Survival Bias on Injection Molding Data Analysis

More and more companies emphasize the importance of data collection and analysis, saying that the decisions across their business operations are data-driven. In an easy way, this article introduces and explains Survival Bias on data collection in the data analysis field, how it has been existing silently in the injection molding industry, and how to avoid it.

Figure 1 might be the most popular and visualized presentation to illustrate the Survival Bias, or Survivorship Bias, in data analyses for various applications and purposes, including those in the injection molding industry.

Figure 1: Survival Bias – Data analysis on the distribution of bullet holes

?

It tells about a project during World War II, which aimed to minimize military aircraft losses to enemy fire. The distribution of bullet hole positions and damaged zones on the aircraft that returned from their missions were analyzed so that the decision on where an aircraft should be added armor and reinforced to increase its survival rate could be made.

?

The project ended up with a recommendation from the Statistical Research Group at Columbia University represented by the statistician, Abraham Wald, saying that the armor and reinforcement should be added to the areas that showed the least damage (no data therein) on the surviving aircraft, as shown in Figure 2. Regardless of the data observed on the surviving aircraft and the hard work on the related data analyses, he logically inferred that once one of those no-data areas had been hit, the damage to the aircraft would be so fatal that it could not survive and return being among the sample group and giving data.

Figure 2: Survival Bias – Recommended zones to reinforce

?

This story tells us that the answer or solution might not be obtained by only analyzing the observed (surviving) data. We might bark up the wrong tree.

?

Does Survival Bias exist in the injection molding industry? It does, and it might exist more than you perceived. Thinking about the following scenarios, how often have you experienced in your injection molding career so far?

?

<1> Making trial-and-error process condition adjustments based on the individual’s data experienced in other cases.

?

<2> Making trial-and-error process condition adjustments based on the company’s data experienced in other cases.

?

<3> Making trial-and-error process condition adjustments based on the observed data right from the trial-and-error samples on hand.

?

<4> Doing DOE (Design of Experiments) in which only two or three parameters you feel critical are involved, and, for each parameter, only two setting values you feel comfortable with are assigned to produce the samples and data.

?

<5> Introducing an AI (Artificial Intelligence) or machine learning solution being asked to create and feed your data for improving the solution’s performance.

?

How to avoid Survival Bias on an injection molding company’s data analysis? I believe that educational injection molding knowledge helps. As the domain size of the educational injection molding knowledge increases, the scope size of the sample data increases as well and with confidence, bringing the company a higher probability of finding a superior solution than other companies trapped by Survival Bias error unconsciously. ??

?

***Bridge between Knowledge and Practice of Scientific & Robust Injection Molding @ Effinno Technologies***