Congratulations to Technomics’ Pat McCarthy for winning the 2020 International Cost Estimating and Analysis Association (ICEAA) Best Paper Award in the Analytical Methods & Strategies category!
The paper sought to answer four critical questions relevant to the utilization of learning curve theory in cost estimates:
- What are the different types of learning?
- Does learning ever stop and if so, how does it impact a cost estimate?
- How can the production steady state be defined?
- How can the production steady state be identified and utilized in cost estimates?
Types of Learning
After providing an overview of learning curve theory, including the differences between the Cumulative Average (Wright) and Unit (Crawford) theories, two distinctly different types of learning were introduced – Individual Learning and Organizational Learning:
A critical component of Individual Learning was the identification of the requirement that the product being produced and the environment in which it is produced both remain constant. Individual Learning is most closely associated with what is traditionally being estimated using the Cumulative average and Unit Curve theories. Alternatively, Organizational Learning is the result of organizations, program offices and systems modifying aspects of the product design or the production environment.
Dispelling The Myth That Individual Learning Continues And Impacts The Estimates
When learning stops, the steady state of the production run has been reached. If a product and environment remain constant, achieving this state is inevitable due to cognitive and physical limitations of the workforce performing the work and technological limitations of the tools and equipment being used. Organizational Learning can mislead analysts into thinking that Individual Learning is still occurring, when in fact, it is the modifications to the product or environment driving the variability in requirements from unit-to-unit. Accounting for and planning for the production steady state can have an immense impact on the cost estimate:
Defining The Production Steady State
Despite the simplicity of conceptualizing the steady state, the high amount of variability in resource requirements from unit-to-unit in low-volume production environments makes defining the production steady state relatively challenging when compared to the traditional definition of a steady state in discrete time:
In the scenario above, the resource requirements for each product P within the steady state are the same. Due to the unique nature of low volume production environments, an alternative definition in the form of discrete state probabilities was proposed and utilized:
- Pn+1,h = Pn+1,l = 0.5, for:
- Pn+1,h = Probability of Unit n+1 requiring the same amount or more direct labor hours than unit n
- Pn+1,l = Probability of Unit n+1 requiring the same amount or less direct labor hours than unit n
Identifying The Production Steady State
Through an example, the paper then developed a three-step process for identifying when a production environment has reached a steady state process and how that information can be utilized in future estimates:
- Plot the data to look for “flattening” of production requirements over time
- Bin the data and analyze variation in descriptive statistics such as mean, variance and standard deviation across the bins
- Perform statistical testing (e.g. Dickey-Fuller Test) to assess whether or not the difference in resource requirements from unit to unit has become a stationary process
Using the point at which the system steady state commences, as well as other tangible information such as learning curve slope and budgeted work standards, future resource requirements can be estimated with a much higher degree of confidence.
Beware Of False Alarms
The paper concluded with a warning to be on the lookout for false alarms. Production facilities can often be highly dynamic, rapidly evolving environments, with Organizational Learning occurring frequently. In order to get an accurate assessment of what is occurring in that environment, as well as its impact on the data, communication is highly encouraged with other subject matter experts to get an accurate assessment of what the data represents.
Pat McCarthy is an experienced Industrial Engineer, Cost Analyst, and Program Manager with over 18 years of experience in Federal and private industry.