TABLE OF CONTENTS
The Death Penalty in Texas: A State of Crisis
Official Misconduct (National Implications)
Racism in Deciding Who Should Die (National Implications)
The Crisis in Representation in Texas (National Implications)
Clemency in Texas (National Implications)
Texas Crime and the State's Response (National Implications)
Executive Summary Texas is the nation's foremost executioner. It has been responsible for a third of the executions in the country and has carried out two and a half times as many death sentences as the next leading state. Death warrants are being signed at an unmanageable pace, yet the Texas death row is bulging with unprecedented numbers of inmates. But this accelerated form of justice comes at a price. The rest of the country should heed the warning of the Texas experience before it embarks on a wholesale expansion of the death penalty.
The death penalty in Texas is in a state of crisis. Numerous death penalty convictions have been tainted by overzealous prosecutions and the use of perjured testimony. State paid medical "experts" make unreliable predictions about defendants' future dangerousness while other doctors simply lie about tests they never performed. Six innocent people have been sentenced to death and later released since 1987. The race of the defendant and victim play a major part in which cases are selected for the death penalty. Legal representation of indigent defendants at trial is frequently incompetent, and representation for appeals is often non-existent. And the costs of the death penalty in Texas are in the hundreds of millions of dollars with no end in sight.
And yet, Texas has little to show for all this expense and the sacrifice of judicial due process. During the period when Texas rose to become the nation's leading death penalty state, its crime rate grew by 24% and its violent crime increased by 46%, much faster than the national average. Texas leads the country in numbers of its police officers killed and more Texans die from gunshot wounds than from car accidents.
But Texas's death penalty problems are certainly not unique. Many states with large death rows have also been plagued by prosecutorial misconduct, innocent defendants sentenced to death, racism in the application of justice, inadequate representation, and the high costs of the death penalty. Forty-eight defendants have been released from death row since capital punishment was reinstated after evidence of their innocence was discovered. Half of the nation's death row is made up of minorities and almost all capital cases involve white victims.
Many in America are pushing for a faster pace and a wider use of the death penalty on both the state and federal levels. Texas is a paradigm of what can happen under such an expansion.
Some politicians and law enforcement officers in Texas are beginning to have second thoughts about their state's practice of the death penalty. While people want to address the problem of crime, they also want solutions that really work. Nationally, there should be a careful examination of death penalty justice in Texas before we embrace an expansion of executions as an answer to crime.
When in Gregg v. Georgia the Supreme Court gave its seal of approval to capital punishment, this endorsement was premised on the promise that capital punishment would be administered with fairness and justice. Instead, the promise has become a cruel and empty mockery. If not remedied, the scandalous state of our present system of capital punishment will cast a pall of shame over our society for years to come. We cannot let it continue.
--Justice Thurgood Marshall, 1990  From this day forward, I no longer shall tinker with the machinery of death. . . . I feel morally and intellectually obligated to concede that the death penalty experiment has failed.
--Justice Harry Blackmun, 1994 
Introduction Between 1967 and 1977, executions in the United States were halted as evidence of racial injustice and arbitrariness in the use of the death penalty mounted. When most states revised their capital punishment laws, the Supreme Court allowed the death penalty to resume in 1976. But this approval inaugurated a new period of experimentation regarding the application of the death penalty. In the eyes of many, including Justice Blackmun who oversaw this entire critical period of death penalty history, that experiment has failed to meet even the minimal standards of fairness and justice. And nowhere are these failings more evident than in the state of Texas.
No other state comes close to the number of executions being carried out in Texas. It has put to death more than twice as many inmates as any other state since the death penalty was reinstated. In 1993 alone, Texas accounted for more than three times as many executions as any other state and carried out almost half of the death sentences in the entire country.
The accelerated pace of executions and the disturbing number of inmates facing death without legal representation in Texas has drained both the state's resources and the ability of the defense bar to adequately respond. At the same time, the political pressure to achieve even more death sentences and more executions has frequently given due process a back seat.
But the size and problems of capital punishment in Texas are not unique to that state. The United States is perched on the precipice of a wholesale expansion of the death penalty. Before it takes the plunge, the country should look at Texas' experience. The warehousing of hundreds of people awaiting execution, half of whom are minorities, the constant signing of death warrants, the grisly spectacle of weekly executions, and the erosion of due process by the relentless press to execute will be much more common in the years ahead if the United States chooses to follow the Texas model. The problems which Texas has been experiencing in its rise to the position as the nation's foremost executioner are already emerging in other states throughout the country.
This report will look at various dimensions of the death penalty crisis in Texas:
- The examples of official misconduct and resulting mistaken convictions;
- The evidence of racism infecting the application of the death penalty;
- The crisis in death penalty representation which serves to perpetuate Texas' death penalty problems.
- The absence of clemency as a realistic remedy to prevent wrongful executions; and
- The way in which Texas' emphasis on the death penalty interferes with addressing the larger problem of crime.
The United States is perched on the precipice of a wholesale expansion of the death penalty. Before it takes the plunge, the country should look at Texas' experience.
At each step of the way, the report will look at the national implications of what is happening in Texas. It will identify the extent to which the Texas death penalty is likely to be mirrored in the rest of the United States in the near future. Finally, the report will point to the signs of official disillusionment with the death penalty in Texas. The death penalty crisis in Texas should be a warning to our entire country as we struggle to respond to the national problem of crime.
The Death Penalty in Texas:
A State of Crisis To get an idea of the size of the death penalty in Texas, it is instructive to look at what the death penalty in the entire country would be like today if every state had proportionately followed Texas' lead.
"[T]he scandalous state of our present system of capital punishment will cast a pall of shame over our society for years to come."
--Justice Thurgood Marshall In 1993 alone, there would have been 250 executions, one for every business day of the year and the largest number of executions in the country's history. In the last ten years, the U.S. would have executed over 1,000 people. The nation's death row would house more than 5,300 condemned individuals. The national cost for the death penalty would be at least two billion dollars, with much more expense to come. The courts of appeals and the Supreme Court would be deluged with petitions from condemned inmates, while at the same time hundreds of inmates would have no attorney as their execution dates approached. Meanwhile, all of these executions would have done nothing to lower the nation's murder rate. 
In achieving this proficiency in executions, Texas has sacrificed the pursuit of justice. It was probably no coincidence that Justice Blackmun chose a Texas case to condemn the death penalty. Justice Thurgood Marshall had earlier warned that the entire country was in similar danger because of the death penalty: "[T]he scandalous state of our present system of capital punishment will cast a pall of shame over our society for years to come."  As the nation moves toward an even greater expansion of this practice, it should consider whether capital punishment is worth the mantle of such a pall of shame.
1. Official Misconduct: The Death Penalty With A Vengeance Texas has pursued the death penalty with a vengeance. Prosecutors and politicians have staked their careers on getting people executed. Unfortunately, such political grandstanding results in more than rhetoric -- individual rights have been sacrificed and innocent people have been sent to death row.
This phenomenon is certainly not unique to Texas, but Texas politicians have campaigned shamelessly on the strength of their commitment to ever more executions. In the 1990 gubernatorial race, former Governor Mark White portrayed his "toughness" by walking through a display of large photos of people executed during his term, while Attorney General Jim Mattox insisted that he was the one who should be given credit for the many executions under his watch. And the Republican candidate, Clayton Williams, claimed that his proposed laws to expand the death penalty were "the way to make Texas great again." 
The end result of all this political posturing was a spoof of Texas on "Saturday Night Live" and the election of Ann Richards as governor. Governor Richards was the least vociferous of the candidates on the death penalty but has nevertheless presided over a dramatic increase in the pace of executions in Texas.
Convicting the Innocent
Among elected state prosecutors, death penalty rhetoric has sometimes spilled over into serious abuses in order to secure a death sentence. Two of the most famous Texas examples of this misconduct involved Randall Dale Adams and Clarence Brandley, both of whom were released from Texas' death row after years of struggle to prove their innocence. Adams' story was eloquently told in the award winning movie, The Thin Blue Line, and Brandley's struggle with Texas racism is related in Nick Davies' book, White Lies.
The original convictions of Adams and Brandley were not simply the product of honest prosecutorial mistakes. When Randall Dale Adams had his murder conviction unanimously overturned by the Texas Court of Criminal Appeals, Judge M. P. Duncan sharply castigated the prosecution: "[T]he State was guilty of suppressing evidence favorable to the accused, deceiving the trial court during the applicant's trial, and knowingly using perjured testimony." 
"The State was guilty of suppressing evidence favorable to the accused, deceiving the trial court during the applicant's trial, and knowingly using perjured testimony."
--decision overturning Randall Dale Adams' conviction
Similarly, when Texas Special District Judge Perry Pickett reviewed Clarence Brandley's conviction in 1987, he concluded that the state's investigative procedure was "so impermissibly suggestive that false testimony was created, thereby denying . . . due process of law and a fundamentally fair trial." Furthermore, the state had "wholly ignored any evidence or leads to evidence that might prove inconsistent with their premature conclusions that Brandley had committed the murder. The conclusion is inescapable that the investigation was conducted not to solve the crime, but to convict Brandley."
In their zeal to obtain capital convictions, Texas prosecutors have made wide use of medical "experts" selected because of their willingness, in case after case, to parrot the exact words the prosecutor needs to get a conviction. One such "expert" is Dr. James Grigson -- or "Dr. Death," as he came to be known.
Dr. Death, I
In Texas, jurors are required to determine "whether there is a probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society." 
Not only is it difficult for a lay person to make such a judgment, it is also impossible for professionals. Naturally, a jury would give considerable weight to a state psychiatrist who unhesitatingly predicts with scientific certainty that the person sitting in front of them will invariably kill again if he  is allowed to live.
Dr. James Grigson offered just such predictions in at least 124 death penalty cases, 115 of which resulted in death sentences.  Dr. Grigson traveled the plains of Texas offering his testimony in exchange for sizable fees. At first, Grigson would personally examine the defendant, perhaps for 90 minutes. Based on this cursory interview, Dr. Grigson would then be asked by the prosecutor in court:
Can you tell us whether or not, in your opinion, having killed in the past, he is likely to kill in the future, given the opportunity?Grigson would reply:
He absolutely will, regardless of whether he's inside an institutional-type setting or whether he's outside. No matter where he is, he will kill again.Grigson made these same predictions about Randall Dale Adams, despite Adams' having no history of violence. The fact that Adams was exonerated of all charges and was freed from prison a number of years ago has done nothing to sway Grigson's certainty about his predictions. 
In later cases, Grigson would offer his absolutely certain view of the future without even interviewing the defendant. He would simply listen to the prosecutor's description of the defendant's crime and background and then offer the conclusion that such a person would certainly kill again, no matter what the setting.
The American Psychiatric Association has unequivocally condemned the process that Dr. Grigson has used so liberally. "[P]sychiatric testimony of future dangerousness impermissibly distorts the fact-finding process in capital cases,"  they said in a brief to the Supreme Court.
Moreover, empirical studies have shown the inaccuracies of predicting future dangerousness. One study in Texas examined 92 former death row prisoners whom juries had sentenced to death because of their future dangerousness. For various reasons, these inmates had their sentences changed from death to life imprisonment. The study concluded:
Overall these former death row prisoners were not a disproportionate threat to the institutional order, other inmates, or the custodial staff. Indeed, their total rate of assaultive institutional misconduct was lower than those of both the capital murder offenders who were given a life sentence [to begin with] and the general prison population.
