Too Many Warnings?

At 5:35 P.M. on September 17, 2018, the University of Maryland Police Department (UMPD) sent a tornado warning by email and text message to university students, faculty, and staff: "A Tornado Warning has been issued for the UMD campus. The sirens will be activated. Seek shelter immediately, avoid windows."  This message did not mention that the warning was issued by Accuweather.  The National Weather Service (NWS) did not issue a tornado warning.  At 6:04 P.M., the UMPD sent the following message: "The Tornado Warning that was issued by AccuWeather is now cancelled as of 6:00 PM."

Part of Accuweather's business is generating severe weather alerts for its clients. Its website touts the benefits of Accuweather's warnings and its "null tornado notifications" that a NWS tornado warning will not affect a client's facility.  The website discusses the benefits of knowing that a tornado warning is not a direct threat, so a shutdown is not necessary, and claims that Accuweather's false alarm rate is lower than the NWS false alarm rate (10% to 80%).

When one organization issues a warning while the other does not, the inconsistency could decrease trust in warnings, increasing the likelihood of ignoring warnings.
An article in The Washington Post about the false alarm (Angela Fritz and Sarah Larimer, "Red flags at U-Md. over false alarm for tornado," September 19, 2018) quoted Gary Szatkowki, a former NWS meteorologist, who asked "What do we want to do about a weather organization issuing a tornado warning when the Weather Service does not?"

Although the University of Maryland is a large institution (approximately 40,000 students), its cost of a false alarm is smaller than the same cost for a factory (where a shutdown directly impacts throughput and revenue).  Moreover, many students and others are outdoors, and the university has a strong desire to be viewed as a safe place, so the potential impact is large.

In that case, the university should be willing to tolerate more warnings.  Too many warnings?  No, in this case.

Improving Decision-Making Processes

The WomenCorporateDirectors Foundation’s Thought Leadership Commission and the KPMG Board Leadership Center issued a report that describes some of the problems that can cause poor decision making and recommends improving decision-making processes.  The report discusses incomplete information, groupthink, overconfidence, and other poor practices as causes of decision-making failures. 

The report presents five decision-making styles (first presented in an article by Dan Lovallo and Olivier Sibony):

• Visionary,
• Guardian,
• Motivator,
• Flexible, and
• Catalyst.

Each style has its strengths and weaknesses.  I would add that a decision-maker needs to use the right decision-making style for the decision-context that is present (see more about this at this blog post). 

The report also discusses ways to create a more inquisitive, risk-based decision-making process by considering multiple viewpoints, identifying the pros and cons of every alternative, and discussing the associated risks (what could go wrong).  The resulting process will resemble the "discovery decision-making process" described by Paul Nutt.

Finally, the report recommends evaluating the decision-making process, not its outcomes, as a way to identify opportunities for improvement.

HT: ISE Magazine

Scientists describe. Engineers decide.

The May, 2018, issue of Mechanical Engineering magazine included the article "Without Engineering, Civilization Does Not Exist" by Adrian Bejan.  I wrote the following letter about this article; my letter appeared in the July 2018 issue.

To the Editor:

I wholeheartedly concur with Adrian Bejan's conclusion that engineering has transformed our society for the better.  Unfortunately, Professor Bejan's definition of engineering as type of science was inaccurate, which contributes to the public's misunderstanding of our profession.

The article stated that "engineering is a science of what is useful" and "engineering is [a] body of science."  It concluded by repeating that engineering is "the science of useful things."

Although engineers need to learn science and some scientists do some engineering, engineering is not a type of science, for it has a different objective.  Scientists study the world in order to understand it, but engineers design new products and systems.  Scientists describe; engineers decide.  Henry Petroski discussed this distinction and provided more ways to view it in An Engineer's Alphabet.  Professor Bejan was closer to the truth when he described engineers as those who "are developing new contrivances and improving old ones."

Jeffrey W. Herrmann

Visualizing a distribution

Understanding how the students in my course did on an exam includes visualizing the distribution of their scores.  In the past I used two standard visualizations: a histogram of the letter grade and a scatter plot that showed the distribution (similar to a cumulative distribution function).

For example, consider a final exam that 49 students took.  The exam was worth 35 points, so the thresholds for letter grades (90% = "A," etc.) were 21, 24.5, 28, 31.5.  After converting each student's score to a letter grade, I generated a histogram of the letter grades (Figure 1), which clearly shows that the most common grade was a "B."  To get more details, I also generated the full distribution of the exam scores (Figure 2).

Figure 1.  The histogram.

Figure 2. The full distribution (each dot is one score).

