As described in Assignment Seven, the exam will have both open book and closed book questions.  Be sure to bring your copy of The Crime Drop in America.  For that portion of the test, you will be able to refer to both the graphs and the text.

The chapters to be covered in The Crime Drop in America are:  1, 2, 4, 5, 7 and 9.  In reviewing for the closed-book part of the exam, you do not need to try to memorize any numerical findings.  You should be able to answer questions about the authors' main conclusions.  For example, don't try to commit the arguments on pages 101-107 to memory, but try to remember the conclusions from the econometric analyses as summarized on page 108 (the prison expansion added a little to the drop in crime, but most of it would have occurred anyway).

There will also be questions on ethical analysis, so you should review the Introduction to Ethical Analysis  and the excerpt from the Belmont Report.  There will also be some questions on What's So Bad About Hate? and questions about the Forum on the Ethics of Capital Punishment.  The other readings on the Course Web Page are mostly illustrative and will be be covered only occasionally if at all.

Questions on my article on Myths of Murder and Multiple Regression will be limited to general points.  These are summarized in the Abstract and Conclusions, which are reproduced here:

Abstract:  Multiple regression has consistently failed to provide definitive answers to policy controversies in criminal justice, yet researchers continue to attempt to use regression techniques for this purpose.   This review of multiple regression analyses of trends in homicide rates suggests that the method fails because researchers overfit their models to one data set, then fail to test them with fresh data.   The lack of progress in regression modeling of homicide trends over several decades suggests that the trends may be chaotic.  Studies that disaggregate trends and combine qualitative with quantitative data have been much more successful.

5. Conclusions.  We have examined several cases when apparently quite rigorous multiple regression studies, done by highly qualified researchers, were soon debunked by other equally qualified specialists using equally rigorous techniques.  We found several reasons why these studies failed:
? The researchers often overfitted their models to a single data set, then failed to test them with fresh data.  This testing was then done by critics, who found that the models did not hold up when tested with different time periods or units of analysis.
? The available data often did not meet the formal mathematical requirements for a regression analysis.  In some cases, this led to serious problems, e.g., the lack of any variation in the "shall issue" variable in America's largest cities, which happened also to be the location of an epidemic of killings by well armed drug gangs.  This is often a problem in social impact analyses because policies are often implemented only in a few jurisdictions.  Multiple regression is probably not an appropriate tool for the analysis of data sets, such as John Lott's concealed weapons data set, that depart so much from the mathematical assumptions of the technique.
? The amount of data available was often simply too small, allowing chance variations too much weight in the equations.  This is an inherent problem using regression to study historical trends.  The accidents of history become reified into mathematical principles.  The quality of the statistics also often varies significantly over time.  Again, multiple regression may not be an appropriate tool with such limited data sets, certainly it should not be the only tool used in a study.
? Many of the trends were markedly non-linear, yet multiple regression assumes linear relationships.  The techniques the analysts used to correct for nonlinearity introduced other distortions or obscured important findings.  If trends are markedly non-linear, it would probably be better to use other techniques instead of make heroic efforts to fit the data into a linear mold.
? Important variables were often omitted, either because they were overlooked or because quantitative measures were not available.  The apparent impact of the legalization of abortion was left out of all regression analyses of homicide rates prior to 1999, simply because no one thought of it.  It may continue to be left out because of its sensitive political implications, even though statistically it seems to explain a great deal.  Other important variables, such as the importance of the cocaine gang wars on homicide rates and of the war against drugs on imprisonment rates, may have been left out because good quantitative measures were not readily available.
5. Conclusions.  We have examined several cases when apparently quite rigorous multiple regression studies, done by highly qualified researchers, were soon debunked by other equally qualified specialists using equally rigorous techniques.  We found several reasons why these studies failed:
 * The researchers often overfitted their models to a single data set, then failed to test them with fresh data.  This testing was then done by critics, who found that the models did not hold up when tested with different time periods or units of analysis.
 * The available data often did not meet the formal mathematical requirements for a regression analysis.  In some cases, this led to serious problems, e.g., the lack of any variation in the "shall issue" variable in America's largest cities, which happened also to be the location of an epidemic of killings by well armed drug gangs.  This is often a problem in social impact analyses because policies are often implemented only in a few jurisdictions.  Multiple regression is probably not an appropriate tool for the analysis of data sets, such as John Lott's concealed weapons data set, that depart so much from the mathematical assumptions of the technique.
 * The amount of data available was often simply too small, allowing chance variations too much weight in the equations.  This is an inherent problem using regression to study historical trends.  The accidents of history become reified into mathematical principles.  The quality of the statistics also often varies significantly over time.  Again, multiple regression may not be an appropriate tool with such limited data sets, certainly it should not be the only tool used in a study.
? Many of the trends were markedly non-linear, yet multiple regression assumes linear relationships.  The techniques the analysts used to correct for nonlinearity introduced other distortions or obscured important findings.  If trends are markedly non-linear, it would probably be better to use other techniques instead of make heroic efforts to fit the data into a linear mold.
? Important variables were often omitted, either because they were overlooked or because quantitative measures were not available.  The apparent impact of the legalization of abortion was left out of all regression analyses of homicide rates prior to 1999, simply because no one thought of it.  It may continue to be left out because of its sensitive political implications, even though statistically it seems to explain a great deal.  Other important variables, such as the importance of the cocaine gang wars on homicide rates and of the war against drugs on imprisonment rates, may have been left out because good quantitative measures were not readily available.