Despite the unreliability of such predictions, Dr. Grigson's testimony has been used by the prosecution in one-third of Texas' death sentences. The problem of manipulating juries with fear is compounded by Texas law which forbids telling the juries what the alternative to a death sentence really means. A life sentence in a capital case in Texas now means that the defendant must serve 40 years before even being considered for parole. But jurors are told only that their alternatives are the death penalty or a life sentence. They are left with their erroneous assumptions that a life sentence will allow a dangerous murderer to be released in 10 years or less. 
Dr. Death, II
Another critical element of the prosecution's case in a capital trial is proof that the victim's death resulted from the defendant's violent actions. To tie that knot, many prosecutors in Texas have utilized a pathologist by the name of Ralph Erdmann, who has also earned the name "Dr. Death." Erdmann received his medical degree in Mexico in the 1950s and traveled to 40 Texas counties supposedly performing 400 autopsies a year in capital and non-capital cases. Lubbock County alone paid Dr. Erdmann $140,000 a year for his work. Now the verdicts in at least 20 capital murder cases and dozens of other prosecutions are being appealed because Erdmann lied, falsified reports and even neglected to perform some of the autopsies he testified about. 
Erdmann's word began to be doubted when one family read his autopsy report indicating that the deceased's spleen had been examined and weighed as part of the examination. However, the family knew that the dead man's spleen had been removed years earlier. As a result of the family's intervention, the body was exhumed and no incision marks from an autopsy were found. At that point, attorney Tommy Turner of Lubbock was appointed special prosecutor to look into Erdmann's deceptions. Turner concluded that Erdmann was a liar and a con man: "If the prosecution theory was that death was caused by a Martian death ray, then that was what Dr. Erdmann reported."
"If the prosecution theory was that death was caused by a Martian death ray, then that was what Dr. Erdmann reported."
--Special prosecutor Tommy Turner
Killing the Messenger
When Erdmann's methods and testimony came under increasing criticism in death penalty cases, some prosecutors retaliated by prosecuting Erdmann's critics. Two police officers, Patrick Kelly and William Hubbard from Lubbock County, who had testified about Erdmann's misdeeds, were indicted for alleged perjury. And nationally famous death penalty defense attorney, Millard Farmer of Atlanta, was indicted for supposedly tampering with a witness. However, this effort to cover-up the growing scandal around Dr. Erdmann fell apart.
A federal District Court judge ordered a halt to the prosecutions and stated that those being attacked "have offered substantial evidence that the prosecutions were brought in bad faith and for purposes of retaliation." 
A suit brought by the police officers and Mr. Farmer against the prosecutors who indicted them was settled in favor of the plaintiffs with the agreement that the prosecutions be permanently stopped, that the policemen be restored to their jobs with full back pay, and that they be awarded $300,000 in damages.
Dr. Erdmann had earlier pleaded no contest to seven felony charges. He was sentenced to 10 years probation, and fined $17,000 for botched autopsies and exhumation expenses. He also surrendered his medical license and moved to another state. 
As disgraceful as the behavior of these medical "experts" has been, the real scandal is that prosecutors were willing to repeatedly utilize such witnesses in order to get convictions and death sentences. In other instances, prosecutors failed to investigate cases thoroughly and allowed defendants, later found innocent, to be sentenced to death. Besides Randall Dale Adams and Clarence Brandley, at least four other Texas death row inmates have been found innocent in recent years (Muneer Deeb, 1993; Federico Macias, 1993; John Skelton, 1990; and Vernon McManus, 1987) and that number could increase as further abuse is examined. Unfortunately, this pattern of prosecutorial misconduct in capital cases is not unique to Texas.
Official Misconduct The pressure on prosecutors and police to succeed in death penalty cases has resulted in miscarriages of justice all over the country. Representative Don Edwards, Chair of the House Judiciary Subcommittee on Civil and Constitutional Rights, released a staff report in October, 1993, recounting 48 cases since 1970 in which the defendants were sentenced to death but later exonerated and released.  In many of these cases, the prosecutors or police illegally withheld vital information from the defense, encouraged witnesses to lie, and deceived the court in a variety of ways. In other cases, prosecutors pushed for the death penalty in headline cases in which they lacked sufficient evidence even to sustain a conviction.
For example, when Attorney General Janet Reno was a prosecutor in Dade County, Florida, she helped uncover a pattern of official abuse in the death penalty conviction of James Richardson. Richardson had been sentenced to death for poisoning his own children in 1968. He was spared the electric chair when the Supreme Court overturned all existing death sentences in 1972, but he remained in prison. Reno's 1989 investigation affirmed what had long been claimed by the defense: the state had "knowingly used perjured testimony and suppressed evidence helpful to the defense." Richardson was released in 1989.
Just last year, five people were released after years on death row for crimes they did not commit.  In the case of Walter McMillian in Alabama, prosecutors admitted that the case had been mishandled. Evidence was improperly withheld from the defense, the state's three main witnesses all admitted that they had lied, and the "eyewitness" said that he had been pressured to pin the blame for the murder of the young white woman on McMillian, who is black. 
In the case of Kirk Bloodsworth in Maryland, prosecutors improperly withheld evidence of a different suspect who bore a striking resemblance to the police sketch in the rape and murder of a young girl. The other suspect had been found in the woods near the murder scene, had a blood-like spot on his shirt, and was very dirty except for his hands, which were meticulously clean. Moreover, the police found a young girl's underwear in this suspect's car. The suspect had a prior conviction for indecent exposure and had failed a polygraph test.  Nevertheless, prosecutors sought and obtained a death sentence against Bloodsworth. Fortunately, he was completely cleared in June, 1993, when a new DNA test confirmed that someone else had committed the crime. 
Federal prosecutors are also not immune from such practices. If Congress passes a greatly expanded federal death penalty in 1994, U.S. Attorneys will be responsible for a much larger number of death penalty cases. However, recent investigations into abuses in the El Rukin gang prosecution in Illinois and a major racketeering case in Los Angeles in which an appellate court described the government's conduct as "intolerable,"  have shown that some federal prosecutors also engage in misconduct to obtain convictions.
Attorney General Janet Reno has promised much swifter investigations into allegations of abuse by federal prosecutors. But experience has shown that evidence of prosecutorial abuse, if discovered at all, may come only long after the defendant's conviction. In capital cases that may be too late. This problem raises the almost certain specter that innocent people will be executed, especially if capital punishment is expanded. 
2. Racism In Deciding Who Should Die Judge (Roy) Bean opened the session by spending two hours reading the Texas statutes aloud to the courtroom spectators. He then closed the law book, and, dropping it on the bench, declared: "That's . . . the complete statutes of this here state from the Alamo on ahead, and there ain't a damned line in it nowheres that makes it illegal to kill a Chinaman. The defendant is discharged."
--Judge Roy Bean, presiding at the trial of his son for murdering a Chinese laundryman who overcharged him  The cop paused and stared at the two of them, the black man in his white T-shirt and shabby jeans, the little white man with the thick glasses and the ballooning belly.
"One of you two is gonna hang for this," said the cop. Then he turned to Brandley. "Since you're the nigger, you're elected."
--Nick Davies in White Lies, quoting testimony leading to Clarence Brandley's release 
In 1993, national attention was drawn to two murder cases in Texas. In one case, the defendant was given the death sentence; in the other, he was placed on probation. Although, the cases differ in some respects, the most glaring inequity is that a young white man was given leniency for the murder of a black man, while a young black man was condemned to death for the murder of a white man. This disparity is symptomatic of broader inequities in Texas depending on the race of the defendant and the race of the victim.
Both of those convicted of the crimes were 17 years old at the time of the murders. In the first case, a white supremacist skinhead, Christopher Brosky, was given 10 years probation for the murder of Donald Thomas, a black man. One of the jurors responsible for the sentence commented: "We just felt like this might be a man who might be able to turn his life around . . . . If we had sent him to Huntsville, he might have come back in worse shape." 
In the second case, a young black man, Gary Graham, was given the death penalty for the murder of Bobby Lambert in Houston back in 1981. Graham has been on death row in Huntsville ever since. Both cases have split the community and resulted in demonstrations raising issues of race and the administration of justice. Graham's case gained particular attention in 1993 because of new evidence pointing to his innocence.  His execution has been stayed three times, but he remains on death row.
Racism in Texas' Earlier Use of the Death Penalty
Outcomes based on race in death penalty cases have a long history in Texas. From the time of the first state executions, the race of the defendant played a large role in who was given the death penalty. For example, between 1924, when centralized state executions were begun, and 1972, 361 people were put to death in Texas. About 70% of them were either African- or Mexican-American, with blacks constituting 63% of those executed. Of the whites sentenced to death during this period, 34% had their sentences commuted. Only 20% of the blacks received clemency. 
The race of the victim was an even more certain predictor of which cases would receive the death penalty. Prior to 1972, 80% of the victims in Texas death penalty cases were white.  In rape cases where the death penalty was applied, 95% of the victims were white. When a black man was convicted of raping a white woman, the sentence was virtually always death. No white man, however, was executed for raping a black woman. 
Before 1924, central state records on executions were not compiled, since the death penalty was carried out locally. However, Texas' part in the history of lynchings in the U.S. reveals an even more severe practice of racial bias. In post-Civil War Texas, lynchings were often used as a form of punishment and intimidation. Not surprisingly, almost all of those who suffered this illegal form of vigilante justice were black. From 1889 to 1899, over 95% of the recorded lynchings in Texas were of blacks.  The geographical pattern of lynchings in Texas closely followed those areas where slavery had been most prevalent. 
As part of a response to the embarrassment of racial lynchings, state legislators voted to move executions to a central state location in Huntsville and to change the method of execution from hanging to the electric chair. Interestingly, the warden of the Huntsville prison, Captain R.F. Coleman, resigned over this unwanted duty, saying: "A warden can't be a warden and a killer too. The penitentiary is a place to reform a man, not to kill him."  Coleman was replaced by a more accommodating warden, and four days later on Feb. 8, 1924, the State of Texas electrocuted its first five prisoners, all black.
"Coleman was replaced by a more accommodating warden, and four days later on Feb. 8, 1924, the State of Texas electrocuted its first five prisoners, all black."
Racism in the Current Use of the Death Penalty
Racial discrimination in the application of capital punishment was one of the factors that led the U. S. Supreme Court in 1972 to throw out virtually all existing death penalty statutes and sentences. Their ruling required states to more carefully craft new statutes that narrow the class of defendants who can receive the death penalty. Texas was one of the first legislatures to approve new death penalty laws, less than one year after the High Court's decision. In 1976, when the Court allowed the death penalty to resume, the Texas statute was one of three such laws that the Court approved.
The racial composition of Texas' death row has improved only slightly since the death penalty resumed. The percentage of minorities on Texas' death row has decreased from 70% to 55%, still a large disproportion. However, the racial disparities with respect to victims has changed little. In capital cases, if you murder a white person in Texas, you are over five times more likely to receive the death penalty than if you murder a black person. In none of the 74 Texas executions was the victim black and the defendant white. In fact, a recent Texas study of homicide cases between 1980 and 1988 found that no white offender who killed a black victim has even been charged and convicted with capital murder. 
Racism in the death penalty does not fully explain the pace of executions or the size of death row in Texas. However, it is a recurrent and largely untreated sore which skews the use of the death penalty in Texas and eats away at the hope for better relations among the races. A Texas governmental report showed that racial disparities are evident in other areas of criminal justice, as well. For example, the incarceration rate for blacks in Texas is over eight times the rate for whites.  Almost half of black offenders are sentenced to prison, but less than one-third of white offenders are so sentenced. 
But in many respects, the racial problems in other states are as severe as they are in Texas.
