to be significantly related to homicides. Fourth, and finally, to test for the possible nonlinear relationship between the certainty and severity of punishment and homicide rates, log transforms were performed on the homicide variable, and the same analysis as shown in Tables 1-4 was repeated. The results of this analysis, which are not reported in the findings section of this paper, almost exactly parallel the findings shown in Tables 1-4. Accordingly, we find no evidence of either a significant linear or nonlinear relationship between the certainty and severity of imprisonment and the certainty of the death penalty and homicide rates. 21 While these findings are clearly contrary to what deterrence theory would predict, the generalizability of these findings remains as an important question. As discussed earlier, our analysis has been confined to 1950 and 1960 for these are the only two years where the required execution, imprisonment, sociodemographic, and homicide data are all available. Accordingly, it remains unclear whether the pattern of relationships reported above also holds for other years. Unfortunately, a direct examination of this question must await the analysis of imprisonment data that are currently unavailable. (As noted earlier, relatively complete imprisonment figures are only available for 1950, 1960, and 1964, with the required sociodemographic data only being available for the census years.) It is possible, however, to examine the deterrent effect of imprisonment and executions separately

21While these findings for severity of imprisonment and certainty of execution are similar to those reported by Forst (1977), his analysis shows a significant inverse relationship between certainty of imprisonment and homicide rates. In addition to the methodological limitations of the Forst analysis noted earlier, this difference may at least in part be a result of: (a) the different time periods considered in the two analyses, and (b) the different measure of certainty of imprisonment (and for the severity, execution, and sociodemographic variables) used by Forst and in this investigation. As noted earlier, Forst considers the relationship among changes (1960-1970) in his variables, while we examine the deterrence question crosssectionally for states for two years, 1960 and 1950. To further explore this question, an investigation similar to Forst's is currently under way by this investigator to examine the correspondence between levels of change in the sanction, homicide, and sociodemographic variables from 1950 to 1960, and 1950 to 1970. Hopefully, this analysis will shed additional light on the factor(s) responsible for our different findings for the certainty of imprisonment.

IMPRISONMENT V. DEATH PENALTY

255

for the years where data are only available for one type of sanction. By comparing the results of this analysis with the findings for 1950 and 1960, we will be able to at least speculate about whether the 1950 and 1960 findings may be generalized to other years. To extend our analysis of the deterrent effect of the severity of imprisonment, prison admission and release figures and homicide rates were examined for 1964 and 1970. Due to problems of missing and incomplete prisoner data, this analysis was restricted to 32 death-penalty jurisdictions for 1964 and 22 states for 1970.23 In contrast to the findings for 1950 (r = -.486) and 1960 (r = -.457), examination of the bivariate relationship between sentence length and homicide rates reveals a positive association for both 1964 (r = .277) and 1970 (r = .430). However, when certainty of imprisonment was introduced into this analysis as a control variable (as we did in the analysis for 1950 and 1960), the results almost exactly parallel the findings for 1950 and 1960. As with the two earlier years, the partial regression coefficients (B values) for 1964 (-.001) and 1970 (.000) show the severity of prison sentence and homicide rates to be unrelated. Put differently, for neither year does length of imprisonment account for as much as 1% of independent explained variance in homicide rates when the certainty of imprisonment factor is held constant. Also similar to the results of our earlier analysis for 1950 and 1960, only a slight positive relationship was found bewtween the certainty of imprisonment and homicide rates for 1964 (B = .012) and 1970 (B = .022). These coefficients compare to B values of .851 for 1950 and -.024 for 1960. In sum, although we did not introduce execution rates and sociodemographic factors into the 1964 and 1970 analysis due to the problems of missing data (the required sociodemographic data are not available for 1964, and there were no executions in 1970), the results for these two years almost exactly parallel the findings for 1950 and 1960, again suggesting neither the certainty nor severity of imprisonment to be effective deterrents to homicide. To extend our analysis of the deterrent effect of the certainty of the death penalty, the execution rates, homicide rates, and the same sociodemographic factors considered earlier were fitted into a multiple regression analysis for the three census years 1920, 1930, and 1940. (This analysis is simply a replication of that performed for 1950 and 1960 but with the exclusion of the imprisonment variables for these years.~3)