Although the vertical gridlines in the full distribution are on the thresholds for letter grades, there is a great deal of wasted space, and it is not obvious on this chart by itself which letter grade was most common.  (In a larger data set, the vertical distance between points would have to shrink and could become too small to distinguish the markers.)  To overcome this limitation, I changed the cumulative count to restart at each threshold, which yielded the chart (which I am calling a "distrogram") in Figure 3.

Figure 3. The distrogram.

There are now four "curves," one for each letter grade, and each curve shows the distribution of scores in that letter grade.  The height of each curve shows the total number of scores in that letter grade (as the histogram does).  Thus the distrogram clearly shows that the most common grade was a "B."  It also shows that the scores that correspond to "C" (between 24.5 and 28) were near the threshold for a "B"  and that no one earned a perfect score (a 35).  Thus, this distrogram provides the same information as the histogram plus additional information in a layout that is easier to navigate than the full distribution (the second chart).   Because there are multiple curves, there is more vertical space between markers.  Compared with the full distribution, however, the distrogram does require more operations to answer a distribution question such as "How many students earned at least 30 points on the exam?" because one would have to add the counts for multiple bins.

A distrogram should be useful for numerical data where the individual values and their grouping into categories (such as letter grades) based on these values are important.  The key feature is that the simple bar in the histogram is replaced by a scatter plot showing the distribution of the values in that bin.  Markers are needed to show the individual values; lines connecting the markers are not necessary; if they are used, they should be light so that the markers are easy to see.  Horizontal and vertical gridlines should also be light if used.

To create a distrogram, define the thresholds for the bins and sort the values in ascending order.  Determine the bin for each value.  Add a cumulative count that starts at 1 and increases by 1 at each value (even if this value and the previous one are equal).  Reset the cumulative count to 1 when the new value and the previous value are in different bins.  Create a scatter plot, with the values on the horizontal axis and the cumulative count on the vertical axis

Japanese food that minimizes the risk of choking

NPR posted a piece about food that minimizes the risk of choking.  In Japan, more people die from choking than from traffic accidents, and the difficulty that elderly person have swallowing is a leading cause of choking.  The cooked food is pureed and then re-formed (with a thickener) into a dish that looks like regular food but is easier to swallow (no chewing required).

Meanwhile, The Washington Post had an article about the importance of knowing the Heimlich maneuver and CPR, both of which can help someone who is choking. 

These highlight both sides of managing risk: (1) preventing a potential problem (choking) by eating foods that are less likely to cause choking, and (2) following a contingency plan (the Heimlich maneuver) if someone does start choking.

Bill Murray, who played a weatherman who saves someone from choking in the 1993 movie Groundhog Day, saved a man from choking in a Phoenix restaurant in 2016 by using the Heimlich maneuver, which he learned while making the movie.

Educating Future Engineers by Learning about Design Processes

I had the pleasure of attending the Clive L. Dym Mudd Design Workshop at Harvey Mudd College earlier this month. There were many good talks about design research and some discussion of the  skills that engineering students in the future need to learn.

I was given the opportunity to present a poster based on an abstract that I submitted.  You can find the extended abstract here.  The short abstract follows: 

Engineers in the near future should have strong analytical skills, practical ingenuity, and creativity. These engineers should also be dynamic, agile, resilient, and flexible; that is, they should be able to adapt. Consistent with this view, this paper presents a specific vision of what these engineers should be able to do in the years ahead to adapt to the ever-changing needs of society and a vision of how design education should adapt as well.

Article in Wiley StatsRef

My article Rational Decision Making was published in Wiley StatsRef:Statistics Reference Online.  The article discusses the following topics:

• decision making;
• decision theory;
• decision analysis;
• game theory;
• multicriteria decision making;
• risk;
• uncertainty; and
• rationality.

Here is the abstract:

Rational decision making requires executing an appropriate decision-making process to select the best alternative. This can be challenging when information is uncertain or when time is limited. This article describes three important perspectives on decision making: (i) the problem-solving perspective, (ii) the decision-making process perspective, and (iii) the decision-making system perspective. This article describes important concepts from these three perspectives, including rationality, multicriteria decision making, group decision making, decision making under uncertainty, game theory, contexts for decision making, decision-making processes, and techniques for modeling and improving decision-making systems. Understanding and applying these concepts can improve decision making. The article will first consider the challenge of selecting the best alternative, which is the problem-solving perspective. Then, the article will discuss the decision-making process perspective: how people make decisions. Finally, the article will describe decisions from the decision-making system perspective by considering the decision-making behaviors and information flow within organizations and how to improve those decision-making systems. This article also provides numerous references to sources that provide additional details about rational decision making.