Source: NAACP Legal Defense & Educ. Fund, Death Row USA (1/94)
Race and the Death Penalty Racism is also apparent in national death penalty statistics. Half of those on death row are from minority populations that make up only 20% of the country's population. Blacks are represented on death row at three and a half times their proportion in the population as a whole. As is the case in Texas, however, the form of racial discrimination which is most directly attributable to capital punishment concerns the race of victims.
Blacks constitute about 50% of the victims of homicide in this country. One might expect, therefore, that the percentage of death penalty cases involving black victims would approximate 50%. That has never been the case, and all the reforms instituted at the insistence of the Supreme Court in 1976 have done nothing to alleviate the problem.
Since 1976, 84% of the victims in the cases resulting in an execution were white. In 1993, the numbers were even worse: 89% of the cases resulting in an execution involved white victims. Only one out of the 226 executions between 1976 and 1993 involved a white defendant who had killed a black victim. This represents a consistent pattern since the founding of this country.
In all, only 31 of the over 18,000 executions in this country's history involved a white person being punished for killing a black person. In 1990, the U.S. General Accounting Office reviewed the existing studies on racism and the death penalty in the United States and concluded:
Our synthesis of the 28 studies shows a pattern of evidence indicating racial disparities in the charging, sentencing, and imposition of the death penalty after the Furman decision.
In 82% of the studies, race of the victim was found to influence the likelihood of being charged with capital murder or receiving the death penalty, i.e., those who murdered whites were found more likely to be sentenced to death than those who murdered blacks.
The federal government's use of the death penalty has been even more racially disproportionate than the states. Under a new 1988 statute aimed at murders by drug "king-pins," almost 90% of those approved by the Attorney General for capital prosecutions have been either black or Hispanic defendants. 
"[T]hose who murdered whites were found more likely to be sentenced to death than those who murdered blacks."
--U.S. General Accounting Office
The continuation of racial disparities in the use of capital punishment is an embarrassment for the entire country. The riots following the verdict in the first Rodney King beating case in California indicate the serious repercussions possible when a jury appears to ignore the facts and decide a case based on the status of who committed the crime and who was the victim. There have been some Congressional attempts to rectify the problem in capital cases, but these were defeated when prosecutors argued that any law that prohibited racially disproportionate death sentencing would mark the end of the death penalty in the entire country.
This ongoing problem of racial disparities was addressed by Supreme Court Justice Harry Blackmun in his dramatic dissent to a death penalty ruling: "Even under the most sophisticated death penalty statutes," said Blackmun, "race continues to play a major role in determining who shall live and who shall die." He announced that he would no longer "tinker with the machinery of death" because he had concluded that "the death penalty experiment had failed." 
As the number of people on death row and the number of people executed in this country continue to grow, the patterns of racial disparity will become clearer and more disgraceful. What has been a perennial problem in Texas' administration of the death penalty will become apparent as a national problem as well.
"Even under the most sophisticated death penalty statutes, race continues to play a major role in determining who shall live and who shall die."
--Justice Harry A. Blackmun
3. The Crisis of Representation in Texas You are an extremely intelligent jury. You've got that man's life in your hands. You can take it or not. That's all I have to say.
--entire defense offered by a Texas attorney for his client, Jesus Romero, at a capital sentencing 
The state [of Texas] paid defense counsel $11.84 per hour. Unfortunately, the justice system got only what it paid for."
--U.S. Court of Appeals overturning Federico Macias' death penalty conviction
The enormity of the death penalty in Texas has overtaken the state's willingness to mete this punishment out with even a modicum of fairness and due process. Of all the factors which determine whether or not a particular defendant will ultimately receive the death penalty, probably the most important is the quality of representation he or she receives. In Texas, death penalty defendants are frequently given inexperienced and underpaid attorneys at trial. For some critical stages of their appeal, the defendants are given no attorney at all.
As in other states, almost everyone who is charged with a capital crime in Texas cannot afford his own attorney. The state is therefore required to provide him with one. Texas delegates that responsibility to the local county which is trying the case. The State of Texas itself provides no funds for the representation of those charged with a capital crime. The selection and qualifications of the attorney, the fee he or she will be paid, and the amount of resources which will be made available for investigation and expert witnesses are totally in the hands of the 375 local judges, who have widely varying economic resources.
In the larger counties, such as those encompassing Houston or Dallas, the judge might select from more experienced defense counsel and pay them a higher rate. In poorer counties, a general practitioner might be chosen and paid as little as $800 for an entire case.  In Randall County, for example, defense counsel Mallory Holloway was told that he had better not ask for investigation funds since he had already drained the county's budget by insisting on co-counsel. 
Recently, the State Bar of Texas commissioned a study of the system of representation in death penalty cases. A comprehensive report prepared by the Spangenberg Group of Massachusetts was released in March, 1993. The report found that capital representation in Texas was plagued with tremendous problems at both the trial and appellate level. It described the lack of counsel and the inadequacy of funding as "desperate" and "urgent" and concluded:
We believe, in the strongest terms possible, that Texas has already reached the crisis stage in capital representation and that the problem is substantially worse than that faced by any other state with the death penalty.Representation At Trial The problem of representation in capital cases in Texas is multi-layered, beginning at the trial stage. Texas is the only death penalty state which makes practically no use of a public defender system to provide attorneys. Instead, Texas allows each county to secure counsel through the private bar, often on a contract basis. The county judges can individually determine what they believe to be a "reasonable attorney fee" and compensation for "reasonable expenses." Until 1987, the statute regarding payment of attorneys for such work made no mention of compensation for the investigation, research and consultation with experts before trial. The statute did provide minimum payments for in-court appearances and these often became the de facto maximum paid to attorneys for the entire case. Although the statute was changed in 1987, the rates paid in many counties did not change, and Texas' compensation for court-appointed attorneys remains near the lowest in the country.  The prosecutors, on the other hand, represent a team of salaried state employees with ample resources and ready access to other law enforcement agencies for investigating and pursuing their cases. 
The rate of compensation often determines the quality of representation. The Spangenberg Report concluded that defending death penalty cases in Texas is frequently a losing financial venture for attorneys: "The rate of compensation provided to court-appointed attorneys is absurdly low and does not cover the cost of providing representation."  Without adequate compensation, it would be unrealistic to expect the consistent provision of a thorough defense. The consequences of poor representation can be disastrous. Federico Macias, for example, came within two days of execution in Texas because his trial attorney did almost nothing to prepare for trial. Today he is a free man, thanks to volunteer counsel from a large Washington law office that intervened just before Macias' execution. With qualified counsel and ample resources, Macias was not only granted a stay of execution but was eventually cleared of all charges in 1993. The federal court's order overturning the conviction noted that the first attorney had missed evidence of Macias' innocence:
We are left with the firm conviction that Macias was denied his constitutional right to adequate counsel in a capital case in which actual innocence was a close question. The state paid defense counsel $11.84 per hour. Unfortunately, the justice system got only what it paid for.Another man who was freed from death row this past year in Texas was Muneer Deeb. Deeb said he was poorly represented at his first trial. At his re-trial, however, he was represented by one of Texas' best known criminal attorneys, Dick DeGuerin, and was acquitted of all charges.  Other death row inmates who may also be innocent are not so fortunate.
Post-Trial Representation: A Desperate Situation
Access to the appeals process is critical to sparing the lives of those who are mistakenly sentenced to death. Former Texas death row inmates like Randall Dale Adams, Clarence Brandley, Federico Macias, and Muneer Deeb were extremely fortunate that others became interested in their cases and helped them attain freedom. But Texas has severely limited that access by not providing attorneys during critical appeal stages. And most recently, the state has even pushed for executions prior to the completion of a defendant's appeals.
Death row inmates are entitled to representation for only one direct state appeal of their conviction or sentence. After that appeal, Texas generally provides no attorney for subsequent appeals. Unless some court grants a stay, execution warrants can be signed and carried out despite the fact that an inmate might have significant issues requiring state and federal review.
With respect to this period of post-conviction representation, the Spangenberg Report found that:
[T]he situation in Texascan only be described as desperate. The volume of cases is overwhelming. Presently no funds are allocated for payment of counsel or litigation expenses at the state habeas level.Whereas most other states have a system for appointing and compensating attorneys after the direct appeal is over,  Texas leaves this important step to the discretion of the local judge. In almost every case, no attorney is appointed for state post-conviction relief, and those who do the legal work do it without pay. Similarly, funds for expert witnesses and expenses are almost never approved. 
With the defendant unrepresented, local prosecutors have recently begun to push for executions, and some local judges are no longer granting stays until an attorney can be found. The dangerous consequences of this procedure were seen in the recent case of Lesley Lee Gosch, who was scheduled to die just after midnight on September 16, 1993 and who had no attorney. Despite the fact that the Texas Attorney General's office acknowledged that Gosch still had legitimate appeals to pursue, the prosecution persuaded federal District Judge Hippo Garcia to refuse a stay of execution, which was just hours away.
As time was running out, the Texas Resource Center told Judge Garcia that an attorney had been found to represent Gosch. The Judge still refused to stay the execution and instead appointed the Resource Center to represent Gosch. Finally, just 25 minutes before the execution, with the inmate already being prepared for the lethal injection, Judge Garcia relented and granted a stay because, he said, new unresolved legal questions deserved review. Jay Jacobson, Executive Director of the ACLU of Texas, sharply criticized this unnecessarily close call: "Texas justice is in mortal danger of reverting back to the speedy vigilantism of Roy Bean; a rush to judgment in place of justice." 
In response to this crisis, the Texas Resource Center recently brought a case before the U.S. Supreme Court to clarify the federal courts' authority to stay executions while attorneys are being found to properly prepare death row appeals.  In an amicus curiae brief filed supporting the Resource Center's position, the American Bar Association called Texas' attempted denial of an opportunity to appeal a "perverse process" which "effectively nullifies the Great Writ" of habeas corpus. 
Thus, the death penalty in Texas is caught in a spiraling crisis:
- The volume of cases in Texas which has reached the post-conviction stage surpasses that of any other state and is exhausting the supply of volunteer attorneys from Texas and around the country.
- No state funds are available for carrying on the appeals necessary to prevent an inmate's execution, thus discouraging attorneys who might represent death row inmates. 
- With the defendant unrepresented, the state pushes ahead for executions. Death warrants are signed, putting prospective volunteer attorneys under more pressure and creating even more reluctance to take these cases.
In response to the inadequate system of representation for death row inmates in Texas, representatives from the University of Texas School of Law and a committee of attorneys concerned about the crisis proposed a resource center to recruit and train volunteer attorneys to handle death penalty cases after the direct appeal. The Texas Resource Center was created in 1988 and receives the bulk of its funding from the federal government's Administrative Office of the U.S. Courts.
With over 360 people on death row and with new cases being constantly added to the list, there is no way that the 16 attorneys of the Resource Center can represent even a significant proportion of the appeals. The Resource Center recently estimated that more than 75 death row inmates in Texas had no representation, many of whom were scheduled for execution within 5 weeks. Much of the Center's efforts go into recruiting and assisting counsel from other
states who agree to represent Texas' death row inmates. Despite Texas' relatively high rate of executions, the situation would be much graver without the Resource Center. The Spangenberg Report concluded that the Resource Center's staff provides "an invaluable array of services under truly unique pressures and circumstances."  They are not staffed, however, to fill all the gaps created by Texas' failure to appoint and pay counsel in capital cases.
