Data Analysis - When enough data is enough?

An old boss of mine once referred to the pace of a project as "glacial speed". Sometimes I feel that's how were moving with this data collection process.

When is enough data enough? Well this is something I've always struggled with on improvement projects. The perfectionist in me says get every bit of data you can and keep getting it until you feel you have the answers to all the questions. One project I worked on, we analyzed a full year of machine throughput numbers and the distribution of each piece across a couple of thousand possible outputs. Millions of points of data (thank God for excel and import files). The lazy side of me says "it's good enough, move on". I have to recheck my patience when that happens.

The answer to the question of how much data is enough is really a method or approach. First, you need a data collection plan before you start. You have to answer a few simple questions starting with how your data will be displayed when you get it all. This basic question really helps you focus before completing the rest of the plan. In our case we decided to display the data we discussed previously in dot plots showing variation from the mean. That way we could explain away data outliers and come to a reasonable conclusion of what the rate of production capability really is.

Other questions to ask is what data is available and in what format. Old data may not be in the same format as is newer data. This was the case with our data collection, having several different collection formats that we needed to adjust for. You will aslo need to know where to get the data and if you plan to have others collect data you must be very clear in your requirements and maybe even run some test collection to ensure everyone is clear. You also need to address conditions that introduce variability in data such as proficiency of operators or process performance at the time. For example, in our case, some of the data we're using was collected before Kyran and others created lisp routines cutting out many key strokes.

The answer to the question then is.....develop a complete data collection plan and execute it. You will be sure to get the data you need in the right amounts. Don't skip steps and be patient in getting the right data you set out to get. It is the foundation for creating the future state map. Try creating the future state with bad data and see what you'll get.

Search some six sigma and lean sites to find more information on data collection planning and how to lay out a plan.

Lean Guy at Genoa