Discrete Data vs. Continuous Data

Once we’ve decided to control something (think of our prevention and detection controls), we then need to decide how to measure it. Different controls may need different measuring requirements, which can give us discrete or continuous data.

We treat these data types differently when collecting it, determining sample sizes, and analyzing it for results. Tune-in to learn more about how to take the next step in defining controls: figuring out how to measure it and considering the data.

View the Episode Transcript

When we’re creating design specifications, we’re thinking how it links back to our controls. And we think forward to how someone is going to measure it.

No matter which method that we choose to collect data, we need to verify that the controls are judged or measured consistently, that equipment and tools are capable with the correct level of significance, and – if it’s a qualitative measure – different people will come to the same conclusion.

Another QDD episode you might like:
Designing Specs for QA

Episode Transcript

You’re listening to an installment of the “Quality during Design Versus Series”. In this series, we’re comparing concepts within quality and reliability to better understand them and how they can affect product design engineering. We have eight episodes in this series, which means we’ll be reviewing at least 16 topics. Let’s get started. Hello and welcome to Quality during Design, the place to use quality thinking to create products others love for less. My name is Dianna Deeney. I’m a senior level quality professional and engineer with over 20 years of experience in manufacturing and design. Listen in and then join the conversation. Visit qualityduringdesign.com and subscribe.

Hi, welcome to quality during design for products, others love for less. I am your host, Dianna Deeney. We are in a “Quality during Design Versus Series” where we comparing at least two different quality topics and learning how we can apply those to product design. In the previous episode, we talked about controls, specifically prevention and detection controls the kind of things that we want to implement within our design in our users’ process that will help detect or prevent potential failures and reduce the risk that our product has on its own performance on usability and other safety measures. Well, once we’ve defined a control, we then need to figure out how to measure it. So today we’re talking about two different types of data that we can use for measurement to verify the effectiveness of our controls. Today, we’re specifically talking about discrete data and continuous data.

Now as design engineers, we need to handle and deal with and translate a lot of different data. Our data could be coming from our own formative or summative studies about our users with our product, or it could be coming from a third party about our users. We need to be able to analyze that data and translate that into customer needs or requirements. We also have data from bench-top testing and from our test lab. We could also be looking at data that our suppliers are giving us. If we’re evaluating a component for our design and they give us data, we need to be able to interpret and look at it properly so that we understand if it’s going to be appropriate for our device. And, also, we are setting specifications and limits and tolerances that are important to the functionality and safety and performance of our design. What type of data we might need to properly monitor our design controls needs to be considered because we also need to communicate that to others.

Now, why do we need to categorize our data, the type of data that we’re going to be collecting? I asked you to categorize your controls for your designs last week. And now I you to think about what category your data lies in Well categorizing these things helps us plan ahead, and it makes us pay attention to the details that matter. If not now, when we’re working on our design specs, then definitely later when we’re deciding on next steps or passing it off for somebody else to measure for control. Choosing the right kind of data, be it discrete or continuous – it affects a few things about our data. It affects how, and when we collect it. And it also determines sample sizes: how many do we need to be able to make a certain confidence statement about the results? And it’ll also dictate what type of analysis that are capable of being performed.

So let’s go ahead and define what discrete data is versus continuous data. Discrete data can be thought of as counts of categories, like number of failures in a given time period, number of cycles until the first failure, or number of defects on a part. Discrete data could be proportion, like proportion nonconforming. It can also be binary: yes/no, good/bad, pass/fail. Attribute data is a type of discrete data. Discrete data is also known as qualitative because we’re collecting information about the “quality” of our product. Usually, discrete data is easy and quick to collect. Think of check sheets. Continuous data, on the other hand, is measured using a continuous scale, like time to failure, length, weight, or diameter, or even temperature. Continuous data is also known as variable or quantitative because it’s a measure of a quantity of something. Continuous data usually requires the use of measurement tools or equipment.

When we’re collecting data to measure the effectiveness of the controls that we put in place to control a risk or a failure, we need to take a closer look at our control and its purpose. Are we preventing something or detecting it? And that might determine when it is going to be measuring our control. What’s the criticality associated with that control? Is it really severe or is it high? And that could affect the precision and the confidence that we want in the results. Discrete data is usually associated with low precision measurements: think of visual standards and visual inspections for a quality of a product. Visual quality standards are those things that are pictures or diagrams that explain to all users and all inspectors what’s acceptable and what’s not. Anything that is qualitative is usually a low precision data measurement. Because of that, it also requires a higher sample size so that we can achieve the desired confidence and statistical significance that we want when we’re analyzing the data. On the other hand, continuous data is considered high precision. We are using equipment and tools to measure something about our product. Because of its high precision nature. We usually need less samples in order to claim the confidence and statistical significance that we’d want when we’re analyzing the data. No matter what we’re measuring, if it’s discreet or continuous, we need to ensure that our test methods are validated and that any gauge R and R studies are performed.

Now, those are some of the things we can start to think about as we’re designing measures for our controls. But now let’s look forward a little bit into how we want to analyze the different data types. To visualize discrete data, we can plot it on a bar chart. A bar chart are categories on the x-axis and frequency on the y-axis. A Pareto chart uses a bar chart. Our discrete data can be further analyzed using a probability mass function, or PMF. The output of this is a probability at a specific value. Binomial and poison distributions are common with discrete data. For continuous data, we can visualize it using a histogram. A histogram is different from a bar chart. A bar chart has categories. A histogram has intervals of data. We can further analyze our continuous data with the probability density function to get a probability of an outcome. We can calculate the area under a curve between two points. If you remember calculus and integrals, that’s the method that we use for continuous data. Normal, Weibull, lognormal, and exponential are types of distributions with continuous data. No matter if we’re looking at discrete or continuous data, both can use a cumulative distribution function. It plots a cumulative probability from zero to one along the y-axis for any given value along the x-axis.

Something to watch for is if our discrete data happens to have digits: it would not be appropriate for us to treat that as continuous data. For example, we have a survey from one to five. We would not want to report the analysis as a mean and standard deviation. We could report it as 13% of respondents answered one and 20% respondents answer two and et cetera.

So, what is today’s insight to action? And what can you do with what we’ve been talking about today? When we’re creating design specifications, we’re are thinking how it links back to our controls. And we think forward to how someone is going to measure it. No matter which method that we choose to collect data, we need to verify that the controls are judged or measured consistently that equipment and tools are capable with the correct level of significance. And if it’s a qualitative measure, different people will come to the same conclusion.

If you like the content in this episode, visit qualityduringdesign.com, where you can subscribe to the weekly newsletter to keep in touch. This has been a production of Deeney Enterprises. Thanks for listening!