Public Defenders > Legal Research > Papers by Public Defenders

Dealing with DNA in Court: its use and misuse

 

Andrew Haesler SC
Deputy Senior Public Defender

This paper is an updated version of papers prepared for the Judicial Commission of New South Wales, and later published in the Judicial Review, (2008) Vol 8 No 1.

Introduction

DNA Deoxyribonucleic acid matching is a wonderful investigative tool. DNA evidence can provide powerful evidence in support of a prosecution case. DNA evidence can also provide convincing evidence of a person’s innocence.


A profile taken from the DNA of a suspect can be compared with the profile of a sample of DNA taken from a crime scene. It may, depending on the other evidence, be compelling evidence of guilt. Profiles can be stored on a computer database. They can be easily cross-checked and any linkage investigated. A statistically validated “match” or “link” between these profiles is evidence that they come from the same source.

DNA is corroborative evidence. It is particularly useful if supported by statistics that the chance of someone else other than a defendant leaving the crime scene stain is highly unlikely or highly improbable, by a ratio of one in ten billion.

DNA evidence can, however, be misunderstood. If unchallenged, DNA evidence can be “significantly persuasive”. M Findlay and J Grix, “Challenging forensic evidence” (2003) 14 Current Issues in Criminal Justice 269 at 273. It appears to offer a degree of certainty that is often missing from a criminal trial. It also has the cache of being the topic of the moment on popular television.


DNA’s apparent certainty can be deceptive. It can be misused and misapplied. It will not of itself solve the crime problem although the chance of discovery may act as a deterrent. Jacobs J noted in Griffith v The Queen (1977) 137 CLR 293 at 327: “The deterrent to an increased volume of serious crime is not so much heavier sentences as the impression in the minds of those who are persisting in a course of serious crime that detection is likely and punishment will be certain”. DNA evidence is part of a prosecution case; it is not a panacea.


We all want evidence and results that make our already difficult jobs easier. We would love some expert to ease the burden of judgment by saying, “this is the answer”. If only it were that easy. There is still a lot we don’t know about DNA. “As we know, there are known knowns. There are things we know we know. We also know there are known unknowns. That is to say we know there are some things we do not know. But there are also unknown unknowns. The ones we don’t know we don’t know.” Donald Rumsfeld, US Secretary of Defence, Department of Defence Briefing, Washington DC, 12 February 2002. We need to be acutely conscious of the limitations of DNA evidence. Often the question is not, “Whose DNA is it?” but “How did the DNA get there?”


This paper aims to address some of the issues that can arise when DNA evidence is put before the court at trial. I cover six key areas:

· DNA science and technology;
· Statistics;
· DNA reports;
· Problems with the science and technology;
· Problems with statistics; and
· Evaluating DNA evidence in context.



DNA science and technology

Deoxyribonucleic acid (DNA) is found in all cells, except red blood cells. DNA is said by some to be unique to all but identical twins. However the fact, crucial for any understanding of DNA evidence, is that there is no scientific proof of such uniqueness.


Only a few cells, invisible to the naked eye, can be enough to obtain the DNA profile of an individual. DNA can be extracted from the sorts of things regularly left at crime scenes, such as traces of skin, blood, sweat, saliva or other bodily fluids. Most testing in Australia focuses on the DNA found in the nucleus of a cell, although DNA can also be extracted from the mitochondria. This article deals with DNA extracted from the nucleus of a cell.


A crime scene sample is taken either by swab or by collecting an exhibit for sampling. Usual police crime scene or hospital sexual assault forensic exhibit procedures should be followed. The exhibits and/or swabs are then transferred to a government analytical laboratory where the swab is processed. A whole exhibit is examined visually and samples taken from areas where it is presumed that DNA might be found. These samples and crime scene swabs are then tested to see if it is possible to determine the type of material found, for example, blood, semen, skin (epithelial) cells or saliva. It is not always possible to determine the type of material. Sometimes only a presumptive test can be done.


The item is then sent for processing. At this point the machines and technicians take over. In all Australian jurisdictions the processing of evidence for DNA follows a fairly standard procedure based on commercially available kits:


· The DNA is first extracted.


· It is then measured. Too much DNA can skew a sample. Too little can lead to a “no result”. Some products including cloth dyes or cleaning agents can inhibit DNA analysis. Ultra violet light, heat, humidity or bacterial action can destroy DNA. Ideally only a very small amount is needed. Ideally between 0.5 and 1 nanogram of DNA per 20 microlitres A microlitre (ul) is 1 millionth of a litre. There are about 15 drops of water in a millilitre (1 thousandth of a litre). is used. A nanogram is one thousand millionth of a gram, which gives some indication of the sensitivity of DNA analysis!


· The extracted DNA is then amplified. The process is akin to a photocopier, which not only turns out identical copies, but also does so exponentially. After a few repeats we get millions of copies. The amplification process used in Australia, known as PCR (Polymerase Chain Reaction), also enables multiple points/loci on a person’s DNA to be analysed at the one time (multiplexing).


· Those copies are then placed in a machine and analysed. A process known as electrophoresis is applied and a series of graphs and readouts obtained. These are then interpreted.


The beauty of the science and technology of DNA testing is that the process results in visual charts and computer readouts that describe what cannot be seen.


What can’t be seen is this:


Every cell Except red blood cells which have no nucleus; and sex cells (sperm and egg) which have only 23 chromosomes but combine to form 23 pairs. contains 23 pairs of chromosomes. Those chromosomes are made up of and act as a storage mechanism for our DNA. The DNA molecule or genome carries the entire genetic code of an individual. One half is inherited from the mother, the other from the father. Each chromosome is made up of genes, and spaces between them known as introns. Both in turn are made up of a series of bases Adenine, Thymine, Guanine and Cytosine. Adenine always binds with Thymine and Guanine with Cytosine: A-T & G-C attached to a backbone of phosphate sugar. These bases occur in sequence Configured in the famous double helix. and are depicted by the letters A, G, T and C.


The sequence of bases follows regular patterns. It is the differing combinations of bases, which make each person’s DNA potentially unique, as at some points (loci) the sequences vary markedly from person to person.


By locating a predetermined point on a gene (a locus) and measuring how many times a sequence of bases (say ACTG, ACTG, ACTG, and ACTG) occur, a DNA profile is obtained. Because they are less liable to deterioration the shorter sequences are used. These are known as Short Tandem Repeats (STRs).


