Confronting Crawford V. Washington in the Lower Courts

Keenan, Dylan O., The Yale Law Journal

II. EMPIRICAL EVIDENCE

In order to assess the potential ambiguity of the Crawford line, I randomly sampled and coded lower-court decisions (188) and quantitatively analyzed the facts and outcomes of those decisions. This Part details the empirical methodology I employed and the results that this investigation produced.

Previous work has evaluated Crawford in the lower courts by closely analyzing select decisions. (189) While this is the standard method of legal reasoning, there are important reasons to prefer systematic empirical analysis of judicial opinions when surveying an area of law. First, empirical methods help to counteract the cherry-picking phenomenon of conventional case analysis. This problem arises when scholars develop theories from a "small series of cases." (190) While those theories can be thought provoking, they rarely provide a "robust explanation of how the law works." (191) A handful of deliberately chosen cases cannot be considered a representative sample of the law. Conclusions are at best tentative or uncertain when drawn from an insufficiently large sample. (192) By randomly sampling cases, or by analyzing the entire pool of available cases on a subject, researchers can negate the biases of traditional case-analysis methods. (193)

Second, rigorous empirical methods help to discipline the researcher in reading selected cases. Proper quantitative analysis requires the researcher to develop a coding scheme before she reads the opinions. (194) A researcher develops a set of variables and criteria for valuing each variable for a given case. (195) When coding the cases, the researcher is forced to consider each variable as it applies to each case. Thus, the researcher is less likely to suffer from tunnel vision--picking and choosing the factors that subjectively stand out in a given opinion (196)

A. The Data Set

I compiled a random sample of cases from Westlaw, decided before October 17, 2011, that cite Davis. (197) These cases could be from any federal district court, federal court of appeals, or state court. The sample includes both reported and unreported decisions, as well as decisions that had been overturned on appeal. Of the 300 cases I sampled, I excluded 77. Those 77 were nonbinding magistrate judge recommendations and cases in which the court did not evaluate a Confrontation Clause challenge under Crawford's testimonial framework. The remaining 223 cases included several instances in which a single court evaluated multiple statements under the Crawford framework. Thus, the final sample consisted of 278 statements from 223 cases.

For each statement, I coded nineteen independent variables and one dependent variable--whether the reviewing court found the statement to be testimonial or not. The coding rules are described in detail in Appendix A. Table 3 lists the name of each variable and gives a description. I chose variables that both captured the important factors cited in the Crawford line and could be coded objectively.

The variable for measuring an injured party involved some subjective judgment on my part about whether an injury was serious enough to warrant medical attention. To address this, I included an alternate, more objective definition of an injury. The results of the regression analysis are almost identical between the two definitions, but the descriptive statistics do show some differences between the definitions, which are discussed below.

B. Descriptive Statistics

This Section presents descriptive statistics--statistics that summarize and explicate the data. Two conclusions were immediately obvious from the data. First, the government won the majority of cases. Overall, courts held that only 90 of the 278 challenged statements were testimonial. (198) Even this figure--that the defendants prevailed in only one-third of challenges--overestimates the success of defendants. In many of those 90 cases, courts ultimately gave judgment for the government on the basis of harmless error. (199) Second, the data showed a stark contrast in how courts evaluated statements made to different categories of recipients. For all of the graphs and data reported below, I did not include statements that were contained in forensic laboratory reports because the doctrine surrounding forensic analyses has become more refined than the general Crawford doctrine. This exclusion should have the effect of exaggerating any ambiguity or uncertainty in Confrontation Clause cases. Figure 1 plots the frequency with which statements to certain recipients were held to be testimonial.

Figure 1 shows that statements are more likely to be held to be testimonial when made to police officers or other government officials. They are less likely to be held to be testimonial when made to health professionals or 911 operators and very unlikely to be held to be testimonial when made to private citizens. Whereas half of the challenged statements made to police officers and nonpolice government officials were held to be testimonial, only around five percent of statements to private citizens were held to be so. The most curious feature of the data is the relatively high frequency, almost forty percent, with which statements to health professionals were held to be testimonial. This result is partly driven by statements to Sexual Assault Nurse Examiners (SANE nurses) and other health practitioners who specialize in treating victims of sexual assault and sexual abuse. These individuals are mandatory reporters who are trained to collect evidence as well as provide medical treatment. As one court explained, SANE nurse questioning is the "functional equivalent of police questioning." (200) Statements made to SANE nurses thus draw great scrutiny from reviewing courts.

