The appellant was convicted in April 1998 of capital murder.⁽¹⁾ Pursuant to the jury's answers to the statutory special issues,⁽²⁾ the trial judge sentenced the appellant to death.⁽³⁾ Direct appeal to this Court is automatic.⁽⁴⁾ The appellant raises eleven points of error. We shall affirm the judgment of conviction, reverse the judgment of death, and remand the case to the district court.

In point of error four, the appellant contends that the evidence is legally insufficient to show a kidnapping, the underlying offense that aggravates this murder into a capital murder.

The appellant confessed the following facts to police officers: The victim, Larry Nelson, made sexual remarks concerning the appellant's cousins, Jennifer and Lynnette Ovalle.⁽⁵⁾ The appellant, Nelson, and the girls then went into Alfredo Chavez's house. After Nelson commented that he intended to have sexual intercourse with the appellant's cousins, a fight ensued. Nelson punched the appellant; the appellant punched Nelson. The appellant dragged Nelson outside, kicked him in the head three or four times, helped drag him into a truck, and drove away in the truck with Jennifer. The appellant and Jennifer stopped at the appellant's house, where he retrieved a knife. They then drove to Lake Halbert, where the appellant pulled Nelson out of the truck and stabbed him several times, and Jennifer ran over Nelson with the truck.

The appellant's cousin Lynnette testified that, before arriving at Chavez's house, she had told the appellant that Nelson had been flirting with her. Inside the house, Nelson asked Lynnette if she thought "he had a chance with" Jennifer "because he was going to ask her out." The appellant punched Nelson in the face, and he fell to the ground. Lynette later saw the appellant outside the house stomping Nelson. Nelson was unconscious at that point. The appellant and Chavez loaded Nelson into the truck, and the appellant and Jennifer drove the truck away.

Jamie Rendon testified that he saw Nelson on the ground, and he saw the appellant drag the victim outside. Chavez tried to restrain the appellant, who was hitting Nelson. The appellant, Jennifer, and Chavez loaded Nelson into the truck, and the appellant and Jennifer got into the truck and drove away.

Chavez testified that the appellant punched Nelson, who fell to the floor, and the appellant stepped on him. Then the appellant pulled Nelson outside and started kicking him. Chavez attempted to restrain the appellant and eventually succeeded in persuading the appellant to stop hurting Nelson. The appellant and Chavez loaded Nelson into Nelson's truck, and the appellant and Jennifer got into the truck and drove away.

Nancy Rodriguez, a close friend whom the appellant called "Mom," testified that the appellant and Jennifer appeared at Rodriguez's house on the morning of the murder. The appellant told Rodriguez that he had killed someone and that he had used a knife to slit the person's throat. Rodriguez noticed blood on the appellant's hands.

Finally, a medical examiner testified that Nelson was stabbed with an instrument that was consistent with a knife, that he was alive at the time the stab wounds were inflicted, and that the stab wounds were lethal.

In evaluating legal sufficiency, we view the evidence in the light most favorable to the verdict and determine whether any rational trier of fact could have found the elements of the offense beyond a reasonable doubt.⁽⁶⁾ The question here is whether a rational jury could have concluded that the appellant kidnapped Nelson. The evidence shows that the appellant beat Nelson into unconsciousness, loaded him into the truck, and drove away with him. Instead of driving Nelson to a hospital for medical treatment, the appellant drove him to a lake and slit his throat. The evidence is sufficient to show that the appellant kidnapped Nelson. Point of error four is overruled.

In point of error one, the appellant contends that the trial court erred in refusing to grant his motion to set aside the indictment due to the composition of the grand jury. The appellant contended before the trial court, as he does on appeal, that persons of Hispanic origin have been systematically excluded from grand juries in Navarro County in violation of the Equal Protection Clause of the Fourteenth Amendment. He relies heavily upon Castaneda v. Partida, 430 U.S. 482 (1977).

In Partida, the Supreme Court held that an equal protection violation occurs when the government purposefully excludes certain identifiable groups from serving on a grand jury.⁽⁷⁾ A rebuttable presumption of purposeful discrimination arises if the defendant makes out a prima facie case, which consists of showing: (1) that a particular group constitutes "a recognizable, distinct class, singled out for different treatment under the laws, as written or as applied," (2) "the degree of underrepresentation by comparing the proportion of the group in the total population to the proportion called to served as grand jurors, over a significant period of time," and (3) "a selection procedure that is susceptible of abuse or is not racially neutral."⁽⁸⁾ Once a prima facie case has been made, the burden shifts to the State to rebut that case.⁽⁹⁾

We will assume, without deciding, that persons of Hispanic descent constitute a recognizable, distinct class singled out for different treatment.⁽¹⁰⁾ Texas law permits grand jurors to be selected using either a commissioner-based, "key-man" system or the random selection system used to select civil trial juries.⁽¹¹⁾ Navarro County uses the commissioner-based system. The Supreme Court has held that the commissioner-based system, while facially constitutional, is susceptible to abuse.⁽¹²⁾

We focus, then, on the second element of the prima facie case -- the degree of underrepresentation, determined by comparing the proportion of the population composed by the group to the proportion called to serve as grand jurors over a significant period of time. In Partida, the Supreme Court characterized the disparity as a 40% difference.⁽¹³⁾0.Id. at 486-87, 495. We note that not all persons of Spanish descent are Mexican-American. Apparently, the parties in Partida focused upon Mexican-American heritage although the Supreme Court's analysis in fact embraces a larger group of Spanish-speakers.⁽¹⁴⁾ The Court found this difference to be greater than that in many other cases in which disparities were found to be significant.⁽¹⁵⁾