At each locus, or point analysed, there should be two readings for the number of Short Tandem Repeats of a particular sequence. These alternative sites on the chromosome are known as alleles. One allele comes from the person’s mother, the other from their father. If the number of STRs is the same, the individual is said to be homozygous at that point/locus. If the number of STRs are different, the individual is said to be heterozygous at that locus. There are two readings because each chromosome is made up of two linked strands, one from the father the other from the mother. If the parents share that number of STRs a single result may be obtained at that locus. For example, if Mum is 12-14 and Dad 11-14, the possible combinations are 14-14, 11-12, 11-14 and 12-14.


Multiple points, preferably from different chromosomes, are tested. In Australia nine points/loci are tested as part of the Profiler Plus system generally in use. In addition, a test is done of a gene known as Amelogenin, which enables the sex of the sample’s donor to be determined. Each of the 10 points/loci (that is nine loci plus Amelogenin) used in Profiler Plus is given an internationally recognised identifier symbol: D3 S1358, VWA, FGA, Amelogenin, D8 S1179, D21 S11, D18 S51, D5 S818, D13 S317 and D7 S820. The ‘D’ stands for DNA. The next number is the number of the chromosome. The ‘S’ stands for Single Copy Sequence and the following number notes the order in which the particular sequence was discovered.


A typical worksheet from the Department of Analytical Laboratories notes the date, operator and method of extraction: and the date, operator and result of amplification: and analysis. All these tasks are undertaken by technicians under the direction of a supervising biologist, who then undertakes the interpretation of the results.


Separate readout graphs and charts noting peak heights and areas are also produced. The next worksheet will show the summary of case results and the profile of the individual or sample taken from a crime scene. That profile has two numbers corresponding to each of the known locations, for example, D3 S1358 - 15 and 16, meaning at point S 12358 on the 3rd chromosome there were 15 and 16 repeats of a particular sequence of bases. Further readings are given for the other nine loci, and a reading for Amelogenin, which can only be either XX (female) or XY (male).


These numbers can be computer coded and placed on the DNA database. When the same series of numbers comes up on another part of the database, for example, with a crime scene, suspect or convicted offender, a “match” is called and the two results further interpreted to see if the provisional match is justified. In other words, in the system used, no differences between the two samples could be found.


If the numbers do not match a suspect is excluded. If a suspect cannot be either matched or excluded the result will be reported as “not excluded”.


An inconclusive reading can result in the expert going further and giving either a “not excluded as a contributor” or a “not excluded as a source” finding. “Not excluded as a contributor,” means there was no match found, as there was simply insufficient material either to match or exclude the person. “Not excluded as a source,” means that it is possible a match may be made but the DNA is too weak or too complex a mixture to reach a reliable conclusion. That a person is “not excluded” can have no real relevance to proof of their guilt or innocence.


Statistics


A match between a crime scene sample and an offender is reported if the same series of allele numbers appears in the results for both samples. This is generally expressed in terms of a match probability, “The suspect has the same profile (in the Profiler Plus system) as the DNA recovered from point A at the crime scene. This profile is expected to occur in fewer than 1 in 10 billion individuals in the general population”. The statistics can also be expressed as a likelihood ratio, “It is about 10 billion times more likely the suspect left the sample than if a random person left the sample”. R v Karger (2001) 83 SASR 133 at 140.


There are only six billion people on earth. How is such an extraordinary figure arrived at, and what does it mean?


The first step in calculating the probability of a match is to find out what the chances are that another person chosen a random will have the same alleles at a certain locus. If there’s a match at a point the search can be further narrowed by testing another locus, and then another until all nine are examined. (In practice, a technique known as multiplexing allows all loci to be tested at once.) As matches are found at progressive loci, the argument becomes more convincing that the match is not by chance, and that the two samples come from the same source. Of course, if at any point the figures do not match, the suspect can be excluded.


How is this assessment turned into a calculation of probability? What is done bears some similarities to an opinion poll. Samples of DNA from hundreds of people are taken and an estimate made of the percentage of people in the general population who have a certain allele at a specific point/locus. Thus at locus D 13 S317, it may be found that 28% of people have allele 12 and 32% have allele 11. The chances of a random match between a sample and a person from the general population at that locus can then be calculated. An expert in population genetics should prepare these estimates.


The statisticians then apply a statistical formula known as the product rule. The product rule involves multiplying the relative frequencies of each item matched. When the results of each allele and loci are combined with the chances of a random match at the remaining eight loci, the combination of probabilities, or more correctly improbabilities, can become enormous. In Australia an adjustment is made to the product rule to make allowance for the fact that the suspect may come from a subpopulation whose usual genetic profiles may differ from that found in the general population. The Balding-Nichols formula. A profile that is common in the general population may be extremely rare in a subpopulation and visa versa, thus skewing the statistical results. Balding and Nichols worked out a mathematical formula that is said to correct for this problem.


In some jurisdictions, the United States for example, three different population databases are used for Caucasians, Hispanics, and African-Americans. In New South Wales, a single population database is used. The DAL uses a relatively large population database of 739 samples. A New South Wales Aboriginal database has recently been compiled using a population base of over 2500 profiles. In other Australian jurisdictions, Asian and Indigenous databases are sometimes used.


JM Butler has noted: Forensic DNA typing, biology, technology and genetics of STR markers, 2nd ed, 2005, Elsevier Academic Press, Burlington, USA, p 500.

 

“It is important to realise what a random match probability is not. It is not the chance that someone else is guilty or that someone else left the biological material at the crime scene. Likewise it is not the chance of the defendant being guilty or the chance that someone else in reality would have that same genotype. Rather, a random match probability is the estimated frequency at which a particular STR profile would be expected to occur in the population. This random match probability may also be thought of as the theoretical chance that if you sample one person at random from the population they will have the particular profile in question.”

In New South Wales, the DAL biologists do not usually calculate their own match probabilities. They use an Excel spreadsheet and a commercially available program. They simply add in the figures, including the theta value (see below) and up pops the match probability. It is generally so high for a nine loci match that even allowing for an error range, figures well in excess of 1 billion are generated.


DNA reports

The standard report prepared by the NSW Department of Analytical Laboratories is usually brief and to the point: “I’m not into this detail stuff, I’m more concepty”, Donald Rumsfeld again.

 

“The accused has the same profile as the DNA recovered from the crime scene sample. This profile is expected to occur in fewer than one in 10 billion individuals in the general population.”


 

The explanatory note that accompanies the report contains these important qualifications:

 

“The final statistical calculation … does not prove uniqueness, but provides strong support for the hypothesis (without taking other evidence into account) that the DNA from the evidence sample originates from the matched individuals. The profile frequency calculation does not apply to closely related individuals”. (Emphasis added).


