From: Madan Gopal Kundu <madan4331@...>
Subject: [Statisticians_group] Re: Testing Normality of biological variables
To: Statisticians_group@...
Date: Saturday, May 9, 2009, 2:01 PM

Hi Ramesh,

The problem you stated regarding normality test by KS test is genuine and this is true for other normality tests as well. The moral is we should not decide normality of a data only just looking at p-value of normality tests.

To overcome this we should follow any of the following strategies:
1. First increase the significance level i.e. instead of using 5% use 10% significance level.
2. Use Quantile-Qunatile (Q-Q) plot to decide normality.

I think the second option is better and this is available with standard softwares such as SAS. A nearly straight line in Q-Q plot indicates normality. A significant deviation from straight line indicates non-normality. An alternative of Q-Q plot is Probability - Probability (P-P) plot.

Coming to your second question, I strongly discourage you to use Bonferroni methods or other similar methods to adjust multiplicity issue. Use of Bonferroni method ultimately decreases p-value and which in turn increases the possibility of declaring data as non-normal. So DON'T use it.

Hope this helps you.

Thanks & Regards
Madan Gopal Kundu

--- In Statisticians_ group@yahoogroup s.co.in, "Ramesh S.Ve." <optoramesh_ gp@...> wrote:
>
> Dear Friends
> I was working on one of our data related to anatomical measurements of the nerve size & area in the eye... Our sample size is 450 subjects, surprisingly when each parameter was tested for normality showed siginificant (p<0.05) non normal distribution as results in Kolmogrov Smirnov. Though the central limiting theorem suggests that a sample greater than 30 would be normal, bio statistical variables are usually not so.... (I would like to know how many of you would agree).... Shall we continue performing a parametric test such as t test, anova or should we opt for non parametric alternatives in analysisng these data sets.... Kindly advice:
>
> 1. Better methods to test for normality
> 2. Like bonferroni correction for multiple t test is there a correction factor (adjusting the significance level of p value) when 4 or 5 variable are tested together using independent KS test
>
> Thanks in advance
>
> Regards
> Ramesh.S.Ve
> PhD student
> Sankara Nethralaya
> 18, College Road
> Chennai-600006.
> Tamil Nadu
> India.
> Tel: 91-44-28271616
> Fax: 91-44-28254180
> Mobile: 91-09444073160
> Email: optoramesh_gp@ ..., sver@...
>
>
> Confidentiality Notice: This transmittal is a confidential communication. If you are not the intended recipient, you are hereby notified that you have received this transmittal in error and that any review, dissemination, distribution or copying of this transmittal is strictly prohibited. If you have received this communication in error, please notify this office immediately by reply and immediately delete this message and all of its attachments, if any.
>

From: ramesh vijayalayan <ramesh_vijayalayan2002@...>
Subject: Chi Square Test
To: Statisticians_ group@yahoogroup
Date: Saturday, 9 May, 2009, 9:40 AM

Please clarify my doubt. I have cross tabulation 4x5 with total frequency 100, but several cell frequencies (say 7) less than 5.  How can I use chi square test for this table?  Can I use Microsoft Excel for chi square test? Ramesh

---

Now surf faster and smarter ! Check out the new Firefox 3 - Yahoo! Edition * Click here!

From: Khassoum Diallo <kdiallo@...>
Subject: Re: [Statisticians_group] (unknown)
To: Statisticians_group@...
Date: Wednesday, May 6, 2009, 12:21 AM

 

Thanx for answering me i checked the link and got cleared

regards

 

Dear Hasan,   You need a number of data points or values in order to estimate a parameter (e.g. the mean). The degrees of freedom is the minimum number of these data points.   A good explanation is given at the following link.   http://www.statisti cs.com/resources /glossary/ d/degsfree. php   Cheers, Khassoum  --- On Wed, 5/6/09, Hasan ALi <street192005@ yahoo.com> wrote: From: Hasan ALi <street192005@ yahoo.com> Subject: [Statisticians_ group] (unknown) To: Statisticians_ group@yahoogroup s.co.in Date: Wednesday, May 6, 2009, 7:43 AM i had read out almost all the books of statistics but i am still dont clear about degrees of freedom.is there any body who can answer so simply that could be digestable regards