The Supreme Court also conducted a statistical analysis, using statistical decision theory, to explain why the disparity gave rise to an inference of discrimination.⁽¹⁶⁾0.Partida, 430 U.S. at 496 n.17.⁽¹⁷⁾ The Court explained that "if the difference between the expected value and the observed number is greater than two or three standard deviations, then the hypothesis that the jury drawing was random would be suspect to a social scientist."⁽¹⁸⁾ The Court found that the result in its case exceeded 29 standard deviations -- a far greater variation than what would be expected by chance.⁽¹⁹⁾ The Court concluded that the defendant had established a prima facie case of discrimination against Mexican-Americans in Hidalgo County.⁽²⁰⁾

The appellant in this case contends that no person of Hispanic origin served on the grand jury that indicted him and that only eight persons of Hispanic origin have served on grand juries in Navarro County during the past ten years. In support of his position that systematic discrimination occurred, he submitted to the trial court: (1) United States Census Bureau statistics from the 1990 Census showing the percentage of persons of Hispanic origin in Navarro County and (2) 41 grand jury lists, including the grand jury that indicted him and the preceding 40 grand juries (encompassing a ten-year period).

The Census figures show that persons of Hispanic origin composed 7.2% of Navarro county (at the time of the 1990 Census).⁽²¹⁾ The appellant has made no effort to show what percentage of persons eligible for grand jury service are of Hispanic origin.

The appellant does not explain how he arrives at his number of eight Hispanic persons. During trial, the prosecutor and the trial judge discussed which persons might be Hispanic. The prosecutor remembered that "Mr. Gillen" was Hispanic, even though his surname did not appear to be of Hispanic origin. Noting surnames, the trial court listed Andy Garcia, Thomas Averia,⁽²²⁾ Ricardo Ferrer, Emmitt Gonzales, and Edmond Flores as "probably Hispanic" and Glenda Ruiz as "either Hispanic or married to a Hispanic." The trial court also listed the following as people who might be Hispanic: David Commiato, Billy Flores, Kenneth Uresti, and Joe Alan Honea. But except for "Mr. Gillen," the trial court's only basis for determining possible Hispanic origin was the person's surname. The court commented that although Spanish surnames could give one an indication of whether a person was of Hispanic origin, it could not do so with finality. The court also questioned what "Hispanic" meant, noting that he had heard that in some countries where Spanish was the language, some of the residents spoke an Indian language rather than Spanish.

The prosecutor then testified that he had presented every case to the grand jury since 1975 and that surnames were not always a reliable indicator of whether a person was of Hispanic origin:

There are many people that have served on the Navarro County Grand Juries that don't have Hispanic or Spanish sounding last names, many women who are married to Anglos that have served on the jury, and many Hispanics that, what we might think of as a Spanish sounding name is not that at all. Many people that have Spanish sounding names that aren't Hispanic.

The trial court overruled the appellant's motion to quash.

We observe initially that use of general population figures and Spanish surnames involves a degree of uncertainty. The population of persons eligible to serve on grand juries is a better baseline figure from which to gauge discrimination than the general population.⁽²³⁾ Using general population figures requires making the assumption that they are similar enough to the eligible population figures to yield an accurate comparison. In essence, one must assume that groups ineligible for jury service, such as children and resident aliens, do not compose a disproportionate percentage of the Hispanic population as compared to the non-Hispanic population. Given that Texas borders on Mexico, however, there is at least some reason to believe that Census figures for Hispanics may include more resident aliens than would be true of non-Hispanics, causing the figures for Hispanic population to overstate the percentage of the population that is actually available for grand jury service. Whether such is the case in Navarro County is unknown to us since the statistics in that regard have not been presented. Uncertainty also inheres in the use of Spanish surnames to determine Hispanic origin.⁽²⁴⁾0.Ibid.⁽²⁵⁾0.Id. at 2. Word and Perkins defined "householder" as male householders or never-married female householders plus any other male or never-married female in the household not related to the householder. Id. at 2 n.2. Narrowing the inquiry to householders was designed to mitigate the effects of clustering (people in the same household having the same surname and ethnicity) and the problems with calculating the ethnicity of never-married females. Id. at 2 and 2 n.2.⁽²⁶⁾

Thus, the inquiry that the appellant wishes us to make is filled with uncertainty. We are not faced with an established proportion of Hispanic persons serving on grand juries and the mere problem of attempting to determine whether that proportion could likely have occurred by chance. We are faced with the additional problem of determining just how many Hispanic persons actually served and how many we should expect to have served. What surnames are Hispanic? To what degree do Hispanic surnames identify persons who actually are Hispanic? What proportion of the eligible population is Hispanic?