Figure 2 combines the categories listed in Figure 1 to show the dramatic difference in how courts treat statements to state actors and statements to nonstate actors.

Without controlling for other factors, a statement is over three times more likely to be held to be testimonial when it is made to a state actor. Section II.C shows that the results are not quite so pronounced when one controls for other variables. Nonetheless, considering that the Crawford line has not placed a great deal of explicit emphasis on state action, these results are striking.

Less strikingly, the data also reflect different outcomes in cases where an injured party is involved in the events surrounding the statement. Figure 3 shows the outcomes in injured-party cases relative to non-injured-party cases.

Figure 3 shows that statements are, without controlling for other factors, less likely to be held to be testimonial when an injured party is involved. The results are most pronounced under Specification A, where an injury is defined as a shooting, stabbing, or other injury requiring immediate medical assistance. The presence of an injury should matter if courts are following the Crawford line. Davis stressed response to an ongoing emergency, (201) presumably in many cases a medical emergency. Bryant made this point explicit-emphasizing that physical injury plays a significant role in evaluating the testimonial status of statements. (202)

The reader might notice that the results in Figure 3 vary depending on how an injury is defined. When one includes not only shootings, stabbings, and other injuries requiring medical attention, but also kicks and punches, the results are weaker. Figure 4 explores the result further, showing the results by type of injury.

Crawford provides some reason to view different injuries differently. If the police see a victim with a black eye, they might respond more calmly and with greater investigatory emphasis than if they had found a victim with a gunshot wound. Statements made under those circumstances look more like substitutes for live testimony produced through the kind of ex parte examinations that troubled the Framers and the Crawford Court.

The results in Figure 4 suggest something different--when there is a slight injury, statements are more likely to be held to be testimonial than when there is no injury at all. This seemingly odd result is plausibly explained by the role of state action. Statements made around a slight injury are more likely to be made to state actors than statements made when there is no injury. (203) The role of state action may be overwhelming the effect of an injured party. To more fully disentangle the relative effects of state action and injury requires something more than description. To control for state action and injury--to fully evaluate the interaction of all the variables at play--I employed logistic regression.

C. Stepwise Logistic Regression Analysis

Regression analysis is a method for sorting and analyzing data. (204) The researcher assumes that a very general form of relationship exists between variables and attempts to estimate the parameters in that relationship. For example, a researcher might believe that an individual's salary depends on her level of education, gender, and height. Regression analysis allows the researcher to estimate, from a set of data about individuals' salaries, education levels, genders, and heights, how a change in one independent variable will affect salary.

Logistic regression is a particular type of regression. (205) Logistic regression is useful when the researcher hopes to explain the relationship between several explanatory variables and a binary dependent variable. (206) A binary variable is one that takes only two states. Whether someone has completed high school, for example, is a binary variable. A person either completes or does not complete high school. There is no third option. I employed logistic regression because the dependent variable in nay analysis--whether a statement is found to be testimonial--takes only two values.

Stepwise regression adds a final nuance to regression analysis. Imagine a scenario where a researcher is addressing a new problem without background knowledge. For example, the researcher has collected a great deal of data about death penalty defendants and wishes to know which of several hundred characteristics affect the probability of an execution. The researcher may not have enough data to run a useful regression analysis with all the variables. Stepwise regression takes several iterative steps that narrow down the set of variables to those explanatory variables which have a statistically significant relationship with the dependent variable. There are several different methods of stepwise regression. Some methods select variables into the final specification. These methods, known as forward selection, add variables to the specification in descending order of significance. The selection process ends when all statistically significant variables have been added. Other methods select variables out of the specification until all statistically insignificant variables have been eliminated. Stepwise regression thus helps avoid sample size problems by reducing the number of independent variables in the final regression specification. With fewer variables, fewer observations are required to have confidence in the resulting estimates. (207) Stepwise regression has been used frequently in empirical legal scholarship. Without using stepwise regression, the sample size here would be relatively small. There would be fewer than five events per variable--i.e., fewer than five occurrences of a statement being found testimonial for each independent variable. A sample that small would dramatically reduce confidence in the results. Stepwise regression eliminated statistically insignificant variables, resulting in more events per variable--more occurrences of a testimonial statement for each independent variable--in the final regression specification, thus allowing me to more accurately estimate the true relationship between the variables.