Not surprisingly, the Resource Center's pursuit of legal representation and their success in stopping many executions has drawn reactions from prosecutors and politicians intent on an expeditiously functioning death penalty. There have been attempts to discredit the Resource Center in the media and to have Congress withdraw its funding.  These challenges have, in turn, been met by prominent members of the Texas Bar, some of whom serve on the Board of Directors of the Resource Center. The dispute illustrates the highly political nature of the death penalty in Texas.
The Pressure to Execute: A Chronology February 17 Federal District Court judge grants a stay to hear constitutional claims regarding the execution of an innocent person.
February 18 Texas Attorney General's office obtains a Circuit Court order vacating the stay on the grounds that a claim of innocence is irrelevant in federal court.
9:30 PM Defense attempts to obtain another stay of execution from the state or federal courts.
10:00 PM Request for stay filed with U.S. Supreme Court.
12:05 AM Defense checks with local weather station regarding the exact time of sunrise: Texas law requires that the execution take place before sunrise on the appointed day.
1:00 AM Supreme Court rejects stay by vote of 5-4.
4:30 AM Supreme Court again refuses to stay execution but indicates it will entertain a request to review the issue of executing the innocent.
4:35 AM State of Texas informs attorneys that it will begin the lethal injection of Herrera in 30 minutes if a stay is not in place.
4:35-6:20 AM Frantic efforts to obtain a stay from state and federal courts; two members of the Texas Court of Criminal Appeals and a federal District Court judge agree to stay execution so the Supreme Court can hear the case; Texas Attorney General attempts to have stays vacated; State court stay upheld.
6:20-6:55 AM Silence regarding Herrera's status.
7:00 AM Clerk of Supreme Court announces that state court stay is valid.
7:08 AM Texas sunrise: no execution. 
Leonel Herrera's case was eventually argued before the Supreme Court eight months later, in October 1992. He argued that he should be given a hearing to review new evidence of his innocence and that it would be unconstitutional to execute someone who was innocent. Witnesses, including a former Texas judge, revealed that Herrera's brother had actually confessed to the crime. A decision was reached on January 25, 1993, with Herrera losing on a vote of 6-3. Herrera was executed on May 12, 1993.
The Crisis in Representation The crisis in death penalty representation is starting to spread to other death penalty states as well. Because of the number of people on Texas' death row and the rate at which those people are now being processed for execution, the problem in Texas is more acute than in other places. But the size of the national death row is also increasing rapidly: at
least 250 people are sentenced to death each year and other states are experiencing both a shortage of attorneys and a shortage of funds to pay for the death penalty.
A six month study by The National Law Journal of death penalty representation in the south concluded:
Southern justice in capital murder trials is more like a random flip of the coin than a delicate balancing of the scales. Who will live and who will die is decided not just by the nature of the crime committed but equally by the skills of the defense lawyer appointed by the court. And in the nation's Death Belt, that lawyer too often is ill-trained, unprepared and grossly underpaid.The study found high disbarment rates for attorneys who represented death row inmates, widespread inexperience among those appointed to capital cases, and wholly unrealistic caps on the funds available for defense. With limits on attorneys' fees of $1,000 in states like Alabama, Louisiana, and Mississippi, lawyers offering even minimal representation were working for $5 an hour. Such meager pay obviously can affect performance.
In Tennessee, a state not included in the Law Journal study, it is not uncommon for trial attorneys to spend less than 100 hours preparing a capital case, while it typically takes over 1,000 hours in other states. In two death row cases, the attorneys spent 10 and 16 hours respectively preparing for trial. In 17 Tennessee cases, no mitigation evidence whatsoever was offered during the penalty phase of the trial. Under Tennessee law, if no mitigation evidence is presented, the court is compelled to direct a sentence of death, assuming the prosecutor has presented aggravating circumstances.Tennessee has one of the lowest compensation rates for indigent defense in the country: $20/hr. for out-of-court time and $30/hr in-court. 
In contrast, the state of Ohio allowed $40,000 for two attorneys in capital cases.  In California, attorneys are paid about $75 an hour and total fees often exceed $100,000 just for the appellate work.  But even in California, which recently surpassed Texas as the state with the largest death row, nearly a third of those on death row lack lawyers for their appeals. 
In Georgia 60 of the 80 people on death row who have gone beyond their direct appeals are being represented by lawyers from outside the state. "Many [Georgia] firms view defending a person on death row as politically unpopular, bad public relations and bad business," said Robert Remar, who heads a state bar committee to correct the problem. 
Ronald Tabak, chair of the ABA's Individual Rights and Responsibilities death penalty committee, said that the situation is "getting materially worse because demand for lawyers is growing substantially as the number of inmates moving into state post-conviction and federal habeas proceedings is increasing." States like California, Ohio, Pennsylvania and Illinois, with bulging death rows but few executions so far, are a warning that the crisis in death penalty representation will soon be spreading.
4. Clemency in Texas The U.S. Supreme Court recently ruled in the Herrera case that a defendant with a claim of innocence still has the opportunity to apply for executive clemency.  Even though Texas has by far the most death row inmates who have reached the end of their appeals and whose last chance for relief lies with the governor, there have been no commutations granted at a defendant's request since the death penalty was reinstated.  Texas has refused clemency in one case where it was requested by the prosecutor and by the father of the victim,  and in another case where it was even requested by the Pope.  Clemencies have been rare in other states as well, but most of those states have had few inmates who had exhausted all their appeals and sought clemency.
The case of Gary Graham, discussed above, is testing the seriousness of Texas' clemency procedure. Graham was convicted and sentenced to death on the basis of one eyewitness who viewed him only from a distance at night. New evidence indicating that Graham may be innocent has emerged, but it has been barred by Texas procedural rules which forbid introducing new evidence more than 30 days after a conviction. 
Graham was denied clemency and the Pardon Board did not even meet to hear his evidence. He has filed a suit claiming that his due process rights have been violated because he was not given a hearing by the Pardon Board. The Graham case tests whether there is any substance to Texas' clemency process. Texas courts are still considering whether the Board will be required to hold the hearing and possibly spare Graham's life. But regardless of the outcome in Graham's case, clemency in Texas has not been the safety-valve recommended by the High Court.
NATIONAL IMPLICATIONS: Clemency With respect to clemency, the extreme hesitancy of governors to utilize this process in death penalty cases is also a national problem. Clemency used to be granted more liberally by governors in capital cases. Prior to the Furman decision in 1972, commutations were granted in approximately one in five death sentenced cases. The current rate is roughly one out of forty.  The increased politicization of the death penalty has meant that a governor could suffer a sharp decline in popularity for granting a commutation. Indeed, of the 31 clemencies granted since 1972, more than half were by governors as they were leaving office. 
Thus, despite the Supreme Court's assurances that clemency exists as a protection against executing an innocent person, it has never been used in Texas, or most other death penalty states, since 1972. As long as the issue of capital punishment is thought of as a litmus test for politicians to attain and retain office, the prospect of clemency for any death row inmate will remain dim.
5. Texas Crime And the State's Response The word 'crisis' is used far too often in politics and government -- but a crisis is precisely what Texas faces today.
The Texas criminal justice system is failing.
--Report from the Texas Office of the Comptroller 
Closely intertwined with the death penalty is the broader response which a state makes to the problem of crime. Not surprisingly, the turmoil exhibited in Texas' administration of the death penalty is reflected in an even larger crisis with crime. In the same period in which Texas moved from its first execution in 1982 to become the undisputed leader in the use of the death penalty, the state also experienced a tremendous growth in its violent crime rate. From 1982 to 1991, the national crime rate rose by 5%. In the same period, the Texas crime rate rose by 24%, and the violent crime rate in Texas rose by nearly 46%. In 1990, Texas earned the dubious distinction of being the first state in which more people died from gunshot wounds than from traffic accidents.  In 1991, Texas' overall crime rate was third in the nation, and its murder rate was the second highest.
But the problems in Texas go far beyond mere crime statistics. A recent report from the Texas Office of the Comptroller pointed to a larger crisis in the state's response to crime:
[D]espite the need for real solutions, public debate over crime in Texas revolves around hollow calls for the state to become "tougher." In fact, this is a call for the status quo -- for more of the same, only more so. It is a call for a continuing cycle of cynical quick fixes and stop-gap measures, for costly prison construction that cannot keep pace with the demand for new prison space -- for a constant drain on state and local treasuries that make Texas taxpayers poorer, not safer.
The death penalty is precisely one of those "quick fixes" that drain the taxpayers' money. A 1992 study by the Dallas Morning News reported that each death penalty case, followed through to the federal appeal, is costing taxpayers $2.3 million. That is in line with the costs that other states have projected. New York estimated that each capital case would cost $1.8 million, without including costs past the direct appeal. Florida calculated the cost of each execution to be about $3.2 million. 
With over 70 executions since 1976 and close to 400 other people waiting on death row, Texas has likely spent several hundred million dollars on the death penalty, far more than it would have if there were no death penalty and people were sentenced to life imprisonment.
As a response to crime, then, the death penalty is exceedingly expensive and focuses on only a tiny fraction of the problem. Nevertheless, politicians throughout Texas have consistently seized on the death penalty as an answer to violence. They have pushed the death penalty at every possible turn and have lashed out at anyone opposing them. But when the causes of crime are rooted in guns, gangs, drugs, and the deterioration of the social fabric, capital punishment offers nothing in the way of a solution.
The Winds of Change
Crime was recognized as a paramount problem in Texas well before national media attention began to focus on crime. In the 1990 gubernatorial race, the candidates tripped over each other in an effort to look tougher in their responses to violence. The death penalty became the leading symbol of toughness. In fact, populist Democrat Jim Hightower described the campaign as "a race to see who could kill the most Texans."  The rapid rise in the pace of executions in Texas also began in 1990, but now dissatisfaction with both the process and the results is starting to emerge.
Jim Mattox, the former Attorney General of Texas who oversaw 36 executions in the state, was one of the candidates for governor who campaigned on his support for the death penalty. But cases like Gary Graham's and Clarence Brandley's, which raised the prospect of innocent people being executed, have given him second thoughts.
For one thing, Mattox doesn't believe the death penalty is a deterrent to crime: "It is my own experience that those executed in Texas were not deterred by the existence of the death penalty law," he said. "I think in most cases you'll find that the murder was committed under severe drug and alcohol abuse." 
As an alternative to the death penalty, he suggests a sentence of life without parole, which other Texas prosecutors have resisted so far: "Life without parole could save millions of dollars," said Mattox. "It currently costs three times as much -- more than $2 million per inmate -- to carry out the death sentence than to keep an inmate in prison for 40 years."
"In other words," he wrote, "it's cheaper to lock 'em up and throw away the key . . . . As violent crime continues to escalate, it's something to consider." 
Others in law enforcement agree. Norman Kinne, First Assistant District Attorney of Dallas County, praised a new Texas law which allowed sentences for life with no possibility of parole for 35 years (now 40) : "I think we can take more violent offenders out of society for longer periods of time with less expense to the taxpayers."
He pointed out that the new law can also bring a sense of finality to the victim's family: "On a death penalty case, I can't ever tell them they won't have to come back and live it all over again. This can go on ad nauseam." But under the new life sentence law, "there's a finality to all this." 
On another occasion he said:
"Even though I'm a firm believer in the death penalty, I also understand what the cost is. If you can be satisfied with putting a person in the penitentiary for the rest of his life . . . I think maybe we have to be satisfied with that as opposed to spending $1 million to try and get them executed."Dr. George Beto, who headed the Texas prison system for ten years, also favors the death penalty in theory but opposes it in practice. He has clearly recognized some of the problems with the application of capital punishment: "[I]n a democratic society like ours, the death penalty is capriciously and inequitably administered. Whether a person is convicted depends on the quality of his defense, the hysteria of the moment in the community and the culture."