In Partida, the Supreme Court accepted the legitimacy of using general population statistics and Spanish surnames to establish a prima facie case of discrimination. That acceptance must be considered, however, in light of the facts of that case -- a county with a very high population of Hispanic persons and a very large disparity between the percentage of Hispanic persons in the population and the percentage of Spanish surnamed individuals serving on grand juries. At the time Partida was litigated, one would have expected, based upon general population figures, that a grand jury in Hidalgo County would contain, on average, nine Hispanic persons out of the total of the twelve that serve.⁽²⁷⁾ Even if the eligibility figures were different from the general population figures and even if surnames underrepresented the Hispanics that served on the grand juries, those differences were unlikely to account for the massive 40% gap that was observed. For Navarro County, however, one would not expect that any particular grand jury contain even a single Hispanic person because the average expected number of Hispanic persons per grand jury, based upon general population figures, is less than one.⁽²⁸⁾ In such a county, even small variances produced by the uncertainties mentioned earlier could significantly affect the observed disparity between the proportion of Hispanics in the population and the proportion that served on grand juries. We must take into account not only the statistical probability that an unequal distribution could have occurred by chance, but we must also consider the potential inaccuracy of the statistics we are using as a basis for comparison.

With the above observations in mind, we turn to the question of whether the appellant has made a sufficient showing of statistical disparity to satisfy the prima-facie-case requirement under Partida.

We begin with determining the appropriate size of the grand jury sample. The appellant claims that we should use the 500 names provided. An examination shows, however, that every grand jury list is typed and contains twelve names except for the list relating to the appellant's indictment -- which list is handwritten and contains twenty names. But a statute has required, during the entire ten-year period in question, that the commissioners select fifteen to twenty persons to be summoned as grand jurors.⁽²⁹⁾ From that number, the first twelve qualified persons are impaneled on the grand jury.⁽³⁰⁾ We will assume that the usual practice was followed, and that more jurors were summoned on the first forty lists, but only the jurors who served were included in the typewritten lists. It follows also that eight of the twenty names on the handwritten list relating to the appellant's indictment did not in fact serve on the grand jury. Because we have no record for any of the prior forty grand juries of the persons who were summoned but did not serve, consistency dictates that we disregard the last eight names on the list for the appellant's grand jury. Striking those eight names brings the total sample size to 492 names.⁽³¹⁾

Using the general population figures, one would expect about thirty-five Hispanic persons to have served on the forty-one grand juries for which lists were provided.⁽³²⁾ The standard deviation is approximately six.⁽³³⁾

If the discrepancy between the expected number and the actual number of Hispanics serving on Navarro County grand juries is less than three standard deviations, as Partida suggests, the appellant fails to establish a prima facie case. That is, if the actual number of Hispanics who served on grand juries is eighteen or more, the appellant has failed to establish a prima-facie case of purposeful discrimination.⁽³⁴⁾

We find at least nineteen grand jurors who were probably Hispanic.⁽³⁵⁾ The difference between that figure and the expected value of thirty-five is less than three standard deviations. The appellant has failed to establish a prima-facie case of purposeful discrimination. Point of error one is overruled.

In point of error two, the appellant contends that the trial court erred in denying his motion for mistrial after the State improperly impeached the appellant's mother. The State questioned the appellant's mother concerning a prior theft conviction, but she had been placed on probation for the theft and had successfully completed her probation:

Q. Mrs. Ovalle, isn't it a fact that on February the 28th of 1990, you were sentenced to four years in the Texas Department of Corrections for felony theft?

A. No. I was put on probation. Thank you.

Q. Four years probation for felony theft?

A. Yes.

[DEFENSE COUNSEL]: Excuse me, what was the date on that?

THE COURT: 1990 is what she said.

[DEFENSE COUNSEL]: I'm going to object to that your Honor. It was four years probation and this individual has finished out the probation, then that would be inadmissible.

[PROSECUTOR]: She didn't, our records show--

[DEFENSE COUNSEL]: Excuse me. I'm going to object to that, your Honor.

THE COURT: Ladies and gentlemen, go back in the jury room.

Further questioning outside the presence of the jury revealed that the appellant's mother had been placed on probation, that her probation had been completed, and that her probation was never revoked.

The trial court sustained the appellant's objection on the basis of Rule of Evidence 609(c)(2). The court gave the jury an instruction to disregard:

Ladies and gentlemen of the jury, that last witness was asked a question by the District Attorney which was based on good faith on information, good faith information on their part, but it turned out to be incorrect. So I'm instructing you that you cannot consider the last question of Vicki Ovalle for any purpose, don't consider it for any purpose.

The trial court denied the appellant's motion for mistrial.

The State concedes that the question was improper but contends that the trial court did not err in denying the appellant's motion for mistrial because the State's error in asking the improper question was harmless. We agree that the trial court did not abuse its discretion by refusing to grant a mistrial. Ordinarily, a prompt instruction to disregard will cure error associated with an improper question and answer, even one regarding extraneous offenses.⁽³⁶⁾ In a case similar to the present one, in which the State improperly impeached a witness with a crime for which probation had expired, we held that a trial court's instruction to disregard, coupled with other circumstances, rendered the improper impeachment harmless.⁽³⁷⁾ In the present case, the trial court promptly instructed the jury to disregard the State's question and even told the jury that the State's information was incorrect.

Moreover, other circumstances support the conclusion that the improper impeachment was harmless. The only disputed issue in the appellant's mother's testimony was her statement that the appellant had requested an attorney before signing his written confession. If the appellant had requested an attorney, and the State failed to honor that request, then the confession might have been inadmissible. Two police officers testified at trial that the appellant shook his head "no" when he was asked if he wanted an attorney. The appellant's father and another relative testified that, while the appellant's family urged him not to sign the statement and the appellant's mother told officers that he had an attorney, the appellant said nothing during the discussions except to ask whether he should sign the statement.