Table 4 presents the results of the regression analysis. I did not include in the regression analysis statements contained in forensic laboratory analyses for the same reasons that I excluded those statements from the descriptive statistics. The stepwise regression process eliminated several additional variables that were not statistically significant. (208) The middle column is the most important. (209) It shows how a change in the associated variable changes the probability that a statement will be found to be testimonial by the court, holding other variables constant. For example, other factors equal, the fact that a statement is made to a police officer makes it about 75% (between 60% and 89%, with 95% confidence) more probable that the statement will be found to be testimonial. Similarly, other factors equal, the fact that a statement is made while one of the parties is injured decreases the probability that it will be found to be testimonial by about 18% (between 8% and 28%, with 95% confidence).

What conclusions can we draw from the regression results? First, state action matters. A lot. A given statement is almost twice as likely to be found to be testimonial, all else equal, when it is made to a state actor. The fact that the recipient of a statement is either a police officer or a nonpolice government official (for example, a fire marshal, prosecutor, or judge) substantially increases the likelihood that a court will find that statement to be testimonial. This is consistent with the descriptive statistics in Section II.B. It is also consistent with the scholars and lower courts who have emphasized the Confrontation Clause as a protection against abuses of state power. (210) It is even consistent with the subtext of Crawford and the history of confrontation abuses in England. (211) State action, however, has only been explicitly emphasized by the Court in Bryant. (212) And even Bryant only made a comparative statement of priority; it did not state that the Confrontation Clause is exclusively or nearly exclusively concerned with statements made to state actors.

State v. Parker, a case decided by Tennessee's Court of Criminal Appeals, demonstrates the important role of state action in Confrontation Clause cases. (213) There, the victim of a home invasion was sexually assaulted and fled to her neighbor's house. (214) She recounted her attack to her neighbor, who dialed 911. (215) The police arrived several minutes later and questioned the victim, whose statements were "parallel[]" to those she had initially made to her neighbor. (216) Hours later, the victim told a nurse at the hospital that her son's friend had attacked her. (217) The Parker court, after reciting an eight-factor test for testimonial statements, concisely noted that "[s]tatements describing past events to law enforcement officers are testimonial" except "when [those] statements ... are meant to assist the officers in meeting an ongoing emergency." (218) On that basis, the court held that the victim's statements to her neighbor, a nonstate actor, were nontestimonial while nearly identical statements made by the same victim minutes later were testimonial because they were made in response to police questioning. (219)

Second, some of the results are quite consistent with Davis's emphasis on emergency response. For example, statements made during a crime--in the midst of an emergency with uncertain outcomes--are slightly less likely to be found to be testimonial. The same is true of statements held to be excited utterances and those made while a party is injured.

Several results are more ambiguous. For example, statements made during 911 calls are forty percent more likely to be testimonial. This makes some sense, because 911 operators are state agents, although they are not police officers. (220) Without controlling for other factors, 911 operators confront courts with conflicting indicators. They are state actors, yet they act, almost by definition, during emergencies. Regression isolates the effect of a 911 operator, independent of the emergency during which the operator acts. The data suggest that courts recognize 911 operators as state actors. This fact can be hidden because so many 911 calls occur during medical emergencies, but when one controls for that effect, the emphasis that courts place on state action becomes clear. (221)

The results for health professionals are also somewhat ambiguous. The abuse-counseling results seem to fit the state-action theory. Courts are often critical of social workers' interviews of children because they view these interviews as means to collect evidence, (222) and because the interviews sometimes occur at the instigation of the police. That skepticism might also partly explain the result for other health professionals. Health professionals include not only EMTs and doctors but also SANE nurses. As explained above, these professionals draw greater scrutiny from courts. (223)

State v. Steele, decided by the Ohio Court of Appeals, illustrates the complex interplay of medical emergencies, police interrogations, and discussions with health professionals. (224) Police arrived at the apartment of Virginia Austin to find her lying on the ground, partially disrobed and crying. (225) She had also been struck on the head. (226) Austin told the officers that that she had been assaulted. (227) The police called an ambulance, and Austin told the arriving EMT that she had been attacked (228) and repeated that statement to the police. (229) Finally, Austin gave a more detailed report to police at the hospital. (230) The court applied the "primary …

Questia, a part of Gale, Cengage Learning. www.questia.com
Publication information: Article title: Confronting Crawford V. Washington in the Lower Courts. Contributors: Keenan, Dylan O. - Author. Journal title: The Yale Law Journal. Volume: 122. Issue: 3 Publication date: December 2012. Page number: 809+. © 2009 Yale University, School of Law. COPYRIGHT 2012 Gale Group.

This material is protected by copyright and, with the exception of fair use, may not be further copied, distributed or transmitted in any form or by any means.