Can any one of you colleagues help me find a Chart that has approximate comaprison of all the units used to measure SURFACE ROUGHNESS. i.e. comaprison chart of Ra/Rz/N*/um/u"/delta etc.??
I know this has to be an approximate one.
Yes, I am sure. Here's my latest example. I have a 30 question test. I have 2600 people who have taken the test and have answered all 30 questions. When I plot the number of correct answers per student, I get an in-control process. But, when I plot their incorrect answer count (per student) I have 118 results sets beyond the limits.
Sure. I'm trying to determine if our training processes are in a state of control, or if there are some trainers who aren't quite as adept as others. So, I'm plotting the students correct answers (or incorrect answers) on a c-chart. The sample size is the number of questions, so it is a constant sample size. So, my data points may be a sample size of 60, with incorrect answers anywhere from 3 to 15. If I plot by incorrect answers I'll recieve an out of control chart. But if I plot corrects, I receive an in control chart. Does that make sense? Am I just going about this all wrong?
I appreciate the additional information you provided, but now I'm more confused as to what you're charting. If you have 30 questions and 2600 individuals who've completed the test, how can you have a sample size of 60? In an earlier post you mentioned there were 118 out of control points. How many subgroups do you have?
You also wondered if there are some trainers who aren't quite as adept as others. If different trainers are training different groups, how do you know it's the trainer and not the trainees who aren't as adept as others? I assume that the test is always the same. If you're charting the test as a subgroup, are certain questions much more difficult than others, leading to differences from one group to another?
I believe your original post asked why you have more out of control points when you chart incorrect answers, than correct ones. Take a look at the spread of the control limits between the 2 charts. I would guess that the chart for correct answers has a spread between the control limits about twice that of the chart for incorrect answers. A point beyond the limits on the chart for incorrect answers can easily be in control on the chart for correct answers.
I may be wrong, but from what I've read and the questions your information have raised, I get the strong feeling you may be using an incorrect tools to serve your intent.
"Take a look at the spread of the control limits between the 2 charts. I would guess that the chart for correct answers has a spread between the control limits about twice that of the chart for incorrect answers. A point beyond the limits on the chart for incorrect answers can easily be in control on the chart for correct answers"
But the control limits are calculated based on the data so the spread does not really matter. the fact that some of the points in first control chart are in and they are out of limit in second chart may be because of two reasons
1. While calculating the first control chart, you have removed some data points for limit stabilization and while drawing the second chart you have not removed the above points from it.
The spread can make a difference. The control limits on a c-chart are cbar +/- 3(sqrt of cbar).
For example:
1) Incorrect answers
n = 60 cbar = 10
cbar +/- 3(sqrt of cbar) = 10 +/- 3(sqrt 10)
= 10 +/- 9.5
UCL then is 19.5
2) Correct answers
n = 60 cbar = 50 (since 10 were incorrect)
cbar +/- 3(sqrt of cbar) = 50 +/- 3(sqrt 50)
= 50 +/- 21.2
LCL then is 28.8
If the next subgroup has 20 incorrect, it is beyond the UCL of 19.5 and out of control. Conversely, the same subgroup has 40 correct, which is above the LCL and in control.
I don't have any of the original data, but you can see from this hypothetical example the chart for incorrect answers can be out of control while the chart of correct answers is in control.
Hope this clears my earlier statement. With the potential of other noise factors contributing to the differences between subgroups, the charts in question will not be able to distinguish between the abilities of different trainers.
Side note: though the sample size is the same, the number of incorrect answers (no-go's) is much smaller than the number of correct answers (go's). Could this be why the results are so varied? If so, which analysis is correct?
mbpradeep 5/27/2003
are you sure that (no of correct answers + no of in correct answers)= constant
vgriffith 5/30/2003
Yes, I am sure. Here's my latest example. I have a 30 question test. I have 2600 people who have taken the test and have answered all 30 questions. When I plot the number of correct answers per student, I get an in-control process. But, when I plot their incorrect answer count (per student) I have 118 results sets beyond the limits.
qdigest 5/30/2003
To begin, what are you trying to accomplish by plotting this?
