Registered: 1440068891 Posts: 2
Reply with quote #1
First of all, I have just read Stephen's new book Signal and found it incredibly interesting.
I am now trying to apply XmR to my current assignment which is working for a provider of Health Clubs here in the UK. Like most retail operations, they have a sales cycle which has variation in sales for each weekday i.e. sales are weaker at weekends and also variation throughout the year i.e. sales are stronger in January and various other points in the year. My question is whether XmR charts can be used to measure such a process. So to remove the daily cyclicity I could plot weekly sales on my charts, but then because weekly sales naturally vary throughout the year, the average would be meaningless in terms of control. We don't have several years of history to calculate averages throughout the year to take into account this natural annual variation either. I cannot find anything on google so am wondering if anyone here has experience of applying XmR in this common scenario. I think it would make an interesting blog post. __________________ Steve
Registered: 1135986598 Posts: 838
Reply with quote #2
To accommodate seasonality in your data for SPC purposes, you have two choices: 1) you can adjust your sales data to account for seasonality, or 2) you can adjust your the center line and process limits in your XmR charts to account for it. The former approach is usually taken. Don Wheeler talks about these adjustments in his work. Because you and one other reader of Signal has asked about this, I'm planning to cover this topic in my Signal course from now on. I won't be able to add this content to the book immediately, but I will try to write about this in my blog in the next few months. In the meantime, I would like to join you in inviting others who have experience with this to respond with suggestions. __________________ Stephen Few
Registered: 1440068891 Posts: 2
Reply with quote #3
Thanks option 1 makes sense, the budgets have factored in this seasonality so it should be readily accessible, alternatively we can just calculate it from actuals as I'm told that even though the business is expanding the underlying seasonality stays pretty much the same. I'm hoping to take the client on the process control journey so will report back. Currently the budget is king and all performance comparisons start there
Registered: 1348995178 Posts: 190
Reply with quote #4
Two more possible investigations I would consider in your case. 1. Use cycle plots and apply the XmR to the resulting seven graphs. This way you may have seven consistent XmR charts for Mondays, Tuesdays, ... besides the mentioned weekly chart. 2. Use XmR chart over time not on absolute values of sales, but on derived contextual calculated measures. For instance you can calculate the average (mean, median or mode) of the sales values for all Mondays within one year and divide the absolute values for each Monday to that average. Repeat the same logic for the rest of the days. Investigate these relative values over one year instead of absolute sales values. Statistical process control requires much more than a trivial calculation algorithm applied on raw (observed) data. Investigating a measure variation suppose to reveal not only patterns of change and exceptions, but also the degree of stability and predictability of a certain process. It is not always enough to apply correct statistical rules to calculate central lines and control limits to available values. Careful business logic related data filtering and possible derived calculations might describe better a process than default retrieved data.