 

Other States follow a similar format. For example In South Australia a Report from Forensic Science South Australia might say:

 

“The DNA profile obtained is approximately 990 million times more likely to match the DNA profile of the accused if he is the source of eh DNA rather than an unknown related male”


 

Standard information relating to calculations of likelihood ratios, match probability and the type of Database used is attached.


As with all legal documents it pays to read the fine print!


The reports generally note only the results of the analysis they do not describe the process by which the result was obtained. In South Australia provision is made in the Criminal Law (Forensic Procedure) Act 2007 s 49 for tender of an evidentiary certificate of the analyst, which in the absence of evidence to the contrary is proof of matters certified. In R v Sing, (2002) 54 NSWLR 31: see also R v Ryan [2002] VSCA 176. the NSW Court of Criminal Appeal held that, unless admitted by consent, the evidence from the expert who prepared the final report itself was not admissible. To make the evidence admissible, the chain of custody and handling of the exhibit from which the DNA was taken and analysed must be proved — from collection of the sample to final analysis.


This means that, unless the defence do not require them, everyone who handled, processed and analysed a sample should be called for cross-examination. The rationale for the decision was the danger of unfair prejudice that might arise if relevant witnesses were not called. A related issue was whether the expert’s conclusions, reliant as they were on hearsay, were admissible in any event.


In most cases the defence will allow the report to be tendered without objection. There should be more challenges. Evidence, particularly relating to the collection and handling of DNA samples or exhibits from which they are taken, should not be accepted uncritically. The opinion rule precludes the report’s admission unless it can be shown that the opinion in it is substantially based on the author’s training study or experience. Evidence Act 1995 (NSW) requirements were dealt with comprehensively by Justice Heydon in Makita (Australia) Pty Ltd v Sprowles (2001) 52 NSWLR 705 at [85]. For non-uniform Evidence Act States also Clark v Ryan (1960) 103 CLR 486, Wolper v Poole (1972) 2 SASR 419 at 421 per Bray CJ and R v Karger (2001) 83 SASR 1.


In many cases it is not the technology or the science but the supervising biologist’s subjective interpretation of the results that is the crucial factor in assessing whether a suspect sample and a crime scene sample “match”. What she/he is doing is looking at the computer read-outs and coming to a conclusion. In some cases the read-outs will be clear and conclusive, in some the readings will not be so clear and in others they will be far from clear at all. Where professional judgement and expertise are required to be exercised that evidence should be tested, as there is often fertile ground for doubt.


Problems with the science and technology involved in DNA analysis


A lot of time and effort is wasted on testing and challenging the unchallengeable. The science backing DNA analysis is good and getting better all the time. The technology has been tested and cross-tested. There are protocols in place designed to ensure that results are validated and possible contamination of results, or errors in analysis are detected. Mistakes have occurred but they are so rare as to be notorious. “UK DNA Mismatch” <http://www.scoop.co.nz/stories/HL0002/S00053.htm>; and “Murder DNA tests botched”, The New Zealand Herald, 26 May 1999. The police excused the error as “procedural difficulties in investigative analysis.” The incident led to the establishment of the Eichelbaum-Scott Inquiry, which, although it failed to find the cause of the error, stated that the most probable explanation was accidental lab contamination. Deliberate corruption of results has been noted in the United States, as has institutional bias. However, the Australian analytical labs are justifiably jealous of their reputations.


Nevertheless in any contested DNA matter it is often essential that the analyst’s expertise be established and the laboratory file examined, if only to check that all procedures and protocols were followed. “The cogency of DNA makes it particularly important that the DNA testing is rigorously conducted so as to obviate the risk of error in the laboratory. The method of DNA analysis and the basis of subsequent statistical calculation should — so far as possible — be transparent to the defence. The true import of the resultant conclusion [should be] accurately and fairly explained to the jury” R v Doheny & Adams [1997] 1 Cr App R 369; and R v Karger (2002) 83 SASR 135. Most DNA reports do not comply with the Director of Public Prosecution’s duty of disclosure

http://www.odpp.nsw.gov.au/guidelines/guidelines.html>.


The following areas require close scrutiny.


Partial match
A partial match reduces the opportunity for the full application of the statistical equation used to calculate the likelihood of a “match”, known as the “product rule”. In Australia the product rule is modified by the application of the Balding-Nichols Formula to make allowance for genetic similarities within some isolated communities, population substructures and the fact that random breeding does not take place. See “Assumptions” below. A partial match creates the chance that the missing portion may yield a result that would exclude the suspect. At a certain point the match probability figures become so low as to be meaningless.


Weak readings


A weak reading has similar problems to that of the partial match. It is sometimes impossible to tell the difference between a true reading at a locus and a glitch on the graph brought about by the testing process. As a result alleles may be wrongly counted or missed altogether. Most labs have a minimum peak height below which they will not hazard an assessment. On occasion, a match will be given despite a low peak height. Examination of low peaks can also disclose a potential extra contributor to a sample, raising the possibility that this person may be the true culprit, or the possibility of secondary transfer (see below).


The ACT Court of Appeal recently examined the practice by which a lab “overloaded” a mixed DNA sample and so enhanced minor peak heights, which would not otherwise fall into interpretable limits. The retrial subsequently ordered will now have to examine if this is a valid process. Hillier v R [2008] ACTCA 3, Madgwick, Weinberg and Dowsett JJ.


There is also a phenomenon known as a stutter, where an artefact of testing appears as a peak, mimicking an allele’s graph peak. Trained analysts claim to be able to ascertain the difference between an allele and a stutter. Stutters have been, and will continue to be, be interpreted as peaks with the consequence of a false match or false exclusion.


Similar problems can arise if only a single reading is found at one locus. A single reading can mean the alleles at that point are the same (known as ‘homozygosity’). It can also mean something has dropped out or not shown up on the graph (known as allelic drop-out or a null allele). A false positive or false negative reading can result.


Mixtures


Three or more alleles at a locus indicate the presence of more than one contributor. It is often difficult to tell whether the sample originated from 2, 3 or 4 people. Statistical models can help analysts work out the probabilities of more than two contributors however a number of possible combinations could be consistent with the findings. Despite this, unless more than five alleles are found at a locus, the government analyst’s generally assume that only two people contributed to a mixture. This presumption has been proved correct in most cases. However, if this assumption is false it can lead to a part of a mixture being matched wrongly to a suspect.


An assumption is then generally made that the taller peaks are associated with the primary contributor and the shorter peaks the secondary. If all the alleles can be matched to the crime scene or victim sample, these are then taken out and the assumption made that the remainder match only the suspect. Peak heights are often used in interpreting mixtures, but peak height imbalances occasionally occur. Peaks with the same height are presumed to be from the same person, but the reading of peak heights is far from an exact science. Even in a single person sample, peak heights may vary, so assumptions of regularity may be false.


