Q: My company is looking for ways to measure customer satisfaction. We would like to use a system that includes a bar chart in some way. How should we approach this?
A: Customer satisfaction can be measured in many ways, most of which can incorporate a bar chart, or Pareto chart. I’ll outline the benefits here, but you also can find resources to help you on the ASQ website1 or in this month’s cover article, "Building From the Basics."
Designing a measurement and data collection system for customer satisfaction can be difficult. Your customers will likely have many touch points with your company where they may experience satisfaction or dissatisfaction (sometimes called moments of truth).
Product use, delivery of service, calls to a customer service center, billing or using your company’s website are just a few of the ways a customer can interact with your company. Multiple products or multiple service offerings also complicate the design of your data collection efforts. Because customers’ opinions about your products and services can change over time, satisfaction measures should be conducted on a regular or even continuing basis.
Direct measures of customer satisfaction include interviews, focus groups and surveys. Indirect measures of customer satisfaction and dissatisfaction can include the number of referrals and recommendations made by customers, market share, sales, repeat sales and customer retention. The number of complaints and compliments received, warrantee claims and product returns will also reflect customer satisfaction. Data from these measures can be segmented, summarized and plotted on a bar chart.
Some companies can make use of customer satisfaction data provided by external organizations, such as the American Customer Satisfaction Index (www.theacsi.org) or JD Power (www.jdpower.com). This data can easily be trended and summarized in a bar chart. Companies that take measurement seriously use multiple methods for tracking customer satisfaction trends. Some companies have even created their own customer satisfaction indexes using multiple measures.
Before you get started, make sure your executive management team is committed to reviewing the data on a regular basis and implementing any needed changes based on the results of your measures. It would be a waste of your customers’ time and your company’s resources to collect this information and just summarize it in a report or PowerPoint presentation.
Sr. manager, performance management
- "Pareto Chart," www.asq.org/learn-about-quality/cause-analysis-tools/overview/pareto.html.
For more information
- Hayes, Bob E., "The True Test of Loyalty," Quality Progress, Vol. 41, No. 6, pp. 20–26.
- Westcott, Russ, "Your Customers Are Talking, But Are You Listening?" Quality Progress, Vol. 39, No. 2, pp. 22–27.
Q: I would like to know the difference between a 95% confidence interval and 2-sigma value for any parameter. I often come across these terms in situations where both are expressed as percentages. But, at the same time, they have their own meanings.
Indian Space Research Organization
A: Most of us are familiar with the bell-shaped normal distribution curve. If we know the mean and the standard deviation of the process, then we can calculate various probabilities associated with the distribution.
For example, we can compare the process distribution to the specifications and estimate the percentage of the output that is defective. By estimating the annual cost of the defects generated by this process, we can go one step further and compare it to a more expensive process that produces fewer defects.
For example, suppose a wholesaler has a contract with a beverage maker. The wholesaler wants a 12-ounce energy drink. The contract requires that 99% of the cans contain at least 11.9 ounces. The contract also requires the average fill to be greater than or equal to 12 ounces, with 95% confidence.
The beverage maker does not have a statistician, so it is not sure if it is in compliance. The company asks for help, and we tell them to collect a random sample of 20 cans and measure the fill level. The company gets the following results, which are illustrated in Figure 1:
X=12.117; σx= 0.0607
First, we check to see if 99% of the population is greater than 11.9 ounces. We use a normal probability table to look up the value of the "Z score" associated with having 1% of the distribution in the tail, and apply the Z score to this equation:
X-Zα(σx) = 12.117 – 2.33(0.0607) = 11.97 oz.
The calculated statistic is greater than 11.9 ounces, so the first requirement has been met. Then, we look at the second requirement. We know the mean of the sample (because we calculated it), and it is greater than the 12-ounce requirement. But are we confident in the result? We don’t know the mean of the population. Perhaps we got lucky with the sample. How much sampling error is present?
We can create a confidence interval to account for the sampling error by using the equation:
Where X is the sample average, Zα/2 is the Z score from a normal probability table, with α/2 probability, σx is the sample standard deviation and n is the sample size
We will use a = 0.10, because α/2 will give us 5% error in each tail, and we are really only concerned about the lower tail. Plugging in the values, we get:
12.117+ − 1.645(0.0607) = (12.095,
Because the lower boundary of the confidence interval is greater than 12, we can be at least 95% confident the average fill level meets the second contract requirement.
What does the confidence interval really mean? The lower and upper bounds of the confidence interval give us a range that is likely to contain the true population mean. We will never know the true population mean unless we measure the entire population.
But we can take a relatively small random sample, calculate the confidence interval and infer from the associated level of confidence that the population mean is somewhere between the lower and upper bounds of the interval. We could be wrong, and we will never know the truth unless we measure the entire population.
The equation for the confidence interval allows us to change the confidence level by inserting a different value for a. The confidence is simply (1 – α) x 100, so if α = 0.05, then the confidence level is (1 – 0.05) x 100 = 95%. Use a/2 for two-sided confidence intervals and a for one-sided confidence intervals.
One more important point: The width of the confidence interval is a function of two things—the alpha (which is dictated by the desired confidence level) and the sample size. Therefore, the higher the confidence level, the wider the confidence interval. Since n is in the denominator, the bigger the sample size, the tighter the confidence interval.
Principal consultant, Master Black Belt
For more Information
- Breyfogle III, Forrest, Implementing Six Sigma, second edition, John Wiley & Sons, 2003.
- Meeker, William Q., Gerald J. Hahn and Necip Doganaksoy, "Planning Life Tests for Reliability Demonstration," Quality Progress, Vol. 37, No. 8, pp. 80–82.
Q: My organization is planning to be certified to ISO 9001:2000 in February 2009. Are we going to need to implement ISO 9001:2008 instead? If we don’t need to, should we anyway?
Al Khobar, Saudi Arabia
A: Two basic points: First, the changes to ISO 9001 with the 2008 standard will be relatively minor and are not intended to add new requirements. Second, if a system meets the 2000 version, it should have no problem meeting 2008.
Certainly my advice is to get the 2008 version as soon as possible and confirm the system meets it. If the question relates to a certification audit in February, I would think it should be to the 2008 version. The best thing to do is contact the registrar and have a conversation about this.
The Woodlands, TX
For more information
- Hunt, Lorri, "Energize Your QMS," Quality Progress, Vol. 41, No. 10, pp. 20–25.
- West, Jack, "What’s Really Important," Quality Progress, Vol. 41, No. 10, pp. 67–69.