Convenience sampling is a non-probability sampling technique where subjects are selected because of their convenient accessibility and proximity to the researcher.

Given the subjectivity of choice, there is no "a priori" dimension for the convenient sampling and there is not a technique to determine the exact size of the sample, in such way that it is representative of the study population.

Therefore, it is not scientific.

Nitin -

The basic problems with convenience sampling are (1) the potential for a biased selection of respondents, and (2) the dearth of tangible information on accuracy, most notably an estimator for standard error, for quantitative data.

In general, what I note here, is for quantitative data.  However, most concepts should, I think, generally carry over to all types of data in some analogous manner. 

In the first case above, one might note the fact that almost any well-planned probability- and/or model-based survey methodology can be plagued with "nonignorable nonresponse," meaning nonresponse which cannot straightforwardly be imputed nor covered by inflating survey weights, without work to determine how the missing data might differ from the other data in general.  Data may need to be stratified or regrouped in some way. (See "response propensity" groups.)   So, "representativeness" can always be a problem.  Even for random sampling, you could draw an 'odd' sample. (That is at least partially why model-assisted design-based sampling and estimation can be so superior to unaided design-based methods.) 

One way to improve any survey is to do some kind of stratification. The concern with convenience sampling is that this "representativeness" problem can be at an extreme.

In the second case above, no reliable standard error estimate, this just emphasizes problem (1).  Note also that a standard error is based either on the selection probabilities, or, as in other areas of statistics, on a (regression) model, or both.  If you do not have randomized selection, nor regressor data on your population, then inference to your population is problematic.   

However, if you have auxiliary/regressor/predictor data for your entire population, and thus can use model-based estimation (prediction), then that might be an option for you, depending upon your circumstances. Perhaps there might be administrative data available that may serve as regressor/auxiliary data on the population of interest for each variable of interest.

For estimating finite population totals, especially from a skewed population, with a regressor, the following may be helpful: 

 https://www.researchgate.net/publication/261947825_Projected_Variance_for_the_Model-based_Classical_Ratio_Estimator_Estimating_Sample_Size_Requirements

Regressor data are generally available for official statistics, where there are many annual censuses of data, and monthly or weekly samples, especially from highly skewed establishment surveys.  However, in most cases such regressor data may not be available. An exception might be administrative data available on your entire population. But I think most people interested in convenience sampling have little or no information on the frame. 

The links below cover a variety of nonprobability sampling methods, and may help you decide what to do:

 http://www.statcan.gc.ca/sites/default/files/media/14252-eng.pdf

 https://www.researchgate.net/publication/273561892_Summary_Report_of_the_AAPOR_Task_Force_on_Non-probability_Sampling?ev=prf_pub

And you may want to look at other papers under Mike Brick's contributions on ResearchGate:

 https://www.researchgate.net/profile/J_Brick

Perhaps you might try a pilot study first to help decide what is feasible under your circumstances. 

Cheers - Jim

Conference Paper Projected Variance for the Model-based Classical Ratio Estim...

Article Summary Report of the AAPOR Task Force on Non-probability Sampling

Thank you James & Gioacchino for quick and knowledgeable response to my question....

Why this question came in to my mind? 

I work in the field of human Geography. I am discussing 2 cases below...

1. If I have to find satisfaction index of tourist at a particular place. I visit there 3 times and ask the questions ( or field the questioner) to the tourist present there. Then i calculate the responses for eg. If my question is.. what is the quality of hotels at this tourist place ? 

Answers: Very good - 55%, good - 25% average - 5%, Not good - 15%

In this situation whether the convenient sampling is sufficient?

2. If i have to study the problems of seasonal migrants.... then....

I have to interview the migrants available. in this case whether the convenient sampling is applied?

Nitin,

I have not worked with likert scales nor qualitative data, and convenience sampling (with no auxiliary data) is generally agreed to be quite problematic.  However, on page 227 of Blair &  Blair (2015), they say that if a population is fairly homogeneous (they mention people at a shopping mall asked opinions on dish detergent scents) then a convenience sample might be "useful," but I have seen nothing about sample sizes.

However, sample size needs are dependent upon the variability of your population in continuous data with random and/or model-based sampling, and logically in other data as well, so I'd say that hopefully the population is not only well mixed, but also not highly variable, and perhaps after you started collecting data, you might soon note a consistent pattern of responses.

My advice would be to consider a pilot study to work out details. 

You may have to just keep sampling until you reach consistent results.

Also you could revisit that AAPOR paper I sent, as I noticed they had some information on convenience samples, though I saw something about "uncontrolled" convenience sampling being the least accurate of sampling methods.

Blair, E. and Blair, J(2015), Applied Survey Sampling, Sage Publications.

 https://www.researchgate.net/publication/273561892_Summary_Report_of_the_AAPOR_Task_Force_on_Non-probability_Sampling

But my area in surveys was continuous data, and I used probability design-based and also model-based sampling and estimation. I don't think convenience sampling can do very much in the way of inference. 

Best wishes - Jim

PS - As your tourists are "in a particular place," as Blair & Blair seem to have indicated, maybe convenience sampling can help.  For the immigrants to be interviewed, I would think you would at least need to 'stratify' (really just 'group' here), by locations.  I suggest sampling until results become consistent.  

Article Summary Report of the AAPOR Task Force on Non-probability Sampling