In mixed samples alleles may simply be undetectable or indistinguishable from background “noise”. Alternatively, as mentioned above, alleles can simply “drop out” and not appear on the graph or stutters can be confused with alleles from a minor contributor.


Contamination


Most matches are correct, but errors do occur. There may be innocent, accidental or malicious reasons for a false match or a false exclusion. No matter how good the laboratory procedures are, errors do occur. Protocols are in place to pick these up. It is rare, but not unusual to find a worksheet noted “possible contamination”. This finding will result in the re-testing of the sample. In every case I have seen a second clean result was obtained. For most cases of potential contamination in the laboratory the innocent explanation is just that. However, the lab will not know if handling errors have occurred in the course of collecting or delivering the exhibit. Nor will it know if the sample has been deliberately tampered with before it arrived at the lab. In R v Lisoff [1999] NSWCCA 364, the Court of Criminal Appeal ordered a re-trial. Lisoff was subsequently acquitted, presumably because of doubts that the victim’s blood found on L’s trousers may have been planted. That blood was said to contain transfusion products and thus had arguably been taken from the victim after the assault and after he went to hospital.


Bias in the lab?

 

“Forensic scientists may become partisan. The very fact that the police seek their assistance may create a relationship between the police and the forensic scientists. And the adversarial character of proceedings tends to promote this process. Forensic scientists employed by the government may come to see their functions as helping the police. They may lose their impartiality”. R v Ward (1993) 96 Cr App R 1 at 51.

 

An example can be found in R v Button, [2001] QCA 133. in which the forensic scientist looked only for evidence that would implicate the accused and missed crucial evidence pointing to the real culprit, because they did not do the tests. Justice Williams described the various failures in the case as resulting in “…a black day in the history of the administration of criminal justice in Queensland.” Deliberate failure to investigate is rare but we must be alert. The more likely cause of a failure to investigate alternatives is pressure of work and a focus on output rather than using the genuine forensic expertise of the analysts. The more procedures are automated the less the analyst has to do with analysing the samples themselves. See S Walsh, “Is the double helix a double edged sword?”, paper presented to UTS Speaks public lecture, May 2005. The quest for volume can mean only one exhibit or part of an exhibit is analysed. Sometimes as R v Button shows this is simply not good enough. When an expert or technician is cross-examined about what was tested it is sometimes prudent to find what was not analysed.


Inadvertent or secondary transfer


Recent developments in DNA processing have enabled readable DNA to be obtained from tiny samples, unimaginable even a few years ago. DNA can now be recovered from a single cell and it is possible for as few as 30 cells to be processed in order to give a readable result. Similarly, DNA can now be recovered from objects where no bodily fluids are apparent, samples so small they can be obtained from a fingerprint impression and from items such as knife handles or spectacles. R van Oorschot et al (1997) 387 Nature 767. These finding have, however, not always been reproducible, C Ladd et al, “A systematic analysis of secondary DNA transfer”, (1999) 44 Journal of Forensic Science 1270–1272. See also J Raymond et al, “Trace DNA: An underutilised resource or Pandora’s box?” (2004) 56 Journal of Forensic Identification 668–686. In some cases, enough DNA can be recovered for analysis by conventional techniques. P Gill, “Biological Evidence”, paper presented to 13th Interpol Forensic Science Symposium, Lyon, France, October 2001; P Van Renterghem et al, “Use of latent fingerprints as a source of DNA for genetic identification”, (2000) 8 Progress in Forensics Genetics 501–503.


Given that we shed 40,000 skin cells each day, a lot of our DNA can be left lying around. It appears that some of us are “good shedders,” and some not. Experimental studies on Low Copy Number DNA have shown that a simple series of handshakes can transfer DNA from the original source to a third party.


Some experts are reluctant (if not dismissive) of suggestions that the DNA found by their tests got there by way of secondary transfer. They will say the secondary transfer tests involved smaller samples than are regularly tested for. They will say that transfer generally requires more than mere skin to skin contact. Although some studies support this conclusion others do not. See notes 16 and 17 above. At the same time they will triumph in their ability to get a sample from the smallest trace of material left at a scene.


DNA is highly mobile and secondary transfer does occur. Every time we speak and release spittle we have transferred our DNA. If, while still moist, this spittle comes in contact with another object, transfer of a few cells can occur. And only a few cells are enough to get a profile. After all, the taking of a swab from a crime scene is but an example of secondary transfer. We are dealing in such small quantities of material and such new areas of science that assumptions, which presume against secondary transfer, must be put aside.


The Inquest in to the death of Corporal Kovco who was shot with his own weapon while serving in Bagdad provided an incentive for the NSW DAL to review aspects of secondary transfer. Recent studies reveal:

 
That while “normal handshakes” do not appear to transfer DNA. This can be achieved under experimental conditions.
Sharing of caps can result in secondary transfer of a good shedder’s DNA onto a cap they have never worn;
DNA can be transferred to the surface of latex gloves while searching personal effects and then transferred from the gloves to a new surface with a sticky surface such as a balloon containing drugs: transfer by latex glove to oily surfaces such as the stock of a gun was not demonstrated. Poster presentations by Clayton Walton, Sharron Neville & Vivien Bielby- NSW DAL 2007. Note: these were not peer reviewed papers.


 

The trial of Barnes Unreported NSWSC trial (Wollongong), February 2004, Buddin J. provides an example of secondary transfer. A young woman was found dead in a park in Dapto, her discarded clothing covering her naked body. The accused’s DNA was recovered from her bra. Other evidence pointing to Barnes’s involvement inn the death was equivocal. Evidence established that about an hour before her death the two had met outside a club. Both were drunk, and the accused in particular was in a jolly mood shaking hands with a number of complete strangers. The problem posed for the defence was, how did his DNA come to be on the bra strap of a women who when they met was wearing a vinyl coat and a singlet over her bra? The DAL analyst was dismissive of suggestions that spittle sweat or skin from Barnes’s hand had got onto either the victim or her jacket and then been transferred to the bra. The jury as evidenced by their not guilty verdict were more accepting of the possibility of secondary transfer!


Problems with statistics


Statistics determine the probability of an event occurring by looking at possible successes and dividing them by possible outcomes. Predictions can be made in a general sense, but no statistical analysis can say what the next outcome will be. What statistics can do is give a model of expected behaviour. Those models can be independently tested and validated for consistency and rates of error but they are tools and models — they are not real. They must, of necessity, be based on a number of assumptions, which in turn rely on statistical rules, and the developing science of population genetics.


