Sampling Crime: On sampling when testing  multiple packets

By:  Frederic Whitehurst, J.D., Ph.D. [1]

If we accept that forensic crime laboratories are completely overwhelmed, understaffed, underequipped, and exhausting environments in which to function, we see reality.  Despite political rhetoric to the contrary, there has been little, if any, relief for those folks in crime labs who are toiling everyday under impossible workloads.  The US War on Drugs is brought to the doors of crime labs hundreds of times a day because, finally, powders and vegetable materials must be chemically and physically analyzed in order to charge offenders properly. If we believe the Chemical Abstracts Registry, which tells us that there are over 56,000,000 known chemical compounds, and if we realize that a large portion of those materials can occur as powders, then we see the daunting task faced by forensic chemists. White powders seized by law enforcement personnel are not proven to be controlled substances simply because they are white powders which give positive results for presumptive field tests. Those powders must be subjected to rigorously validated scientific analytical protocols in order to test the hypotheses of seizing officers that the materials are actually controlled substances. These tests take an enormous amount of time, time that crime labs do not have.  In order to address the issue of limited resources, the North Carolina State Bureau of Investigation (NC SBI) Crime Laboratory in the past has turned to case law, State v. Brenda Harding, and sampling protocols which allow one to determine the whole from an analysis of a part of that whole.  This article will discuss the scientific and legal implications of the sampling protocols which the SBI Laboratory has utilized to achieve determinations of seized multiunit samples of alleged controlled substances.

When law enforcement officers seize suspected controlled substances, often those seizures are in many small units of powders. An investigation may end in the seizure of hundreds of small bags or bindles. The chemical analyst cannot realistically analyze every small bag; therefore, for many years the NC SBI Lab utilized a sampling protocol referred to as the “square root of n plus 1” protocol. That protocol in its entirety is produced here:

Name of Procedure:

Random Sampling

Random Sampling of Multiple Packages or Units

Suggested Uses:

Random sampling is a procedure that is used when analyzing an item of evidence that consists of multiple packages or units.  This procedure allows a chemist to determine the composition of the evidence by analyzing some randomly selected packages or units and extrapolating the results.  Random sampling is an accepted procedure used in forensic science and has been upheld by the Appellate Courts of North Carolina (see literature references).

Random Sampling Procedures:

1. Visually examine all of the packages or units in the item of evidence, as well as the contents, for differences in size, weight, color, packaging, markings, signs of tampering, labeling or other characteristics.  If there are no appreciable differences, all of the packages or units should be considered together for the selection of random samples.  If there are appreciable differences, segregate the packages or units into individual groups, based upon such observed differences.

2. To determine the number of random samples to be selected from a total number of packages or units, where n equals total number of packages or units:

a. If n is less than or equal to 4, then random sampling is not done.

b. If n is greater than or equal to 5, then the number of random samples selected is equal to the square root of n plus 1, expressed as:

random samples = √n + 1

c. Weight determination – the total weight of all packages or units may be extrapolated from the weight of a random sample of the packages or units.

d. Weight count – the total number of all packages or units may be extrapolated from the weight of a random sample of the packages or units.

Classification of Evidence:

There are three main forms of controlled substances:

1. Plant material.

2. Controlled substances consisting of marked dosage units from legitimate pharmaceutical manufacturers.

3. Controlled substances derived from clandestine manufacturers.

a. Packages or units containing powder or solids.

b. Packages or units containing liquid.

c. Packages or units consisting of any substance which is used as a median to absorb or contain a controlled substance (plastic bags, glassine bindles, paper bindles, blotter paper, gelatin, sugar cubes, tea leaves, parsley, etc.)

Application of Procedure on Evidence:

1. Random sampling of plant material:

a. Visual examination of all packages or units and a complete analysis of one package or unit is required to confirm identification (minimum requirements).

2. Random sampling of marked dosage units from legitimate pharmaceutical manufacturers:

a. The visual examination and the markings on the dosage units provide identification of the controlled substance and a complete analysis of one dosage unit is required to confirm identification (minimum requirement).

3. Random sampling of controlled substances derived from clandestine manufacturers:

a. Random samples of packages or units must be selected and subjected to at least one screening test.  A complete analysis of a portion of the random samples is required to confirm identification (minimum requirement).

Safety Concerns:

Not applicable.

Literature References:

Colon, Rodriguez, and Diaz, Representative Sampling of Street Drug Exhibits, Journal of Forensic Sciences, Vol. 38, No. 3, May 1993, pp. 641-648.

Siegel, and Saferstein, Forensic Identification of Controlled Substances, Forensic Science Handbook, Vol. 2, Prentice Hall, 1988.

Tzidony and Ravreby, A Statistical Approach to Drug Sampling: A Case Study, Journal of Forensic Sciences, Vol. 37 November, 1992, pp. 1541-1549.

