Question

Topic: Advertising/PR

Sampling - What Are Sufficient Numbers?

Posted by stacey on 2/15/2007 at 11:04 PM ET 125 Points
What type of numbers are needed to take a viable sample?

Our company allows businesses to get their products directly into the hands of a very specific, niche market. Since we are so specific (including family status, location, income, age, etc.) we do not and will not reach the masses.

I'm wondering if a sample test of 500 would warrant enough information or is 1000 more significant, or if anything below 10,000 is worthless, etc.

It's late, I hope I'm explaining myself well enough.

Thank you in advance,

Stacey
To continue reading this question and the solution, sign up ... it's free!
SUBSCRIBE
Or, SIGN IN

RESPONSES

  • Posted by wnelson on 2/15/2007 at 11:23 PM Member
    Hi, Stacey,

    Please see this website.

    https://www.berrie.dds.nl/calcss.htm

    In most cases, for tests involving proportions like a marketing survey where you want to know how many people approve of a product, you need only 385 samples to be 95% confident that your error will be no more than 5%. That means if 70% of the people approve in your sample, you can be 95% sure that the world will approve at a rate between 65% and 75%. This is true if the sample is random and representative of the population.

    You can use that website to change the population size, the confidence, or the allowable maximum error to come up with sample sizes that fit other requirements. For instance, for a population of 10,000, you would need 371 samples for 95% confidence of no more than 5% error. Or for a population of 10,000, 99% confidence that your error is no more than 1%, you need a sample of 6,240.

    If you take specific measurements - like the length of a manufactured part, you can get by with sample sizes much less. But in almost all marketing cases, proportion data is used.

    I hope this helps.

    Wayde
  • Posted on 2/16/2007 at 3:01 AM Member
    However specific the niche is , sampling is fine. Any data set of more than 30 is amenable to be fitted to statistically relevant Normal Analysis. To be on the safe side- take a number of 300- that shd suffice in making prediction related to product/ service acceptance/ pricing etc.
  • Posted by MajorHal on 2/16/2007 at 5:22 AM Member
    Good Morning Stacey,
    To elaborate a little on the reply from 'shghosh': Yes any sample of 30 or greater will produce a disribution worth review. It may not however turn-out to be Normal. That is shaped like a bell curve. The website that 'wnelson' referred to is worth following.

    However, you must follow these rules in any sampling program:
    1) The sample must be unbiased. You can't exclude anyone by chose from the population you sample. Once the sample responses have been taken you can plot them on a run chart with control limits too see if any of them fall outside these limits, for exclusion. These are called 'outliers' and don't reflect the population.
    2) The samples must be chosen completely at random from the population you defined. Assign each customer a number. Then use a calculator with a random number generator and equate the generated number to the corresponding customer.
    3) You must be willing to accept any 'sampling error' that occurs from the data collection and analysis.

    I know this seems extreme, however sampling is an exact math science. If the rules aren't followed the answer you get is worthless and more importantly it can lead you to take harmful action rather than beneficial.

    Good Luck,
    MajorHal
  • Posted by MajorHal on 2/16/2007 at 5:26 AM Member
    Hi Again Stacey,
    You may want to checkout the 'Research/Metrics' section of KHE. The link is shown at the top of this page.
    Good Luck,
    MajorHal
  • Posted by wnelson on 2/16/2007 at 9:50 AM Accepted
    Stacey,

    Just to clarify, it may seem like you have received four different conflicting answers on this thread. However, I assure you that each are saying elements of the same thing.

    Benjamin (benwung) tells you that sampling is not good for niches. What he is purporting is that for niche marketing, you should be in very good touch with your small group of customers and fine tune your product/service features to conform. If your niche is small - like 1000 people, you need to take 278 samples to get significant data. That seems ridiculous - 278 samples is almost all! For many marketing research projects, you would do a "focus group" of five to a dozen customers and talk indepthly to them. In this case, because the data is in more "story form." The purpose of these sessions are to develop detailed information and relationships. From this, if you do several focus groups, if you reach a number above 30, your "parametric data" is significant - as Sharatee and Hal (shghosh and MajorHal) discuss when you are looking at statistics concerning the population statistics - mean, variance, etc. If you are looking at statistics from proportions, you need more samples to be significant because you are using less information. Knowing a number like "how often do you buy these kinds of products" is more data than "Very often, often, seldom, rarely." And with your focus group data, IF you want to run surveys, then you have valuable information up front on which to base your survey. And in this case you can use the link I sent you to the sample size calculator, you can calculate the exact sample size you need for your size population and the accuracy you want to achieve in your survey. As a rule of thumb, a sample of 300 is a good size, giving you at least 95% confidence of no more than 5% error in your answers for populations up to 1350. Or, alternately, you can have 91.65% confidence that your error is no more than 5% for populations of 1,000,000. Or you can have at least 95% confidence that your results will be no more than 5.66% for a population of 1,000,000. You can use the calculator to optimize the sample size based on your needs for accuracy and confidence.

    I hope this clarifies the seemly diverse answers we have given you. They all fit together - but just from different directions.

    Wayde
  • Posted on 2/16/2007 at 1:28 PM Accepted
    Hi Stacey -

    Just one more think to add to the really good answers you've already received - note that the samples to which people refer need to be qualified as "completes". A "complete" is what you determine ahead of time to be a full response - if you have five questions, and two are unanswered by a respondent, and they happen to be very key questions, this might not be a "complete". The sample is also, very oftenm, a small proportion of the outgo. In order to get 300 completes, you would have to send out a lot more surveys. Of course, this all assumes that it's a survey that you're talking about.

    Good luck!

    Joy
  • Posted by stacey on 2/16/2007 at 4:38 PM Author
    Wow, thank you all for your answers. I've said it before, but this website is just a tremendous resource for someone like me who has a lot to learn.

    Essentially, what I'm hearing is that the number - 600 - is sufficient for a 'sample test' but that the fact that our service is so specifically targeted to a niche audience actually makes it useless for sampling, is this accurate?


  • Posted by wnelson on 2/16/2007 at 5:30 PM Accepted
    Stacey,

    No. Not correct. 600 COMPLETED answers is MORE than adequate and sufficient. The targeted niche market DOES NOT make sampling useless. You can take samples from ANY population as long as the samples are representative from the target population. A niche represents a target population. For niche markets, all of the same benefits apply for sampling. However, if you have a small niche and can afford to contact everyone and get a response from everyone, then do it. Then you have 100% confidence that the answers reflect the "voice of the niche." How many customers are in your niche?

    Wayde
BACK TO TOP

Post a Comment

Related Discussions

Community Info

Top 25 Experts
(Advertising/PR)