Assumptions


The first step in calculating the probability of a match is to find out what the chances are that another person chosen at random will have the same allele at a certain locus. Before any match probability or likelihood ratio can be calculated the range of possible outcomes must be found. This involves an understanding of population genetics and the making of some basic assumptions in formulating the model or database against which a suspect sample can be compared.


The basic model assumes that there is an infinitely large population in which no one selects a mate on racial or ethnic lines, a population in which there is no mutation, no migration and the biological process known, as natural selection does not apply. As none of these things are true, a basic model can overstate the value of the statistical evidence. Allowance is now made for what we now know about human genetics. Real populations do not breed randomly. Rather groups of people (subpopulations) tend to breed within the group. Thus within that group certain alleles may be more or less common than in the general population. A formula worked out by Balding and Nichols is applied to make allowance for this fact. H Roberts, “Statistical evaluation in forensic DNA typing”, in I Freckelton and H Selby, Expert Evidence: Law practice, procedure and advocacy, 3rd Ed, 2005, Law Book Co, Ch at para [80A].77380] –[80A.480].

 

In addition, it is assumed that:


· There is no genetic predisposition to crime, which might skew the results

.
· The samples used to compile the database are taken from a random selection of the population and contain no close relatives.


· The databases are as good as a whole population survey. They can’t be, as we don’t test everyone (yet!). The databases cannot be more than a model.


· The statistics take no account of other evidence. Such as “I have six brothers all of whom live in the area”, R v Watters [2000] EWCA Crim 89.or, “I have an evil twin”. If there is a chance a close relative or same sex sibling has left the sample the probabilities of a match are again quite different. For example, there is a 1 in 10,000-match probability with a same sex sibling and a 1 in 100 million chance of a match with a first cousin. Butler, op cit n 14, p 511.


· Variations among populations are truly random. They are not. (See discussion on theta/FST below.)


· There is no link between the alleles tested. An example of linkage is the fact that blonde hair normally occurs in conjunction with blue eyes. Linkage can skew the product rule calculations.


· We can tolerate a degree of uncertainty. As lawyers, this is where our finely-tuned doubt detection meters start to operate. How can there be uncertainty and not doubt?


Is a 9-locus profile unique?


A high match probability or likelihood ratio carries with it an implication that no one else has the same profile and that another match cannot exist. But statistical probability cannot predict the next outcome. The circumstances that led to one person’s genetic code may happen again by chance. A survey of DNA databases by the National Institute of Forensic Science examined 33,858 profiles, and found 206 matching pairs! At each of the nine markers used on the Profiler Plus system! This was significantly more than the statistical model predicted. S Walsh and J Buckleton, Report on Duplicate Detection accompanying; J Buckleton, S Walsh and R Mitchell, “Autosomal microsatellite diversity within the indigenous Australian sub-population”, 2004, National Institute of Forensic Science, Melbourne. New Zealand reviews of their database have also turned up an unusual number of matches. See J Buckleton, S Walsh and C Trigg, Forensic DNA evidence interpretation, 2005, CRC Press, Boca Raton, FL, USA, p 463. Matching pairs can be explained as twins, brothers or duplicates, because offenders use aliases or unrelated coincidence matches. Only by investigating each match can the real reason be known. To date the investigation needed to explain the matches has not been done. Unexplained matches illustrate the proposition that the statistics are simply another tool.


Random man not excluded


This method of calculating the likelihood ratio from mixed samples is generally more conservative than the product rule and is regularly used by some laboratories, for example, in Tasmania but not New South Wales. It involves looking at the sample as a whole and not first separating out “known” components, such as the victim’s profile. That different methods of calculation give different match probabilities illustrates the hypothetical nature of any “match” conclusion.


Theta/Fst


We tend to mate with people of similar genetic backgrounds, for example, it is unlikely for someone from Brazil to have a child with someone from Iceland. In some societies there is a cultural tendency to inbreeding, such as first cousin marriages. Small or isolated population groups tend to have similar patterns of DNA. Allowance is made for both direct and underlying relatedness in the population in the DAL calculations by adding in an error factor, called theta or Fst. Although calculated differently theta and Fst have approximately the same effect.


Studies have shown that for European and Caucasian communities these values are very low — less than 1%. Databases for the general Australian community still use a conservative figure of 0.03 or 3%. This is incorporated into the standard programme for calculating match probability.


Research into isolated communities genetically distant from the Caucasian model has shown however that the 3% figure is not conservative. Because of Australia’s long-term isolation prior to 1788 and restricted breeding practices in some Aboriginal groups a theta value of 6% has been recommended. NIFS study, Evidence given by J Buckleton in R v S (unreported, April 2006, NSWDC (Sydney), Norrish DCJ). The research looked at the maximum genetic variation between surveys of tribal groups and the error factor needed to allow a data base, based, for example, on the Tiwi Islands to be used for assessing probability ratios for someone from a mainland Aboriginal tribe. A variation or difference between people from Northern China and South China including Vietnam has been assessed as high as 6%, Roberts, op cit n 31, 80A–621, citing Cavalli-Sforza et al. Even this figure has been said to be too low. See R v Bropho [2004] WADC 182.


Relatives


Where a close relative could be a suspect or where the suspect or suspects come from a genetically isolated population, many of the assumptions on which the mathematical calculation of a probable match do not apply. Close relatives are more likely to have the same profile as a suspect than the randomly selected person used as a basis of match probability calculations because we share a much more limited genetic heritage than the population at large. If a brother cannot be excluded specifically and may have been involved then the match probability or likelihood ratios given by application of standard formulae must be very conservatively revised. If the brothers or cousins come from a small and genetically isolated group the figures must again be reduced.


The problems were graphically illustrated in the Western Australian case of R v Bropho. [2004] WADC 182; see also R v Watters [2000] EWCA Crim 89. This was a judge-alone trial of an old rape allegation. The complainant was not a reliable witness but a child had been born at the time of the alleged rape. DNA evidence of paternity pointing to the accused was crucial to the prosecution case. Initially the likelihood of the accused being the father compared to a random person using a theta/Fst value of 3% was assessed at 1:3134. However, a theta of 13% reduced this to 1:358. Once problems with a number of the loci used and the fact that close relatives were also suspects were considered the judge could not use the DNA evidence as “reliable corroboration”. Mr Bropho was acquitted.


