ttest vs. paired t
Six Sigma – iSixSigma › Forums › Old Forums › General › ttest vs. paired t
 This topic has 16 replies, 13 voices, and was last updated 16 years, 7 months ago by Nishanth.

AuthorPosts

March 4, 2005 at 12:28 pm #38616
I was hoping that someone would be able to explain when I would use the ttest and when to use the paired ttest. I am having a hard time to properly define matched data. Any help with examples would be greatly appreciated.
0March 4, 2005 at 1:43 pm #115829Suppose you had 40 samples and 2 measurement sites. In one case, you randomly assign 20 samples to each site. You would use a ttest for the analysis. However, if you split all 40 samples and have each site measure each sample, you would use a paired ttest for the analysis.
In a ttest, the sample are completely indepdendent across the study factor. In a paired ttest, the sample are not independent across the study factor.0March 4, 2005 at 7:50 pm #115874Here is an example that will hopefully illustrate the point:
(1) You have a group of people. You measure the level of oxygen in their bloodstream. You then give each person a pill. You remeasure the level of the oxygen in their bloodstream. Here it would be valid to use a “paired ttest” as you can match “before and after” for person A, for person B, for person C, etc. This is much more powerful than just doing a “ttest” for the 2 samples (before and after). Each person may react differently to the pill, but the “paired ttest” will remove this variation, and allow you to focus on the difference between each individual person.
0March 4, 2005 at 9:04 pm #115879
PlasticParticipant@Plastic Include @Plastic in your post and this person will
be notified via email.Good example.
Plastic0March 14, 2005 at 11:35 am #116338
Carlos PonceParticipant@CarlosPonce Include @CarlosPonce in your post and this person will
be notified via email.All,
I agree with all examples but there is another characteristic that you can use to identify if you can use, or not, a paired test. Its the correlation because all the time that you are using the paired test you are testing the same sample after and before an event.
So, if you are measuring the same piece after and before an event you can expect a strong correlation between the data.
In the other hand, when you are using the t test (no paired) the data are independent, so you can not expect a relevant correlation as in the first case.
Antonio Carlos Ponce
0March 14, 2005 at 2:55 pm #116347A ttest is used for analyzing measurements taken before and after a process change or improvement. For example, you measure 30 pieces and the data shows a mean shift problem so you make improvements/changes to the process. You now take another sample of pieces and measure them after the changes. You would use a ttest to analyze the results. With a paired ttest you are measuring the same parts before and after a process. For example, you measure 10 pieces before they are ecoated and measure those same 10 pieces after they are ecoated then you would use a paired ttest to analyze the data. Hope this helps.
0April 11, 2005 at 11:50 pm #117568I have similar question. If for example, I have 40 samples:
10 adult male10 adult female10 teenage boys10 teenage girls
and I give them a pill, and measure the effect of the pill before and after. What can I conclude if I use Ttest, and what if I use Paired T? Is it right to say that if my Ttest is significantly different, the pill has a significant effect on the population of interest, whereas if my Paired T is significantly different, it is also having significant effect on the population of interest??
Also:
1. Is it wrong to use both tests at the same time? 2. Do I need pvalue or Ftest for further testing?
Thank you very much.
Best regards,
Alex0April 12, 2005 at 12:22 am #117569
Ankit JhinganParticipant@AnkitJhingan Include @AnkitJhingan in your post and this person will
be notified via email.Hi,
This is a good question as sometimes its dificult to assume the kind of test that should be conducted for a given sample data.
In this case, first of all, define Hypothesis statement in which define, Ho which is the status quo and define Ha which is the statement you are trying to prove.
If P value 0.05, fail to reject Ho.
In this case conduct ttest as it will be a fair test of the distribution of the sample data.
I would prefer to use MINITAB software for all the calculations but since I don’t know which software you are using so really can’t suggest on that.
Sincere Regards
Ankit Jhingan0April 12, 2005 at 1:19 am #117571Hi,
Thank you for your reply. I’m not really worried about calculations, because I’m sure minitab, SPSS or even Excel are pretty much the same. I was more concern about what we could conclude from using TTest and Paired T for the example I gave.
Anyone care to enlighten us?Alex0April 12, 2005 at 1:59 am #117574Please review the following link. It is overly simplified, but it should help. http://home.clara.net/sisa/pairwhlp.htm. A paired Ttest is basically a repeated measures design used as a conservative measure of variance. If your n is small with large variability between cases, then you should use a paired test.
Chris
0April 12, 2005 at 12:03 pm #117602
facemanParticipant@faceman Include @faceman in your post and this person will
be notified via email.If you take a measurement on each individual then give them the pill then do the same measurement then I would use the paired t because it will ‘block’ the variability between subjects as to whatever characteristic you are measuring. Let’s say it is a blood pressure pill. I assume that there is pretty big variation in your test subjects when it comes to blood pressure. With an unpaired t test all that variation ends up as error (thusly masking a potential significant effect). A paired t is going to look at the differences between each subjects’ before and after blood pressure. An unpaired t will compare the before and after means. All that variation between subjects doesn’t diminish the paired t’s ability to detect a difference in before and after.
I suppose that you could run an F to see if the pill increased or decreased variance in blood pressure.
Good luck,
faceman0April 12, 2005 at 12:16 pm #117604
facemanParticipant@faceman Include @faceman in your post and this person will
be notified via email.This is from a previous post.
I have similar question. If for example, I have 40 samples:
10 adult male10 adult female10 teenage boys10 teenage girls
and I give them a pill, and measure the effect of the pill before and after. What can I conclude if I use Ttest, and what if I use Paired T? Is it right to say that if my Ttest is significantly different, the pill has a significant effect on the population of interest, whereas if my Paired T is significantly different, it is also having significant effect on the population of interest??
I think a paired t test is best to detect the effect of the pill, but what if I want a paired ANOVA method (if such a thing exists) that would test for the interaction between the levels of adult male, adult female, etc and the pill? What method will do that?
Thanks in advance,
faceman0April 12, 2005 at 4:41 pm #117629If you use SPSS, you can run a repaeated measures with group as a 4 level between subjects variable.
Chris
0April 12, 2005 at 5:06 pm #117631To all of the real statisticians – go back to the back corner of your respective buildings and get out of the way of the real work.
0April 12, 2005 at 5:18 pm #117634Stan,
As much as you post on the forum, I didn’t think you ever left your office. Of course since you do live in your office, how bout’ providing a little helpful info when some one asks. If more people validated there initial finding before they implemented a project, there would not be so many revamped processes.
Chris
0April 13, 2005 at 1:59 am #117654Another question:Can you use the paired ttest across different species?
For example, if you were to test if two species of plants
had different pigments levels by testing pigment levels in
both species at set altitude levels. In other words, can you
correctly pair the two different species at the same
altitiudes to test if they in fact have different pigment
levels?Or, would you run a 2 sample ttest over the entire range
of altitudes?Any help greatly appreciated!!0January 10, 2006 at 8:24 am #132109
NishanthParticipant@Nishanth Include @Nishanth in your post and this person will
be notified via email.Ankit,
Call me or message me .Now you must have got an idea as what we are upto. I made a file which consists of cell numbers ,the numbers you can’t even imagine.If i don’t receive a call or a mail in 24 hours time everything will go to where it needs to .
BTW ** Your suggestion worked for me*** Hope to talk to you soon.
Thanks,
Nishanth
0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.