- Email: [email protected]
Movies and smoking
CORRESPONDENCE
The dark side of the UK National DNA Database Sir—Adrian Linacre (May 31, p 1841)1 is an enthusiastic proponent of the UK’s National DNA Database (NDNAD). We wish to offer a more critical perspective of the database. In 2002, a young girl was murdered in a Tuscan village by a foreigner. From a bloodstain left at the crime scene, the suspect’s DNA profile was produced and circulated on the Interpol net. In February, 2002, P N Hankin, a barman in England was identified from the NDNAD and charged with the murder. He denied the charge and protested that he had been at his workplace on the day of the murder, which was corroborated by several eye-witnesses. He was, however, arrested. Eventually, British officers reported a fresh profile from a sample volunteered by Hankin, with one previously unseen mismatch. He was subsequently cleared of all charges. This case highlights some of the limitations of DNA profiling. Inferring a person has committed a crime on the basis of a profile match alone is problematic. The rest of the evidence should not be overlooked in favour of the DNA match. The use of DNA databases does not prevent forensic scientists from misrepresenting the population associated with a crime (the forensic population). A forensic population is not an abstract category, but a small group of individuals identified by background evidence. In the past, concerns have been raised about the existence of genetic isolates and the forensic use of gene frequency multiplication rule.2 These hypotheses have been strenuously and effectively criticised.3 But informing a person’s guilt probablility by comparison with the national (or whatever other abstract) population is a flawed approach that has little to do with population genetics axioms and their exceptions. Linacre’s statement that “such high probabilities lead the investigator to place the person at the crime scene” reiterates an old cliché—technically known as the fallacy of transposed conditional4—here embodied in the desultory idea that every man on earth can pass by a Tuscan village on a certain day and commit a crime. Databases contain undetected errors—the larger the database, the more prone it is to errors. As it turns out, hidden errors are part of forensic evidence, in the sense that we can modify our beliefs according to the chance of their occurrence. We believe
834
national DNA databases offer an opportunity to study errors and their real-scale effects. Why overstate the importance of these databases when the real population of criminal suspects is so small and the risk of misclassification with DNA from a national database so tangible. The forensic community needs to adopt a more circumspect attitude towards the use of DNA evidence from national databases in criminal cases and recognise the problems inherent in such an approach. *Vince L Pascali, Giampietro Lago, Marina Dobosz *Istituto di Medicina Legale, Università Cattolica, largo F Vito, 1 00168 Rome, Italy (VLP, MD); and Reparto Carabinieri Investigazioni Scientifiche, Rome (GL) (e-mail: [email protected]) 1 2
3
4
Linacre A. The UK National DNA database. Lancet 2003; 361: 1841–42. Lewontin RC, Hartl D. Population genetics in forensic DNA typing. Science 1991; 254: 1745—50. Lander ES, Budowle B. DNA fingerprinting dispute laid to rest. Nature 1994; 371: 735–38. Aitken CG. Statistics and the evaluation of evidence for forensic scientists. Wiley: Chichester, 1995: 36–37.
2
a cohort study. Lancet 2003; 362: 281–85. Glanz SA. Smoking in movies: a major problem and a real solution. Lancet 2003; 362: 258–59.
