Survey Sample Size MKTG 3342 Fall 2008 Professor Edward Fox Sample Size Determination Convenience – Say … about 100. Rule of Thumb - At least 30 per each subgroup (e.g., males/females) that will be analyzed. Budget Constraint - Have a $300 budget for sampling. On average it costs $2 per returned questionnaire. Then go for sample size of 150. Comparable Studies or Industry Average Typical Sample Sizes for Studies of Human and Institutional Populations Number of Subgroup Analyses People or Households Institutions National Regional or Special Regional or Special National None or few 500-1500 200-500 200-500 50-200 Average 1500-2500 500-1000 500-1000 200-500 Many 2500+ 1000+ 1000+ 500+ Determining Sample Size Using Statistical Methods There are statistical formulas for computing sample sizes. These consider three aspects: Precision: Percent of sampling error deemed acceptable by the researcher Confidence:How confident is the researcher that the true average value lies in the interval [lower, upper] estimated Variance: Dispersion of the true value across the population Determining Sample Size In general: If you desire greater precision in your estimate, you need a larger sample size, other things being equal If you want greater confidence in your estimate, you need a larger sample size If the estimated variance in the population is high, then you need a larger sample size Determining Sample Size from Web There are many sample size calculators on the web. Two such websites are: http://www.steinermarketing.com/calc_sample_size.htm http://www.dssresearch.com/toolkit/sscalc/size.asp The formula on the first website is for estimating proportions. Use 95% and 99% confidence level, and confidence interval (allowable error) to be 2% to 10%. Use any population size such as 100, 1000, 10,000. See what happens to sample size as you vary the parameters. What About Response Rates? The calculated sample size is the number of desired actual responses, or completed questionnaires In the real-world not all surveys sent out are completed (response rates are less than 100%) You must incorporate the expected response rate when deciding how many questionnaires to send out or how many people to call Response Rate Calculation For Mail Surveys: Response # Usable Surveys Returned = # Surveys - # Surveys Returned Rate Mailed " Not Deliverable" Mail Surveys Estimating the number of surveys required to achieve given sample size: n Surveys Required = [(1-U) RR] n = required sample size U = estimated proportion “not deliverable” RR = estimated response rate (proportion) Mail Survey Example You have determined that sample size of 200 will allow reasonable precision and confidence for your estimates of important population parameters. You will be conducting a mail survey of households in Highland Park. You expect that about 5% of mail will be undeliverable and the expected response rate is 10%. How many mail questionnaires should you send out? Response Rate Calculation For Telephone Surveys: #Completed Interviews RR = #Completed + #Refusals + #No Answers Interviews Telephone Survey Estimating the number of calls required to achieve given sample size: n Total Calls = [(1-NE) (1-R) (1-NA)] where: n NE R NA = = = = required sample size estimated proportion of non-eligibles estimated proportion of refusals estimated proportion of no answers Telephone Survey Example You have determined that a sample size of 200 will allow reasonable precision and confidence for your estimates of important population parameters. You will be conducting a telephone survey of university students ages 20 and older. After checking with university registration officials you know that 50% of all university students meet this criterion. Further, you expect about 20% of the people you contact not to participate in the survey and about 15% not to be reachable even after trying at several different times on different days of the week. How may total calls should you expect to make for this project?