The trial court admitted the appellant's written confession, but a jury issue concerning its legality was submitted. The jury was instructed that it should disregard the confession if it found that the appellant had not been given one of the required warnings (including the right to counsel) or if it found that the appellant had not knowingly, intelligently, and voluntarily waived his rights.

Because the trial court sustained the appellant's objection, we will presume the inadmissible impeachment did not influence the trial court's decision to admit the evidence.⁽³⁸⁾ Of course, the jury was free to believe or disbelieve any of the conflicting evidence presented concerning the confession's admissibility. For several reasons, we think it unlikely that the impeachment had any impact upon the jury's decision in that regard.

First, to the extent that the jury was informed that Mrs. Ovalle had committed a crime, they were also informed that she had been placed on probation and the probation had been completed. The discussion of this offense was brief, the details of the offense were not given, and the trial court instructed the jury not only to disregard the State's question, but also that the State's information had been incorrect. The impeachment evidence does not seem so inflammatory that the jury could not have followed the trial court's instruction to disregard.

Second, given the evidence, Mrs. Ovalle's testimony was comparatively weak. She was the only witness who testified that the appellant asked for an attorney, and as his mother, she was a very interested witness at that. The State also impeached Mrs. Ovalle by showing that, at the time the police were attempting to persuade the appellant to sign the confession, she was angry with the authorities -- and walked out of the room when things were not going the way she wanted. And, the appellant's other two relatives to testify on the matter -- both defense witnesses, including the appellant's father -- contradicted Mrs. Ovalle's version of the events. A witness's testimony may be found to have negligible impact upon the trier of fact when it is found to be inconsistent with most of the other evidence, including evidence from the defense, and there are other reasons to doubt the witness' credibility. Burks v. State, 876 S.W.2d 877, 906 (Tex. Cr. App. 1994), cert. denied, 513 U.S. 1114 (1995) (proposed defense witness' testimony would have had negligible impact upon the trier of fact given inconsistencies with the testimony of other witnesses and the statements of the defendant, as well as questionable credibility of the declarant who was drunk at the time statements were made).

Finally, even if the jury had concluded that the appellant's confession should be disregarded, the appellant's guilt was supported by evidence that was so overwhelming that we can conclude that the jury's verdict would not have been affected. We have previously held that a defendant's confession to a relative who was emotionally close to the defendant, coupled with strong physical evidence, could be so overwhelming as to render harmless beyond a reasonable doubt a confession to law enforcement authorities that was erroneously admitted.⁽³⁹⁾ In the present case, three eyewitnesses, including one of the appellant's relatives, testified that the appellant beat the victim into unconsciousness and then drove away with him in a truck. The appellant confessed to a friend so close to him that he called her "Mom" that he had killed the victim with a knife. During this confession the appellant had blood on his hands. He was with Jennifer, who had accompanied him as he drove away with the victim. Furthermore, the medical examiner's testimony showed that the victim was killed by stab wounds from an instrument that could have been a knife. And here the jury was entitled to hear the confession. Since the jury would have heard the appellant's confession regardless of the mother's testimony, any inflammatory effect the confession might have had on the jury would not have been avoided even if the erroneous impeachment had never occurred. Point of error two is overruled.

In point of error three, the appellant contends that the trial court erroneously refused a requested punishment charge concerning the consideration of evidence as it relates to the future dangerousness special issue.⁽⁴⁰⁾ The jury charge provided:

In deliberating on Special Issue No. 1 you shall consider all the evidence at the guilt or innocence stage of this trial, including evidence of the defendant's background or character or the circumstances of the offense that militates for or mitigates against the imposition of the death penalty.

At trial, defense counsel objected to the instruction as being incomplete:

Your Honor, on the first page, the last paragraph where it says (reading) "you shall consider all of the evidence at the guilt or innocence stage of this trial," I object to that, and would request that it say, "you shall consider all of the evidence admitted to you in this whole trial."

The trial court overruled the objection.

The appellant contends that the jury instruction in question did not conform to the requirements of the relevant portion of the statute that provides:

The court shall charge the jury that:

In deliberating on the issues submitted under Subsection (b) of this article, it shall consider all evidence admitted at the guilt or innocence stage and the punishment stage, including evidence of the defendant's background or character or the circumstances of the offense that militates for or mitigates against the imposition of the death penalty.⁽⁴¹⁾

The appellant is correct; the trial court erred by leaving out the italicized portion of the statutory instruction. The appellant's requested phrase "the entire trial" was an adequate synonym for the statutory language and would have corrected the error in the charge. We find that the appellant's objection, coupled with his suggested revision in the charge, was sufficient to place the trial court on notice of the error.

We next must ask if the error was harmless. Our standard is set, by act of the legislature, in the first sentence of Article 36.19 of the Code of Criminal Procedure:

Whenever it appears by the record in any criminal action upon appeal that any requirement of Articles 36.14, 36.15, 36.16, 36.17 and 36.18 has been disregarded, the judgment shall not be reversed unless the error appearing from the record was calculated to injure the rights of defendant, or unless it appears from the record that the defendant has not had a fair and impartial trial.

