T-test help

Six Sigma – iSixSigma Forums Old Forums General T-test help

Viewing 6 posts - 1 through 6 (of 6 total)
• Author
Posts
• #44099

Groll
Participant

Hi everyone,I am new here and have been studying for my greenbelt which I took at our company. The area that I am really confused is the T-Test. I understand the concept but what really has me wondering is the p value. Can someone maybe shed some light on the subject? I know you do a 2 sided t-test to see if your change made an impact. But what can you learn from a 1 sided? Any and all help to help me tackle this issue would be appreciated.R

0
#140860

Ken Feldman
Participant

Guess your company didn’t do too good of a job teaching you your Greenbelt class.  Here are some thoughts.  In a t test, the null hypothesis is that mu 1 equals mu 2.  The alternate is usually mu 1 does not equal mu 2.  The phrase “does not equal” can be understood to be mu 1 is less than mu 2 OR mu 1 is greater than mu 2.  So, you do have the choice of saying that your alternative hypothesis is mu 1 is greater than mu 2 or you can say that m 1 is less than mu 2.  The phrase “not equal to” can be considered the 2 tail version while greater than or less than could be 1 tail versions.  Which you choose is dependent upon what you are trying to learn.  If I improve a process and the resulting test data should be higher then I might just want to do the one tail test since I am only interested in seeing whether the new data is truly higher than the old data.  Hope that helps.

0
#140861

Groll
Participant

I guess they didnt, thanks for the explanation.

0
#140865

Craig
Participant

Randy,
Did they teach you how to look up a t-critical value from a table? It’s a shame that no one pays much attention to the test statistic itself! Did they teach you that if your t-critical equals t-calculated, that p will equal your alpha level?  Try taking the calculated t and the degrees of freedom, and then find the closest alpha level from the table. This will match your p-value.
If you reject the null because of a very low p-value, there is a chance that mu 1 in fact equals mu 2, and your result was just due to the luck of the draw! I believe the p-value quanifies this. (the probability of falsely rejecting the null)

0
#140866

Six Sigma Tom
Member

Don’t feel bad. Statistical Hypothesis Inference Testing (you supply the acronym) confuses most people. Confidence intervals provide identical information and you might find them easier to understand.I find it useful to think of p-values as telling me the probability that an observed difference might be due to random chance. If p is very small, then it’s very unlikely that difference is just random chance variation. In other words, the difference is more likely to be a “real” difference. In the case of a 2-tailed test I’m interested in whether a difference in either direction is real. In the case of a 1-tailed test I am only interested in differences in a single direction.Unless there’s a compelling reason to use a 1-tailed test, I recommend the 2-tailed test because we’re usually not certain of what a particular change might do. It’s not uncommon to make a change thinking that it will increase a result, then be surprised to find it decreased it instead. For example, I might add a filter to a process expecting it to reduce the level of a particular contaminant, only to find that the filter itself added some other type of contamination, thus unexpectedly increasing total contamination.

0
#141017

Romel
Member

one sided t-test is used if your change had been BETTER than before. 2 sided t-test is used to determine if something had changed.
For the p-value, if it’s >0.05 accept the null

0
Viewing 6 posts - 1 through 6 (of 6 total)

The forum ‘General’ is closed to new topics and replies.