Q: I’m looking for some advice on how to create quality awareness in my organization. All other department heads, including the general manager, do not seem to understand quality. For example, the production manager and the general manager will blame only the quality inspectors for not being able to detect a defect during in-process inspection.
I cannot deny that the quality inspector has responsibility, but the production operators also share responsibility to make sure that only quality material is sent to quality inspection. When any problems happen, the only thing on their minds is to blame quality control. We are supposed to share responsibility and perform as a team.
A: Having worked at a company with a similar quality culture, I can empathize with the frustration you are feeling. In your situation, many people would choose to leave such a company, so kudos to you for wanting to stay and make a positive change.
Your company’s management doesn’t seem to understand that inspection is not the way to manage product quality. Also, it seems as though management does not want to spend the time or effort to implement an effective quality assurance program.
ASQ has many resources on its website to address the issue, including an area titled "Making the Case for Quality."1 There, you can find an online community of members who may be able to provide you with some assistance and support, as well as case studies and a white paper on the topic. Other resources on the website can be found by searching for "quality and the bottom line."
Your management may not be aware of how poor quality affects the bottom line. Work with your accounting department to determine how much value is added to the product at each stage in the process from beginning to end.
Examine and calculate the rework and scrap costs. Look through your inspection records and determine the point in the process at which each defect was introduced (root cause analysis). From there, you can calculate how much money is being spent using the inspection department to find defects rather than eliminating them at the source. Hopefully, this number will be significant enough to get management’s attention.
Take a look at the types of customer complaints you are receiving. Look for recurring complaints and the types of defects associated with them. Again, work with accounting to determine the costs for replacing and repairing products returned from customers.
Talk with your sales department. Has the company lost any business due to product quality issues? Has the company made concessions or offered lower pricing to smooth over quality problems with customers?
Who are your major customers? Do they have a quality culture? Would someone at one of those companies be willing to talk with your management about their experiences in managing quality assurance? Would your management be willing to listen? If not, then you have a very serious issue.
The road ahead is not an easy one. But, if you are willing to do the work and make a convincing case to management, it will likely have a positive effect on the company and on your career.
Navis Pack & Ship MD-1106
Annapolis Junction, MD
- ASQ, "Making the Case for Quality," www.asq.org/economic-case.
For more information
- Sherman, Peter J. and James G. Vono, "All Ears," Quality Progress, July 2009, pp. 16-23.)
Douglas C., "Blurred Vision,"
Quality Progress, July 2008, pp. 28-33.
Q: Which is the best test to use when comparing two sets of data (for example, two sample preparations for the same lot, or supplier results vs. company lab results): Mann-Whitney, Mood’s median, paired t-test or two-sample t-test?
Arecibo, Puerto Rico
A: Of the four tests you mention, the paired t-test is the most powerful. It can detect small differences between samples even if the sample sizes are small. There is a catch, though: The data need to be in logical or dependent pairs.
For example, suppose you want to test a diet pill. You recruit a handful of study participants, distribute the pills and weigh the participants at the beginning and the end of the study. In this case, the pair of data is each individual’s starting and ending weight.
To analyze the results, you would subtract the starting weight from the ending weight for each participant. Some people call this a paired difference t-test, because you must calculate the difference between the pairs. If the distribution of differences is significantly different than zero, then the test result is significant. If the result is favorable, then you may want to perform another test and include a placebo in a double-blind study to make sure the weight loss is due to the drug rather than the motivation of the participants.
In Statistics for Experimenters, the authors describe a test comparing two different materials for the soles of boys’ shoes.1 Some boys are more active than others, and the study designers did not want the study results (wear rate of the shoe) to be dominated by the activity level of the boys. The solution was very simple and quite clever.
The study designer created pairs of shoes with material A on one foot and material B on the other foot. The pairs were randomly assigned to the boys so some of them had material A on the left foot, and some boys had material A on the right foot. At the end of the study, they compared the wear rates using a paired t-test.
There is a lot of variability between boys. There is a lot less variability within each boy, because pairs of feet move together. This example illustrates why the paired t-test is so powerful. It focuses on the variability within pairs and effectively eliminates the variability between pairs or samples.
Let’s get back to your question. Suppose you want to compare your lab data with your supplier’s data. If you want to use the paired t-test, you could label and measure several samples, and then send them to the supplier. The supplier would measure the same samples, keeping track of which measurement goes with which sample. Analyze the data with a paired t-test for a powerful, efficient comparison of the measurement systems.
If the test is destructive, consider whether you can make a large homogenous sample and randomly split the samples (half for you and half for the supplier). This will usually work for things such as chemical assays and bulk materials if the sample is thoroughly mixed before it is split. There is one restriction on the paired t-test: The differences must be normally distributed.
If you can’t get homogeneous samples or repeat measurements on a single sample, then you should consider a two-sample t-test. This test is used to compare the population means of the two samples, but it has restrictions: Both sampled populations must be approximately normally distributed with equal variances, and the samples must be collected independently of each other. It is not as sensitive as the paired t-test, but you may have existing data that you could analyze immediately without the delay or expense of conducting a round-robin test.
If your data are not normally distributed, then you could try the Mann-Whitney two-sample rank test. The Mann-Whitney test compares medians and does not require a normal distribution, but it does require the two populations to have a similar shape and approximately equal variances.
The Kruskal-Wallis test is more general than the Mann-Whitney test because it does not require equal variances. It’s also similar to the Mood’s median test, in that both require the two populations to have the same shape. Mood’s median test is generally less powerful than the other tests I mentioned, but one advantage is that it is more robust against outliers.
Consultant, Master Black Belt
- George E.P. Box, William G. Hunter and J. Stuart Hunter, Statistics for Experimenters, John Wiley and Sons, 1978.