This sentence was explained in Almanza v. State, 686 S.W.2d 157, 171 (Tex. Cr. App. 1984):

Article 36.19 actually separately contains the standards for both fundamental error and ordinary reversible error. If the error in the charge was the subject of a timely objection in the trial court, then reversal is required if the error is "calculated to injure the rights of defendant," which means no more than that there must be some harm to the accused from the error. In other words, an error which has been properly preserved by objection will call for reversal as long as the error is not harmless.

On the other hand, if no proper objection was made at trial and the accused must claim that the error was "fundamental," he will obtain a reversal only if the error is so egregious and created such harm that he "has not had a fair and impartial trial" -- in short "egregious harm."

In both situations the actual degree of harm must be assayed in light of the entire jury charge, the state of the evidence, including the contested issues and weight of probative evidence, the argument of counsel and any other relevant information revealed by the record of the trial as a whole.

Accordingly, we look first at the entire jury charge. The first paragraph of the charge told the jury it was necessary to determine the answers to the special issues "from all the evidence in the case."

The erroneous paragraph told the jury that in answering the first special issue, it should consider "all the evidence at the guilt or innocence stage of the trial."

A subsequent paragraph of the charge told the jury that they would be "considering all of the evidence before you in answering Special Issue No. 2." This instruction, which comports with the general statement in the introductory paragraph, creates a contrast between the instructions on the special issues. In answering issue number one the jury was to "consider all of the evidence at the guilt or innocence stage of the trial," while in answering issue number two the jury would be "considering all of the evidence before you." It would be reasonable to infer that the difference in language was significant.

Next we consider the evidence. We have recited the facts of the offense above. We also consider the evidence at the punishment stage. The State's punishment evidence was that the appellant was a known member of a "gang" called Vato Locos and had a gang tattoo that could signify a drive-by shooting or other things. The appellant, who was 21 years old at the time of trial, had three years previously been found in the bedroom of a 12-year old girl, who had had consensual sex with him. Two years previously he had broken into some cars and stolen some tapes and stereo equipment. The harmful effect of the error was not to exclude consideration of the State's punishment evidence and arguments; it was to exclude consideration of the appellant's punishment evidence. This included testimony from relatives and friends on his good traits of character. This evidence was in the form of opinion and specific acts of good conduct. The witnesses also gave evidence of the appellant's low intelligence, epileptic seizures, background of poverty, and mistreatment at the hands of his father.

We consider the arguments of counsel. The State's jury argument that the appellant would be "a continuing threat to society in the future" (a reference to issue number one) was based almost entirely on the appellant's "past," that is, the crimes that had been proved at the punishment stage. The appellant's argument as to issue number one was that the offense was not "a death penalty case" like more serious murders, the unadjudicated offenses were neither many nor violent, the appellant had no other record, and he would not be eligible for parole for forty years.

We do not resolve the issue by asking whether the appellant met a burden of proof to persuade us that he suffered some actual harm, as the dissent would have it. No party should have a burden to prove harm from an error, and there ordinarily is no way to prove "actual" harm. Burdens and requirements of proving actual facts are appropriate in the law of evidence, but they have little meaning for the harmless-error decision.⁽⁴²⁾

In evaluating what effect, if any, an error had on the jury's verdict, the appellate court may look only to the record before it. The function of a party carrying the burden is simply to suggest, in light of that record, how prejudice may or may not have occurred. At that point, the court makes its own assessment as to what degree of likelihood exists as to that prejudicial or non-prejudicial impact and then applies to that assessment the likelihood-standard of the particular jurisdiction.⁽⁴³⁾

We do not agree with the dissent that "the jury could not have been misled" by the erroneous charge.⁽⁴⁴⁾ The notion seems obviously false. If the jury read the charge and followed it, it is at least reasonably possible that the jury would have considered only the evidence from the guilt or innocence stage, which was all the Court told them to consider.⁽⁴⁵⁾ The dissent's speculation -- that the jury did not need to be told that it could consider the punishment-stage evidence, since it could well have used "common sense,"⁽⁴⁶⁾ and relied on a general statement in the introductory paragraph, and been influenced by the State's argument, and remembered what was said before and during voir dire -- is not impossible. But it is remarkably contrary to the established law on the function of the charge.⁽⁴⁷⁾

The most the record supports is the possibility that, if the jury ignored the charge, or knew that the correct law was an unspoken given, or read the charge to say something that it did not, they might have considered the appellant's punishment-stage evidence on special issue number one. Our law does not support a supposition that juries treat charges in such a fashion.

After the Almanza analysis, we cannot say with confidence that the error did not cause some harm. The statutory standard for reviewing error in the charge requires us to set aside the sentence and remand the case for a new punishment hearing.

Our disposition of this point of error makes it unnecessary for us to consider the appellant's remaining points.

The judgment of guilt is affirmed. The judgment assessing the death penalty is reversed, and the case is remanded to the district court.

En banc.

Delivered March 8, 2000.

Publish.

1. ¹"A person commits an offense if he commits murder as defined under Section 19.02(b)(1) and...the person intentionally commits the murder in the course of committing or attempting to commit kidnapping...." Tex Penal Code 19.03(a)(2).

2. ²See Tex. Code Crim. Proc. art. 37.071, 2(b) & (e).

3. ³See id. 2(g).

4. ⁴See id. 2(h).