And in Washington, some of Texas' Congressional delegation have been leading the way towards alternatives to the death penalty. Rep. Craig Washington (D-TX) has spearheaded the effort to present an alternative federal crime bill which excludes the death penalty and emphasizes a range of positive responses to crime, and Rep. Henry Gonzalez (D-TX) is the perennial sponsor of a constitutional amendment to end the death penalty altogether.
Meanwhile, juries in Texas are also beginning to see things differently, especially with the availability of longer guaranteed sentences. Formerly, criminals in Texas were serving only 20% of their sentence and some of those with life sentences were released after only five years.  Now that life can mean no parole for 35-40 years, juries have real alternatives to a death sentence. Dallas County District Attorneys, for example, used to have a perfect record when seeking the death penalty. But three of the past six capital cases have ended in life sentences. "Sometimes it makes you think the public isn't 100 percent with you," said Assistant District Attorney Hugh Lucas, who recently "lost" a capital case when it ended in a life sentence for Anthony Hampton. 
The Crime Problem The issue of violent crime has now reached national prominence as well. Politicians all over the country have been using the headlines of crime to promote the death penalty as a quick fix solution. If the people buy this promotion as they did in Texas, then it is likely that other states will match Texas' high rate of executions. The federal government, for example, has increased death penalty prosecutions and is seeking ways to greatly expand their role as a response to the national problem of violence. States like New York, Kansas, and Alaska have all been considering reinstating capital punishment.
On the other hand, states that have used capital punishment extensively, like Texas, have been beset with its problems. The death penalty has failed to reduce the number of murders, it has proved enormously expensive, and there continues to be the uncomfortably present danger of executing an innocent individual. As a result of these problems, some states are relying more on the alternative of life sentences with severe restrictions on parole. The political tug of war between more and faster executions on the one hand, and more efficient and effective ways of reducing crime on the other, is a battle raging in the entire nation, as well as in Texas.
Foreshadowing A National Crisis The death penalty in Texas is in a state of crisis. Even more alarming, however, is the prospect that what is happening in Texas will be happening across the country if the U.S. expands its use of the death penalty. The size of the national death row, the willingness of the courts to accept the practices utilized in Texas, the increasing pace of executions, the public's concern about crime -- all indicate that the use of the death penalty could become as common nationwide as it is in Texas.
On the other hand, the problems in implementing the death penalty in Texas are a warning to the rest of the country that it is wading into a swamp that it should avoid. The death penalty skews the process of prosecution and leads to official abuse. The death penalty has also been a symbol of racial division. As the numbers of executions begins to rise, the impact of these injustices will force itself more clearly into our consciousness.
Similarly, the costs of the death penalty are not a problem only in Texas. As thousands of cases move into the later stages of appeal and as more and more people are added to death row every year, the costs will become greater and the strain on other crime fighting programs will become more severe. It is clear even to proponents of capital punishment that this expansion of the death penalty will mean that hard choices must be made between preventive methods of law enforcement and more costly and ineffectual executions.
Furthermore, the crisis in death penalty representation, which is closely related to the problem of costs, augurs poorly for the country as a whole. What is a crisis in Texas because of the numbers involved and the scarcity of qualified counsel willing to take these cases, will become a national problem as the number of inmates approaching execution continues to grow.
Such a death penalty may not be acceptable to the American public. Moreover, such a death penalty may not meet the standards of the High Court, which set this experiment in motion 18 years ago. That experiment, as Justices Marshall and Blackmun have pointed out, has so far established that the death penalty remains arbitrary and capricious. Texas has been the nation's crucible for this experiment with the death penalty, and the results of this experiment should speak volumes to those who choose to listen.
References . Speech at Annual Dinner in Honor of the Judiciary, American Bar Association, 1990, quoted in The National Law Journal, Feb. 8, 1993.
. Callins v. Collins, No. 93-7054, slip opin., at 4 (Feb. 22, 1994) (Blackmun, J., dissenting).
. Projections based on Texas population of 17.66 million and United States population of 259 million. The cost projection was based on a cost of $2 million per execution and a national projection of 1,027 executions. See, e.g., P. Cook & D. Slawson, The Costs of Processing Murder Cases in North Carolina (Duke University, May, 1993), at 1 ("the extra cost . . . per execution exceeds $2 million.")
. See note 1.
. See Millions Misspent: What Politicians Don't Say About the High Costs of the Death Penalty, The Death Penalty Information Center, October 1992, at 13.
. M. Radelet, H. Bedau, & C. Putnam, In Spite of Innocence 71-72 (1992).
. Id. at 133.
. Id. at 134 (emphasis added).
. Texas Penal Code Art. 37.071(b)(2) (Vernon's 1985).
. Since executions were centralized at one state facility in 1924, Texas has not executed any women. There are, however, four women currently on Texas' death row. Since defendants in capital cases are almost exclusively male, the male pronouns will sometimes be used in a generic sense.
. See R. Rosenbaum, Travels With Dr. Death, Vanity Fair, May, 1990, at 142.
. Rodriguez v. Texas, trial transcript, p. 2136 (1978), quoted in J. Marquart, et al., Gazing Into The Crystal Ball: Can Jurors Accurately Predict Dangerousness in Capital Cases?, 23 Law & Society Review 449, 458 (1989).
. See R. Rosenbaum, note 11, at 142.
. Barefoot v. Estelle, 463 U.S. 880 (1983) (amicus curiae brief).
. Marquart, et al., note 12, at 464.
. See id. at 457.
. Although the defense can offer their own experts to refute such claims that the defendant will be a danger to society, they are at a distinct disadvantage for two reasons. First, the defense is severely limited in the funds available for such experts. Secondly, even if a psychiatrist could be found who would predict with absolute certaintly that someone would not be a danger in the future, the jury is less likely to believe that claim since it has already convicted the defendant in the underlying crime.
. R. Suro, Impact of a Pathologist's Misconduct Ripples Through West Texas Courts, The New York Times, Nov. 22, 1992, at 22.
Executions in Texas from 1994–2005 do not deter homicides, contrary to the results of Land et al. (2009). We find that using different models—based on pre-tests for unit roots that correct for earlier model misspecifications—one cannot reject the null hypothesis that executions do not lead to a change in homicides in Texas over this period. Using additional control variables, we show that variables such as the number of prisoners in Texas may drive the main drop in homicides over this period. Such conclusions however are highly sensitive to model specification decisions, calling into question the assumptions about fixed parameters and constant structural relationships. This means that using dynamic regressions to account for policy changes that may affect homicides need to be done with significant care and attention.
In the last decade there is a resurgence of academic studies estimating the possible deterrent effect of capital punishment on homicide rates [2–6] With few exceptions [7, 8] these recent deterrence studies employ non-experimental fixed-effects panel designs that span the period since the reinstatement of the death penalty in the U.S. after the 1976 Supreme Court decision in Gregg v. Georgia, and use ordinary least squares (OLS) or instrumental variables (IV) estimators. Joana Shepherd, author of several of these studies, summarizes the latest econometric findings in her congressional testimony as follows: “The modern studies have consistently shown that capital punishment has a strong deterrent effect, with each execution deterring between 3 and 18 murders” . Another leading researcher in this area, Nai Mocan, the co-author of two of the recent panel studies is quoted in an Associated Press report about the robustness of the deterrent effects of executions, saying “Science does really draw a conclusion. It did. There is no question about it. The conclusion is there is a deterrent effect” .
Like their national, aggregate time-series predecessors , these pro-deterrence death penalty papers have been subjected to considerable academic scrutiny, with critics highlighting numerous conceptual and methodological problems. These include the failure of the research properly mitigating omitted variable bias (e.g., prison population growth), using possibly dubious deterrence ratio variables as proxies for potential homicide perpetrators’ perceptions of the expected costs of committing capital murder, assuming that executions are simultaneously determined (i.e., executions are endogenous events) through the application of the instrumental variables (IV) estimator (and even if that were the case using invalid and unreliable instrumental variables to instrument for execution risk), failing to correct standard errors for the presence of serial correlation, and sampling fragility of the results to alternative periods and outliers [12–16]. Despite several studies addressing these methodological shortcomings, the literature confirms critics’ suspicions that the recent findings were largely a byproduct of the aforementioned criticisms. When these issues are addressed, there is little or no robust empirical evidence of a significant relationship between the presence of the death penalty or increases in execution risk and the homicides rate. The pro-deterrence authors counter that the problems highlighted were either incorrect or inconsequential to their original conclusions [2, 17, 18].
Given this conflicting body of evidence, and the fact that many of the pro-deterrence death penalty research findings have found their way into the policy-making arena, it is not surprising then that the National Research Council (NRC) convened a panel of scholars to reconcile the latest scientific evidence on the deterrent effects of the death penalty. The NRC committee concluded that “research to date on the effect of capital punishment is not informative about whether capital punishment decreases, increases, or has no effect on homicide rates. Therefore, the committee recommends that these studies not be used to inform deliberations requiring judgments about the effect of the death penalty on homicide” .
 recently tackle another limitation of the modern panel studies of the death penalty: How to reliably identify the causal effect of executions on homicide given the sporadic, infrequent, or nonexistent application of executions in the majority of states with active capital punishment statutes. Fig 1 shows that of the 36 states with active death penalty statutes in 2009, only 13 have carried out more than twenty executions since the Gregg decision and 19 states have recorded 10 or fewer executions during this period . What can also be inferred from Fig 1 is that most death penalty states in most years do not execute a single death row inmate. Berk’s  reanalysis of  reveals the highly skewed nature of the execution distribution in state panel data. Specifically, 86 percent of the state-year observations on the execution variable were equal to 0 with another 8 percent of the observations equal to 1. Importantly, only 5 percent of the observations contained execution values larger than 5. Given the relative infrequency of executions in the post-Gregg era, it is no surprise then that one of the main concerns raised by those assessing the sensitivity of the modern statistical evidence on the death penalty has been the suitability of panel studies when only a handful of states account for the lion’s share of the nation’s executions [1, 12, 14]. This is the main conclusions drawn by  after sifting through the recent death penalty literature: “Our key insight is that the death penalty at least as it has been implemented in the United States since Gregg ended the moratorium on executions is applied so rarely that the number of homicides it can plausibly have caused or deterred cannot be reliably disentangled from the large year-to-year changes in the homicide rate caused by other factors.” Berk’s  reanalysis of the state panel dataset of  shows that a few leverage points have unusually large effect on the conclusions about deterrence. Berk shows that their pro-deterrence findings were largely driven by 11 of the most influential observations (just 1 percent of the 1,000 state-year observations), with most of those observations occurring in Texas.
Count of Post-Gregg executions by state as of 2009.
Not surprisingly, Land et al.’s  solution to this low dosage or insufficient treatment problem is to zero in on Texas, a state accounting for 464 of the 1,231 (38 percent) of all persons executed since the Gregg decision [20, 21]. They apply autoregressive integrated moving average (ARIMA) intervention time-series methods  to first-differences of monthly data on executions and homicides in Texas from 1994–2005. They build a series of seasonal linear transfer function models, using first and seasonal differences of both executions and homicides based on the assumption that “short-run” effects of executions needed to be separated from any “long-run” trends (or non-stationarity). Their decision to focus on the post-1993 period is based on U.S. Supreme Court decisions in McCleskey v. Zant and Herrera v. Collins which greatly limited habeas corpus petitions brought on by death row inmates in federal courts. They claim that these court decisions marked the beginning of a regime change in Texas since the number of death row inmates executed from 1994 through 2005, compared to the 1980 to 1993, grew by 300% (from 71 to 284) or from an average of 0.42 executions a month to nearly 2 executions per month. Therefore, the increased frequency and consistency of executions in the post-1993 period makes Texas a fertile test site for examining the deterrent effects of executions.  apply Granger causality tests to assess whether one or two-way causality exists between the two series and find no evidence of Granger feedback from homicide to executions. They build a series of seasonal linear transfer function models. Results for their best fitting model (see, their Table 2, Model 7) reveal a statistically significant reduction of 1.3 homicides in the first month following an execution with additional reduction of 1.2 homicides occurring 4 months later—for a total deterrent effect of 2.5 homicides per execution.
