Question

Topic: Research/Metrics

Is It A: Sample Or A Census Survey

Posted by Anonymous on 1/14/2010 at 12:03 AM ET 125 Points
This is my first posting on this site and I am hoping you will be able to help clarify a fuzzy concept for me regarding sampling.

If I am dealing with a small population where I will have to practically survey all individuals to obtain the neccessary number of respondents to achieve a certain confidence level and margin of error, then is it a sample survey or a census survey?

Example scenario: Simple Random Sample Method

Population: 200
Confidence: 95%
Error: +/-5%
Respondents Needed: 132
Estimated Response Rate: 65% (known from previous experience)
Total Number to Survey: 203

Since I am still short 3 individuals, I would then survey everyone in my population. So with the conditions listed above, would this be a probability sample survey or census survey? This would then lead to the question of whether I can apply probability statistics to make inferences. If it is census survey, I would assume I do not need to make inferences but merely describe he data as is. Please help!
To continue reading this question and the solution, sign up ... it's free!
SUBSCRIBE
Or, SIGN IN

RESPONSES

  • Posted by koen.h.pauwels on 1/14/2010 at 7:05 AM Accepted
    Hi,

    If your population is only 200, and you are able to interview them all, please do so! Your calculation of required sample size is correct, but it is based on inferential statistics; i.e. how to infer information about the population from a smaller sample of that population.

    So you are right this is census survey, and that you should simply report the data as is. The additional benefit is that you will capture specific information on each individual or business you are interviewing. For instance, Boeing and Airbus can and should get information on each potential customer (there aren't too many airlines left in the world), as each airline may have specific needs.

    Cheers
  • Posted by steven.alker on 1/15/2010 at 1:40 PM Accepted
    I am in total agreement with Koen and that the confidence, error, respondents and total numbers are only needed if you are looking at surveying a percentage of the total population.

    In your scenario, the only source of error will be if your questions are leading or ambiguous or if your respondents hate you and decide to lie.

    You can’t handle those problems with stats but you can ensure that your questions and the words which you use are appropriate. If you can’t afford a market research consultant to check them over for you, then publish them here and we can have a go.

    The only danger with the latter approach is that with 340,000 subscribers some of your targets might read this and skew their eventual answers when to carry out your census!

    Steve Alker
    Xspirt
  • Posted on 1/15/2010 at 5:57 PM Author
    Thank you Koen and Stevea! Some additional details that I left out would be that it is not possible to interview all respondents and actually the survey will be conducted electronically by email with an invitation only link.

    There is one thing that still has me unsure about this. I would be in complete agreement that it is a cenus survey if I received ALL 200 responses (or very close to it) but that is not the case. Remember that I will have non-response. I only expect 65% of the audience surveyed to respond therefore leaving me with 132 responses. Wouldn't this take the project back to being a sample survey since I originally needed to have 132 sample respondents??? If I survey everyone, aren't I just over sampling to ensure my response rate???

    The confusing part is that when I survey everyone it becomes a census survey even though I have non-response which could be as great as 50%. However, when using the sampling method and you know there is going to be non-response, from my understanding, you are suppose to oversample to achieve the target rate. Which type of survey do I have? Does the rule of randomness still apply since I am surveying everyone?
  • Posted by koen.h.pauwels on 1/21/2010 at 4:10 AM Member
    Wow, I am impressed with your thorough thinking about this issue!

    You raise an important additional point: non-response bias. In academia, we aim to minimize this by:

    1) the design of the survey itself (eg length, layout, where you put sensivite questions,...), the medium (face-to-face yields more response than internet), the motivation to complete (money for some, access to the report for others,..) and the follow-up (e.g. two waves of phone call follow-up for non-responders), etc

    2) examining how non-responders differ from responders: how do they differ in measurable criteria (e.g. age, gender, business size, position within the company,current vendor,...) and how do the answers of late responders (eg after a follow-up reminder) differ from early responders. If these differences are not statistically significant, we conclude there is no strong evidence of non-response bias.

    As a final note: non-response bias is unrelated to your original question of how to label your survey (with which you seem overly concerned :-)): if you sample the full popultation, no matter whether they respond or not, you have a census survey (even the US census has to deal with non-response....)

    Cheers,
BACK TO TOP

Post a Comment

Related Discussions

Community Info

Top 25 Experts
(Research/Metrics)