5. ⁵The appellant's confession referred to "Roxanne" and "Annette." Other witnesses established that Jennifer was sometimes known as Roxanne. No similar testimony appears for Lynnette, but a review of the testimony of all the witnesses in the case concerning who was present at the events relevant to this prosecution leads to the inescapable conclusion that Lynnette is the "Annette" referred to in the appellant's confession.

6. ⁶Cf. Jackson v. Virginia, 443 U.S. 307, 319 (1979).

7. ⁷Partida, 430 U.S. at 494-95.

8. ⁸Id. at 494.

9. ⁹Id. at 495.

10. ¹⁰See id. (Mexican-Americans found to be a clearly identifiable class).

11. ¹¹See Tex. Code Crim. Proc. art. 19.01.

12. ¹²Partida, 430 U.S. at 497.

13. ¹³The Court used the most recent census figures (1970) to determine the proportion of Mexican-Americans in the total population in Hidalgo County and examined grand jury lists for that county over a period of eleven years. Id. at 486, 495-96. The Court found that Spanish-surnamed and Spanish-language individuals (which it assumed were all Mexican-Americans) constituted 79.1% of the county but that Spanish-surnamed individuals composed only 39% of the persons summoned for grand jury service.

14. Id. at 486-87, 495. We note that not all persons of Spanish descent are Mexican-American. Apparently, the parties in Partida focused upon Mexican-American heritage although the Supreme Court's analysis in fact embraces a larger group of Spanish-speakers.

15. ¹⁵Id. at 496.

16. ¹⁶The Court first compared the actual number of Mexican-Americans on grand juries over the eleven-year period with the expected number (based on percentage of population).

17. Because a total of 870 persons had been summoned over that period of time, the expected number was 688 (79.1% of the total). The observed number on the lists, using surnames as the method of determining Mexican-American origin, was 339. The measure of the predicted fluctuations from the expected value is the standard deviation, defined as "the square root of the product of the total number in the sample (here 870) times the probability of selecting a Mexican-American (0.791) times the probability of selecting a non-Mexican-American (0.209)." The standard deviation thus derived was approximately 12. Id. at 496 n.17.

Other courts have used different methods of determining whether there is underrepresentation. See, e.g., United States v. Shinault, 147 F.3d 1266 (10th Cir. 1998) (comparative disparity); Ramseur v. Bever, 983 F.2d 1215 (3d Cir. 1992) (absolute disparity).

18. ¹⁸Partida, 430 U.S. at 496 n.17.

19. ¹⁹Ibid. ([688 - 339] / 12 = 29.083).

20. ²⁰Id. at 501. The Supreme Court recognized that we had held in Partida's appeal that the defendant "should have shown how many of the females who served on the grand juries were Mexican-Americans married to men with Anglo-American surnames, how many Mexican-Americans were excused for reasons of age or health, or other legal reasons, and how many of those listed by the census would not have met the statutory qualifications of citizenship, literacy, sound mind, moral character, and lack of criminal record or accusation." Id. at 489; see Partida v. State, 506 S.W.2d 209, 210-11 (Tex. Cr. App. 1974). The Court discussed the possible effect of limiting the inquiry to the eligible population. Partida, 430 U.S. at 488 n.8. The Court found that the disparity was still too great although it refused to rely upon eligibility figures due to uncertainties in the statistics. Ibid. The Court did not discuss the failure to take into account Mexican-American women with Anglo surnames due to marriage.

21. ²¹The figures show that persons of Mexican origin constituted 6.6% of the population. We will not pursue the issue of whether persons of diverse Hispanic origins (e.g., Spain, Mexico, Central America, South America, Cuba, Puerto Rico) possess such similar interests that they constitute a cognizable class.

22. ²²On the grand jury lists, the surname is spelled "Avera."

23. ²³United States ex. rel. Barksdale v. Blackburn, 639 F.2d 1115, 1124 (5th Cir.), cert. denied, 454 U.S. 1056 (1981); State v. George, 503 S.E.2d 168, 170-71 (S.C. 1998), cert. denied, U.S. , 119 S. Ct. 1050 (1999).

24. ²⁴ While there is a rough correlation between Spanish surname and Hispanic origin, having a Spanish surname does not mean one is Hispanic, nor do all persons of Hispanic origin have Spanish surnames. A Hispanic woman who marries an Anglo man and takes his surname (a prevalent custom) will not have a Spanish surname. Likewise, an Anglo woman who marries a Hispanic man could have a Spanish surname. In fact, the prosecutor testified that such a phenomenon had occurred among people who had been selected for Navarro County grand juries. See United States v. Esparsen, 930 F.2d 1461, 1466 (10th Cir. 1991)(making similar observations).

Moreover, using Spanish surnames as a yardstick for Hispanic origin raises the question of what precisely is a Spanish surname. While some surnames are clearly of Spanish origin, other surnames present a closer question. In Partida, the Supreme Court referred to a list of 8,000 Spanish surnames compiled by the Immigration and Naturalization service. Partida, 430 U.S. at 486 n.5. In 1980, the Census Bureau constructed a "Spanish Surname List," containing 12,497 different "Spanish" Surnames. See David L. Word and Colby Perkins, Jr., Building a Spanish Surname List for the 1990's -- A New Approach to an Old Problem, Technical Working Paper No. 13, Population Division, U.S. Bureau of the Census 1 (1996). The surnames appearing on the 1980 list "were culled from a database of 85 million taxpayers filing individual federal tax returns for 1977." Ibid. The 1980 list determined that surnames were "Spanish" by comparing the similarity of the surnames' geographical distribution with the geographical distribution of the Hispanic origin population in the United States and by performing a complex mathematical function for determining the likelihood that a name was Spanish. Ibid.