From: Madan Kundu <madan4331@...>
Subject: Re: [Statisticians_group] (unknown)
To: Statisticians_group@...
Date: Thursday, May 7, 2009, 12:53 AM

Let me try!   Suppose you have 4 variables, say, x1, x2, x3, x4. Now you know their sum i.e. x1+x2+x3+x4= n, where n is some known number.   Now here you can vary the value of only 3 variables, becasue the value of 4th variable will be calculated by substracting the sum of the 3 other variables from n. Since here you can vary only 3 variables at your will, degrees of freedom will be 3, not the 4.   for example, if x1+x2+x3+x4= 120, and the value of x4 must be 120-x1-x2-x3. See here you can only vary the value of x1, x2 and x3. It won't allow you to vary the value of x4 as it will be calculated from 120-x1-x2-x3.   In general, we can say that degress of freedom will be equal to no. of the variables minus the no. of constraints.   In the above example, we had 4 variables and one constraint that is there sum will be 120. So in this case degrees of freedom will be 4-1=3.   Hope this helps.   Regards ------------ -- Madan Gopal Kundu Biostatistician, Ranbaxy Labs. Ltd. Gurgaon, Haryana India Web: http://www.freewebs .com/madanstata mobile: 91-9868788406 e-mail: Madan.Kundu@ ranbaxy.com   --- On Wed, 6/5/09, Hasan ALi <street192005@ yahoo.com> wrote: From: Hasan ALi <street192005@ yahoo.com> Subject: [Statisticians_ group] (unknown) To: Statisticians_ group@yahoogroup s.co.in Date: Wednesday, 6 May, 2009, 12:13 PM i had read out almost all the books of statistics but i am still dont clear about degrees of freedom.is there any body who can answer so simply that could be digestable regards mr. kandu thanx for ur answer, i already know about the theory that u mentioned but the aim of my question is ask about what are the benefits while knowing the sum of all variables and to exclude on of them. what is the significance of removing the variable.if we remove the variable so is there any specific variable that we can remove or just to subtract from the total. regards

---

Now surf faster and smarter ! Check out the new Firefox 3 - Yahoo! Edition * Click here!

From: Hasan ALi <street192005@...>
Subject: [Statisticians_group] (unknown)
To: Statisticians_group@...
Date: Wednesday, 6 May, 2009, 12:13 PM

i had read out almost all the books of statistics but i am still dont clear about degrees of freedom.is there any body who can answer so simply that could be digestable regards

From: biotechbs <gedoardo83@...>
Subject: [Statisticians_group] Calculation of Confidence Interval
To: Statisticians_group@...
Date: Wednesday, 6 May, 2009, 9:13 PM

Hi everybody,

This is my question. I have two set of data: set A (from A1 to A8) and set B (also from B1 to B8). Each value of both sets are calculated as average of a distinct group of measures and has its associated standard deviation.

Then I have another set of value (set C, the one I want to plot as final results) which is calculated as the ratio between each A value and correspondent B value (for example A1/B1, A2/B2, ecc.).

How can I calculate a confidence interval for set C values to use to plot error bars?

Thanks a lot.
Edoardo

From: biotechbs <gedoardo83@...>
Subject: [Statisticians_group] Calculation of Confidence Interval
To: Statisticians_group@...
Date: Wednesday, May 6, 2009, 11:43 AM

Hi everybody,

This is my question. I have two set of data: set A (from A1 to A8) and set B (also from B1 to B8). Each value of both sets are calculated as average of a distinct group of measures and has its associated standard deviation.