ARIMA transfer function models of Monthly Texas Homicides, 1994–2005.
In our view, Land et al.’s  decision to first difference the execution and homicide series without formally testing for the presence of trends (be they deterministic, stochastic, or seasonal) leads to incorrect inferences about their key results. We argue and demonstrate that differencing possibly stationary or deterministically trending series such as executions and homicides induces false dynamics, making it wrongly appear that a causal relationship exists between the two series because of the false correlations. Once we test for and properly model the data, the subsequent specifications are more parsimonious than those reported by  and strongly resist the interpretation that executions in Texas significantly deter homicides. Perhaps more importantly, we show that the deterrent effects of executions are inflated and largely spurious because of the authors’ failure to control for other important factors coinciding with Texas’ execution binge and decline in homicides—the incapacitative effects of prison growth generally, and an improving economy (proxied with unemployment rates). The results of our analysis do not support the conclusions drawn by . When controlling for changes in the number of persons incarcerated and unemployed, the magnitude of the effects of executions on homicide are less than half the published estimates and are statistically indistinguishable from zero. Consistent with previous research, we find that changes in incarceration were primarily responsible for reducing homicide after the mid-1990s. We employ the latest econometric time-series methods for interpreting time-series results, a set of impulse response functions with error bands that quantify the relative uncertainty of the effects of an execution on the number of subsequent homicides. Finally, we show that these results are robust to different, possibly endogenous changes in parameters using a series of changepoint models as a robustness test.
Monthly Texas Homicides Time Series, 1994–2005
The Land et al.  claims about the deterrent effects of the death penalty on homicides in Texas relies upon a monthly analysis of the relationships of the executions on homicides. Based on data from 1994–2005, they conclude that the use of the death penalty in Texas reduces the short term number of monthly homicides by 2.5 based on ARIMA models with multiplicative seasonal processes. (We thank Ken Land and Hui Zheng for providing us their data.)
Like in most of the modern time series literature, we start with an initial characterization of the time series properties of monthly homicides in Texas. We do this so that one can then know the empirical regularities that need to be captured in building any model that purports to explain or predict aggregate homicides in Texas over recent decades. Despite the necessity to first analyze the dynamic properties of the time series of interest and then incorporate the effects of covariates,  rely on what we consider to be multiple ad hoc model specifications to assess the deterrent effects of executions on homicides. The end result is a set of atheoretical models that fail to account for the dynamic properties of the Texas homicide series and more importantly, induce spurious correlations with executions.
We start with an analysis of the presence of the deterministic trends, stochastic trends and seasonal trends or cycles in the homicides. This is part of the standard Box-Jenkins or dynamic regression framework. The initial goal of this analysis is to capture those parts of the time series of interest so we can separate out any potential correlates of homicide.
Critical to the specification analysis of these models are the determination of the presence of either trend or difference stationarity, with and without drift in the data. A deterministic trend stationary variable, Xt is a function of time, where the regression Xt = βt leads to a description of the trend. A difference stationary series is one where the first differences of the series of interest, say Xt is rendered non-trending by first differencing or using the Xt − Xt − 1 = Yt values of changes as the dependent variable of interest. One then proceeds to model these first differences as a function of their lags and a constant, such as , where c would be the drift or the rate of change in the Xt series and the ϕℓ coefficients capturing the remaining short run dynamics. If the trend model includes a constant, c, this is interpreted as a drift term which indicates the secular shift in the trend over time. Additionally, a time series may include both deterministic and stochastic trends or drift.
Determining the presence of trends is a critical part of ARIMA modeling. The analysis of  focuses on the short term effects, under the presumption that there was a stochastic trend in Texas homicides that needed to be removed. Here we show this is not the case. Specifying a trend or differencing the data when no stochastic trend is present leads to an incorrect ARIMA model and an incorrect estimate of the effects of executions on homicide reduction. We provide details on why this is the case below. We show how this can be corrected and provide proper inferences about this relationship. The results show that executions are not meaningful predictors of the reduction in the number of homicides in Texas. Further, we show that the  findings are a methodological artifact that is not robust to alternative dynamic model specifications and the inclusion of relevant correlates of homicide to mitigate omitted variable biases.
Testing for trends, 1994–2005
To test for trends in the Texas monthly homicide series one can consider any of several tests in the literature. We start with the two of the more common: the augmented Dickey-Fuller (ADF) and the KPSS tests . The ADF test assumes that the time series model is a random walk, or that the model of a time series yt is
yt = c + τtrendt + ρyt-1 + ϵt
where the coefficient ρ = 1 and the error process is Gaussian white noise. If we reject that ρ = 1, then we conclude that the series is stationary. The test is augmented by additional lagged values of the series to account for residual serial correlation. These additional lags are selected by a model fit statistic such as the Akaike Information Criteria (AIC). Further versions of this test can be constructed that test whether the deterministic trend coefficient τtrend or c differ from zero. These versions allow one to rule out stochastic and deterministic drift in the monthly homicides time series.
The KPSS tests are parallels to the ADF tests . Instead of evaluating the null hypothesis that ρ = 1, the null hypothesis is that the series is stationary or ∣ρ∣ ≠ 1. This test also has variants for whether a time trend is included in the model. As with the ADF test, there are short (4 lags) and long (13 lags) corrections to account for serial correlation.
Table 1 presents the results of a battery of ADF and KPSS tests applied to the 1994–2005 Texas monthly homicides series that are the main evidence in . These tests have non-standard test statistic distributions that depend on the sample size and model specification; these are reproduced here for easy exposition. In total nine tests were computed to determine whether there is a trend in the Texas homicide data.
ADF and KPSS tests for unit roots in the Texas monthly homicides, 1994–2005.
Test 1 assesses the null of a unit root with no drift or deterministic trend using the ADF specification with no drift—or that c = 0 and τtrend = 0. Based on the critical values in the right-most columns of the table, the test statistic of -0.94 does not reject the null hypothesis that the series is a unit root or trending variable. Note however that this model includes no drift or deterministic component. These are added in the ADF tests 2 and 3. In both tests the null hypothesis of a unit root is strongly rejected. The drift terms are significant as is the trend term. So we can reject the unit root hypothesis if we include a deterministic trend and/or a constant when modeling the series.
The KPSS results, Tests 4–9, use a different null hypothesis (stationarity and not a unit root, or ρ = 0 under the null), but come to the same conclusions. Tests 4 and 5 reject the null of stationarity, but with only a short lag correction for serial correlation. When longer lagged corrections for serial correlation are added, as in Tests 6 and 7, there is less evidence for rejecting stationarity. Finally, in Tests 8 and 9 where stationarity is assumed under the null with no corrections for serial correlation, the tests strongly reject the null of stationarity (in part because of the unmodeled serial correlation).
There is strong reason to suspect that differencing the data (both first and seasonal differencing) is incorrect. There is only weak to no evidence for non-stationarity or stochastic trends. In fact, the ADF and KPSS tests (Tests 2, 3, and 6) offer scant evidence for non-stationarity once a trend and/or constant are included in the model. So across the most general of the tests (2, 3, and 6) one sees equivalent results: there is no need to difference the data and there may be evidence of a deterministic trend. The reason to prefer these test results is that they are for more general null specifications that include different possible trends and serial correlation corrections. Failing to account for these alternative trends leads to potentially conflicting and erroneous inferences in the ADF and KPSS tests.
We can graphically examine the data and some autocorrelations to see if there are trends in Texas homicides between 1994 and 2005. Fig 2 plots the raw data, the first and seasonal differenced data used in the  analysis and their associated autocorrelation functions (ACF) and partial autocorrelation functions (PACF). Column 1 (2) shows the data and correlations for the raw monthly (differenced) homicides time series. The ACFs show the raw correlations over different lags (measured in terms of the monthly period of the data) and the partial autocorrelations. The ACFs capture the pattern of raw correlations over lags t − 1, t − 2,…, and are used to assess the patterns of decaying autoregressive lags and moving averages. The PACFs capture the lags at higher lags controlling for those at lower lags, and are used to identify patterns of decaying and seasonal serial correlation. Data that are overdifferenced will exhibit serial correlation and moving averages at the period of the differencing .
Texas homicides, first and seasonal differences and associated autocorrelation functions, 1994–2005.
The left (right) column of Fig 2 is the raw (differenced) data and correlation functions. The undifferenced series appears near stationary and has an ACF that decays slowly toward zero across 1.5 periods (18 months), indicative of a seasonally stationary process. The PACF for this series confirms this, with no significant lags after the first 12 months (period 1.0). The differenced series in the second column shows weaker to no autocorrelation but has statistically significant moving average spikes at lags 1 and 12 (at periods 0 and 1.0). These ACF and PACF results for the differenced series are consistent with those for a stationary series that has been overdifferenced. To see this, consider a stationary white noise time series process, or yt = ϵt where ϵt ∼ N(μ, σ2) (where the distributional requirement can be relaxed). If we first difference this series (regardless of the stationary dynamic process for yt) we see:
where we have assumed that θ = 1 and ∇ is a first differencing operator. So we have taking a stationary time series that may or may not have had a moving average component term and included one where θ = 1. This implies an autocorrelation function where the first order autocorrelation is . So even if there is truly first order autocorrelation, it is not eliminated by taking this difference—it is just replaced by a spurious one.
The easiest way to see this time series identification point is to generate a series of independent and identically distributed or uncorrelated random variates and then apply a first and seasonal differencing to them. The resulting ACFs and PACFs will look like those in the second column of Fig 2 with PACF spikes of -0.5 at period 1. Since the generated data are serially uncorrelated, the resulting patterns are the result of overdifferencing the data. This is the same approach used to determine the order of differencing in non-stationary data, since when too many additional differences are taken, the result are excess moving average processes looking like those described here.
Based on these unit root tests, and the ACF and PACF results, it is very unlikely that Texas homicides over the 1994–2005 sample period should be modeled using first and seasonal differences, since there is no evidence of unit roots. In what follows we examine models for the undifferenced homicides’ time series since there is little to no evidence to support the approach taken by Land et al. Our assumption is that the series is stationary with possible changes in deterministic trends or regime.
Identification, specification, and estimation, 1994–2005
We begin, with possible transfer functions for the the dynamics of the explanatory or input variables , analyzing the homicide series. This analysis begins with an investigation of the dynamics of the 1994–2005 Texas monthly homicide series. Based on Fig 2 and the earlier unit root tests, the initial model is a multiplicative seasonal ARIMA model with p autoregressive lags, d differences, q moving average terms, P seasonal, D seasonal differences, and Q seasonal moving averages. This is written as an ARIMA(p, d, q)(P,D,Q)s model where s is the period of the seasonality, in this case s = 12. Since the last section offered no evidence to support differencing, we set d = D = 0.