Absence of DNA


Just because no DNA is found doesn’t prove that the defendant was not there. “The absence of evidence is not the evidence of absence. It is basically saying the same thing in a different way. Simply because you do not have evidence that something does exist does not mean that you have evidence that it doesn’t exist.” Donald Rumsfeld, US Secretary of Defence, again. Although DNA can be obtained from a sample of less than 0.5 of a nanogram A nanogram is 10 to the -9 of a gram, that is, 1/1,000,000,000th of a gram. of human cells extracted from crime scene material, testing of samples has varying success rates. When taken from fresh blood there is a 90% chance of the profile being obtained. Saliva on a cigarette butt gives a 67% chance of the profile being obtained. Saliva or sweat from a balaclava a 43% chance. Sweat left on an object from a hand a 17% chance. Fallen hair with dead roots has only 25% chance. The success rate is much higher for recently plucked hair roots. N Cowdery, “DNA innocence panel”, College of Law Paper (O3/45), June 2003.


The failure to find DNA may simply mean that the testing processes did not succeed. It helps neither the prosecution nor defence case. Finding someone else’s DNA can, however, exclude a suspect. If the suspect and crime scene samples differ at any of the markers measured on the Profiler Plus system then they are from different individuals.


The significance of others


There is a song by the rock band The Whitlams with the line, “She was one in a million. So there’s five more just in New South Wales.” “Up against the wall”, from the album, Eternal nightcap. Let’s be realistic, if the singer was silly enough to let her go, it is most unlikely that there are five more available and interested women in New South Wales at all. The same fallacy applies in a DNA case when the probability ratio is not at an extremely large level: It is wrong to say of a 1: X ratio, “Well there are X more persons who it could be and they have not been eliminated. So there must be a doubt, as it has been statistically proved that the real offender could still be out there.


The fallacy arises because this conclusion ignores all other factors personal to the accused, which make him a suspect above those other X persons; things such as opportunity, motive, proximity to crime scene, age, sex, and physical description.


If, on the other hand, there is simply no other evidence than the DNA “match”, the reasoning is not fallacious at all.


The prosecutor’s fallacy

 

“Two different questions may be asked regarding evidence of a match between the defendant’s profile and the profile obtained from a crime scene. First, what is the probability that the defendant’s DNA profile matches the crime scene sample profile, given that he or she is innocent? Second, what is the probability that the defendant is innocent given that his or her DNA profile matches the crime scene profile?” The first question assumes the innocence of the defendant and asks about the chances of getting a match; the second assumes that the defendant’s profile matches and asks about guilt or innocence. The “prosecutor’s fallacy” consists of mistakenly giving the answer to the first question as the answer to the second”. Australian Law Reform Commission, Essentially yours: Protection of human genetic information in Australia, Report 96, 2003, Vol 2 at [44.29]. See also D Balding and P Donnelly, The prosecutor’s fallacy and DNA evidence (1994) Criminal Law Review 711, 716–717; R v Keir (2002) 127 A Crim R 198; and R v Wakefield [2004] NSWCCA 288.

 

There is a clear danger of being overawed by statistics and falling into fallacious reasoning. “If your experiment needs statistics you ought to have done a better experiment”, Lord Rutherford quoted in NTJ Bailey, The mathematical approach to biology and medicine, 1967, Wiley & Sons, New York; “Statistics are like bikinis. What they reveal is suggestive, what they conceal is vital”; Aaron Levinstein; “Statistics mean never having to say you’re certain”, American Statistical Association T Shirt. An example comes from R v Keir. In evidence the DAL analyst said: (2002) 127 A Crim R 198 at [14].

 

“It is approximately 660,000 times more likely to obtain this particular DNA profile found in the bones if it comes from a child of CS and GB than from a child of a random mating in the Australian population”.

 

The error was in the Crown, and later the trial judge, transposed this statement and directed the jury: Ibid at [16].

 

“There is a 660,000 times to one chance that those are the bones of G K (the daughter of CS and GB) as distinct from any other person”.


 

Evaluating DNA evidence in context


Impressive statistics can indicate that it is unlikely that another person would have the same DNA profile as the defendant and the crime scene stain. Where that statistical certainty is challenged careful attention needs to be paid to the nature of the challenge. If the prosecution has not excluded the possibility of the accused having a sibling or other relative who may have left the sample, or where the expert called cannot support the statistical or population genetics behind their opinion, then the apparent certainty of the model used may be undermined.


So too, if there is other evidence that contradicts the DNA results. Examples recently accepted by juries (or the DPP in no-billing matters) include alibi and where the victim’s description of the offender simply failed to match in any reasonably acceptable way the description of the person matched by DNA linkage to the crime.


Where DNA is the only evidence or is critical to the case against the defendant, significant care must be taken when evaluating its efficacy in proving the prosecution case.


When the Crimes (Forensic Procedures) Act 2000 (NSW) was introduced the Police Minister Paul Whelan was explicit: Hansard, NSW Legislative Assembly, 31 May 2000, p 6293.

 

“It is important to note that DNA will be only one tool in the police officer’s kit. They will still need to assemble a brief of evidence against the offender; DNA alone will not convict!”

 

In R v Pantoja, (1996) 88 A Crim R 554 at 559, Hidden J agreeing. Justice Hunt made the same point. Although our understanding of DNA has advanced since 1996, the point still remains valid. A DNA link or match between the accused and a crime scene stain demonstrates only that the accused could be the offender. It does not establish that he was the offender.


In the same case Justice Abadee J Ibid at 583 and 584. put it more empathically. He held that the tribunal of fact must first be satisfied beyond reasonable doubt that there is a match between the two profiles. That means only that the defendant cannot be excluded and therefore it is possible he left the crime scene stain. Further, the matching results could not, in the absence of other evidence, prove beyond reasonable doubt the defendant was responsible for the crime scene stain.


In R v Watters, [2000] EWCA Crim 89. it was emphasised that, there was no rule, that when the statistics reached a certain level a prima facie case could be established. Rather, it was emphasised that the DNA evidence must be evaluated in the context of the other evidence in the case.


Pantoja was decided in 1988. Courts have subsequently expressed greater confidence in DNA and how it is presented. In R v Galli, the Chief Justice said: (2001) 127 A Crim R 493 at [98].

 

“The courts have approached DNA evidence with caution. However that caution is naturally abating as experience with the use of such evidence has grown.”

 

There have been strong statements in support of the proposition that DNA should be treated like fingerprint evidence.

 

“The evidence was the subject of expert opinion and close scrutiny. The reliability and accuracy of the DNA analysis is beyond reproach. It appears much safer than eyewitness identification”. R v Rowe [2004] SASC 427 at [40] (SA CCA).

“The use of DNA evidence appears to have reached a similar stage to fingerprint evidence in 1912. Like fingerprint evidence, Fingerprint evidence alone is good enough to convict: R v Parker [1912] VLR 152 (Vic CA); and in the High Court (1912) 14 CLR 681. it is of particular power, both to establish innocence and guilt.” R v Galli (2001) 127 A Crim R 493 at [106] per Spigelman CJ.