Then I have another set of value (set C, the one I want to plot as final results) which is calculated as the ratio between each A value and correspondent B value (for example A1/B1, A2/B2, ecc.).

How can I calculate a confidence interval for set C values to use to plot error bars?

Thanks a lot.
Edoardo

From: Hasan ALi <street192005@...>
Subject: [Statisticians_group] (unknown)
To: Statisticians_group@...
Date: Wednesday, May 6, 2009, 7:43 AM

i had read out almost all the books of statistics but i am still dont clear about degrees of freedom.is there any body who can answer so simply that could be digestable regards

David,

Yes I hope someone in the group has worked with your type of spatial correlation data. True, you would need computer software to permute even an appreciable number of those possible combinations of 10 pairs of nests.

I still don’t see how the correlation could be exactly zero; everything is correlated with everything else in this world. You know the saying: A butterfly flapping its wings in North America can start a chain of events that leads to a hurricane in China. But anyway that’s not the most important issue here. Rather I want to alert you that a dataset could actually support the null hypothesis Ho over the alternative Ha, no matter how small is the p-value. So, if you insist that Ho is a priori possible, then Ho could have probability >50% even with a very small p-value. This result is known as Lindley’s Paradox. Therefore, it is not correct to “infer that dispersal is probably not random” using the p-value. The problem is that a p-value is a statement about data assuming this or that hypothesis, rather than a statement about any hypothesis given data. --- On Sun, 5/3/09, slop badgerd <slopbadgerd@...> wrote:

From: slop badgerd <slopbadgerd@...>
Subject: RE: [Statisticians_group] pearson correlation test with permutation
To: statisticians_group@...
Date: Sunday, May 3, 2009, 2:05 PM

thanks for your your advice andrew, though im not sure how i would go about permuting all the possible combinations of 10 nests without some sort of computer program, as it would surely take a long time to do it manually.
 
"David, if you obtain a p-value, what would it tell you about your proposed null hypothesis (H) of absolutely no correlation between distance D & relatedness R? A p-value will only indicate the probability of data at least as extreme as what you observed, assuming H; it’s a statement about data, rather than about H. Besides, I would have a difficult time accepting a priori that H could be true in the first place, & since I’m already (almost) sure H is false, I have little reason to test it."
 
the null hypothesis of no correlation between distance and relatedness represents what would be expected if the bees dispersed to a random site within their patch (which is certainly very plausible). if the pearson correlation coefficient was negative and the p-value was low (less than 0.05 for my purposes) i could infer that dispersal is probably not random, but that the bees disperse only short distances and set up nests close to their sisters (which is also plausible). conversely, if there was statistically significant positive correlation, that would suggest the bees actively disperse as far away as possible from their sisters, which again is ecologically interesting and plausible (e.g. to avoid competing against their relatives).
 
"If I was in your position I think I would rather determine the probability that the correlation coefficient (rho, say) is within some practically meaningful range, [a,b]. viz., the approach should rather be one of estimating the correlation, not seeing whether the correlation is exactly zero."
 
if there is a real correlation i have no reason to expect that the correlation will be a particular value (or within a particular range of values) and in fact it isn't particularly important in my case as (for what i am researching at this stage anyway) it doesnt really matter if the correlation is 0.3 versus 0.5 say - the ecological implications would be similar. in contrast, the ecological implications of any correlation (whatever its exact value) would be different from no correlation. hence at this stage im really only interested in testing whether my observed correlation is statistically different from zero. however certainly in the future testing whether any correlation is within a particular range of values could be an interesting follow up question and i thank you for the suggestion.
 
but at this stage im still just looking for a way of conducting a permutation- based significance test (2-tailed) for the pearson correlation between the 2 variables. if anyone has any suggestions it really would be a huge help.
 
david

 

