One thing to add: you asked the necessary response rate. I'd encourage you to focus less on the response rate and more on evaluating specifically for the presence of nonresponse bias. See this book: http://books.google.com/books?id=3LK6AAAAIAAJ&q=survey+nonresponse+in+design+data+collection&dq=survey+nonresponse+in+design+data+collection&hl=&cd=1&source=gbs_api. and this article: Public Opinion Quarterly, Vol. 72, No. 2, Summer 2008, pp. 167–189. (All that said, people quote 60% without a lot of good evidence behind that.)

Good luck!

i think that could be ideal but impractical!!!

Acharya is correct. It depend on what are you objectives, population size and resource constraints.

Thank you Muhammad Mohiuddin ,Suwash Chandra Acharya and Subrata Chakraborty for your valuable answers.

Of course it would be implausible to survey large population..

in my case it is limited population and I have access to all of them. 

for updating today i came across with a term of "census ". generally, it refers to survey every one in the population... i would go back and read more about it and see.

Good Day to you all,

New update!

I found very useful book that give a full guideline for researcher to choose between Census and Sampling:

SAMPLING ESSENTIALS , Johnnie Daniel (Sage Publication) 2012

link for the book on Google Books.. :)

http://books.google.com.my/books?id=qKV7XQXQOZgC&dq=sampling+essentials+johnnie+daniel&lr=&source=gbs_navlinks_s

Yes you have said it. When we survey the entire population it is known as Census. Definitely you can do it. and it is the ideal. When we are doing sampling and then generalizing, it is a short cut method as surveying the entire populations  involves more time and money

If the total population is less than 300 it is better to go for Census though we can still do a  10% sampling of 30. This is because to be relevant statistically the sample should be minimum 30 ( If the total population is  less than 30  we will go for census)

 For Example you want to study the  Learning behavior of  a class of 30 or 60 students you need to go for census. But suppose you want to study the learning behavior of students of  a particular school- say with a population of 600 - spread across 10 classes, We need to go for a minimum sample of   30 students. If we are able to make to sample to say 50 or 60 our study will be more accurate.

It has been found that  most of the time things behave like a normal distribution and the above is based on that assumption.

Abdullah, your sample frame will be framed based on your research question/s and research focus. To me sampling is the representative of total population, it is, for some important reasons better than whole population to take into account. Such as time, money, analysis (to avoid complexity of huge data). So, considering all these you can choose simple random or restricted random also probability or non probability. all will depends on your research focus.

You can survey an entire population apart from a census if the population is small, reachable and financially feasible.

Thank you Ramachandran,  Khan  and Raimi 

Thank you all for your valuable answers, I really appreciate your cooperation.  

One thing to add: you asked the necessary response rate. I'd encourage you to focus less on the response rate and more on evaluating specifically for the presence of nonresponse bias. See this book: http://books.google.com/books?id=3LK6AAAAIAAJ&q=survey+nonresponse+in+design+data+collection&dq=survey+nonresponse+in+design+data+collection&hl=&cd=1&source=gbs_api. and this article: Public Opinion Quarterly, Vol. 72, No. 2, Summer 2008, pp. 167–189. (All that said, people quote 60% without a lot of good evidence behind that.)

Good luck!

Hi Andrew,

indeed, non-response bias test is crucial.

thank you a lot for the profound contribution :)

To reduce nonresponse bias, you may want to investigate the use of "response propensity" groups. (You could start with an Internet search.)  It is a way of isolating subsets of your population that may be biased in different directions to reduce the influence of less relevant data when imputing or reweighting for each such group. You might be able to implement with software designed for poststratification, I think. 

I assume you are referring to continuous data.  If you have a related, reliable census, you could use that as regressor data to impute for missing data in your current near-census. The variance of the prediction error could be useful for you when studying accuracy.   Cheers!  

 Hi James,

this was a cross sectional research. I performed a non-response bias by analyzing early and late responses separately. very grateful to your valuable and thoughtful answer.

Regards.

Offcourse you can