This indirect approach to determining Hispanic surnames was discarded for the 1990 list. Used instead was a fi0 Census records. This file was originally created for the purpose of estimating undercount in the 1990 Census. Ibid.

25. The analysis was further narrowed to 1,868,781 "householder" records that included valid responses to both surname and Hispanic origin.

26. Id. at 2. Word and Perkins defined "householder" as male householders or never-married female householders plus any other male or never-married female in the household not related to the householder. Id. at 2 n.2. Narrowing the inquiry to householders was designed to mitigate the effects of clustering (people in the same household having the same surname and ethnicity) and the problems with calculating the ethnicity of ever-married females. Id. at 2 and 2 n.2.

Within that sample, surnames were categorized by their probability of being Hispanic. Id. at 6. A householder was considered Hispanic if he or she answered the Hispanic origin question affirmatively in the sample -- Hispanic origin was, in essence, self-reported. Id. at 6. A limitation on the effectiveness of self-reporting as establishing Hispanic origin is response variance. Id. at 9. Word and Perkins observed that seven per cent of persons saying they were of Hispanic origin in the 1990 Census decided that they were non-Spanish at a later date. And eleven percent of persons saying they were Hispanic on reinterview had indicated that they were non-Spanish on their 1990 census forms. Ibid.

For frequently occurring surnames, the proportion of respondents identifying themselves as Hispanic was used to determine the status of the surname. For rarely occurring surnames, the following factors were considered, in order of importance: (1) the proportion Hispanic, (2) orthographic structure, and (3) appearance on the 1980 surnames list. Orthographic structure refers to five rules relating to the structure of the surname that were formulated by the linguist Robert W. Buechley. According to the five rules, a Spanish surname should contain: (1) no "K," (2) no "W," (3) no double letters other than "L" or "R," (4) certain combinations of three letters at the beginning of the name, (5) certain combinations of three letters at the end of the name. (Valid three-letter "starts" and "ends" are found in id. elec. app.

The work that resulted from these calculations is not really a "list," but a set of categories, based upon the probability that a person with a given surname would be Hispanic. Id. at 13. Surnames that appeared in the sample were divided into five categories: (1) Heavily Hispanic -- over 75 per cent, (2) Generally Hispanic -- over 50 per cent but not over 75 per cent, (3) Moderately Hispanic -- over 25 per cent but not over 50 per cent, (4) Occasionally Hispanic -- over 5 per cent but not over 25 per cent, and (5) Rarely Hispanic -- not over 5 per cent. Id. at 6. Names such as "Smith" and "Jones," for example, fell into the Rarely Hispanic category. See id at 13 & elec. app., Splong.dat (the "long list"). One could construct a sixth category for names that did not appear in the study sample -- such names would be classified as "indeterminate," since there is no information from which to determine whether or not the name could be classified as Hispanic. Id. at 6.

Finally, we observe that Hispanic status may also be indicated by first name as well as surname, especially for women, who may be married to men with non-Spanish surnames. Esparsen, 930 F.2d at 1468. As discussed above, the prosecutor in the present case testified that there were women falling within that category who had served on grand juries in Navarro County.

27. ²⁷79.1% of 12 = 9.49.

28. ²⁸7.2% of 12 = 0.864.

29. ²⁹Tex. Code Crim. Proc. art. 19.06.

30. ³⁰Id. art. 19.26.

31. ³¹41 x 12 = 492.

32. ³²7.2% of 492 = 35.424.

33. ³³Determined by finding the square root of the product of the total number in the sample (492) times the probability of selecting a Hispanic person (0.072) times the probability of selecting a non-Hispanic person (0.928), and rounding to the nearest whole number.

34. ³⁴35.43 - (5.73 x 3) = 17.22.

35. ³⁵We determine that the following persons are probably Hispanic based upon their surnames: Ben Aldama, Ben Aldama, Jr., Bernard Camirillo, Ricardo Ferrer, Billy Flores, Edmond Flores, Andy Garcia, Emmaline Gonzales, Emmett Gonzales, Lee Guillen (served twice), Gloria Guillen, Tomas Llinas, Walter Palma, Jr., Glenda Ruiz, and Kenna Uresti. Ferrer, Flores, Garcia, Gonzales, Guillen, and Ruiz are all included in the 639 most common Heavily Hispanic surnames. See Word & Perkins, supra note 22, elec. app., Splong.dat, (the "long list"). Though "Camirillo" does not appear in the electronic appendix, we assume that it is a misspelling or variation of "Camarillo," which is included among the 639 most common Heavily Hispanic surnames. Aldama, Llinas, and Uresti, while not common surnames, are also classified as Heavily Hispanic. Palma is classified as Generally Hispanic, with a Hispanic hit rate of 61%. But in the eleven states with large numbers of Hispanics -- of which Texas is one -- the name was found to be 71% Hispanic, very close to the Heavily Hispanic category. Using surnames by their probability of being Hispanic, then, yields sixteen Hispanic grand jurors (Lee Guillen counting twice).