Sir—Madeline Dalton and coworkers1 show a direct correlation between exposure to smoking in movies and increased initiation of smoking in adolescents; a causative connection is assumed. However, the study design does not exclude the possibility that underlying factors—eg, familial background, peer behaviour— have a direct effect on smoking habits and on choice of movies. We believe group behaviour is the actual cause for their results. A considerable amount of empirical evidence identifies peer relationships as the primary cause of increased frequency of smoking in adolescents.2 Furthermore, the effect of peer pressure could influence the choices of movies and consequently lead to a higher cumulative exposure to smoking in movies. Maybe Dalton and colleagues’ findings indicate an effect rather than a cause? *Hans Peter Colvin, Joshua Colvin, Bernhard Steger
Movies and smoking Sir—In their article (July 26, p 281),1 Madeline Dalton and colleagues state in that “52·2% (30·0–67·3) of smoking initiation can be attributed to exposure to smoking”. However, their results show that 52·2% of the variance attributable to exposure to smoking in movies is apparent only after all other covariates have been controlled for. Many of these covariates are greatly associated with the dependent variable, smoking initiation. What proportion of the variance is explained by the other covariates and, hence, what amount for variance left for exposure to smoking in movies remains? 52·2% of what is explained by the variable? In his accompanying Commentary,2 Stanton Glanz (p 258) uses the figure of 52·2% to estimate the number of adolescents in the USA that begin smoking as a result of watching smoking in movies. I think this extrapolation from the findings is inappropriate, given the concerns raised above. Bob Volkmer Child and Youth Health, 295 South Terrace, Adelaide 5000, Australia (e-mail: [email protected]) 1
Dalton MA, Sargent JD, Beach ML, et al. Effects of viewing smoking in movies on adolescent smoking initiation:
School of Medicine, Leopold Franzens University of Innsbruck, A-6020 Innsbruck, Austria (e-mail: [email protected]) 1
2
Dalton MA, Sargent JD, Beach ML, et al. Effects of viewing smoking in movies on adolescent smoking initiation: a cohort study. Lancet 2003; 362: 281–85. Kobus K. Peers and adolescent smoking. Addiction 2003; 98 (suppl 1): 37-55.
Authors’ reply Sir—We appreciate the careful reading of our article by Bob Volkmer, and by Hans Peter Colvin, Joshua Colvin, and Bernhard Steger. Volkmer questions the 52·2% statistic and the contribution of other covariates. To be precise, 52·2% is the attributable risk of movie exposure and not the proportion of variance explained, as Volkmer suggests. The attributable risk was calculated on the basis of the overall rate of 9·96% for smoking initiation in our sample, which was reduced to 4·71% when we limited movie-smoking exposure to the lowest quartile for all adolescents. Had we not controlled for any other covariates, limiting movie-smoking exposure to the lowest quartile would have reduced the rate of smoking initiation by 66·0% to an overall rate of 3·4%.
THE LANCET • Vol 362 • September 6, 2003 • www.thelancet.com
For personal use. Only reproduce with permission from The Lancet.
The dark side of the UK National DNA Database Sir—Adrian Linacre (May 31, p 1841)1 is an enthusiastic proponent of the UK’s National DNA Database (NDNAD). We wish to offer a more critical perspective of the database. In 2002, a young girl was murdered in a Tuscan village by a foreigner. From a bloodstain left at the crime scene, the suspect’s DNA profile was produced and circulated on the Interpol net. In February, 2002, P N Hankin, a barman in England was identified from the NDNAD and charged with the murder. He denied the charge and protested that he had been at his workplace on the day of the murder, which was corroborated by several eye-witnesses. He was, however, arrested. Eventually, British officers reported a fresh profile from a sample volunteered by Hankin, with one previously unseen mismatch. He was subsequently cleared of all charges. This case highlights some of the limitations of DNA profiling. Inferring a person has committed a crime on the basis of a profile match alone is problematic. The rest of the evidence should not be overlooked in favour of the DNA match. The use of DNA databases does not prevent forensic scientists from misrepresenting the population associated with a crime (the forensic population). A forensic population is not an abstract category, but a small group of individuals identified by background evidence. In the past, concerns have been raised about the existence of genetic isolates and the forensic use of gene frequency multiplication rule.2 These hypotheses have been strenuously and effectively criticised.3 But informing a person’s guilt probablility by comparison with the national (or whatever other abstract) population is a flawed approach that has little to do with population genetics axioms and their exceptions. Linacre’s statement that “such high probabilities lead the investigator to place the person at the crime scene” reiterates an old cliché—technically known as the fallacy of transposed conditional4—here embodied in the desultory idea that every man on earth can pass by a Tuscan village on a certain day and commit a crime. Databases contain undetected errors—the larger the database, the more prone it is to errors. As it turns out, hidden errors are part of forensic evidence, in the sense that we can modify our beliefs according to the chance of their occurrence. We believe
834
national DNA databases offer an opportunity to study errors and their real-scale effects. Why overstate the importance of these databases when the real population of criminal suspects is so small and the risk of misclassification with DNA from a national database so tangible. The forensic community needs to adopt a more circumspect attitude towards the use of DNA evidence from national databases in criminal cases and recognise the problems inherent in such an approach. *Vince L Pascali, Giampietro Lago, Marina Dobosz *Istituto di Medicina Legale, Università Cattolica, largo F Vito, 1 00168 Rome, Italy (VLP, MD); and Reparto Carabinieri Investigazioni Scientifiche, Rome (GL) (e-mail: [email protected]) 1 2
3
4
Linacre A. The UK National DNA database. Lancet 2003; 361: 1841–42. Lewontin RC, Hartl D. Population genetics in forensic DNA typing. Science 1991; 254: 1745—50. Lander ES, Budowle B. DNA fingerprinting dispute laid to rest. Nature 1994; 371: 735–38. Aitken CG. Statistics and the evaluation of evidence for forensic scientists. Wiley: Chichester, 1995: 36–37.