“If the results indicate that it is at least 10 billion times more likely that the crime scene sample originated from the accused than from any other unrelated person selected at random from the population, then if the evidence is sound then beyond any doubt the accused is the offender. To paraphrase Anderson J in R v Shaheed [2002] 2 NZLR 377 at [207].”

 

In most cases I have reviewed there is some other evidence, insufficient of itself to prove guilt, which the DNA evidence corroborates. Examples include: R v Gum [2007] SASC 311, where there were similarities in appearance between the accused and the alleged rapist; R v Fitzherbert [2000] QCA 255, where there was evidence of animosity and contact between the accused and the victim; R v Butler [2001] QCA 385 where the evidence was DNA and opportunity and R v Weetra [2004] SASC 337 where the accused lived nearby and stolen property was found near his home.


I have yet to find a superior court decision where DNA alone has been used to convict: eg where the Crown could not prove the defendant was in Australia at the relevant time. There are however cases where the barest of other evidence and a DNA link between a defendant and a crime scene, to a very high probability, has been used to convict. An example is R v Rowe: a police officer could at best give a general description of his assailant but where blood from which was recovered DNA linked to the accused was found on the officer shirt. As such it was not strictly a DNA only case, although it came very close. Doyle CJ said:

 

“The appellant sought to argue that because the DNA evidence was the only evidence of identification of the appellant, the conviction was somehow unsafe and unsatisfactory. The evidence was the subject of expert opinion. It was subjected to close scrutiny by the trial Judge who directed the jury that they must be satisfied beyond reasonable doubt as to the reliability and accuracy of the DNA analysis. It probably founded a safer basis for a conviction than the frailty often attending the evidence of a single eyewitness who gives evidence of identification of an offender. There is no substance in this ground.” (At [40])


 

I note that despite this view where DNA is the only evidence, the usual, but not exclusive practise, of the NSW Director of Public Prosecutions is to no-bill the matter. Cases where DNA is the only evidence are understandably rare, but are becoming more common as more and more serious offenders are placed on the offenders’ database. It was reported in The Advertiser of 28/8/2008 that South Australia had 41,161 profiles on its database, a 100% increase on a year before. Cold links are now being made, between crime scene stains and this database, with increasing frequency. Examinations of “cold cases exhibits” have turned up nuclear DNA from exhibits over 10 years old, see R v Stone (2004) 144 A Crim R 568. Stone pleaded guilty in 2004 to a 1990 murder. Sometimes cases are presented to court solely on the basis of this link. More often the link leads to further investigations and other evidence such as admissions, opportunity, identification or motive is presented. The DNA link then provides powerful corroboration of that other evidence.


Is DNA evidence sufficiently reliable for it to be used as the sole evidence against a defendant?


When an unadulterated suspect sample is compared with an unadulterated crime scene stain it is highly unlikely that there is error. However, because such errors have been known to occur it would not be wise to presume that the DNA evidence is infallible. There are a number or reasons why it is simply not possible to say that a match between a crime scene stain and a suspect proves he or she is the offender. These include:

· There is still so much we don’t know about DNA and statistics and population genetics, in he particularly areas of linkage between genes The best illustration is that some traits are not independent, for example, blonde hair and blue eyes. and interrelatedness. Illustrated by R v Bropho [2004] WADC 182.

· The potential for human error.

· The potential for contamination of samples. For an excellent review of why courts cannot be complacent about contamination, see K Edwards, “Ten things about DNA contamination that lawyers should know” (2005) 29 A Crim LJ

 

71. Most police forensic officers are well-trained professionals but not all exhibits are collected by trained professionals. Sometimes pressure of work or cost-cutting can lead to unacceptable shortcuts being taken. Examples of improper techniques that can lead to contamination of DNA samples include, improper bagging and storage of exhibits, for example, bundling them all in a bag or back of the car, transfer of DNA by fingerprint brushes, tweezers, or gloves which were used on more than one item.

· Handling errors because of the conditions in which the DNA is kept or stored. It is surprising how many samples, numbers and barcodes do not match with numbers and barcodes recorded in notebooks or exhibit books. DNA profiles can be obtained from such small amounts of body tissue or fluid, it is often hard to avoid cross-contamination of samples. The most recent example arose in R v Murdoch in the NT Supreme Court (the Falconio case) where DNA from the chief of the laboratory doing the testing was found on a crime scene sample.

· As we are dealing with statistical models a chance match simply cannot be excluded (and the possibility of a chance match increases if relatives or those from certain racial groups may be involved).

· There is always a possibility of tampering with the crime scene. This is not to impugn the police or Laboratory staff. Other criminals could have left the sample and it would be naïve to assume that all police can resist the temptation to plant evidence.

· The suspect’s sample could have arrived at the crime scene for a number of innocent reasons — secondary transfer or prior contact with the scene or exhibit.

· DNA technology allows only for comparison of computer-generated profiles of the two samples of only a tiny fragment of DNA, There are over three billion pieces in the DNA code. The Profiler Plus system looks at only 10 segments of that total code. unlike fingerprints, where there is an actual physical comparison between the fingerprint found at the scene and that of the suspect or defendant.

· There is danger that the statistical significance of a DNA match can be overstated. Some of the risks attaching to DNA evidence are only now becoming apparent. “The highly subjective nature of the mathematical process remains concealed behind the apparent certainty of a bald statistic”, Mathew Goode, “Observations on evidence of DNA frequency” (2002) 23 Adelaide Law Review 45 at 66–67. Examples include; the Eichelbaum-Scott report on DNA in New Zealand in 1999 The Rt Hon Sir Thomas Eichelbaum and Professor Sir John Scott, Report on DNA anomalies for The Hon Tony Ryall, New Zealand Minister for Justice, 30 November 1999, the inquiry in Victoria into how a female crime victim’s DNA was found in the Jaiyden Leskie murder, the more recent problem of contamination in the Gesah case The Age 22 July 2008 and 8 August 2008 and the number of what are called “unresolved pairs” often found when the data bases are searched for unexplained matches. Forensic DNA Evidence Interpretation, Buckleton, Triggs & Walsh CRC (USA) 2004 page 463. In 2005 Walsh and Buckleton reviewed Aboriginal DNA data bases for the National Institute of Forensic Science, in an unpublished report “on Duplicate detection” they noted that out of a sample group of 33,858 there were 1,575 matches, 206 occurred at 9 loci or greater. They explained these as being; coincidental matches between unrelated individuals, the same person giving more than one sample underran alias, close relatives matching or identical twins matching. The report also noted that in New Zealand a similar review had found 64 unresolved matches from a database of 50,000 people.