Frank, Hinkley, and Hoffman, Representative Sampling of Drug Seizures in Multiple Containers, Journal of Forensic Sciences, Vol. 36, March 1991, pp. 350-357.

Waggoner, R.W., t Distribution and Prediction Intervals, North Carolina State Bureau of Investigation, 1996.

State v. Myers, 301 S.E. 2d 401, 402 (N.C. App. 1983)

State v. Wilhelm, 296 S.E. 2d 664, 667 (N.C. App. 1982)

State v. Absher, 237 S.E. 2d 1325, 1328 (N.C. App. 1977)

State v. Clark, 197 S.E. 2d 81, 82 (N.C. App. 1973)

Literature References (continued):

State v. Riera, 172 S.E. 2d 535, 539 (N.C. App. 1970)

State v. Harding, 429 S.E. 2d 416 (N.C. App. 1993)

Complete with scientific literature references and case law, the protocol might at first seem to be an unaddressable issue when one is questioning its validity. Indeed State v. Harding seems at first glance to stand solidly behind the protocol. Recall that, Harding stands for the premise that one can determine the whole from an analysis of a part. However, Harding does not address which sampling protocol one must utilize. Furthermore, Harding does not support the use of the protocol listed above. In order to demonstrate this lack of support, let us look at an example of a bindle of heroin seized as evidence.

This bindle is a cellophane container which has been folded twice over and then sealed with a piece of tape. Powder within the bindle cannot be seen. One cannot, therefore, follow the protocol that is repeated in part below by visually examining the contents for differences in size, weight, or color…:

1. Visually examine all of the packages or units in the item of evidence, as well as the contents, for differences in size, weight, color, packaging, markings, signs of tampering, labeling or other characteristics. If there are no appreciable differences, all of the packages or units should be considered together for the selection of random samples. If there are appreciable differences, segregate the packages or units into individual groups, based upon such observed differences.

The opaque nature of the bindle stops an examiner from determining the color and the morphology (size/shape) of the powder crystals inside the individual packages. Furthermore, street samples of heroin found in these bindles are usually in the hundredths of a gram size.

To give us some idea of the size of those samples, notice that a package of sugar substitute that one finds on the table in a restaurant contains 1 gram of material.

Now imagine two to four hundredths of a gram is all the powder within the bindle. Imagine then closing your eyes while someone removes 10% of the powder in that package without your knowledge and being asked to look back at the material. One cannot really detect such an appreciable change in such a small amount of material. Therefore, the protocol set forth above does not accurately function for such samples. Additionally color is not a very probative characteristic of materials when numerous chemicals and combinations of chemicals in the world might very well have the same color.

Because the visual examination through a multilayered semi-opaque material is not possible, the examiner must open all the bindles in order to visually analyze the contents of those bindles. In the experience of the author, Dr. Whitehurst, at the point of opening these packages, very often the analyst will simply mix the contents of the packages, thus destroying information concerning those contents. The undisturbed contents of these packages is evidence which a defendant has a right to review and an act, such as mixing these contents, is a direct violation of N.C. GS 14-221.1 and 15-11.1 which prohibit the destruction of evidence without court authorization.  Mixing contents often accomplishes the aim of anyone wishing to establish guilt of trafficking. However, assuming that the prosecution’s duty is to uphold justice, this mixing circumvents both the role of the prosecution as well as defense counsel.

The random sampling protocol above also neglects to instruct an analyst regarding how to actually conduct a random sampling. Analysts should be instructed that in order to prove a lack of bias in choosing which samples to weigh and analyze one must “randomly” select samples.

One way to achieve this random sample is to assign a number to each sample. Then, utilizing a random number generator computer program, select those samples directed by that program. An inexpensive version of the random sampling consists of putting all the samples into a box or bag and picking out samples without looking at the samples chosen for analysis. Thus, despite being named “Random Sampling” procedure, the randomness of the sampling is not defined to ensure random selection.

Let’s review how this protocol directs us to look at the number of samples to be analyzed:

2. To determine the number of random samples to be selected from a total number of packages or units, where n equals total number of packages or units:

a. If n is less than or equal to 4, then random sampling is not done.

b. If n is greater than or equal to 5, then the number of random samples selected is equal to the square root of n plus 1, expressed as:

random samples = √n + 1

c. Weight determination – the total weight of all packages or units may be extrapolated from the weight of a random sample of the packages or units.

d. Weight count – the total number of all packages or units may be extrapolated from the weight of a random sample of the packages or units.

For the reader who successfully may have forgotten any lessons from math classes in high school the “square root” of a number is another number which when multiplied by itself will equal the number. For example the square root of one hundred is ten. The square root of sixteen is four. Therefore, if there are one hundred bindles in a seizure, one will then randomly select ten plus one and analyze eleven samples. In Dr. Whitehurst’s experience, examiners in the NC SBI lab will sometimes choose the square root of n plus one samples and then immediately mix them together. Once mixed, the court will never know which samples contained what material. Using our common sense we know that if we mix unknown powders together (in order to prove a trafficking weight), this mixture is a newly created mixture. Because NC statutes criminalize the manufacture of powder containing controlled substance x, the state’s analyst is possibly in violation of these manufacturing statutes by mixing together samples of this powder containing x.

