One of the key components in analyzing data for a Lean Six Sigma project is performing a hypothesis test. When doing this, it’s important to carefully choose both the appropriate test and interpret the results accurately.
One of the first steps in choosing the right hypothesis test, and for the Lean Six Sigma Methodology in general, is ensuring whether or not the data is normal. Those new to hypothesis testing tend to assume data normality, as that simply makes data analysis easier. However, there’s danger in these assumptions as certain projects may involve atypical data (like process cycle-time reduction). Normal data has normal variation and generally takes the shape of a bell curve. This curve represents the data’s central tendencies. The properties of this type are such that those working with it may use a probability plot for actual verification that it is distributed in a normal way. This tool is based on statistical testing.
Data produced by this probability plot, if the results are considered normal, will follow a straight line. This indicates that as the value increases, so does the percentage of total data that falls within that particular range. If anywhere between 80-90 percent falls between these lines, it is considered normal. The probability plot serves as the first step in deciding which hypothesis testing is indicated check site.
Once this information regarding the normality of the data is determined, it’s necessary to determine the type of test to use. The Hypothesis Flow Chart is used to give direction regarding the type of test. This chart encourages users to input the normal data from each “line” or “process” examined. If the comparison is being done between two or more groups, the variances of each line are examined. Tests that help determine the variance are the test for equal variance or an F-test. Lines with equal variances should then use the ANOVA for hypothesis testing.
If data isn’t normal, users can try converting it to discrete. This results in a contingency table, which can easily be used in a chi-square test. This sort of test helps to determine which line is not performing like the others. Other options, if converting to discrete include non-parametric hypothesis tests like the Kruskal-Wallis test or the Mood’s median test.
Once this testing is done, the results must be interpreted with a focus on the p-value. This assists in determining whether or not the null hypothesis should be accepted. Once the statistical analysis is out of the way, other Lean Six Sigma tools can then be sued to increase an organization’s profitability.