A DNA match thus shows that it is possible to a very high degree of probability that the defendant is the person responsible for leaving the stain. Despite the power of the statistical analysis that accompanies DNA testing I argue this can never be enough to prove a case beyond reasonable doubt in the absence of some other evidence for the DNA to corroborate. As the court said in Pantoja all the DNA match or link shows is that they could be the offender. In most, if not all, cases there will be good reasons why the mere fact of a match and the giving of a high match probability cannot be viewed with the certainty necessary of itself to prove the prosecution case beyond reasonable doubt.


Where the prosecution can point to no other evidence to support or corroborate their case against the accused, the case cannot be proved on the necessary high standard.


Where there is some other evidence, common sense leads to the conclusion that before a case can be proved beyond reasonable doubt all relevant evidence (DNA and otherwise) must be considered in context.


How should a court evaluate the DNA evidence?


The Supreme Court of British Columbia Australian Law Reform Commission, op cit n 45, Part J, Law Enforcement and Evidence Chapters 39–46, [44.50]. In the United Kingdom suggested guidelines can be found in R v Doheny [1997] 1 Cr App R 369 and in the Northern Territory in Latcha v The Queen (1998) 104 A Crim R 390. has suggested that before DNA evidence is presented to a court it should be made sufficiently clear that:


· the estimates are not intended to be precise;

· they are the products of mathematical and scientific theory not concrete facts;

· they do not purport to define the likelihood of guilt;

· they should only be used to form a notion of the rarity of the genetic profile of the accused; and

· the DNA evidence must be considered along with all the other evidence in the case relating to the issue of identification.


A useful start for considering a direction on DNA evidence comes from the leading South Australian decision of R v Karger: (2002) 83 SASR 135, Doyle CJ at 140–141.

 

“The statistical evidence interpreting the significance of the DNA match is not evidence of the probability that the appellant was the source of the incriminating DNA. To so regard it would be to make an error. However, the statistical evidence interpreting the DNA match is expert evidence that the jury could use in deciding whether it was satisfied beyond reasonable doubt that the appellant was the source of the incriminating DNA. The statistical evidence is undeniably strong evidence pointing to a conclusion that the accused was the source of the incriminating DNA, but is not direct evidence of that fact. And, as is obvious, the statistical evidence must be considered in the light of other evidence in the case.

It was for the jury to consider the evidence and to decide what significance and weight should be attached to the evidence. The jury were not obliged to act on the evidence. Nor should the jury allow any expert opinion put before them to be used as a substitute for their own satisfaction, to the appropriate degree of proof, of a matter required to be proved as part of the prosecution case. The proper approach to the issue of whether the incriminating DNA came from the appellant, and to the issue of guilt of the crime charged, was to treat the statistical evidence as evidence to be considered and weighed along with the other circumstantial evidence, not allowing it to displace or to overwhelm the consideration of all material evidence, but at the same time giving it such weight as the jury thought proper.”

 

The case of R v S unreported, April 2006, NSWDC (Sydney), Norrish DCJ. illustrates the point. A rape victim described her attacker in detail and by name. That person’s DNA did not match the semen found. He was discharged. Later a cold link was made to “S” who had been in custody for unrelated and quite different offences. He was put on trial. He lived in the town where the rape occurred. Apart from his Aboriginality he did not match the description of the attacker. He did, however, have a close relative who not only matched the description but also had the same first name as the originally nominated suspect. That relative was not tested. The prosecution experts said that the probability of “S” providing the sample was astronomically high. 1:370 billion rounded down to 1:10 billion. It was still enormous even if a close relative couldn’t be excluded. Doubt was cast on those figures by the disparity between the descriptions, the untested relative, the acceptance by all experts that the profile could not be said to be unique, disputes between experts about Aboriginal Fst and the revelation of the unexplained matches on databases reviews. The jury had a doubt despite the statistical ‘certainty’ of the match probabilities given and “S” was acquitted. In no DNA case, should the question of guilt be approached strictly on the basis of mathematical calculation. R v Galli (2001) 127 A Crim R 493; and R v GK (2001) 125 A Crim R 315.


Conclusion


DNA is just another piece of physical evidence. The statistical evidence interpreting the DNA match is expert evidence that can be used in deciding whether it has been proved that the appellant was the source of the incriminating DNA. When it comes to prove a case it is suggested that any DNA evidence and its accompanying statistics must be evaluated cautiously with knowledge of its limitations and in context with all the other evidence in the case. If there were no other evidence it is difficult to see how a conviction could result.


A DNA profile and its accompanying statistics are not real. They are scientific and statistical constructs. Even if the DNA evidence and statistics are good, questions can arise about how the match came about. This is an area that we as lawyers know and understand as it involves questions of proof, evidence and weight of evidence. Like all evidence DNA can be used, misused and abused.


Andrew Haesler *
2008


* Andrew Haesler SC is a NSW Barrister and Public Defender
Admitted to practice in 1981 Andrew worked as a Solicitor with the Redfern Legal Centre. He was the Centre’s Principal Solicitor from 1985 to 1989. He also worked in Alice Springs with the Aboriginal Legal Service and in Wollongong with the Legal Aid Commission. Called to the NSW Bar in 1990 he became a Public Defender in 1995. In 1999-2000 he was the Director of the Criminal Law Review Division of the NSW Attorney General’s Department and appointed Senior Counsel in 2004. He has given, and had published, many papers, on a variety of topics concerned with criminal law.



BIBLIOGRAPHY

Forensic DNA Typing, JM Butler, 2nd Ed. Elsevier (USA) 2005
Forensic DNA Evidence Interpretation, Buckleton, Triggs & Walsh CRC (USA) 2004
DNA Profiling on Criminal Investigations, L Wilson-Wilde, in Freckelton & Selby Expert Evidence, Chapter 80.
Statistical evaluation in forensic DNA typing, H Roberts, in Freckelton & Selby Expert Evidence Chapter 80A.
Evaluating Forensic Evidence: Essential elements of a competent defence review, WC Thomson, S. (http Ford, T. Doom, M. Raymer & D. Krane. Available at www.cs.wright.edu/itri/ ://www.cs.wright.edu/itri/EVENTS/SUMMER-INST-2003/SIAC03-Krane2.PDF)
Challenging Forensic Evidence, M Findlay & J Grix, 14 Current Issues in Criminal Justice 269 at 273.
DNA and the Changing Face of Justice, J Goodman-Delahunty & D Tait, Australian Journal of Forensic Science (2006) Vol 38 p.97.