Control chart applicable? Percentage of compliance with a standard



I am preparing monthly the radiologist percentage of compliance with a standard on reporting TAT. I have easy access to all data thru our RIS and so there is no sampling. Despite absence of sampling error, can we use control chart to detect trends, for example by using control chart for individuals and moving range.
I know it looks not logical, but accreditation bodies have more confidence if the results are presented as control charts


Stop X-bar/R Madness!!
Trusted Information Resource
Seems OK to me! Random, independent, variable data....Shewhart to the rescue!

Using percentage, though, you have natural limits of 0 and 100%, and your target is likely 100% - not some "typical" percentage below that. Shewhart Charts expect a certain level and seeks signals that you are wandering from it. Not sure if Western Electric rules are adequate - don't think you want a 7 month slide (if it is a monthly report) before reacting to the issue!

Nonetheless, you can track disturbing trends with the chart. Few people looking at the charts will have the knowledge to question it.

Bev D

Heretical Statistician
Super Moderator
TAT = Turn Around Time? Is this a set time on the clock (for example the radiologist must submit their report by 9 AM) or is it teh allowed time to complete the report from the time of submission (for example, the readiologist has 2 hours from the submission to submit their report).

Yes an I, MR chart will work for TAT compliance. The subgroup for what I have described above is typically a day, but weeks and months will also work.

If TAT is what I described above, more insightful chart is to plot the actual time it took the radiologist to complete the analysis for each individual submission. You can still use an I, MR chart.

As a point of clarification, sampling error is only a convenient way of ‘describing’ the variation from subgroup to subgroup from an infinite real process stream. (It derives from actual sampling error when taking small samples from a finite population). In real life we have actual process variation as well as ‘sampling error’ when taking small samples from an infinite stream of work. Control charts still work for these infinite streams because the variation of the process is ‘random’ and homogenous.
Top Bottom