A common question about hypothesis tests that arise during a Six Sigma project is: Why go to the trouble of running statistical tests to see if things are different? Consider the following example problem from the 1 Sample Test:
A corn-growing agricultural company produced an average of 168 bushels per acre every year for five years. To boost the yield of its fields, the company changed its planting process. Following the seed swap, each year’s crop yielded an average of 175 bushels of maize per acre. Is there a statistically significant difference in yield per acre due to the seeding change?”
Is the mean corn yield per acre statistically higher due to the seed change? From the figures supplied in this scenario, we can see that the mean after the seed change is 175. When compared to 168, it is clear that it is higher.
Why go to the trouble of setting up and conducting a hypothesis test?
When we first introduce the normal curve in the Measure Phase, we’ll point out that it’s not always enough to show that a histogram is symmetric and follows the normal curve. Because raw data can be misleading, we used a normality test to check that the data was normal. When comparing statistics from samples, the same idea applies.
Yes, 175 outnumbers 168. Is it statistically different in the case of corn yield?
Six Sigma teams can’t simply answer, “Is this number different?” They must respond to the inquiry, “Is this number statistically distinct enough that we can act on it?”
The scale has nothing to do with statistical differences. In one process, the difference between 10 and 23 may not be statistically significant, whereas, in another, the difference between 10 and 10.5 maybe.
Learn more about analysis toolset in our signature Lean Six Sigma courses.