---

To: Statisticians_ group@yahoogroup s.co.in
From: khahstats@yahoo. com
Date: Sat, 2 May 2009 21:35:13 -0700
Subject: Re: [Statisticians_ group] pearson correlation test with permutation

David, if you obtain a p-value, what would it tell you about your proposed null hypothesis (H) of absolutely no correlation between distance D & relatedness R? A p-value will only indicate the probability of data at least as extreme as what you observed, assuming H; it’s a statement about data, rather than about H. Besides, I would have a difficult time accepting a priori that H could be true in the first place, & since I’m already (almost) sure H is false, I have little reason to test it.   If I was in your position I think I would rather determine the probability that the correlation coefficient (rho, say) is within some practically meaningful range, [a,b]. viz., the approach should rather be one of estimating the correlation, not seeing whether the correlation is exactly zero.   When all the data (D & R) are i.i.d., this can be done using the posterior probability distribution for rho given in http://www.pubmedce ntral.nih. gov/picrender. fcgi?artid= 155684&blobtype=pdf Your case may be a little more complicated because, as you note, the data between any pair of nests A & B depend on the data between (say) A & C (i.e., whenever pairs of nests share a member). I don’t have any experience with this type of spatial analysis; however, I would guess that since you are trying to infer something about the relation between D & R, rather than D & R themselves, I would guess that taking all 21*20/2=210 pairs of D & R as independent would not bias the sample statistic (rho-hat); it would only complicate the calculation of the variance of rho-hat (conditional on rho) & hence of the spread of the posterior probability distribution of rho.   You are looking for an answer ASAP. Therefore, without researching this in depth, I would say that you could Split the 21 nests into 2 sets of 10 A & B, leaving 1 out. Calculate rho-hat for those 10 pairs of nests. Form 2 sets of 10 nests each in a new way. Calculate rho-hat for those 10 new pairs of nests. Iterate the above until you have done this for all possible setups of 10 pairs. Calculate the mean of all the resulting rho-hats. Compare that mean with the rho-hat you get when treating all the 210 pairs of D & R as independent. You could do something similar, I imagine, comparing 2 ways of calculating the variance of rho-hat (conditional on rho), making an adjustment for the square root of the sample size.   I don’t know how good is this method of handling the dependencies; someone else may be able to provide a better answer. Nonetheless, I do feel strongly that the focus should be on estimation rather than significance testing. Best wishes. --- On Sat, 5/2/09, slop badgerd <slopbadgerd@ hotmail.com> wrote: From: slop badgerd <slopbadgerd@ hotmail.com> Subject: [Statisticians_ group] pearson correlation test with permutation To: statisticians_ group@yahoogroup s.co.in Date: Saturday, May 2, 2009, 9:00 PM hello everyone, im new to the group. just thought i'd say hello and ask for some help with a stats problem. im studying bees and am trying to find if there is a correlation between the distance between a pair of nests and the genetic relatedness between the bees occupying the nests. there is 21 nests in the sample and have i created a pairwise distance matrix showing the distance between each pair of nests, and similarly, a pairwise relatedness matrix for each pair of nests. it is my understanding that you cant use tables to get the statistical significance (p-value) of the pearson's correlation coefficient because the data are not independent (if you change the position of one nest, all the pairwise distance values from that nest to all the others will change) and therefore statistical significance must be calculated by permuting the data to generate a null distribution.   i have used the mantel test in the program zt to calculate the pearson correlation and its significance. however i now need to test for a correlation using only a subset of the data, such that it is not possible to create a full matrix (i.e. there will be gaps in the distance and relatedness matrices), and so the mantel test does not work properly.   the data is in 2 columns and i have done a normal pearson correlation test on it. however to calculate the p-value i assume i need to permute the data to create the null distribution. i have been looking on the internet for a program that can do this but i've had no luck. i downloaded a program called corrperm that looked like it could do the job but i couldn't figure out how to work it. i emailed the guy who wrote it and he replied with a new version of the program for R but i dont have a clue how to use that (i have no programming knowledge or experience with anything of that sort). the guy's away now for a week and i need to do this test asap so if anyone has any advice it would be really appreciated, e.g. is there any simple program out there that can run on windows command prompt with clear instructions how to use it?   thanks in advance,   david Share your photos with Windows Live Photos – Free. Try it Now!

