Interpreting a Data-Driven World

I have been renting out a very small property in the UK for many years now, and it is always a challenge when it is time to find a new tenant. This summer seems to have been particularly difficult, so while in the UK recently, I phoned the realtor to see what was happening.

I could hear her pulling up statistics on her computer. “It’s actually doing really well,” she said. “The number of website hits we’ve had over the last three months is very good. In fact, you are doing the best in this office and are in the top 10 for the region. You have over 10,000 hits.”

I asked why, in that case, we had no takers for the letting. She admitted that was puzzling. I asked how many live visits occurred over that period. There was a long pause, and then I heard her counting aloud.

“Let’s see. One that week. Two that weekend, there is another one …” This continued for a few minutes before the conclusion that 14 people had viewed. Sadly, she had no computer-generated statistics for visits and no insight into what people really thought of the property.

What Does the Data Mean?

There are many examples in the world of analysis and simulation where large amounts of data are output. The difficulty can be in interpreting what all this data really means.

Perhaps we can draw some lessons from my realtor experience. Our problem was in focusing on the wrong data. The website analytics showed the level of general interest in that type of property and concluded that it was high. The trouble was, high relative to what criterion? Our objective was not to maximize the number of hits on a website but to get the property let. The key metric is simply how many live bodies viewed the property. Of even greater importance would be their opinions on the property. What aspects were impeding them from interest in renting? There was no formal data capture for visits. Information on likes and dislikes was purely anecdotal and difficult to recall. It would be very useful for the accompanying agent to fill in a quick questionnaire from the viewers on the spot.

An Industrial Example

One of the most important tasks in any finite element analysis (FEA) simulation is to decide on the objectives of the analysis and how to interpret the results in a physically meaningful way. As I have mentioned in other articles, there is a danger of being a little hazy. I recall an analysis on a nuclear transportation container. The fundamental objective was to ensure breaching of the container could not occur in any accident scenario. One potential failure mode was rupture of the bolts holding the container lid.

Analyses were run across many impact orientations to understand the load levels that the bolts would see. However, it was very difficult to identify exactly what would constitute a rupture of one or more of the bolts, from viewing the analysis results. The first instinct was to look at the stress waves traveling up the bolts and to compare this with the ultimate stress of the bolt material. However, it became clear that these stresses far exceeded the ultimate value. It seemed to indicate that the bolts should not be surviving. In fact, parallel test evidence showed the container to be much more robust than we were predicting, and bolts were not failing at these load levels. It took some time to realize that we should be looking at the energy developed in a bolt under loading and compare that with the energy required to fracture that bolt. Energy accumulation in each bolt was a lot more meaningful than looking at stress waves.

It was lucky for us that we had the test results to bring a level of sanity to our thinking. Without tests, we would have overdesigned containers quite significantly. Please bear in mind this was many years ago before FEA methodologies in this area had really been established.

So, in summary we must make sure we look at the right data for the right reasons. This is largely associated with defining the objectives of the analysis and relating the simulation to the real world.

As a happy ending to my letting experience, it does look like the real-world data combined with some basic practical decisions have produced a positive result.