We should not think as variability as something that must be eliminated or eradicated, because it is impossible. What we can do is to reduce it to the minimum by monitoring our processes and making data-driven decisions.
María Zamora Cereza
Tests and Trials
“The enemy of quality is variability”. How many times have we heard this? Still, It is not entirely true.
Variability comes natural from life. It goes inherent in all processes, meaning that it is not under our control to remove. We should not think as variability as something that must be eliminated or eradicated, because it is impossible. What we can do is to reduce it to the minimum by monitoring our processes and making data-driven decisions.
This is what Statistical Process Control (SPC onwards) is about. Designed by Walter A. Shewhart in the early 1920s and first implemented by Motorola, this analysis is broadly used in businesses for which it is critical to control and continuously improve quality in order to maintain their customers satisfaction and profitability records. Nevertheless, there is a lack of knowledge of this tool in the primary sector and yet many companies do not implement it in their operations system.
Statistical Process Control, a technique to boost company’s profitability
SPC responds questions such as: Am I producing conforming products? How is the performance of my process? Do I have much unexplained variability? Is a change statistically significant? Are my processes efficient enough or may I act? In such case, where should I prioritize?
The tool can monitor and control any process in which the product output can be measured and the factors that influence it can be treated as variables. One advantage that this method gives is that the loss of quality is detected early, and problems can be solved in real time. In other words, teams act proactively and reduce the number of inspections and waste.
Figure 1: Example of a Control Chart of the variable “Average parity of the sows farrowed”
The evaluated process is monitored over time to detect any changes (positive or negative) and the method can predict how much data or replicates are needed in the analysis (control) to prove that the change in the process is statistically significant.
The primary technique to perform an SPC analysis is the use of Control Charts and there are several types which are used depending on the variable nature (continuous or categorical). They can detect both small and large deviations and its representation follows the Six Sigma philosophy.
Control charts are two-dimensional charts, where the horizontal axis is the period (hours, days, weeks, months …) and the vertical axis represents the value of the variable of interest. On this axis, there are three lines, the Center Line (CL) and the upper limit (UCL, Upper Control Limit) and lower limit (CLC, Lower Control Limit).
The CL represents some measure of centralization of the data (mean, standard deviation…). The distance between the limits will be ±3ó, where ó is the standard deviation of the sample, assuming that most processes follow a normal Gaussian distribution.
When our process is stable and the variation is due to random, the process is “under control” and the samples are within the natural limits. Any point beyond the limits is a signal of a special cause of variation that we must analyze to stabilize the process and redirect it to our desired way.
What we do by controlling the processes by means of control charts is a continuous hypothesis test to check if the process is under control.
How do I demonstrate to my clients the quality of my products (under their own conditions)?
We will come back in the next article with a technical explanation of SPC. Now we want to focus on explaining why SPC is also used as a selling point. This is what we call “trial dilemma”.
• There is a limited number of replicates.
• Genetics, diets (that often don’t reflect industry) and health status are defined.
• There are control and treatment(s) group(s).
• It is possible to perform a statistical analysis.
• There is a lower stocking density, and it is time costly.
This leaves some questions unanswered: Is there an accurate prediction? Do the conditions reflect the real world?
For the commercial trials…
• There are more replicates (but limited by commercial realities).
• There are different genetics and conditions (diets, buildings characteristics, drinkers, …). Also, the health status fluctuates in-between batches.
• With all this factors, we find ourselves in a situation in which It is easy to make mistakes and a statistically accurate comparison may not be possible.
So here, the principal unanswered questions are: If there are so many factors influencing the outcome, how do we control them? Is there here an accurate prediction?
This uncertainty does not remain with SPC:
• We get accurate prediction by using real-world data.
• To perform this analysis, we use large populations, and the stocking density is part of the prediction.
• There are control and treatment groups.
• The genetic variability, dietary performance and health status are accounted for in prediction.
• It is also time consuming as we need much data to get representative results, but the information obtained is statistically valid.
Validity of SPC in Animal Production
As said below, with SPC we can use data from any animal production system. We just need to know what our productive variables of interest are, organize them in a database, and record the information in the facility or facilities.
¹Average Daily Gain, ²Average Daily Feed Intake, ³Feed Conversion Rate, 4% of mortality, 5Number of piglets weaned from litters weaned, 6Medications (all health interventions)
What we will know is how product (or any other change you have made in the process: a new dosage, the use of a new bascule, more percentage of protein in the raw material …) works or performs, and how it does in open trials. At the end, the results tell customers how your something works in their own circumstances, and you use evidence instead of impressions or predictions. If you are going to ask our clients to invest money, the least they deserve is to be sure about the benefits they stand to gain by using your product(s).
Let’s see it with a short example
Suppose you have the historical data from a farrow to finish farm and from 156 weeks of production. Here, you are covering all seasonal variabilities (because remember that with SPC inherent variability is accounted for and can be separated from the special variation). This historical database can have any variable. Some examples of productive parameters (the data in the table is fictitious and its purpose is to facilitate the understanding of the article):
After this period of “control”, it is important to calculate the sample size (ARL, Average Run Length) that estimates how many weeks (batches) we need to observe a significant variation of the main objective variable (it is advisable to be at least one year of production under the new experimental condition).
Now, suppose we start introducing our new product of which we want to monitor the process performance and analyze if it is changing (and in which direction). This stage is called the “experimental period” as the animals are supplemented with the “treatment” product. It is critical to make sure we get all productive data from the farm.
Once the ARL is reached and all data has been recorded, is when it is checked whether the system has changed or not. Depending on the results, we have two options: decide based on that information and use it for internal improvement or as a selling point, or continue recording more data for longer.
As we see, SPC is perfectly suitable to be applied in the animal production industry to monitor changes and detect any deviation that may decrease production and profitability. It is just a matter of time that companies begin to implement it in their operations systems.