In risk management, one of the key concepts is analysis. You must be able to analyze a situation in order to prepare the proper mitigation techniques. Ideally, this analysis is based on a scientific approach, which should allow for a series of tests from which data can be gathered and studied objectively. In these studies, the key concept is control—comparing the test condition against a controlled baseline. 

For example, an experimenter investigating impairment caused by mobile phones would compare each driver’s performance with and without phone usage. Nothing (or else very little) differs between the two tests. If there is an effect, it must be the cell phone that makes the difference because it is the only factor that changed. 

And this is where the breakdown between the need for information and the way that information is gained can occur. Epidemiological studies must rely on statistical control because experimental controls cannot be applied—the data already exist. (The researcher generally has no way to assess the data’s accuracy or validity, so there is always the chance of GIGO—garbage in, garbage out—operating.) If an epidemiological researcher wants to show that mobile phones increase accidents, he or she cannot simply count the number of accidents attributed to mobile phone distraction, as this alone provides no measure of risk. The researcher must also compare the number to a set of control data. Finding legitimate control data, however, can be difficult.

 For the comparison to be fair, the researcher must ensure that the result is not due to incidental factors such as age, income, gender or place of residence. This is virtually impossible because the number of potential variables is uncountable. 

In a study published in the New England Journal of Medicine entitled, “Association between Mobileular Telephone Calls and Motor Vehicle Collisions,” researchers Redelmeier and Tibshirani take the cleanest track by using subjects as their own controls. This eliminates many possible differences between accident and nonaccident data points. However, they cannot control all variables and must still make some assumptions—namely that other, unidentified important variables will be random and cancel out between groups. 

In a sense, the epidemiologist creates an implicit model of the world, one that meets certain subjective assumptions. But a model is not reality; it is only like reality. The question, of course, is whether it is like reality in the important ways. Researchers try to anticipate what these important ways are, but unfortunately the world is full of linkages that are not obvious or apparent. For example, no one comparing passenger conversation with mobile phone conversation knew to control for intensity. (See feature  “Adequacy of the Evidence” in November 2001)

Proper application of statistics further requires assumptions about the data that is gathered. Depending on the exact statistical method, the requirements often include observations that must be independent. The variables must fall in normal distributions with the same variance and without skew (for distribution to be symmetric, the data must be linear and have ratio-scale metric properties, which means that it must have a true zero point so that it can be properly  divided and multiplied). These requirements can seldom be proved, so they are just assumed. Redelmeier and Tibshirani attempted to validate one such variable, time of day, and were forced to rely on the accuracy of driver memory—yet another assumption. (There are methods that circumvent some of these difficulties, but they require even more assumptions in turn and produce weaker results.)

Perhaps the biggest potential pitfall is sampling bias. In order for the results to extrapolate to the entire population, the sample must also be random and representative. The classic example of sampling bias that caused an unreliable result occurred during the 1948 presidential election. The famous pollster George Gallup performed telephone surveys and predicted that Republican Dewey would defeat Democrat Truman in a landslide. Gallup’s sample was biased because Republicans were wealthier and more likely to have telephones, so most of the people surveyed were Dewey supporters.

Redelmeier and Tibshirani used subjects who had a specific type of accident and who voluntarily went to a single collision damage reporting center in North York, Ontario, outside of Toronto. The subjects were self-selected, both because they agreed to be in the study and because they bothered to report the accident at all. They were also limited to those whose phone records were available. Is this a representative sample of all mobile phone drivers in the United States, who have had different types of accidents and who drive under different kinds of conditions? Are there other idiosyncrasies of the situation that are not apparent and which may be significant? Of course, biased sampling can also underestimate effect size, which Redelmeier and Tibshirani suggest.

When studies tell us what we want to hear, it is sometimes difficult to stop and consider not only all available sources of information, but also the fundamental mathematical basis behind those results.

—Marc Green

Identity Theft

A report published in September indicates that identity theft is the fastest growing white-collar crime in the United States, with financial institutions predicted to lose in excess of $8 million annually within the next three years.

In the study, Identity Theft and Its Effect on the Financial Services Industry, Sang Lee, an analyst at Celent Communications, examines the rapidly growing cases of identity theft and stresses that more proactive steps must be taken against it. 

Unless financial institutions take a more active role in reversing this trend, financial losses stemming from identity theft will continue to increase at an alarming rate.

 “While identity theft had existed prior to the advent of the Internet, there is no question that in recent years, criminals have taken advantage of all the readily available confidential information on the Internet,” says Lee. “While identity theft cannot be completely eliminated, there are steps that financial institutions and consumers can take to minimize the chances for identity theft-related crimes.” 

These steps include:

• Tightening internal security to ensure confidential customer information is handled properly

• Establishing a central fraud office to handle reports of identity theft from consumers

• Conducting customer service training to prevent information leaks from customer service representatives 

• Educating customers about efforts to prevent identity theft

“It is in the interest of the financial institutions to spend the necessary time and money to implement procedures to limit confidential customer information leaks internally,” say Lee, “as well as to educate the consumers regarding the dangers of identity theft.”

—Stephen Nickson

Internet Policies Sales Strengthen

Reinsurer GE Frankona Re’s most recent survey of the stand-alone term insurance market shows Internet sales are beginning to grow. With overall term insurance sales up by 8 percent in 2000, Internet sales accounted for over thirty thousand policies, or 3 percent of sales.

“We expect sales to flourish as U.K. consumers become more comfortable with purchasing life products over the Internet,” says Peter Elliot, head of sales at GE Frankona Re. “Experience in the Internet sales in general, where U.K. purchasers are leading the way in Europe, points to a promising future.” 

Level term policies remain the favorite plan type, accounting for just over half of total sales. The decreasing term market, which accounts for 36 percent of total sales, continues to stay in line with the mortgage market.  

While growth in North American markets is still sluggish, it may be just a question of time. “It comes as no surprise that Internet sales are beginning to make an inroad into term distribution,” says Elliot. “The product is simple to understand and keenly priced, both of which make it ideal for distribution over the Internet.”

—SN 

Call for Risk Managers

The American Risk and Insurance Association (ARIA) is asking for research presentation proposals for its annual meeting to be held from August 11 to 14, 2002, in Montreal, Canada. 

Specific subject areas include, but are not limited to, insurance law or regulation, public policy, economics, finance, health care, international issues, employee benefits or risk management. 

Submit executive summaries (less than three pages) that focus on the purpose, expected results and importance of the research, with the names and affiliations of all co-authors, and the telephone, fax number and e-mail address of the designated contact person on a separate page.

The deadline for submissions is February 15, 2002. Mail four copies to Robert E. Hoyt, ARIA Vice President and 2002 Program Chair, University of Georgia, Terry College of Business, Brooks Hall 206, Athens, GA 30602, USA or e-mail t-aria@terry.uga.edu. 

RIMS and Business Insurance are asking for nominations for the Risk Manager of the Year Award. The deadline for submissions is November 19. The entry form can be downloaded from the RIMS Web site at www.rims.org.