---

Get the New Internet Explore 8 Optimised for MSN. Download Now

David, if you obtain a p-value, what would it tell you about your proposed null hypothesis (H) of absolutely no correlation between distance D & relatedness R? A p-value will only indicate the probability of data at least as extreme as what you observed, assuming H; it’s a statement about data, rather than about H. Besides, I would have a difficult time accepting a priori that H could be true in the first place, & since I’m already (almost) sure H is false, I have little reason to test it.

If I was in your position I think I would rather determine the probability that the correlation coefficient (rho, say) is within some practically meaningful range, [a,b]. viz., the approach should rather be one of estimating the correlation, not seeing whether the correlation is exactly zero.

When all the data (D & R) are i.i.d., this can be done using the posterior probability distribution for rho given in

http://www.pubmedcentral.nih.gov/picrender.fcgi?artid=155684&blobtype=pdf

Your case may be a little more complicated because, as you note, the data between any pair of nests A & B depend on the data between (say) A & C (i.e., whenever pairs of nests share a member). I don’t have any experience with this type of spatial analysis; however, I would guess that since you are trying to infer something about the relation between D & R, rather than D & R themselves, I would guess that taking all 21*20/2=210 pairs of D & R as independent would not bias the sample statistic (rho-hat); it would only complicate the calculation of the variance of rho-hat (conditional on rho) & hence of the spread of the posterior probability distribution of rho.

You are looking for an answer ASAP. Therefore, without researching this in depth, I would say that you could

1. Split the 21 nests into 2 sets of 10 A & B, leaving 1 out.
2. Calculate rho-hat for those 10 pairs of nests.
3. Form 2 sets of 10 nests each in a new way.
4. Calculate rho-hat for those 10 new pairs of nests.
5. Iterate the above until you have done this for all possible setups of 10 pairs.
6. Calculate the mean of all the resulting rho-hats.
7. Compare that mean with the rho-hat you get when treating all the 210 pairs of D & R as independent.

You could do something similar, I imagine, comparing 2 ways of calculating the variance of rho-hat (conditional on rho), making an adjustment for the square root of the sample size.

I don’t know how good is this method of handling the dependencies; someone else may be able to provide a better answer. Nonetheless, I do feel strongly that the focus should be on estimation rather than significance testing. Best wishes.

From: slop badgerd <slopbadgerd@...>
Subject: [Statisticians_group] pearson correlation test with permutation
To: statisticians_group@...
Date: Saturday, May 2, 2009, 9:00 PM

hello everyone, im new to the group. just thought i'd say hello and ask for some help with a stats problem. im studying bees and am trying to find if there is a correlation between the distance between a pair of nests and the genetic relatedness between the bees occupying the nests. there is 21 nests in the sample and have i created a pairwise distance matrix showing the distance between each pair of nests, and similarly, a pairwise relatedness matrix for each pair of nests. it is my understanding that you cant use tables to get the statistical significance (p-value) of the pearson's correlation coefficient because the data are not independent (if you change the position of one nest, all the pairwise distance values from that nest to all the others will change) and therefore statistical significance must be calculated by permuting the data to generate a null distribution.
 
i have used the mantel test in the program zt to calculate the pearson correlation and its significance. however i now need to test for a correlation using only a subset of the data, such that it is not possible to create a full matrix (i.e. there will be gaps in the distance and relatedness matrices), and so the mantel test does not work properly.
 
the data is in 2 columns and i have done a normal pearson correlation test on it. however to calculate the p-value i assume i need to permute the data to create the null distribution. i have been looking on the internet for a program that can do this but i've had no luck. i downloaded a program called corrperm that looked like it could do the job but i couldn't figure out how to work it. i emailed the guy who wrote it and he replied with a new version of the program for R but i dont have a clue how to use that (i have no programming knowledge or experience with anything of that sort). the guy's away now for a week and i need to do this test asap so if anyone has any advice it would be really appreciated, e.g. is there any simple program out there that can run on windows command prompt with clear instructions how to use it?
 
thanks in advance,
 
david

---

Share your photos with Windows Live Photos – Free. Try it Now!