~2Aswith 1950 and 1960, certainty and severity of imprisonment data for 1964 and 1970 were secured from the National Prisoner Statistics Series issued by the Federal Bureau of Prisons. For 1964, adequate severity and/or certainty figures were not available for the following death penalty states: New Hampshire, Vermont, New Jersey, Pennsylvania, Indiana, Iowa, Nebraska, Florida, North Carolina, Virginia, Alabama, Arkansas, Louisiana, and Texas. 2SFor 1920, 1930, and 1940, sociodemographic data were secured from the following sources: (1)Abstract of the Fourteenth Census of the United States: 1920 (Tables 6, 8, 15, 22, 37, 49, 143); (2) Abstract of the Fifteenth Census of the United States: 1930 (Tables 2, 3, 5, 6, 7, 29, 107, 143); (3) Sixteenth Census of the United States: 1940. Population . . . Characteristics of the Population . . . United States Summary (Tables 5, 21, 22, 23, 26, 31); (4) Sixteenth Census of the United States: 1940. Population... The Labor Force,.. United States Summary (Table 7). Unfortunately, homicide figures for individual states are only available from the F.B.I. Uniform Crime Reports from the year 1933. As a result, to examine 1920 and 1930, homicide figures were secured from the U.S. Public Health Vital Statistics Series for these years.

BAILEY

256 Table 5. Relationship between S t a t e ' s Execution Rates and Homicide Rates for Five C e n s u s Y e a r s ~

Year

Simple r

B coeff.

Beta coeff,

Pct. independent explained variance

1960 1950 1940 1930 1920

-.041 .129 -.017 -.175 -.243

.603 .250 -.152 -.294 -.468

.026 .087 -.031 -.121 -.151

0.1 0.5 0.1 0.5 1.2

aCertainty and severity of imprisonment for homicide are only introduced as control variables for 1950 and 1960.

Our analysis is confined to these three periods for execution, and homicide data are only available for a few states for earlier census years.24Results of this analysis for the certainty of the death penalty, as well as the findings for 1950 and 1960 (from Table 1), are summarized in Table 5. ~s Table 5 reports simple r correlations, standardized (beta weights) and unstandardized regression coefficients (B values), and the percent of independent explained variance in homicide rate accounted for by executions. Examination of the findings for the earlier census years reveals a pattern generally consistent with that found for 1950 and 1960. In line with the deterrence hypothesis, the bivariate correlations fall in the predicted negative direction for 1920, 1930, and 1940. Similarly, but in contrast to 1950 and 1960, after the control variables are introduced into the analysis, execution rates and homicides continue to be negatively associated. As with 1950 and 1960, however, none of the zero-order or partial coefficients are statistically significant for the earlier years. As noted earlier, due to the problems of missing severity and certainty of imprisonment data, it is not possible to directly compare the partial regression results for the earlier (1920-1940) and later (1950-1960) census years. Nor is it possible to directly assess to what extent (if any) the findings for the earlier years were spuriously produced due to our inability to control for the imprisonment variables. Despite this problem, however, it is important to note that even when the certainty and severity of

24For example, for 1910 we were only able to secure execution and homicide data for 13 states, a number too small to conduct a meaningful analysis. For the three later census years, the number of states examined are" (1) 1920 = 20; (2) 1930 = 34; (3) 1940 = 36. 26We chose to compare our results for the earlier census years with the 1950 and 1960 findings reported in Table 1 because: (1) the findings reported in Table 1 are quite similar to the results of the additional regression analyses (Tables 2--4), and (2) the model considered in Table 1 is the most straightforward and parsimonious of the four models examined. To be on the safe side, however, we performed the same type of analysis as shown in Tables 2--4 for the earlier census years. Again, as with 1950 and 1960, the four separate regression analyses produced almost identical results for each of the three years. In addition, an examination of the possible nonlinear relationship between execution rates and homicide rates for the earlier census years produced almost identical results to that shown in Table 5.

