Quo Vadis Training & Consulting

In my last post I extolled the virtues of the histogram and proclaimed March “Hug a Histogram” month. It seems only fitting then that I dedicate another post or two to the benefits of using histograms – and the costs of not doing so.

I recently came across a great, real world example in Michael Lewis’s wonderful book, “The Big Short,” which tells the story of a few, very sharp individuals who foresaw the pending collapse of the mortgage securities market and made a fortune betting on its inevitability.

As Lewis tells the story, mortgages were being moved off the books of banks, packaged together into bonds, then sold to investors. The price of those bonds depended on the rating agencies’ assessment of its risk, which they calculated by taking the average FICO score of all its underlying mortgages (a FICO score is a numerical assessment of the credit worthiness of the borrower). If the average FICO score was 615 or higher, the bond received the highest possible rating (AAA). It didn’t matter the distribution of the individual FICO scores in the bond, the rating agencies cared only about the average. Yes, the AVERAGE.

Under this approach, two bonds with the following distribution of FICO scores were both rated AAA.

Clearly, these bonds have very different risk profiles – the “hump” of mortgages representing low FICO scores in the second makes it much likelier to default. And, of course, many just like it did! Yet the agencies saw average FICO scores of 615 and rated both AAA. And thus more and more mortgages were packaged and sold – wash, rinse, repeat – with investors blind to the risk all the while. What a mess.

Please, make the histogram part of your analytical tool kit. The costs of not doing so can be very painful!

Selecting the right chart type and designing it well is essential in clearly communicating our analytical findings.

I’m Captain Obvious, yes, I know. But we often get this wrong.

Last week I came across the following chart in a Wall Street Journal article describing the dire state of Nevada mortgage holders.

What is the primary message we should take away? Or per the title of my blog: So What?

I think the author’s point is that Nevada is far worse than other states on two important housing measures: home equity and foreclosure percentage. But if so, this would be better communicated using two separate bar charts.

My guess is the author chose the scatter plot to improve the “insight per chart,” if you will. But in doing so, he/she also added complexity and confusion. Scatter plots are used to illustrate the correlation between two variables, usually with the assumption that a cause-and-effect relationship exists. That could certainly be the case here. I imagine negative equity in one’s home increases the likelihood of foreclosure. But if this were the point, the independent variable (negative equity) should be on the x-axis and the dependent variable (foreclosure rate) on the y-axis, as here.

But now we find a low correlation. That’s because we have the wrong variables. Rather than show the percent of homes with negative equity, we should show the actual amount of negative equity. In other words, the foreclosure rate is unlikely to be very different between a state with a small percentage of homeowners with a very small percentage of negative equity and a state with a large percentage of homeowners with the same very small percentage of negative equity. The death grip of foreclosure wouldn’t have yet kicked in. But if the second state had homeowners with a large amount of negative equity, you would expect a higher foreclosure rate. So, we care not so much about how many homeowners are under water, but how far under water they are.

In the end, this chart contains some interesting information. But it’s easily lost because the author – however well meaning – has chosen the wrong chart, (most likely) the wrong variables, and designed it poorly.

Picking the right chart matters. Committing the simple guidance on page 27 of Gene Zelazny’s excellent book “Say It with Charts” to memory would greatly reduce these errors.

In a blog entry last week, Seth Godin shared the following clip illustrating changes in the earth’s temperatures since 1880, and posed the question: If a picture is worth a thousand words, what’s a short video worth?

While video is unquestionably a very powerful communication tool, I believe the most important lessons from this video are analytical in nature, specifically: 1.) Challenge the starting and ending data points in any time series analysis; and 2.) Select a scale that accurately reflects the variation in the data.

To the first point, why start the time series in the year 1880? Would our conclusion change dramatically if instead we chose 1950? How about 1500? What if we went back as far as 1000, or even further? The good critical thinker will ask these questions. In the interest of keeping this discussion analytical instead of political, I’ll leave it to you to explore the answers to these questions. If you’re interested, this site will help.

To the second point, note how the colors on the map fluctuate dramatically from dark blue to brilliant red. What was your reaction viewing this? Time to buy beachfront property on Greenland, perhaps? How much does this suggest earth’s temperatures have varied during this period? The actual answer – found in a simple chart on the same website as the sample video – is about 0.8 degrees. Yes, less than one degree. Again, my intent is not to make this a political discussion, but simply to say that as viewers of the video, and consumers of the underlying data, it’s our responsibility to understand the scale in use and thus the variation in the data.

So, yes, video is indeed more powerful than pictures, charts, and words. And its role in data communication in business is ever increasing. As Problem Solvers, we need to develop an ever more critical eye in response.

A couple months ago I stood in line at the local Apple store to buy the new iPhone 4s (I used to shake my head at people who did this, yet there I was). The purchasing process took longer than expected, so by the time I made it home I was inclined to take the family out to eat rather than to cook (I own the kitchen on the weekends).

“Hey, let’s use Siri to make our dinner plans!” (In case you’ve been on the moon the past few months and missed all the television ads, Siri is the voice-activated assistant on the iPhone).

“Siri, what’s the number for Maggiano’s?” I asked.

“I’m sorry, I couldn’t find the phone number for Mike Geonese,” Siri responded.

Hmmm.

“What’s the PHONE NUMBER for Maggiano’s RESTAURANT?” I tried again.

“I found 25 restaurants, tap the one you want to call,” Siri replied. Maggiano’s was not among them.

“What’s the telephone number for Maggiano’s restaurant on Freedom Drive in Naperville, Illinois?” I said again, attempting to refine my request.

“I found 25 restaurants, tap the one you want to call,” Siri repeated.

Stupid Siri!

When we seek data as part of our work – through Internet searches, database queries, and requests of our colleagues, clients and suppliers – how precise are we? How well are our needs understood? How well does the data we receive satisfy them?

When we provide information to our business partners – through formal presentations, meetings, email, and every day conversation – how well is it understood? Does our audience tell us when it’s not?

Siri’s not stupid. It let’s us know when we haven’t been clear. It let’s us know its limitations. This is incredibly valuable in the process of communication.

Tables are one of two primary means for sharing quantitative information (charts the other).

The first step in designing an effective table is to determine the comparison we want viewers to make. Do we want them to compare sales of a particular product from month-to-month? Or maybe variations in sales across different regions? Or stores? Or perhaps we want them to assess the accuracy of our sales forecast?

The answer determines both the data we include in the table, as well as how we arrange it in rows and columns. We should include no unnecessary data, and the data should be organized to help the viewer grasp our intended point as quickly and easily as possible. Design should always occur with purpose.

It’s amazing how often this obvious idea is violated.

Consider the following table that appeared in the Wall Street Journal last week, showing the budget deficits of European Union members over the past ten years.

There is a lot of interesting information here. But what comparison does the author intend for us to make? What key conclusion are we supposed to reach? The organization of the rows should help us answer these questions. But it doesn’t. As you can see, the rows represent an alphabetic listing of EU countries. Ask yourself: How often do we want to compare variables because of their alphabetic placement? The answer, I think, is never (an alphabetic arrangement helps us quickly find particular data points in very large tables, but that’s needed here).

This table would be far easier to navigate with a simple reordering of its rows. If the author’s main point was to illustrate the relative health (or lack thereof) of the EU countries, the following table would work better.

If the point was to compare the recent plight of countries based on their historic economic health, the following table would be best.

The key point is not to find the ideal look for this particular chart, but rather to emphasize that all visual design decisions must be made for a single purpose: to help our audience quickly grasp our message. Tables organized alphabetically rarely meet this vital requirement.