The "Mr. Gillen" referred to by the prosecutor, could be either Lee Guillen or Lloyd Gillen, Jr. We should probably assume that the prosecutor was referring to Lloyd Gillen, since that was how the name was spelled in the record and since it is (as the prosecutor said) not a Spanish surname, while "Guillen" is a Spanish surname. If we were to make that assumption, both Mr. Gillen and Mr. Guillen would be counted toward the total of Hispanics serving. But we will give the appellant the benefit of the doubt here since the record is unclear, and assume that the prosecutor was referring to Mr. Guillen.

Although "Commiato" was discussed as a possible Hispanic surname by the trial court, we dismiss that possibility because the surname ble to us that would indicate that the name has any Hispanic origin or association. That surname does not appear in the electronic appendix. The surnames Avera and Honea, also discussed, satisfy all of Buechley's rules but show up as Rarely Hispanic.

In addition, we perceive at least three persons without Hispanic surnames who have first names that are often Hispanic: Dolores Carter, Juanita Tankersley, and Nevada Hollifield. The Hispanic nature of these names is supported by our common knowledge concerning the ethnicity of these names. There is some confirmation of the origin of each of these names. For "Dolores," see <http://www.first-name.com/d.html>; for "Juanita," see Esparsen, 930 S.W.2d at 1468; for "Nevada," see <http://www.leg.state.nv.us/General/FACTS.htm> (state of Nevada website). Spanish origin for each of these three names was also indicated on <http://www.babyzone.com/

babynames/namesearch.htm>.

36. ³⁶Fuller v. State, 827 S.W.2d 919, 926 (Tex. Cr. App. 1992), cert. denied, 509 U.S. 922 (1993).

37. ³⁷Adams v. State, 685 S.W.2d 661, 670 (Tex. Cr. App. 1985).

38. ³⁸Cf. Gipson v. State, 844 S.W.2d 738 (Tex. Cr. App. 1992). In non-jury trials appellate courts formerly presumed that the trial court ignored inadmissible evidence, and judgments were not reversed on the ground of incompetent evidence if sufficient proper evidence existed to sustain the judgment. See, e.g., Tolbert v. State, 743 S.W.2d 631, 633 (Tex. Cr. App. 1988) and cases cited therein. We overruled that line of cases in Gipson, holding that Tex. R. App. P. 81(b)(2), with its presumption that errors were reversible, implicitly voided the presumption test. Since Tex. R. App. P. 44.2 became effective, the presumption of former Rule 81(b)(2) obtains only for constitutional errors considered under Rule 44.2(a). The new rule thus gives rise (though not in this case) to the question of whether the rule of Gipson would still require a presumption of harm for non-constitutional errors under Rule 44.2(b).

39. ³⁹Beck v. State, 712 S.W.2d 745, 748 (Tex. Cr. App. 1986).

40. ⁴⁰That issue asks "whether there is a probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society." Tex. Code Crim. Proc. art. 37.071 2(b)(1). The issue was submitted as "Special Issue No. 1" in the present case.

41. ⁴¹Id. 2(d)(1) (emphasis added).

42. ⁴²See Roger J. Traynor, The Riddle of Harmless Error 25-26 (1970).

43. ⁴³Wayne R. LaFave & Jerold H. Israel, Criminal Procedure 1165 (2d ed. 1992).

44. ⁴⁴See post at 3.

45. ⁴⁵If speculation is to be indulged, it would seem proper to imagine what the jury would say during deliberations on special issue one. The dissent seems to believe that something like the following probably happened:

JUROR A: What about the evidence at the punishment stage of the trial? Can we consider that?

FOREMAN Well, the charge doesn't say we can, but we can take it as a given because this issue was submitted at the punishment stage.

ALL: Okay.

The dissent implies that something like the following "could not" have happened.

JUROR A: What about the evidence at the punishment stage of the trial? Can we consider that?

FOREMAN: Let's see. The charge says that "in deliberating on Special Issue No. 1 we shall consider all the evidence at the guilt or innocence stage." But in deliberating on Special Issue No. 2 we consider "all of the evidence before us." If the judge wanted us to consider all the evidence on Special Issue No. 1, surely he would have said so. It looks like we consider only the evidence at the guilt or innocence stage.

ALL: That must be right.

46. ⁴⁶See ibid.

47. ⁴⁷One is reminded of the government's argument in Bollenbach v. United States, 326 U.S. 607, 613-14 (1946) (Frankfurter, J.). A court of appeals had reversed a conviction because of an erroneous jury charge.

The Government contend[ed] that the court below failed to appreciate several factors in regard to the criticized charge. What reason is there for assuming that the jury did not also fail to appreciate these factors which the Government, in an elaborate argument, explains as requisite for a proper understanding of that which at best was dubiously expressed? A conviction ought not to rest on an equivocal direction to the jury on a basic issue. The Government's suggestion really implies that, although it is the judge's special business to guide the jury by appropriate legal criteria through the maze of facts before it, we can say that a lay jury will know enough to disregard the judge's bad law if in fact he misguides them. To do so would transfer to the jury the judge's function in giving the law and transfer to the appellate court the jury's function of measuring the evidence by appropriate legal yardsticks.

The charge in Bollenbach, which was that the jury could rely on an untenable presumption, may have been worse than the one before us, which was only an omission of what was correct. But the Court's criticism of the government's argument is nonetheless relevant.