Complaints time series significant change

@Steve Prevette I have only 2 more questions if you don't mind.
How to treat data that is subject to a continuous growth (the subscriber data) of which one is continuously rising? Is the way to go to correct for the growth factor to get a flat line.
How to adjust the control chart when there is a shift in volume of complaints (i.e. product is fixed) that can be explained. Is there a minimum of data point one would need in order to set the new start point for the calculations?

01mercy said:

@Steve Prevette I have only 2 more questions if you don't mind.
How to treat data that is subject to a continuous growth (the subscriber data) of which one is continuously rising? Is the way to go to correct for the growth factor to get a flat line.
How to adjust the control chart when there is a shift in volume of complaints (i.e. product is fixed) that can be explained. Is there a minimum of data point one would need in order to set the new start point for the calculations?

Click to expand...

For continuous growth, you could plot the first derivative of the data and see if that is stable. It would be similar to a Range chart, except I do not take the absolute value of the difference between two sequential points. Or you could shift to linear regression, but be sure to use the prediction intervals (not a familiar calculation for most) to check for random variation vs signal.

If a shift in the data occurs AND is confirmed by a signal on the control chart, then one should start a new baseline (average and control limits) after the observed (and verified) shift. I do have a fair amount of materials here on the Cove dealing with SPC, including that situation. Since the average and control limits are supposed to represent a PREDICTION of future performance, if there has been a shift in the process (and it is confirmed that it really had an effect) then make a new prediction.

I have run into problems in the past with managers who state - but we made a change to the process! And my response sometimes is, unfortunately, it just looks like continued random noise on the old prediction, so the change must not have had an effect.

Oh, and I did find on the job that if the new data made three changes of direction in the pattern (an "M" or a "W"), usually I had enough data for a valid, new baseline. As far as I know, that is a "rule of thumb" that worked for me, and I tested it, but has not been published elsewhere or independently validated. BUT FIRST make sure you have a signal prior to invoking MW.

Steve Prevette said:

For continuous growth........ signal prior to invoking MW.

Click to expand...

Thank you @Steve Prevette clear. Well anyway for us it will be just the other way around. If they see a shift in the data and it is confirmed to continue outside the CL they will check for an explanation. Unfortunately we are not that far yet to link development changes to the PMS data but this will come in the near future when complaint coding (symptom codes, failure codes, versions etc) will be implemented to be tagged to the customer feedback.

Is there a easy way to find the content you've created? I've been away from this forum for a while it was before the landingpage change and the legal fuss

I think the c-chart gives a better result. Since I checked it for one product of which we know there was an increase in a certain issue and we've taken counter measures. This does show up clearly in the c-chart but not in the I-mr chart (in a sense of gowing over the CLs)
In fact all of the I and mr charts I've created no point goes over the CLs

For the IMR chart, if you have specific shifts caused by known special causes, you should omit the moving range associated with those shifts from the MR-bar calculation. This will tighten the control limits.

Miner said:

For the IMR chart, if you have specific shifts caused by known special causes, you should omit the moving range associated with those shifts from the MR-bar calculation. This will tighten the control limits.

Click to expand...

So basically the reason is it doesn't point out is the fact that I include. So I should take a recent past stable period and that will tighten the control limits and shows this change as significant over the CLs also.
Thanks I will adjust it

All control chart limits should be plotted for a single stable period. If you KNOW that changes have occurred these points should not be included with other points in limit calculations.

I have a tutorial on this but am traveling right now so I will try to post it when I return home….

Bev D said:

All ontrol char...ry to post it when I return home….

Click to expand...

Thank you

01mercy said:

I think the c-chart gives a better result. Since I checked it for one product of which we know there was an increase in a certain issue and we've taken counter measures. This does show up clearly in the c-chart but not in the I-mr chart (in a sense of gowing over the CLs)
In fact all of the I and mr charts I've created no point goes over the CLs View attachment 29149

Click to expand...

If your process knowledge says the c-chart makes more sense, then I'd suggest continuing it on for a few months and see how it performs. If you can find good reasons for the areas above and below, the control limits, then hopefully that will be helpful for process improvement. It may also be necessary to shift the baseline time interval on the c-chart to reflect what is the "stable" process without the anomalies. Good luck! Keep in touch as to how it is going, and if you have a specific subset of the data you want help with, let us know.

@Steve Prevette @Bev D I was wondering while working on the data that it might be of more sense to take the data in to weekly frequencies instead of monthly. In a monthly period total is will be harder to discover trends than in a weekly total. It would give more variation but I'm kinda worried that a total of a month covers what the audience is actually interrested to see. Would that make sense and have you also experience in taking shorter time intervals of measure in order to get more insight? On what base you made that decision?