IMPRISONMENT V. DEATH PENALTY

257

imprisonment were left as uncontrolled factors, execution rates and homicide rates were not significantly associated. ~s In addition, when the 1950 and 1960 analysis was repeated, excluding the certainty and severity of imprisonment variables, the partial regression, results were very comparable to that found for the earlier years. As a result, it would appear that our generally negative findings for the death penalty for 1950 and 1960 may be generalized to at least the three earlier census years. Although the above analysis allows us to at least in part address the question of whether our findings for 1950 and 1960 are unique to these years, there remain a number of years where it was not possible to examine the deterrent effect of either executions or imprisonment. While we see no a priori reason why the years considered here could not be viewed as representative, the fact remains that many years are yet to be considered. Hopefully our request of the Federal Bureau of Prisons will result in additional prisoner data that will allow us to extend our analysis. Until such an analysis is possible, however, the available evidence would suggest that neither severe prison sentences nor the certainty of the death penalty are effective deterrents to homicide. While this conclusion would seem consistent with the results of most previous investigations, and also consistent with the belief held by many criminologists that expressive offenses like murder are extremely difficult to deter, it would still seem premature at this time to conclude that neither imprisonment nor executions has a deterrent effect on homicide. 27As noted above, neither this nor previous deterrence investigations are free of methodological limitations. For example, in the absence of alternative prisoner data, we were forced to operationalize the severity of imprisonment for each year as the average length of prison sentence served by homicide offenders released during the year. For the reasons discussed earlier, an alternative, and possibly better, measure of the severity of imprisonment might be the length of sentences received by convicted homicide offenders. Along somewhat similar lines, it would also seem important to again call attention to the possible difficulties involved in using F.B.I. and U.S. Public Health homicide figures in this type of investigation. As noted, while the death penalty is only typically prescribed for first-degree murder, we were forced to use offense categories that combine all types of homicides as a measure of our dependent variable. As a result, it remains unclear what our finding might have been had we been able to separately examine figures for capital and noncapital homicides. In addition to these two methodological considerations, this investigation also suffers from a few theoretical limitations that should not be overlooked. As discussed earlier, deterrence theory argues that if sanctions are to have a maximum deterrent effect they must be (1) severe, (2) certain, (3) administered promptly, and (4) made

~"Ifanything, introduction of the certainty and severity of punishment variables into the analysis would have the effect of further reducing, and not strengthening, the relationship between execution rates and homicide rates. 27For a general theoretical discussion of the presumed differential deterrent effect of punishment for "instrumental" and "expressive" offenses see Zimring and Hawkins (1973) and Packer (1968).