The initial model for the Texas homicide series ht sample is an ARIMA(3,0,0)(1,0,1)12 with a deterministic linear trend t. This model was selected based on the results of the earlier unit root tests which indicated that there was no stochastic trend (so no need for first differencing), but the possibility of a deterministic trend as a function of time. The ACF and PACF functions provided an initial idea of the serial correlation patterns and lags to be used. These were then refined until white noise residuals were achieved based on Box-Ljung statistics for correlations at various lags up to 16 months. The results of this estimation strategy are (with standard errors in parentheses):
The Box-Ljung statistics for residual serial correlation in indicate no unmodelled serial correlation for lags one through 16. This is an ARIMA model with several interesting features. First, the sum of the first three autoregressive terms is stationary since ∣0.79∣ < 1 and all of the coefficients are positive. The fact that this sum is less than 1 is indicative of a stationary process. The ARIMA results are consistent with the earlier unit root tests and provide evidence that the homicide series should not be first or seasonally differenced prior to analysis. Second, the deterministic trend term t has a p-value of 0.03 or a significant negative trend. This captures the downward trend in the number of Texas homicides over the sample: the mean number of homicides per month is 1994 is 169 while it is 117 in 2005. Third, the seasonal component is stationary with a seasonal AR term of 0.93. A seasonally differenced version of this model is plagued by serial correlation and fits worse than this model. This alternative model’s Box-Ljung statistics indicate serial correlation after lag three. The error variance of the estimate for the seasonally differenced model is 231; it is 224 for the non-differenced data model in Eq (4).
Since this model has no residual serial correlation it is a good basis for inference about the Texas homicides. Any efforts to model this series as a function of other variables has to have them mimic the properties seen in this basic ARIMA model: there is positive serial correlation, strong seasonal effects that decay slowly, and a shift over time in the trend.
Transfer function analysis, 1994–2005
One of the most difficult challenges facing death penalty researchers using either a single time-series or panel designs concerns specifying the timing for when changes in execution risk are supposed to have an effect on the number of homicides. Should changes in execution risk have an immediate or lagged effect on homicides? While it may be difficult in many cases to specify when a change in law or policy is supposed to have an effect on the target variable, in those instances where theory (or an understanding of the intervention) does provide guidance on the matter, it is imperative those candidates be selected and tested . Failure to do so can result in making ex post facto interpretations (which may be true) for unexpected results, thereby making the policy efficacy hypothesis nearly impossible to falsify . As noted over a half a century ago by : “If such time series are to be interpreted as experiments, it seems essential that the experimenter must specify in advance the expected time relationship between the introduction of the experimental variable and the manifestation of an effect” (emphasis added).
The most common intervention point in the single time-series or panel design death penalty literature is a one time period lag of the execution risk measure under the assumption that it takes time for actual changes in execution risk to alter prospective killers perceptions of execution risk. However, we are not aware of any compelling explanation in the deterrence literature as to why it would take several months or up to a year for prospective killers, whether assessing risk consciously or unconsciously, to be reminded of the possibility of being executed for murder. Land et al.  take another approach in the time series policy literature, relying on ARIMA model building methods to identify the best-fitting model (i.e., let the execution and homicide data speak for themselves). They include all lags of executions 1 to 12 in the homicide models and, as noted above, find that the best fitting ARIMA model is one which produces statistically significant coefficients for executions at lags 1 and 4. They interpret the significant coefficients for executions as a deterrent effect of the death penalty and provide a statistical explanation for why the deterrent effect of executions ceases to exist after one month and then picks up again three months later. As far as we know, there is nothing in the deterrence doctrine or any other criminological theory that can even remotely explain the impact patterns they have identified. Put another way, the belief that prospective killers consider execution risk in the subsequent month following an execution, cease to consider risk of execution two and three months later, but then consider risk of execution some four months later is implausible at best, impossible at worst.
We suggest the more theoretically defensible model, within the deterrence framework, is to estimate a dynamic input model that adds contemporaneous and once lagged counts of the numbers of executions as predictors in the homicide model of Eq (4). To the extent that deterrence depends on conscious consideration of risk, more frequent and recent executions should serve as the greatest reminder to prospective killers of the risk of execution. Conscious assessment of risk does not have to be very sophisticated, and thus does not even necessarily entail comparing execution frequency from one month to the next. Conversely, preconscious considerations of execution risk, especially those rooted in fear, might also be influenced by places and times with more frequent executions. While it is impossible to know for sure how execution frequency is treated by prospective killers in risk assessment calculations, assuming they are considered at all, we believe this model offers the most plausible scenario by which deterrent effects via frequent executions is likely to manifest itself.
Table 2 presents the results for the ARIMA model with the contemporaneous and lagged monthly number of executions included in the model. The basic ARIMA model specification is as in Eq (4) and is presented in the first column of Table 2. The only change here is the addition of the regressor(s) for the number of executions. In the table, column 1 reproduces the baseline ARIMA model for the homicide series with ϕj as the autoregressive coefficient for the j′ th lag, t is the deterministic linear time trend and θk is the k′ th lag moving average coefficient.
The contemporaneous effect of executions on homicides is very small. The effect of contemporaneous executions on homicides is −0.93 with a standard error of 0.68. The two-sided p-value for whether this contemporaneous effect in column 2 of Table 2 differs from zero is 0.17. The p-value of the joint effect of the contemporaneous and lagged effects modeled in column three of Table 2 is 0.20. The total effect or cumulative drop in the number of homicides is given by the impact multiplier
So there are four fewer homicides after an execution. The estimate for the model with two lagged values of executions is -7.3 fewer homicides. Repeating the earlier impact multiplier computation for the model with the contemporaneous and lagged executions as regressors yields . These estimated effects are larger than those reported in misspecified ARIMA models in Land et al.
It bears emphasizing, however, that substantively these estimates are very small and uncertain. The standard deviation of the residuals (reported in Table 2) is on the order of 15 homicides. So even the largest effect is within a standard deviation of the residual error of the model. Our argument, however, is that the effects of executions on homicides are highly uncertain and substantively meaningless. There are three reasons for this claim.
First, the p-values for the hypothesis tests for whether the coefficients for the executions variables are different from zero are barely different from the widely and erroneously accepted 5% level of significance. Given the possible model uncertainty and the restriction of the sample to 1994–2005, we should require more significant evidence, even for a one-sided hypothesis test.
Second. these marginal and total effect estimates do not fully account for the uncertainty of the parameters in the ARIMA specification. Accounting for this full uncertainty requires the computation of an impulse response and its error bands. The impulse response traces out the impacts of a one unit change in the number of executions on the dynamic path of homicides. This can be accomplished via a dynamic simulation of the ARIMA model that accounts for the model uncertainty for a single increase in the executions given that the parameter estimates equal those in Table 2 with error variances as reported in the table. This is consistent with the time series econometrics literature described in  and  who document that the best practices for tracing out the dynamic effects of a change to a covariate or a unit shock in a time series model is via a Monte Carlo simulation. Such a simulation (conducted below) accounts for the parameter uncertainty and the sample size of the model when estimating the distribution around the change in executions on the number of homicides over time.
Third, estimating the impulse response function for the effect of a marginal change in executions on the number of homicides per month will allow us to measure and report the possible asymmetry in the effects. The reason this matters is that a decision-maker’s cost function about this research is clearly not symmetric. In the decisions and inferences about deterrence, the cost of a Type I error is not likely to be the same as a Type II error. More bluntly, the cost of an execution relative to a potentially saved life is not the same. The way we can help quantify this tradeoff is to provide information about the full density around the impulse response function which traces out the dynamic impact of a single execution on the homicides series.
To address these concerns, we simulate the impulse responses. This allows us to see both the uncertainty around the effects and over what horizon the responses are observed. An impulse response (function) traces out a path of the homicides assuming one additional execution. With impulse responses one is interested in the size and uncertainty or variance around the response as well as how long it takes for the effects to be realized, based on the estimated time series model. The impulse response effectively shows the time path of the cumulative response or impact multipliers computed above. More importantly, we want to look at the shape and variance of the response. These allow us to assess the likelihood that an execution affects the number of subsequent homicides—an important first step in establishing the deterrent effect of executions.
Impulse responses are best summarized using Monte Carlo simulation since this allows one to construct an error band or standard error estimate around the dynamic path of the effect of an execution . The impulse responses for the effects of executions from the last two models in Table 2 were simulated 5000 times and are summarized in Fig 3. These responses account for both the ARIMA regression parameter uncertainty and the residual variance of the responses over 12 months. Failing to account for the residual variance uncertainty in the simulation yields identical modal impulse response estimates. But it will lead to understating the uncertainty by a by nearly a factor of 20 because it fails to model the uncertainty from the estimate of σ2 for each model. Once this residual parameter uncertainty is included (correctly) in the Monte Carlo simulation, the resulting estimates are those in the figure. The modal estimates are presenting using the solid line while the 68% confidence region, or approximately one standard deviation confidence region of the response is plotted using dashed lines.
Impulse Response Functions for One Execution for Models in Table 2.
These impulse responses trace out the change in the number of homicides for an additional execution based on the models in the final two columns of Table 2. As the figures show, the total estimates over twelve months match those predicted: within 12 months the first transfer function model predicts 3.34 fewer homicides (the cumulative sum over the 12 months, out of a total of 4.28 fewer) and the second transfer function model predicts 7.02 fewer homicides (the cumulative sum over 12 months, out of a total of 7.3 fewer). But these effects are not statistically or substantively meaningful because the error bands on the responses cover zero over the full time horizon. So while there is an effect, it is swamped by the residual variation in the monthly Texas homicide series. We also computed Bayesian-shape error bands that account for the serial correlation in the uncertainty of the impulse response estimates . While these error bands are smaller than those in Fig 3 the same conclusions still hold.
The conclusion here is straightforward: there is no evidence in this analysis or that of  that can be used to offer evidence for executions having a policy-relevant, deterrent effect on homicides in Texas over the 1994–2005 period. As a preliminary conclusion, the results of  are highly sensitive to model specification assumptions.
Based on the estimates in Table 2 and the impulse responses in Fig 3 the estimated effects of executions do not have a deterrent effect on the number of monthly homicides in Texas. Our claim is based on the fact that once we account for the residual variation of the fitted ARIMA models, it is hard to say that the effects of executions on homicides are substantively meaningful. This is because the ARIMA model only accounts for about 58% of the observed variation in the homicides series, per Table 2.
So one might ask, how could a better fitting time series model be constructed? Could the the effects of execution be more clearly revealed? This is mainly a call to advance inquiry beyond the bivariate analysis to the inclusion of relevant controls to reduce the risk of omitted variable bias and spurious inference for the execution variable. Such a criticism has been made recently by .
To evaluate this possibility, we specified additional models that included two additional covariates. The first is the (monthly) number of tens of thousand of prisoners in Texas prisons. These data on the on-hand offender population by month were obtained from the Texas Department of Criminal Justice (TDCJ) through a public information/open records request on September 3, 2010. Several studies such as  and [32, 33] suggest that controlling for prison expansion in the U.S. and Texas is a necessary covariate for analyzing the drop in crimes seen across the U.S. in this period. During the study period covered by  (i.e., 1994 to 2005), the number of people incarcerated in Texas prisons increased dramatically from 62,322 to 151,925 or by 143 percent. A sophisticated multivariate panel analysis of Texas counties finds that most of the drop in violent and property crime in Texas during the 1990s can be attributed to increases in the number of offenders in jail and prison . It is worth noting that  explicitly controlled for executions in both the violent and property crime specifications and found no evidence of a deterrent effect of executions on either crime type at occurring at the county-level. Of course, the results for the execution variable may have differed if homicide had been used as the dependent variable. [32, 33] also cites prison growth as one of the leading causes of the crime decline, especially for homicide, experienced in the U.S. during the 1990s [31, 35, 36].