I'm still not clear exactly what you're charting. Can you provide the information for a couple of subgroups, so I can understand the data?
Thanks
vgriffith 6/4/2003
Sure. I'm trying to determine if our training processes are in a state of control, or if there are some trainers who aren't quite as adept as others. So, I'm plotting the students correct answers (or incorrect answers) on a c-chart. The sample size is the number of questions, so it is a constant sample size. So, my data points may be a sample size of 60, with incorrect answers anywhere from 3 to 15. If I plot by incorrect answers I'll recieve an out of control chart. But if I plot corrects, I receive an in control chart. Does that make sense? Am I just going about this all wrong?
mbpradeep 6/11/2003
Can you please tell me the following
Upper control limit:
Lower control limit:
Center line:
Sample size which u tell is constant:
These details for both the charts
qdigest 6/10/2003
I appreciate the additional information you provided, but now I'm more confused as to what you're charting. If you have 30 questions and 2600 individuals who've completed the test, how can you have a sample size of 60? In an earlier post you mentioned there were 118 out of control points. How many subgroups do you have?
You also wondered if there are some trainers who aren't quite as adept as others. If different trainers are training different groups, how do you know it's the trainer and not the trainees who aren't as adept as others? I assume that the test is always the same. If you're charting the test as a subgroup, are certain questions much more difficult than others, leading to differences from one group to another?
I believe your original post asked why you have more out of control points when you chart incorrect answers, than correct ones. Take a look at the spread of the control limits between the 2 charts. I would guess that the chart for correct answers has a spread between the control limits about twice that of the chart for incorrect answers. A point beyond the limits on the chart for incorrect answers can easily be in control on the chart for correct answers.
I may be wrong, but from what I've read and the questions your information have raised, I get the strong feeling you may be using an incorrect tools to serve your intent.
mbpradeep 6/11/2003
You have said
"Take a look at the spread of the control limits between the 2 charts. I would guess that the chart for correct answers has a spread between the control limits about twice that of the chart for incorrect answers. A point beyond the limits on the chart for incorrect answers can easily be in control on the chart for correct answers"
But the control limits are calculated based on the data so the spread does not really matter. the fact that some of the points in first control chart are in and they are out of limit in second chart may be because of two reasons
1. While calculating the first control chart, you have removed some data points for limit stabilization and while drawing the second chart you have not removed the above points from it.
2. Error in calculation
3. Precision issue (approximation errors)
Let me know your views on this
Thanks
qdigest 6/12/2003
The spread can make a difference. The control limits on a c-chart are cbar +/- 3(sqrt of cbar).
For example:
1) Incorrect answers
n = 60 cbar = 10
cbar +/- 3(sqrt of cbar) = 10 +/- 3(sqrt 10)
= 10 +/- 9.5
UCL then is 19.5
2) Correct answers
n = 60 cbar = 50 (since 10 were incorrect)
cbar +/- 3(sqrt of cbar) = 50 +/- 3(sqrt 50)
= 50 +/- 21.2
LCL then is 28.8
If the next subgroup has 20 incorrect, it is beyond the UCL of 19.5 and out of control. Conversely, the same subgroup has 40 correct, which is above the LCL and in control.
I don't have any of the original data, but you can see from this hypothetical example the chart for incorrect answers can be out of control while the chart of correct answers is in control.
Hope this clears my earlier statement. With the potential of other noise factors contributing to the differences between subgroups, the charts in question will not be able to distinguish between the abilities of different trainers.
vgriffith 5/21/2003
Side note: though the sample size is the same, the number of incorrect answers (no-go's) is much smaller than the number of correct answers (go's). Could this be why the results are so varied? If so, which analysis is correct?
thank you.