From: Riadi <a_riadi_a@...>
Subject: Re: [Statisticians_group] The variance of OLS estimator
To: Statisticians_group@...
Date: Saturday, May 2, 2009, 5:55 PM

Thank you for your advice. I have read from this link: http://connection. cwru.edu/ econ326/mod4/ A/Computing% 20Variance% 20of%20Simple% 20OLS%20Estimato r.doc It presents simple model y=a+bx+e, after that it shows how to find the variance of OLS estimator b. What I want to know is how to find the variance of OLS estimator a ?   Any help will be much appreciated. Thank you very much. --- On Sun, 5/3/09, Andrew Hartley <khahstats@yahoo. com> wrote: From: Andrew Hartley <khahstats@yahoo. com> Subject: Re: [Statisticians_ group] The variance of OLS estimator To: Statisticians_ group@yahoogroup s.co.in Date: Sunday, May 3, 2009, 2:36 AM A Riadi, I advise you to consult a standard textbook on this topic. These variances are almost always presented in introductory statistics texts. --- On Sat, 5/2/09, A. Riadi <a_riadi_a@yahoo. com> wrote: From: A. Riadi <a_riadi_a@yahoo. com> Subject: [Statisticians_ group] The variance of OLS estimator To: Statisticians_ group@yahoogroup s.co.in Date: Saturday, May 2, 2009, 8:23 AM Dear all statisticians_ group member. Can any body help me to show how to derive the variance of Ordinary Least Squares (OLS) estimator? y=a+bx+e I have seen the variance of OLS estimator a and b. var[a] and var[b], but I don't know how to derive those variance. Please help me. Thank you very much.

From: Andrew Hartley <khahstats@...>
Subject: Re: [Statisticians_group] The variance of OLS estimator
To: Statisticians_group@...
Date: Sunday, May 3, 2009, 2:36 AM

A Riadi, I advise you to consult a standard textbook on this topic. These variances are almost always presented in introductory statistics texts. --- On Sat, 5/2/09, A. Riadi <a_riadi_a@yahoo. com> wrote: From: A. Riadi <a_riadi_a@yahoo. com> Subject: [Statisticians_ group] The variance of OLS estimator To: Statisticians_ group@yahoogroup s.co.in Date: Saturday, May 2, 2009, 8:23 AM Dear all statisticians_ group member. Can any body help me to show how to derive the variance of Ordinary Least Squares (OLS) estimator? y=a+bx+e I have seen the variance of OLS estimator a and b. var[a] and var[b], but I don't know how to derive those variance. Please help me. Thank you very much.

From: A. Riadi <a_riadi_a@...>
Subject: [Statisticians_group] The variance of OLS estimator
To: Statisticians_group@...
Date: Saturday, May 2, 2009, 8:23 AM

Dear all statisticians_ group member.

Can any body help me to show how to derive the variance of Ordinary Least Squares (OLS) estimator?

y=a+bx+e
I have seen the variance of OLS estimator a and b.
var[a] and var[b], but I don't know how to derive those variance. Please help me. Thank you very much.

Akib,

Your topic is extremely important; I’m glad that people are working on it. This problem of a lack of rural doctors & nurses affects a large number of countries.

An example of something you might want to study is the proportion of health care professionals (HCPs) who migrate to cities because they enjoy city life. Call this proportion “mu.” It’s a “quantity of interest,” or a “parameter.”

