Hypothesis Tests on Proportions

I

Inspector-71

My question surrounds tests on proportions so looking at Binomial for 1 and 2 proportion tests and chi-sq test for 3+ proportions

Am I correct in saying that these tests are only used to infer the proportions (and differences between them) of populations based on a sample?

So to take minitab?s example, a manager of three customer support centres want to check successful resolution (yes/no) for three different support centres so tallies successful and unsuccessful resolutions per branch in a table

If the manager tallies the yes/no per branch for all the calls and they come out %Yes 34, 35, 36, then surely they are different and we don?t need to run any stats tests?

Or

Are we saying that we do need a stats test because although we have all the calls to date, it is just a sample of the future population (assuming no changes)?

I am preparing training material on this and can imagine this being a tough sell.

Obviously it becomes simpler if we are sampling say 10 bolts per hour out of hundreds per production line and testing them to infer %defective of all the bolts.

Thanks,
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
Am I correct in saying that these tests are only used to infer the proportions (and differences between them) of populations based on a sample?

SSP: Yes, that would be correct. If I "knew" the entire population then I have the answer directly from the results.

>So to take minitab?s example, a manager of three customer support centres want to check successful resolution (yes/no) for three different support centres so tallies successful and unsuccessful resolutions per branch in a table

>If the manager tallies the yes/no per branch for all the calls and they come out %Yes 34, 35, 36, then surely they are different and we don?t need to run any stats tests?

SSP: Depends upon if the manager wants to make a decision related to future performance or not. Also, even if you think you've sampled the entire population, you only did it for one moment in time. Perhaps if you asked the same question to a person 10 minutes from now rather than right now, their answer might be different, depending on how they are feeling, or what they experience in the next 10 minutes.

Or

>Are we saying that we do need a stats test because although we have all the calls to date, it is just a sample of the future population (assuming no changes)?

SSP: Yes, that is a better interpretation. Dr. Deming did say that the job of management is prediction.

I am preparing training material on this and can imagine this being a tough sell.

Obviously it becomes simpler if we are sampling say 10 bolts per hour out of hundreds per production line and testing them to infer %defective of all the bolts.

SSP: I am not so sure the difference is "obvious". By the way, the Red Bead Experiment is a great hands-on tool for introduction of these ideas to students.
 
Top Bottom