258

BAILEY

public. Although we did consider the severity and certainty dimensions of punishment in this investigation, the celerity and publicity factors were not incorporated into the analysis. Nor have these two dimensions of punishment been systematically examined in earlier deterrence investigations of imprisonment or the death penalty. As a result, it is not possible to determine how omitting these factors influenced our findings. This question too will simply have to await the analysis of currently unavailable data. In conclusion, although this investigation fails to add support to claims of the deterrent effectiveness of imprisonment and executions on homicide, the possible importance of the above theoretical and methodological limitations cannot be overlooked. While future research may well show these limitations to be of little or no consequence, this must remain an open question until these considerations have been systematically examined. Research is currently underway by this investigator to address these questions and thus provide a more comprehensive examination of the deterrence issue. REFERENCES Andenaes, J. The general preventive effects of punishment. University o f Pennsylvania Law Review, 1966, 114, 949-983. Andenaes, J. Punishment and Deterrence. Ann Arbor: The University of Michigan Press, 1974. Andenaes, J. General prevention revisited: Research and policy implications. Journal o f Criminal Law and Criminology, 1975, 66, 338-365. Bailey, W, C. Murder and the death penalty. Journal o f Criminal Law and Criminology, 1974, 65(3), 416-423. Bailey, W. C, Murder and capital punishment: Some further evidence. American Journal o f Orthopsychiatry, 1975, 45, 669-688. Bailey, W. C. A multivariate cross-sectional and longitudinal analysis of the deterrent effect of the death penalty. Paper presented at the American Society of Criminology Meetings, Tucson, Arizona, 1976. Bailey, W. C., Martin, D. J., & Gray, L. N. Crime and deterrence: A correlation analysis. Journal o f Research in Crime and Delinquency, 1974, 11(2), 124-143. Barnes, H. E., & Teeters, N. K. New Horizons in Criminology. Englewood Cliffs, New Jersey: PrenticeHall, 1959. Bedau, H. A. The Death Penalty in America. Revised Edition. New York: Doubleday-Anchor, 1967, 56-74. Bedau, H. A. Deterrence and the death penalty: A reconsideration. Journal o f Criminal Law, Criminology and Police Science, 1970, 61(2), 539-548. Bedau, H. A. The death penalty in America. Federal Probation, 1971, 35(2), 32-43. Bowers, W. J. A Causal Framework for the Analysis of Deterrence and Related Processes. In Akers, R., & Sagarin, E. (Eds.) Crime Prevention and Social Control. New York: Praeger, 1974a, 22-38. Bowers, W. J. Executions in America. Toronto: D.C. Heath, 1974b. Bowers, W. J., & Pierce, G. L. The illusion of deterrence in Isaac Ehrlich's research on capital punishment. The Yale Law Review, 1975, 85, 187-208. Dann, R. H. The deterrent effect of capital punishment. Friends Social Service Series, 1935, Bulletin No. 29, Third Month. Ehrlich, I. The deterrent effect of capital punishment: A question of life or death. American Economic Review, 1975, 65, 397-417. Federal Bureau of Investigation. Crime in the United States: Uniform Crime Reports - - 1940. Washington, D.C., 1940. Federal Bureau of Investigation. Crime in the United States: Uniform Crime Reports - - 1950. Washington, D.C,, 1950. Federal Bureau of Investigation. Crime in the United States: Uniform Crime Reports - - 1960. Washington, D.C., 1960.