Then, I suppose a meaningful result you want to present would be in the format “We can be 95% certain that mu is between 0.2 & 0.3.” This type of result identifies the “strength of the findings” & the degree to which we can “be confident about the conclusions” that you mention.

However, when you focus on “strength of the findings” or being “confident about the conclusions” of this or that percentage, you must keep in mind that standard statistical techniques don’t convey that type of meaning. A 95% confidence interval (CI) is one such technique. Most people (& even many statisticians) believe that a CI is a statement about strength of findings or about confidence in conclusions. It’s not. To understand 95% CIs, you must pretend that we are performing an experiment an infinite number of times, & for each experiment we calculate a 95% CI. The 95% implies that 95% of all these CIs will contain the parameter. The 95% is, then, a confidence about an infinite series of CIs, rather than a confidence that the CI we have obtained contains the parameter. Unfortunately, most people cannot resist the temptation to say that “We are 95% certain this CI contains the parameter.” (If you want me to illustrate how we might be more or less than 95% confident of that, I can do so). Furthermore, since they think we can be 95% that a given CI contains the parameter, they continue to use CIs.

You may be looking for quick & ready answers. However, not knowing what you’ve heard concerning standard statistical techniques such as CIs, I’m taking the time to explain this to help you understand the disconnection between CIs & the “strength” & “confidence” you seek. Does what I’ve written so far make sense to you?

--- On Fri, 5/1/09, akib ul <akib_du@...> wrote:

From: akib ul <akib_du@...>
Subject: Re: [Statisticians_group] Percentage - Percentile?
To: Statisticians_group@...
Date: Friday, May 1, 2009, 3:21 AM

Dear Hartley Thanks for the reply. The study was a type of opinion survey trying to identify the causes of why health care professionals do not want stay in rural areas, why they want to migrate to cities, or to the developed countries. To collect the data a questionnaire was used where the questions contained economic, political, social, and professional aspects.  A convenient type of sampling procedure was used to collect date, of course which was neither random nor representative. The researcher used percentage (based on how many respondents answered a question "yes" among the total 300 respondents) to show the importance of different contributing factors. Finally, based on the percentage the researcher drew some conclusions and put some suggestions to stop above mentioned migrations. My question is about the strength of the findings (even if it were a random sample) which are purely based on percentage. To what extent we can be confident about the conclusion of identifying the factors contributing to/responsible for something which is a psychological or sociological phenomenon. It might be that other more important factors are there which are not included in the questionnaire or considered by the researcher.  In a behavioral research what is the scope of using percetage to come to a conclusion. Sorry, it has become a long mail. However, the answer will help me to clear up my confusion. Also, a reference article/book (available in the internet) will be helpful. Regards Akib --- On Thu, 4/30/09, Andrew Hartley <khahstats@yahoo. com> wrote: From: Andrew Hartley <khahstats@yahoo. com> Subject: [Statisticians_ group] Percentage - Percentile? To: Statisticians_ group@yahoogroup s.co.in Date: Thursday, April 30, 2009, 10:08 AM Akib, I have difficulty approaching your question without more information. "Percentage" can mean many different things; are you speaking of "percentile? " Can you please explain more about what you want to do? What are you studying? What scientific conclusions do you want to draw? Are you looking at economic, biological, clinical, educational data or what? Are you trying to infer something about the location of a population? How certain do you need to be about that location? Why can't you take a purposive sample? --- On Thu, 4/30/09, akib ul <akib_du@yahoo. com> wrote: From: akib ul <akib_du@yahoo. com> Subject: [Statisticians_ group] please help me To: Statisticians_ group@yahoogroup s.co.in Date: Thursday, April 30, 2009, 4:43 AM Can anyone please tell me about the strength of "percentage" as a statistical tool to draw conclusion without doing any significance test and using purposive sample. Regards Akib