If you have categories of yes/ no or multiple choice or some variable that receives a set of numerical or alphanumeric data, then there are averages and standard deviations that can be applied to the data set.

To generalize you should show that the average and standard deviations of your sample is sufficiently close to the averages and standard deviations of the general population within the percentage chance you are required to use.

Normally the acceptable probabilities in research are either 99.5% for general questions or 99.95% when health and safety are being questioned.

In design of your experiment you could decide in advance which statistic will be used. Example to use a normal distribution you mist have more than 30 sample size and show that the distribution of results is close enough to normal distribution.

Distributions will not always be normal. They may be binomial, Poisson, or about a hundred different distributions.

One way to decide on sample size is to choose three or more sample sizes and discover the distribution that they all have. Example for averages or central tendency, if the general population is normally distributed, then small samples will all agree in the Student t Test. In the same data type for the standard deviations, if the general population follows a normal distribution, then all of your smaller samples should agree in a Chi Square Test.

For each distribution for large population of unknown size, there is a related statistic for the smaller sample of known size.

Statistical inference is such a large field that you should probably find a graduate level text book in your categories marketing and social research, to help decide what statistical distribution to use. Otherwise there is a huge amount of work to do testing all of the possibilities.

The point of doing inference is that if the population averages and standard deviations are not known, then your sample sizes should continue to increase until the related statics give the same results for averages and standard deviations for different sample sizes within the allowed probabilities.

There is no hard and fast rule , it's varies from research to research. If you will use small set of data, you can use grey system theory.

For example - Exploring socially responsible behaviour of Indian consumers: an empirical investigation

I'm not specialist in social investigations but approach I hope should be the same like in General statistic procedure. The larger amount of people are in volved into yours Quest. the better! For good represetatopn of the data is important waht type of distribution on the area you have to do!!! It depends on concentration of population in large cilies and towns and general area of analized surface. It is bette to achieve some standart type, like Poisson or negative binomial.

As well distribution of the sampling data should follow normal curve. Should check for normality the data obtained. If abnormal should transform. Use a procedure of normalization (like lognormal in Biological cases). After that can compare Cases (eg, in Seasonal aspect, one month before elections, two month before elections etc.). IT IS NOT SO SIMPLE...

Andrey

As Nennli said each test paragraph five individuals, for example, the standard or test on twenty paragraphs, the number of individuals is one hundred individuals, and the sample is consistent with the study community was the results of the study realistic.

Valuable inputs..

If you can manage 1 promil of the sample, it should be ok. For any staristical analysis you need to have over 100 respondents.

Dear Sajda Taha Mahmood - I have already mentioned one of example of rsearch that is -- For example - Exploring socially responsible behaviour of Indian consumers: an empirical investigation

Dear Mirna Leko-Šimić -- Thanks for your reply. You mean to say that results of a research can be generalized if respondents are 100 or more... Can you please provide some reference / literature upon it, validating this fact....

Dear Sir All research market include sampling and sample size and the equations of how you can identify sample size . I think it will help

Dear Mr. Haji - my question is not about how to calculate the sample size. I think u have misinterpreted the question !!

Will anyone help me in this regard ?

Good discussion.................. following

If you have categories of yes/ no or multiple choice or some variable that receives a set of numerical or alphanumeric data, then there are averages and standard deviations that can be applied to the data set.

To generalize you should show that the average and standard deviations of your sample is sufficiently close to the averages and standard deviations of the general population within the percentage chance you are required to use.

Normally the acceptable probabilities in research are either 99.5% for general questions or 99.95% when health and safety are being questioned.

In design of your experiment you could decide in advance which statistic will be used. Example to use a normal distribution you mist have more than 30 sample size and show that the distribution of results is close enough to normal distribution.

Distributions will not always be normal. They may be binomial, Poisson, or about a hundred different distributions.

One way to decide on sample size is to choose three or more sample sizes and discover the distribution that they all have. Example for averages or central tendency, if the general population is normally distributed, then small samples will all agree in the Student t Test. In the same data type for the standard deviations, if the general population follows a normal distribution, then all of your smaller samples should agree in a Chi Square Test.

For each distribution for large population of unknown size, there is a related statistic for the smaller sample of known size.

Statistical inference is such a large field that you should probably find a graduate level text book in your categories marketing and social research, to help decide what statistical distribution to use. Otherwise there is a huge amount of work to do testing all of the possibilities.

The point of doing inference is that if the population averages and standard deviations are not known, then your sample sizes should continue to increase until the related statics give the same results for averages and standard deviations for different sample sizes within the allowed probabilities.

Dear Jerry Decker - thank you very much for your this response. This is really very meaningful for me....

Thanks everyone for your participation and to make this discussion interesting

It is widely suggested in the literature to calculate the sample size based on the population at a 95 confidence level and with a margin of error ± 5% (Saunders, Lewis & Thornhill, 2009:219; Cohen, Lawrence & Manion, 2005). However, it was difficult to determine the sample size using that role in case of big size population as in social media users, and in case of studying unkown clear cut population.

Instead, to determine the adequate size of the research sample, we prefare to adapte the rule of thumb regarding each of the analytical methods and their estimated statistical power as presented by VanVoorhis and Morgan (2007).

· For example if the aim of your study is to determine how a number of factors predict an dependent factor. The author can use the rule of thumb that N > 104 + m (where m is the total number of independent variables) to test the effect of each of the predictors as recommended by Green (1991). For example, if you have five independent variables, a sufficient number of participants should be any number more than 109 participants for assuming a medium-sized relationship between each of the independent variables and the dependent variable.

The second example in case of using ANOVA to identify the power of variability of each of the identified variables between number of tested research samples (different groups), 30:1 is the rule of thumb to achieve a statistical power of 80% (VanVoorhis & Morgan, 2007:48). This means that we needed 30 participants for each variable. If you have five variables, you need to have at least 150 participants. However, more of each of the above cases will be better.

Nice topic....interesting discussion