Furthermore, suppose, for example, that the defendant has 100 bags of powder, produced by an average street pharmacist (consisting of unreliable contents), and some of those bindles contain a controlled substance while others consist of a non-controlled substance. There is no quality control in those street pharmacies. This problem is evidenced by statutes in NC which prohibit the sale of false controlled substances. We will not assume naively that the buyer, upon finding he has been cheated by purchasing a non-controlled substance, will go to the better business bureau. In fact the level of lawlessness in the illicit drug business is legendary.

Complicating this picture further, the US Drug Enforcement Scientific Working Group on Drugs has published its 2008 “Scientific Working Group for the Analysis of Seized Drugs (SWGDRUG) Recommendations” which can be found at www.swgdrug.org. Page 8 of that document describes sampling strategy and notes that the sampling procedures are divided into statistical and non-statistical procedures.  Among the statistical procedures the hypergeometric, Bayesian, and other probability-based approaches are listed and among the non-statistical procedures are the square root n, management directive, and judicial requirements standards. One can infer from this list that the NC SBI lab’s random sampling protocol is a non-statistical sampling procedure because page 9 explains that: “If an inference about the whole population is to be drawn from a sample, then the plan shall be statistically based and limits of the inference shall be documented.”

Page 11 goes on to say: “Depending upon the inference to be drawn from the analysis for a multiple unit population, the sampling plan may be statistical or non-statistical…Statistical approaches are applicable when inferences are made about the whole population. For example:  b) The total net weight of the population is to be extrapolated from the weight of a sample.”

Page 12 also indicates that: “Non-statistical approaches are appropriate if no inference is to be made about the whole population.”

Based on these passages, it seems as though the NC SBI lab has been utilizing a non-statistical sampling protocol from which one cannot determine the total weight of a sample of multiple units. Furthermore, these passages indicate that the total net weight for multiunit samples has been determined without valid foundation. If one reviews the protocol as provided under discovery by the NC SBI lab during past cases, one will see that this protocol has been in use since 1996. The use of this non-statistical protocol for nearly 15 years is troubling and leads us to A.J. Izenman’s “Statistical and Legal Aspects of the Forensic Study of Illicit Drugs in Statistical Science, 2001, Vol. 16, No. 1, 35-57.  On page 47 Izenman writes: “The square-root and other popular rules. A worldwide survey of sampling practices and choices of sample sizes for forensic drug analysis …found that the most popular rule for deciding how many containers or items, whether homogeneous or not, to sample for drug testing was not a statistically motivated one. Instead, the most popular rule was the square-root rule, … used by laboratories in Australia, Austria, Canada, England, New Zealand, Hong Kong, Northern Ireland and the United States and U.S.A. Army-Europe. One would assume, therefore, that the square-root rule would be an accepted part of sampling practice. Yet, in an informal, but extensive, survey of sampling practitioners, we found that most sampling experts had never encountered the square-root rule and no textbook on sampling theory or practice nor review of the field…even refers to it…

The square-root rule apparently originated in the 1920’s from a need to provide agricultural regulatory inspectors (specifically, those who knew how to extract a square root) with a convenient, memorizable rule for sample size determination.”

The historic context of the use of this protocol versus its present application is troubling. Since 1996 has the North Carolina SBI laboratory been deciding weight of total population in drug trafficking cases with a sampling protocol which the community of statisticians opines is not valid for such a use? Have law enforcement, prosecutors and juries decided a defendant possessed a certain amount of material when in fact the NC SBI Laboratory could not accurately make that determination with the protocol that it utilized?

Recently a colleague requested a review of the lab discovery material in a criminal matter. Under Item 1 on the SBI Laboratory report dated September 2010 one can read that a statistical sampling plan that demonstrates 95% confidence was utilized.  That report goes on to call that plan the “hypergeometric” sampling plan. Does this protocol change indicate  that the NC SBI lab is now recognizing that the previous non-statistical sampling plan was indeed fundamentally flawed? If there truly is a new sampling plan—a statistical sampling plan—what consequences will this changed protocol have not only for post conviction relief matters, but also for the many trafficking cases which were analyzed using the square root of n plus one sampling plan, a plan which the NC SBI lab may now admit to itself, if not to defendants in courts of law under Brady, proliferated mistakes and possibly caused miscarriages of justice?

__________

[1] Executive Director, Forensic Justice Project, Washington, D.C., B.S. Chemistry, 1974, East Carolina University, Ph.D. in Chemistry, 1980, Duke University, J.D., 1996, Georgetown University School of Law. (202)342-6980.