IMPRISONMENT V. DEATH PENALTY

259

Federal Bureau of Investigation. Crime in the United States: Uniform Crime Reports ~ 1970. Washington, D.C., 1970. Federal Bureau of Prisons. National Prisoner Statistics: Prisoners in State and Federal Institutions, 1950. Washington, D.C., 1950. Federal Bureau of Prisons. National Prisoner Statistics: Prisoners Released from State and Federal Institutions, 1951. Washington, D.C., 1951. Federal Bureau of Prisons. National Prisoner Statistics: Characteristics of State Prisoners, 1960. Washington, D.C., 1960. Federal Bureau of Prisons. National Prisoner Statistics: Prisoners in State and Federal Institutions, 1963. Washington, D.C., 1963. Federal Bureau of Prisons. National Prisoner Statistics: State Prisoners: Admissions and Releases, 1964. Washington, D.C., 1964. Federal Bureau of Prisons. National Prisoner Statistics: State Prisoners: Admissions and Releases, 1970. Washington, D.C., 1970. Forst, B. The deterrent effect of capital punishment: A cross-state analysis of the 1969's. Minnesota Law Review, 1977, 61, 743-767. Forst, B., Filator, V., & Klein, L. The deterrent effect of capital punishment. Paper presented at the American Society of Criminology Meetings, Tucson, Arizona, 1976. Furman v. Georgia, 408 U.S. 238 (1972). Graves, W. F. A doctor looks at capital punishment. MedicalArts and Sciences, 1956, 10(4), 137-141. Gray, L. N., & Martin, J. D. Punishment and deterrence: Another analysis of Gibbs' data. Social Science Quarterly, 1969, 49, 289-295. Gregg v. Georgia, 96 S. Ct. 2909 (1976). Jeffery, C. R. Criminal behavior and learning theory. Journal of Criminal Law, Criminology and Police Science, 1965, 56, 294-300. Jurek v. Texas, 96 S. Ct. 2950 (1976). Logan, C. H. General deterrent effect of imprisonment. Social Forces, 1972, 51, 64-73. Orsagh, T., & Black, T. New evidence on the efficacy of the death penalty as a deterrent to homicide. Paper presented at the American Society of Criminology Meetings, Tucson, Arizona, 1976. Packer, H. L. The Limits of Criminal Sanctions. Stanford, California: Stanford University Press, 1968. Passell, P., & Taylor, J. B. The deterrent effect of capital punishment: Another view. Discussion Paper 747509, Columbia University, 1975. Proffitt v. Florida, 96 S. Ct. 2960 (1976). Roberts v. Louisiana, 96 S. Ct. 3001 (1976). Savitz, L. D. A study of capital punishment. Journal of Criminal Law, Criminology and Police Science, 1958, 49, 338-341. Schuessler, K. A. The deterrent influence of the death penalty. Annals of the American Academy of Political and Social Science, 1952, 284, 54-63. Sellin, T. Minutes of proceedings and evidence, Joint Commission of Senate and House of Commons on Capital Punishment and Lotteries. Ottawa, 1954, 669. Sellin, T. The Death Penalty. Philadelphia: American Law Institute, 1959. Sellin, T. CapitalPunishment. New York: Harper & Row, 1967, 122-124, 135-160, 239-253. Sutherland, E. H. Murder and the death penalty. Journal of the American Institute of Criminal Law and Criminology, 1925, 51, 522-529. Swartz, C., & Graham, R. Supplement to the Survey of Current Business. Washington, D.C.: U.S. Government Printing Office, 1956. U.S. Department of Commerce, Bureau of the Census. Abstract of the Fourteenth Census of the United States: 1920. Washington, D.C.: U.S. Government Printing Office, 1923. U.S. Department of Commerce, Bureau of the Census. Abstract of the Fifteenth Census of the United States: 1930. Washington, D.C.: U.S. Government Printing Office, 1933. U.S. Department of Commerce, Bureau of the Census. Sixteenth Census of the United States: 1940. Population, Volume II. Characteristics of the Population. Part I, United States Summary. Washington, D.C.: U.S. Government Printing Office, 1943. U.S. Department of Commerce, Bureau of the Census. Census of the Population: 1950. Volume II, Characteristics of the Population. Part I, United States Summary. Washington, D.C.: U.S. Government Printing Office, 1953.

260

BAILEY

U.S. Department of Commerce, Bureau of the Census. Historical Statistics of the United States. Washington, D.C.: U.S. Government Printing Office, 1960. U.S. Department of Commerce, Bureau of the Census. Census of the Population: 1960. Volume I, Characteristics of the Population. Part I, United States Summary. Washington, D.C.: U.S. Government Printing Office, 1964. U.S. Department of Health, Education and Welfare. Homicide in the United States: 1950-1964. Washington, D.C.: U.S. Government Printing Office, 1967. van den Haag, E. On deterrence and the death penalty. Journal of Criminal Law, Criminology and Police Science, 1969, 60, 141-147. Wilson, J. Q. The death penalty. New York Times Magazine, 1973, October 28, 26-27, 30, 34-39, 42-43. Woodson v. North Carolina, 96 S. Ct. 2978 (1976). Zimring, F. E., & Hawkins, G. Deterrence: The Legal Threat in Crime Control. Chicago: The University of Chicago Press, 1973.