Statistics
Six Sigma – iSixSigma › Forums › Old Forums › General › Statistics
 This topic has 20 replies, 8 voices, and was last updated 17 years, 9 months ago by Chwang.

AuthorPosts

March 9, 2005 at 11:22 am #38679
Hi!I want to know about the test to compare two percentages from the same sample.
0March 9, 2005 at 11:36 am #116111
Joseph BanerjeeParticipant@JosephBanerjee Include @JosephBanerjee in your post and this person will
be notified via email.you want to compare the mean percentage of the two samples if yes then depending on the sample size you can go for one sample z or t test again depending if your data is normal.
0March 9, 2005 at 11:46 am #116112Bad call. It is a test of two proportions found under stat – basic stat in Minitab.
Proportions follow a binomial distribution. If certain assumptions are met, the assumption of normality may hold, but why bother with that – do it right.0March 9, 2005 at 11:48 am #116113
Ken FeldmanParticipant@Darth Include @Darth in your post and this person will
be notified via email.WRONG!!!!! You use a 2 proportion test. You use a one sample z to compare one sample to a standard…continuous data and a t test for smaller samples. A two sample t test for comparing the continuous mean of two samples. The 2 proportion test is for two proportions which is the same as two percentages which is what the poster asked about.
0March 9, 2005 at 11:50 am #116115
Ken FeldmanParticipant@Darth Include @Darth in your post and this person will
be notified via email.Stan, u r up early today. Beat me to the answer to this wrong response. Saw that Silicon Valley Stan finally gave his notice. A teary eyed farewell note to the gang….sob sob… Have a good day.
0March 9, 2005 at 12:05 pm #116117
indreshParticipant@indresh Include @indresh in your post and this person will
be notified via email.have a query might be trivial
percentage is a discrete data then why use Ttest
can’t we use chi sqaure to know if there is commonality between the standard and the sample in question ????
rgds,0March 9, 2005 at 12:36 pm #116119
Joseph BanerjeeParticipant@JosephBanerjee Include @JosephBanerjee in your post and this person will
be notified via email.Stan its not necessary that proportions to follow a binomial distribution
0March 9, 2005 at 12:43 pm #116121
Joseph BanerjeeParticipant@JosephBanerjee Include @JosephBanerjee in your post and this person will
be notified via email.Hi Indresh you need to ask yourself whether percentage is a discrete data.
0March 9, 2005 at 1:03 pm #116122Stan
Usually agree with you, but this time you’re aoff base. the “proportion test” is a Z test , plain and simple.
So your comment “If certain assumptions are met, the assumption of normality may hold, but why bother with that – do it right.” makes no sense. What would be doing it riight? Please provide an example. If the binomial approximations are not met the alpha risk is increased, but there is no parametric “proportions” test other than using the Z test.
Reference ” Handbook of Parametric and Nonparametric Statistical Procedures, Sheskin.0March 9, 2005 at 1:34 pm #116125Wrong
0March 9, 2005 at 1:35 pm #116126Wrong
0March 9, 2005 at 1:37 pm #116127Joseph,
Go learn something before you give advice0March 9, 2005 at 1:45 pm #116129That’s right!
You’re wrong!0March 9, 2005 at 1:50 pm #116131Why should a proportion test be a Ztest? You use the binomial distribution to compute the confidence intervals
0March 9, 2005 at 1:52 pm #116132Daves,
Before you get too far down this trail, go to the help menu in Minitab on the test of two proportions under stat – basic stat. Read.
Let’s assume you are right. I have one proportion – I have one defect in a sample of 30. I have another proportion I have 5 defects in a sample of 100. Is there a statistical difference and how exactly do you perform a z test on this?
Humble (or not so humble) apologies will be accepted.0March 9, 2005 at 1:59 pm #116134Staney boy
Please read from MINITAB help, Formulas and methods. I’m not apologizing for squat. It is simply an application of the Z test. It uses the binomial mean and sd, but the test statistic is Z.
Hypotheses
The null hypothesis for a 2 proportion test is: H0: p1 – p2 = d0
where:
· p1 = the proportion for the first population
· p2 = the proportion for the second population
· d0 = the hypothesized difference between the population proportions
You can choose any one of three alternative hypotheses:H1: p1 – p2 > d0
Onetailed test
H1: p1 – p2 < d0
Onetailed test
H1: p1 – p2 ddd0
Twotailed test
Test statistics
The calculation of the test statistic, Z, depends on the method used to estimate p.
By default, Minitab uses separate estimates of p for each population and calculates Z by:0March 9, 2005 at 2:19 pm #116136Now that you have given enough information that the person can doing something with (if they choose to do by hand), I agree.
My original response was correct of where to go and what test to run. If the original poster simply goes to a z test in Minitab, they wouldn’t know what to do.
Peace0March 9, 2005 at 2:26 pm #116137Also note that you get the following warning from Minitab when doing the sample I suggested –
* NOTE * The normal approximation may be inaccurate for small samples.
This is because we are using the normal (z) as an approximation of the binomial and there are certain assumpltions that must be met. The assumprions are simple np and n(1p) must both be greater than 5. There is a reason I gave you the example I did – I knew it did not meet the assumptions.
The correct way to do this is to get the confidence intervals of the two proportions using binomial and see if proportion 1 is inside of the confidence interval for proportion 2 and vice versa.0March 9, 2005 at 3:07 pm #116139Average men working in earnest together complete the building of the temple before the wisest solitary man will place even one stone.
0March 9, 2005 at 11:25 pm #116166Hey Confucius,
Strange to see you show up on a site dedicated to rampant profitering. Last time I checked, you were against such things.
Be that as it may, let me give you my new favorite Confucius type sayings –
1) Be careful when trying to sound like Confucius, people will not know if you are an idiot or a wise man who just sounds like an idiot.
2) Average men, with a little learning that they think is great learning, working in earnest together, are just a group of average men with a little learning.0March 11, 2005 at 7:34 pm #116289A learned man who casts away all who approach him is valued by men as is a large rock atop the highest mountain.
0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.