2
a cohort study. Lancet 2003; 362: 281–85. Glanz SA. Smoking in movies: a major problem and a real solution. Lancet 2003; 362: 258–59.
Sir—Madeline Dalton and coworkers1 show a direct correlation between exposure to smoking in movies and increased initiation of smoking in adolescents; a causative connection is assumed. However, the study design does not exclude the possibility that underlying factors—eg, familial background, peer behaviour— have a direct effect on smoking habits and on choice of movies. We believe group behaviour is the actual cause for their results. A considerable amount of empirical evidence identifies peer relationships as the primary cause of increased frequency of smoking in adolescents.2 Furthermore, the effect of peer pressure could influence the choices of movies and consequently lead to a higher cumulative exposure to smoking in movies. Maybe Dalton and colleagues’ findings indicate an effect rather than a cause? *Hans Peter Colvin, Joshua Colvin, Bernhard Steger
Movies and smoking Sir—In their article (July 26, p 281),1 Madeline Dalton and colleagues state in that “52·2% (30·0–67·3) of smoking initiation can be attributed to exposure to smoking”. However, their results show that 52·2% of the variance attributable to exposure to smoking in movies is apparent only after all other covariates have been controlled for. Many of these covariates are greatly associated with the dependent variable, smoking initiation. What proportion of the variance is explained by the other covariates and, hence, what amount for variance left for exposure to smoking in movies remains? 52·2% of what is explained by the variable? In his accompanying Commentary,2 Stanton Glanz (p 258) uses the figure of 52·2% to estimate the number of adolescents in the USA that begin smoking as a result of watching smoking in movies. I think this extrapolation from the findings is inappropriate, given the concerns raised above. Bob Volkmer Child and Youth Health, 295 South Terrace, Adelaide 5000, Australia (e-mail: [email protected]) 1
Dalton MA, Sargent JD, Beach ML, et al. Effects of viewing smoking in movies on adolescent smoking initiation:
School of Medicine, Leopold Franzens University of Innsbruck, A-6020 Innsbruck, Austria (e-mail: [email protected]) 1
2
Dalton MA, Sargent JD, Beach ML, et al. Effects of viewing smoking in movies on adolescent smoking initiation: a cohort study. Lancet 2003; 362: 281–85. Kobus K. Peers and adolescent smoking. Addiction 2003; 98 (suppl 1): 37-55.
Authors’ reply Sir—We appreciate the careful reading of our article by Bob Volkmer, and by Hans Peter Colvin, Joshua Colvin, and Bernhard Steger. Volkmer questions the 52·2% statistic and the contribution of other covariates. To be precise, 52·2% is the attributable risk of movie exposure and not the proportion of variance explained, as Volkmer suggests. The attributable risk was calculated on the basis of the overall rate of 9·96% for smoking initiation in our sample, which was reduced to 4·71% when we limited movie-smoking exposure to the lowest quartile for all adolescents. Had we not controlled for any other covariates, limiting movie-smoking exposure to the lowest quartile would have reduced the rate of smoking initiation by 66·0% to an overall rate of 3·4%.
THE LANCET • Vol 362 • September 6, 2003 • www.thelancet.com
For personal use. Only reproduce with permission from The Lancet.