Statistical Measures of Complaints


M Greenaway

Can any of you stats guru's out there answer my question.

In just about every company I have worked for there has been a monthly report generated of customer complaints. Often this is simply the number of complaints received ina month. This is often then charted in a simple bar chart showing month by month counts, and sometimes someone will stick a best fit trend line onto the chart.

Is presenting this data in this fashion actually telling us anything at all ?

Or should we treat this like any run chart, and be able to calculate mean and standard deviation of the monthly counts received ?

Would there be any greater value in doing this, or is the data itself so broad and unspecific that we couldnt do anything about it anyway ?

Please advise.

Atul Khandekar

Would depend on what you want to do with the data. I believe a simple Pareto chould be a good starting point.

M Greenaway

The idea is that the graph trend line shows if complaints are going up, or down over time - which personally I think is impossible to determine form such a graph.

i.e. if our process is performing in control then we should get a mean number of complaints month by month - say 10. Obviously our process is subject to causes of variation like any other, as such some months we might get 30 complaints, and some months we might get 2 complaints. Now I think to present this data in the format described in my original post is unsound. It is a bit like drawing a trend line on an SPC chart used in production - you just wouldnt do it would you ?


I am not a Stats Guru but I'll take a crack at it.

To answer your last quesion first, one of the purposes of SPC tracking is to spot trends, so I would have to say that you might insert a trend line on a SPC chart.

As far as how to track and evaluate cutomer complaints I think that the simple graph you mention is not a lot of good without some additional evaluation. In order to determine trends you might average the reject totals quarterly and semi-annually and look for trends there.

On the whole this method of graphing is a place to start, and has some benefit in raising peoples awareness, but isn't going to supply you with the kind of detail needed to select and resolve specific issues.



Maybe I'm not reading this properly, but IMO keeping track of the number of customer complaints serves no useful purpose.
However, keeping track of the detailed nature of each complaint and stratifying the results would allow you to target the appropriate areas in order to bring a process into control or to implement improvement

Aaron Lupo

I agree with Sam. What good does it do to just track the number of complaints you receive. What would be more useful IMHO is something like, looking at the # of complaints and comparing it to the output, what type of complaints are you getting (are they the same every month), are they complaints valid or should it just be considered feedback. Breaking the complaints down and analyzing them is much more useful than saying we had 3 complaints this month, 5 last month etc.. this tells you nothing. Gather the information, analyze it and take action on what you find.


I agree with Sam except that with QS-9000 you have to report the number of customer complaints at each surveillance audit so I would keep track of the number.

I track the number of complaints as a % of shipments as well as by department/plant. I pareto the data by plant - product - type and track the response time.


M Greenaway


Yes we do break down the complaints to Pareto the causes.

Lets add another slant. I am sure those of you that work in companies that use the type of complaints monitoring I describe also perhaps have a target number of complaints, possibly set by upper/corporate management. Should we use this value given as our upper specification limit, and zero as our lower specification limit, and then plot our monthly count and determine our mean (which should be half way between zero and our complaints target) and our process capability (which may be +/- 6 standard deviations within the specification limits) ?

I just think that this is much more meaningful as a simple trend line on such a graph will always be either up or down, and will be randomly up or down depending on where the first count of complaints on our graph came within our normal distribution. i.e. if we plot a rolling twelve months, and on the first month we had a count of complaints that was towards the upper specification limit, then for the rest of the year (assuming the process is in control) the trend line will be downward. If however our count on the first month is towards the bottom specification limit (zero) are trend line will always be upwards.

As such the trend line on such a graph actually tells us nothing about whether the process performance is getting better or not. What would tell us that is if the mean value moved towards the lower specification limit, and/or the value of standard deviation (variance) reduced.

M Greenaway


Yours and others posts are well founded methods in determining where the problems lie, and therefore in deploying useful corrective action.

My question really relates to monitoring if we are getting any better - neither Pareto or measles diagrams will necessarily tell you that as they do not show 'capability'

Top Bottom