The second covariate is the first differences of the Texas unemployment rate. This monthly-level unemployment data for Texas were obtained from the Bureau of Labor Statistics website on April 18, 2012. We use differences for the unemployment data because they have a unit root. Failing to do so would lead to an unbalanced specification with a trend, or a spurious regression. Most macro-structural theories of crime contend that improving economic conditions should result in less crime, although the relationship between unemployment and crime is admittedly tenuous, at best [37–39]. Many of these same theories, however, predict a criminal justice system that is less punitive when the economy is improving as unemployed/marginal workers are perceived as less threatening and needed potential laborers [40–43]. This was clearly not the case in Texas or anywhere else in the U.S. throughout most of the 1990s and early 2000s. Indeed, in most years between 1994–2005 unemployment rates were lower than the previous year’s levels while incarceration rates and the number of persons executed were generally higher than the previous year. This means the deterrent effect identified by Land et al. may be a spurious correlation since Texas’ decision to increase the use of the death penalty coincided with an improving economy.
Again, we restrict the analysis to the critical 1994–2005 time period for comparison to the results in . We use the earlier ARIMA specification and look at the effects of adding these additional covariates on the results. Table 3 presents the results with these new covariates. The effects of an additional execution are smaller here: the estimated coefficient is now −1.31, but the dynamic multiplier is . So the effects of additional executions are mitigated when we control for the number of prisoners (in 10000s) and the change in unemployment. Each additional 10000 prisoners lowers the number of homicides by nearly 9, or a total effect of 10.4 fewer homicides. So changes in incarceration alone predicts a larger drop in homicides than executions. Also, the effect of changes in unemployment is not a predictor of homicides in Texas. Note also that this last model is superior to second column of results in Table 2, since a likelihood ratio test comparing them has a value of 18.2 (p-value < 0.01). We also considered specifications using changes in the number of prisoners month-over-month and year-over-year. Neither of these rival specifications fit as well as that reported here.
ARIMA transfer function models of Monthly Texas Homicides with additional covariates, 1994–2005.
The results of this section confirm a rather obvious result: there are omitted variable biases in the ARIMA specification. That is, we can easily improve the fit of the homicide time series model in Eq (4) by adding strongly suggested covariates such as the change in the number of prisoners. As in most omitted variable problems, the marginal effects of substantively significant variables changes radically when different specifications are examined. So adding a variable that a) is predicted to negatively be correlated with the homicides and b) positively correlated with executions (since more prisons mean there can be more people sentenced) will attenuate the effect of executions. The number of monthly homicides is correlated with the number of 10000 prisoners at −0.7 and the number of executions is correlated with the number of 10000s prisoners at 0.19.
Simply put, bivariate time series studies of the death penalty, no matter how statistically sophisticated, are virtually worthless for evaluating the deterrent effect of executions as they cannot rule out any alternative explanations for changes in homicide, a point recently and cogently made by .
Possible causes of the sensitivity of these results are that there are time-varying or omitted structural changes in the homicides series and the other variables. These could include the sanctions regime discussed in . The issue is that given the discussion in , we are unsure of when the sanction regime changes in Texas. One way to evaluate the possible sanctions regime effect is to employ a model that allows for structural changes in the regression parameters. The Bai and Perron (1998)  model is commonly used to detect such structural changes. The model assumes a regression with m structural breaks (or m + 1 regimes) with the following specification
for j = 1,…,m + 1. Here, the dependent variable time series yt is explained by a set of fixed regressors and a set whose effects change over the m + 1 regimes. The effects of regressors xt are not time-varying, while the effects of zt do depend on the changepoints or regimes. The indices for the breakpoints are T1,…,Tm, and define when the changepoints occur (assuming that T0 = 0 and Tm + 1 = T). The timing of the breaks or the values of the Tj are assumed unknown and need to be estimated.
For each partition of the sample into (T1,…,Tm)—meaning a split of the sample into the m + 1 regimes—we minimize the sum of squared errors,
The estimator then selects the optimal split of the sample into the estimated breakpoints that minimizes the sum of squared errors. The optimal number of breakpoints is selected based the minimum the Bayesian Information Criteria (BIC) that accounts for fit, but penalizes using too many parameter or regimes. We examine dynamic regression model specifications with zero to five breaks and report those that minimize the BIC for a given specification of regressors.
We begin with an admittedly underspecified breakpoint analysis of the full monthly homicides series, covering from 1980(1)–2009(8). For this series, regressed only on a constant (so this is a simple changing means model), the optimal number of breakpoints is three, at 1984(12), 1990(4), and 1994(11) (results not reported). One problem with this initial specification is that it does not capture the autoregressive or seasonal components of the homicides series. Note the Land et al. sample comes after each of these breakpoints. But this sets a simple upper bound on the number. Including covariates for the dynamics of the homicides, the number of prisoners, the number of executions, etc. should explain some of these breakpoints or change the dates of their estimated locations.
Table 4 presents the dynamic regression model with one break. This two regime model optimally splits the sample at 1994(10) and accounts for the autoregressive dynamics of the homicides series (the residuals for this model are white noise). Note that the intercept drops in the latter period (after 1994) when the number of homicides falls. Also, the AR(1) process weakens (the coefficient drops from 0.42 in the first regime to 0.20 in the second regime).
Changepoint model for monthly Texas homicides, 1980(1)–2009(8).
Next, we include the covariates in the breakpoint model. We include here the year-over-year percentage change in prisoners (since the levels of prisoners are possibly non-stationary) and the number of executions at time t and t − 1, as before. We also looked at models with only the change in prisoners or only the executions, which produce similar results. We only have the prisoners’ data from 1985(3), so the sample for the change specification (year-over-year) begins in 1986(3). The optimal breakpoint model with this specification is reported in Table 5
Changepoint model for monthly Texas homicides with covariates, 1986(3)–2009(8).
The optimal number of breakpoints for this specification is one and it splits the sample at 1992(1)—prior to the date of Land et al.’s sample. The dynamics of the homicides series (seen in the AR coefficients) are remarkably similar before and after the breakpoint. Note that the effects of the main covariates are contrary to expectations: executions increase the number of homicides prior to 1992 and are not statistically significant predictors after 1992. This further supports that argument that the specification decisions and lack of structural identification make it difficult, if not impossible to make clear claims about the relationships between executions and homicides in Texas over this sample period.
Our conclusions about the effect of executions on homicides in Texas from 1994–2005, as reported in  are very sensitive to model specification decisions and samples.
First, changing from a model of first and seasonal differences to one with a pure autoregressive specification leads to one that is more parsimonious and better explains the dynamics of the Texas monthly homicides over this stretch. This is critical, since without a correct dynamic specification of the basic ARIMA process, the subsequent analysis of the effects of executions on homicides will be incorrect.
Second, using the improved time series specifications, we see that the results about the effects of executions are very sensitive to model specification decisions vis-a-vis those made in . When we change the dynamic specification we get an estimated effect that is larger than that reported previously, but also more uncertain. Using a standard interpretative method to assess the dynamic effects, an impulse response function with error bands to ascertain the full uncertainty of the model and its estimates, indicates that the effects of executions on homicides are no different than random chance. So while the impact multiplier is negative and weak evidence for the executions deterring homicides in Texas, the results are not unambiguous.
Third, the results are called into question when we add two highly plausible covariates to the analysis: changes in unemployment rates and the number of prisoners in Texas prisons. Adding these covariates to the analysis shows that the incarceration of more criminals predicts a larger drop in the number of Texas homicides than the effect predicted by an execution. But even with these additional controls, the large residual variance swamps the estimated dynamic effects. So even for this updated model, there is little to no evidence that strongly suggests that either incarceration or execution changes substantively the number of homicides in Texas between 1994 and 2005.
Fourth, the changepoint model results in the previous section show that the critical changes come prior to 1994 and in fact do not work as Land et al. suggested. Rather we see that the effects for executions are weak and not significant by any conventional expectation. Further, the lack of identification makes determining the effects of these deterrence and incapacitation variables quite hard.
What substantive conclusions should be drawn from these results? The initial conclusion should be that research purporting to find effects for executions on homicides are likely suspect because of issues involving model underspecification and interpretation. One typically likes to interpret the marginal impacts of single covariates, but this alone does not account for all of the full dynamic effects or the potential model misspecification. Using the impulse analysis methodology employed here does a much better job of this.
Further there is no good identification strategy for disentangling the possible deterrent effects of executions on homicides . At best, what is presented here suggests that there might be some weak correlations. At worst, it shows that the inclusion of different covariates or model specifications lead to dramatically different and highly sensitive results. This is not a good basis for making policy projections or decisions since there is no evidence that the effects of the covariates like executions on the homicides variable are robust to different specification or identification assumptions. While one can continue to suggest new or additional covariates, the limited degrees of freedom for the homicides series means that there need to be more theoretical work ahead that can define better data analysis strategies.
Finally, we caution that these results are based on what can be described as good, but not exhaustive time series data analytic techniques. Employing ARIMA and changepoint models we are making explicit assumptions of the endogeneity of the homicides series and the exogeneity of the other covariates (the input series). This is typically hard to defend in many social science applications . So future empirical examinations should employ methods that look at the possible dynamic, endogenous nature of the aggregate homicides time series with other predictors, rather than making indefensible strict exogeneity assumptions.
Brandt’s research has been funded in part by NSF grant SES-0921051. The authors are responsible for any errors. Replication data and code in R are available as part of the Supplemental Replication zip archive S1 Replication Archive files for this paper.
The authors have no support or funding to report.
All relevant data are within the paper and its Supporting Information files.
1. Land K, Teske R, Zheng H (2009) The Short-Term Effects of Executions on Homicides: Deterrence, Displacement, of Both?Criminology47: 1009–1043. doi: 10.1111/j.1745-9125.2009.00168.x
2. Dezhbakhsh H, Rubin PH, Shepherd JM (2003) Does capital punishment have a deterrent effect? new evidence from postmoratorium panel data. American Law and Economics Review5: 344–376. doi: 10.1093/aler/ahg021
3. Dezhbakhsh H, Shepherd JM (2006) The deterrent effect of capital punishment: Evidence from a “judicial experiment”. Economic Inquiry44: 512–535. doi: 10.1093/ei/cbj032
4. Mocan HN, Gittings RK (2003) Getting off death row: Commuted sentences and the deterrent effect of capital punishment. Journal of Law and Economics46: 453–478. doi: 10.1086/382603
5. Shepherd JM (2004) Murders of passion, execution delays, and the deterrence of capital punishment. Journal of Legal Studies33: 283–321. doi: 10.1086/421571
6. Zimmerman PR (2004) State executions, deterrence, and the incidence of murder. Journal of Applied Economics7: 163–193.
7. Cloninger DO, Marchesini R (2006) Execution moratoriums, commutations and deterrence: the case of Illinois. Applied Economics38: 967–973. doi: 10.1080/00036840500462020
8. Hjalmarsson R (2009) Does capital punishment have a ‘local’ deterrent effect on homicides?American Law and Economics Review11: 310–334. doi: 10.1093/aler/ahn004
9. Subcommittee on Crime Terrorism and Homeland Security (2004). Hearing before the Subcommittee on Crime, Terrorism, and Homeland Security to consider H.R. 2934, the Terrorist Penalties Enhancement Act of 2003, to strengthen penalties for terrorist offenses. 108th Congress, Second Session. 24.