The Difference Between Accuracy And Precision Measurement In Your Machine Shop

almost like darts, when you group them, its precise

You got it.Milo

The other metric that people also frequently confuse is that resolution and degrees of freedom differs from both precision and accuracy. Taking your target example, if the resolution of your position measurement was only the single degree of freedom of the score for which ring one hits on the target, then one cannot discern accuracy from precision. But when one examines the two dimensional image of shooting at these concentric rings then one can determine if one is accurate, precise, both, or neither.

An interesting article - with the common error.

In terms of percentage, you are reporting INaccuracy and IMprecision.

Do you really want a procedure to be less than 30% accurate and manufacturing repeatability to be less than 30%.

I feel not.

Tony

As an ME Intern I came face to face with these two concepts. My task was to perform an audious statistical analysis/comparison of two thermal sensor designs (30 samples each for a total of 60). I subjected the samples to a UL-based oil bath test and used a DAQ to record the activation temperature. These activation tempratures were then visually represented in overlapping continuous distribution curves a.k.a bell curves as a baseline for comparison. The results showed that one design, call it Type A, had more samples within the +/- 5 degF tolerance but were spread sparsely. Type B samples fell outside the tolerance but were concentrated in distribution at 9 degF above activation temp. Conclusion: "Type A is accurate but not precise while Type B is precise but not accurate." Much to the confusion of the good folks from sales! In any case, Type A was revealed to be of better quality.

Thanks for sharing your experience!

Milo

Good question and depends, depends, depends;

Control charts have factors and sample subgroup factors built in; (Note I prefer to use a calculator and calculate an average and standard deviation rather than use the Control chart factors to estimate std deviation, usually eliciting howls of protest from other practitioners but otherwise no noticeable downside) I use my bulk calculation as a check and let the SPC boys finagle polish their charts*;

For capability studies the standard most of us have worked to has been 50 minimum samples in subgroups of 5 (GM manual) /AIAG

For my own personal work I prefer 30 samples and use Student t test if looking for difference between averages;

For DOE I have used the appropriate L Matrix and F test for statistical significance.

I now await the obligatory beatings from The ASQ certified practitioners who will undoubtedly shout "heresy" at my utterances.

Milo

* Why I prefer the statistical calculator rather than the control chart and factors method.

I took statistics and DOE courses in college, and have a calculator as well as an excel spreadsheet. I don't need the ratio of RBar and Sample std deviation to estimate the Standard deviation of all the data; The calculator or spreadsheet can just give me the actual standard deviation of all of my data. Computing power means not having to do estimates, you can crunch all the data. The textbooks will show you how the greater the number of subgroups will give you a more confident estimate of the universe standard deviation, I get it; I prefer to use all the data that I actually have and ignore the possible time series issues that might exist in stratified subgroups. Warts and all. I never lost an order or had to pay a claim based on missed capability because I used the calculated std deviation instead of d2 on the control chart. (Also I caught a lot of statistical liars (operators) throwing out data points they didn't like by calling them "special cause" ) My point was the special cause was in fact LACK OF CONTROL. My calculated Average and std deviation ended up being more accurate and precise than their Control chart factor methods. My 2 cents over 30 years of manufacturing quality practice. Let the beatings begin.

Based on this example, this is how I visualize Accuracy vs. Precision (sorry for the quick and shabby mspaint) :

I disagree as your distributions are all about the same width. I see precise as a very narrow distribution (small std deviation) which may be anywhere related to the target value; I see accurate as a distribution that is symmetric and with average centered on the target. Accurate and precise is a narrow symmetric distribution centered on the target. .

I'm not buying the skewness of what you call "not accurate but precise" as reflective of precision. The long tail argues against precision.
Accuracy is centering of the distribution average; Precision is reduction of the distributions standard deviation.

Milo

Milo